id sid tid token lemma pos fcis-9453 1 1 frontiers frontier NOUN fcis-9453 1 2 in in ADP fcis-9453 1 3 computing computing NOUN fcis-9453 1 4 and and CCONJ fcis-9453 1 5 intelligent intelligent ADJ fcis-9453 1 6 systems system NOUN fcis-9453 1 7 issn issn VERB fcis-9453 1 8 : : PUNCT fcis-9453 1 9 2832 2832 NUM fcis-9453 1 10 - - SYM fcis-9453 1 11 6024 6024 NUM fcis-9453 1 12 | | NOUN fcis-9453 1 13 vol vol NOUN fcis-9453 1 14 . . PROPN fcis-9453 2 1 4 4 NUM fcis-9453 2 2 , , PUNCT fcis-9453 2 3 no no INTJ fcis-9453 2 4 . . NOUN fcis-9453 2 5 1 1 NUM fcis-9453 2 6 , , PUNCT fcis-9453 2 7 2023 2023 NUM fcis-9453 2 8 67 67 NUM fcis-9453 2 9 improved improve VERB fcis-9453 2 10 yolov5l‐based yolov5l‐based ADJ fcis-9453 2 11 detection detection NOUN fcis-9453 2 12 of of ADP fcis-9453 2 13 surface surface NOUN fcis-9453 2 14 defects defect NOUN fcis-9453 2 15 in in ADP fcis-9453 2 16 hot hot ADJ fcis-9453 2 17 rolled roll VERB fcis-9453 2 18 steel steel NOUN fcis-9453 2 19 strips strip NOUN fcis-9453 2 20 zhiren zhiren PROPN fcis-9453 2 21 zhu zhu PROPN fcis-9453 3 1 * * PUNCT fcis-9453 3 2 , , PUNCT fcis-9453 3 3 liming lime VERB fcis-9453 3 4 zhou zhou PROPN fcis-9453 3 5 , , PUNCT fcis-9453 3 6 fankai fankai PROPN fcis-9453 3 7 chen chen PROPN fcis-9453 3 8 , , PUNCT fcis-9453 3 9 chen chen PROPN fcis-9453 3 10 liu liu PROPN fcis-9453 3 11 , , PUNCT fcis-9453 3 12 fanrun fanrun PROPN fcis-9453 3 13 meng meng PROPN fcis-9453 3 14 college college PROPN fcis-9453 3 15 of of ADP fcis-9453 3 16 electrical electrical ADJ fcis-9453 3 17 engineering engineering NOUN fcis-9453 3 18 , , PUNCT fcis-9453 3 19 tianjin tianjin PROPN fcis-9453 3 20 university university PROPN fcis-9453 3 21 of of ADP fcis-9453 3 22 technology technology NOUN fcis-9453 3 23 and and CCONJ fcis-9453 3 24 education education NOUN fcis-9453 3 25 , , PUNCT fcis-9453 3 26 tianjin tianjin PROPN fcis-9453 3 27 , , PUNCT fcis-9453 3 28 china china PROPN fcis-9453 3 29 * * PUNCT fcis-9453 3 30 corresponding correspond VERB fcis-9453 3 31 author author NOUN fcis-9453 3 32 : : PUNCT fcis-9453 3 33 zhiren zhiren PROPN fcis-9453 3 34 zhu zhu PROPN fcis-9453 3 35 ( ( PUNCT fcis-9453 3 36 email email NOUN fcis-9453 3 37 : : PUNCT fcis-9453 3 38 710203401@qq.com 710203401@qq.com X fcis-9453 3 39 ) ) PUNCT fcis-9453 3 40 abstract abstract NOUN fcis-9453 3 41 : : PUNCT fcis-9453 3 42 to to PART fcis-9453 3 43 address address VERB fcis-9453 3 44 the the DET fcis-9453 3 45 problems problem NOUN fcis-9453 3 46 of of ADP fcis-9453 3 47 complex complex ADJ fcis-9453 3 48 background background NOUN fcis-9453 3 49 , , PUNCT fcis-9453 3 50 different different ADJ fcis-9453 3 51 sizes size NOUN fcis-9453 3 52 and and CCONJ fcis-9453 3 53 easy easy ADJ fcis-9453 3 54 to to PART fcis-9453 3 55 miss miss VERB fcis-9453 3 56 and and CCONJ fcis-9453 3 57 mis mi NOUN fcis-9453 3 58 - - NOUN fcis-9453 3 59 detect detect NOUN fcis-9453 3 60 in in ADP fcis-9453 3 61 the the DET fcis-9453 3 62 detection detection NOUN fcis-9453 3 63 of of ADP fcis-9453 3 64 surface surface NOUN fcis-9453 3 65 defects defect NOUN fcis-9453 3 66 in in ADP fcis-9453 3 67 hot hot ADV fcis-9453 3 68 - - PUNCT fcis-9453 3 69 rolled roll VERB fcis-9453 3 70 strip strip NOUN fcis-9453 3 71 , , PUNCT fcis-9453 3 72 an an DET fcis-9453 3 73 improved improve VERB fcis-9453 3 74 yolov5l yolov5l NOUN fcis-9453 3 75 - - PUNCT fcis-9453 3 76 based base VERB fcis-9453 3 77 method method NOUN fcis-9453 3 78 for for ADP fcis-9453 3 79 detecting detect VERB fcis-9453 3 80 surface surface NOUN fcis-9453 3 81 defects defect NOUN fcis-9453 3 82 in in ADP fcis-9453 3 83 hot hot ADV fcis-9453 3 84 - - PUNCT fcis-9453 3 85 rolled roll VERB fcis-9453 3 86 strip strip NOUN fcis-9453 3 87 is be AUX fcis-9453 3 88 proposed propose VERB fcis-9453 3 89 . . PUNCT fcis-9453 4 1 firstly firstly ADV fcis-9453 4 2 , , PUNCT fcis-9453 4 3 by by ADP fcis-9453 4 4 adding add VERB fcis-9453 4 5 the the DET fcis-9453 4 6 simam simam ADJ fcis-9453 4 7 attention attention NOUN fcis-9453 4 8 mechanism mechanism NOUN fcis-9453 4 9 module module NOUN fcis-9453 4 10 to to ADP fcis-9453 4 11 the the DET fcis-9453 4 12 aggregation aggregation NOUN fcis-9453 4 13 network network NOUN fcis-9453 4 14 , , PUNCT fcis-9453 4 15 the the DET fcis-9453 4 16 important important ADJ fcis-9453 4 17 information information NOUN fcis-9453 4 18 is be AUX fcis-9453 4 19 focused focus VERB fcis-9453 4 20 with with ADP fcis-9453 4 21 high high ADJ fcis-9453 4 22 weights weight NOUN fcis-9453 4 23 to to PART fcis-9453 4 24 improve improve VERB fcis-9453 4 25 the the DET fcis-9453 4 26 recall recall NOUN fcis-9453 4 27 rate rate NOUN fcis-9453 4 28 of of ADP fcis-9453 4 29 the the DET fcis-9453 4 30 original original ADJ fcis-9453 4 31 algorithm algorithm NOUN fcis-9453 4 32 ; ; PUNCT fcis-9453 4 33 secondly secondly ADV fcis-9453 4 34 , , PUNCT fcis-9453 4 35 by by ADP fcis-9453 4 36 replacing replace VERB fcis-9453 4 37 all all DET fcis-9453 4 38 c3 c3 NOUN fcis-9453 4 39 modules module NOUN fcis-9453 4 40 in in ADP fcis-9453 4 41 the the DET fcis-9453 4 42 yolov5l yolov5l PROPN fcis-9453 4 43 structure structure NOUN fcis-9453 4 44 with with ADP fcis-9453 4 45 c2f c2f PROPN fcis-9453 4 46 , , PUNCT fcis-9453 4 47 a a DET fcis-9453 4 48 richer rich ADJ fcis-9453 4 49 gradient gradient NOUN fcis-9453 4 50 of of ADP fcis-9453 4 51 information information NOUN fcis-9453 4 52 flow flow NOUN fcis-9453 4 53 is be AUX fcis-9453 4 54 obtained obtain VERB fcis-9453 4 55 to to PART fcis-9453 4 56 improve improve VERB fcis-9453 4 57 the the DET fcis-9453 4 58 accuracy accuracy NOUN fcis-9453 4 59 rate rate NOUN fcis-9453 4 60 of of ADP fcis-9453 4 61 the the DET fcis-9453 4 62 original original ADJ fcis-9453 4 63 algorithm algorithm NOUN fcis-9453 4 64 . . PUNCT fcis-9453 5 1 the the DET fcis-9453 5 2 experimental experimental ADJ fcis-9453 5 3 results result NOUN fcis-9453 5 4 show show VERB fcis-9453 5 5 that that SCONJ fcis-9453 5 6 the the DET fcis-9453 5 7 average average ADJ fcis-9453 5 8 detection detection NOUN fcis-9453 5 9 accuracy accuracy NOUN fcis-9453 5 10 using use VERB fcis-9453 5 11 the the DET fcis-9453 5 12 improved improve VERB fcis-9453 5 13 yolov5l yolov5l NOUN fcis-9453 5 14 improves improve VERB fcis-9453 5 15 by by ADP fcis-9453 5 16 5.3 5.3 NUM fcis-9453 5 17 % % NOUN fcis-9453 5 18 and and CCONJ fcis-9453 5 19 the the DET fcis-9453 5 20 accuracy accuracy NOUN fcis-9453 5 21 rate rate NOUN fcis-9453 5 22 by by ADP fcis-9453 5 23 8.3 8.3 NUM fcis-9453 5 24 % % NOUN fcis-9453 5 25 compared compare VERB fcis-9453 5 26 to to ADP fcis-9453 5 27 the the DET fcis-9453 5 28 original original ADJ fcis-9453 5 29 network network NOUN fcis-9453 5 30 , , PUNCT fcis-9453 5 31 resulting result VERB fcis-9453 5 32 in in ADP fcis-9453 5 33 higher high ADJ fcis-9453 5 34 detection detection NOUN fcis-9453 5 35 accuracy accuracy NOUN fcis-9453 5 36 and and CCONJ fcis-9453 5 37 lower low ADJ fcis-9453 5 38 error error NOUN fcis-9453 5 39 and and CCONJ fcis-9453 5 40 miss miss VERB fcis-9453 5 41 detection detection NOUN fcis-9453 5 42 rates rate NOUN fcis-9453 5 43 , , PUNCT fcis-9453 5 44 meeting meet VERB fcis-9453 5 45 the the DET fcis-9453 5 46 requirements requirement NOUN fcis-9453 5 47 of of ADP fcis-9453 5 48 hot hot ADV fcis-9453 5 49 - - PUNCT fcis-9453 5 50 rolled roll VERB fcis-9453 5 51 strip strip NOUN fcis-9453 5 52 steel steel NOUN fcis-9453 5 53 inspection inspection NOUN fcis-9453 5 54 in in ADP fcis-9453 5 55 industrial industrial ADJ fcis-9453 5 56 manufacturing manufacturing NOUN fcis-9453 5 57 . . PUNCT fcis-9453 6 1 keywords keyword NOUN fcis-9453 6 2 : : PUNCT fcis-9453 6 3 hot hot ADJ fcis-9453 6 4 rolled roll VERB fcis-9453 6 5 strip strip NOUN fcis-9453 6 6 ; ; PUNCT fcis-9453 6 7 simam simam ADJ fcis-9453 6 8 attention attention NOUN fcis-9453 6 9 mechanism mechanism NOUN fcis-9453 6 10 ; ; PUNCT fcis-9453 6 11 c2f c2f NOUN fcis-9453 6 12 module module NOUN fcis-9453 6 13 ; ; PUNCT fcis-9453 6 14 defect defect ADJ fcis-9453 6 15 detection detection NOUN fcis-9453 6 16 . . PUNCT fcis-9453 7 1 1 1 X fcis-9453 7 2 . . X fcis-9453 7 3 introduction introduction NOUN fcis-9453 7 4 hot hot ADV fcis-9453 7 5 - - PUNCT fcis-9453 7 6 rolled roll VERB fcis-9453 7 7 strip strip NOUN fcis-9453 7 8 [ [ X fcis-9453 7 9 1 1 X fcis-9453 7 10 ] ] PUNCT fcis-9453 7 11 is be AUX fcis-9453 7 12 an an DET fcis-9453 7 13 economical economical ADJ fcis-9453 7 14 ‘ ' PUNCT fcis-9453 7 15 green green ADJ fcis-9453 7 16 steel steel NOUN fcis-9453 7 17 ’ ' PUNCT fcis-9453 7 18 commonly commonly ADV fcis-9453 7 19 used use VERB fcis-9453 7 20 in in ADP fcis-9453 7 21 industry industry NOUN fcis-9453 7 22 . . PUNCT fcis-9453 8 1 hot hot ADJ fcis-9453 8 2 - - PUNCT fcis-9453 8 3 rolling rolling NOUN fcis-9453 8 4 is be AUX fcis-9453 8 5 rolling roll VERB fcis-9453 8 6 above above ADP fcis-9453 8 7 the the DET fcis-9453 8 8 recrystallisation recrystallisation NOUN fcis-9453 8 9 temperature temperature NOUN fcis-9453 8 10 and and CCONJ fcis-9453 8 11 is be AUX fcis-9453 8 12 widely widely ADV fcis-9453 8 13 used use VERB fcis-9453 8 14 in in ADP fcis-9453 8 15 mechanical mechanical ADJ fcis-9453 8 16 applications application NOUN fcis-9453 8 17 because because SCONJ fcis-9453 8 18 of of ADP fcis-9453 8 19 its its PRON fcis-9453 8 20 low low ADJ fcis-9453 8 21 energy energy NOUN fcis-9453 8 22 consumption consumption NOUN fcis-9453 8 23 , , PUNCT fcis-9453 8 24 low low ADJ fcis-9453 8 25 cost cost NOUN fcis-9453 8 26 , , PUNCT fcis-9453 8 27 good good ADJ fcis-9453 8 28 vibration vibration NOUN fcis-9453 8 29 resistance resistance NOUN fcis-9453 8 30 and and CCONJ fcis-9453 8 31 high high ADJ fcis-9453 8 32 production production NOUN fcis-9453 8 33 efficiency efficiency NOUN fcis-9453 8 34 . . PUNCT fcis-9453 9 1 in in ADP fcis-9453 9 2 actual actual ADJ fcis-9453 9 3 manufacturing manufacturing NOUN fcis-9453 9 4 , , PUNCT fcis-9453 9 5 the the DET fcis-9453 9 6 hot hot ADJ fcis-9453 9 7 rolling rolling NOUN fcis-9453 9 8 process process NOUN fcis-9453 9 9 will will AUX fcis-9453 9 10 produce produce VERB fcis-9453 9 11 a a DET fcis-9453 9 12 variety variety NOUN fcis-9453 9 13 of of ADP fcis-9453 9 14 different different ADJ fcis-9453 9 15 defects defect NOUN fcis-9453 9 16 on on ADP fcis-9453 9 17 the the DET fcis-9453 9 18 surface surface NOUN fcis-9453 9 19 of of ADP fcis-9453 9 20 the the DET fcis-9453 9 21 strip strip NOUN fcis-9453 9 22 , , PUNCT fcis-9453 9 23 the the DET fcis-9453 9 24 common common ADJ fcis-9453 9 25 ones one NOUN fcis-9453 9 26 are be AUX fcis-9453 9 27 crazing craze VERB fcis-9453 9 28 , , PUNCT fcis-9453 9 29 rolled roll VERB fcis-9453 9 30 - - PUNCT fcis-9453 9 31 in in ADP fcis-9453 9 32 scale scale NOUN fcis-9453 9 33 , , PUNCT fcis-9453 9 34 scratches scratch NOUN fcis-9453 9 35 , , PUNCT fcis-9453 9 36 inclusion inclusion NOUN fcis-9453 9 37 , , PUNCT fcis-9453 9 38 patches patch NOUN fcis-9453 9 39 , , PUNCT fcis-9453 9 40 pitted pit VERB fcis-9453 9 41 surface surface NOUN fcis-9453 9 42 , , PUNCT fcis-9453 9 43 six six NUM fcis-9453 9 44 kinds kind NOUN fcis-9453 9 45 of of ADP fcis-9453 9 46 surface surface NOUN fcis-9453 9 47 defects defect NOUN fcis-9453 9 48 these these DET fcis-9453 9 49 defects defect NOUN fcis-9453 9 50 have have VERB fcis-9453 9 51 a a DET fcis-9453 9 52 serious serious ADJ fcis-9453 9 53 impact impact NOUN fcis-9453 9 54 on on ADP fcis-9453 9 55 the the DET fcis-9453 9 56 qualification qualification NOUN fcis-9453 9 57 rate rate NOUN fcis-9453 9 58 of of ADP fcis-9453 9 59 hot hot ADJ fcis-9453 9 60 rolled roll VERB fcis-9453 9 61 products product NOUN fcis-9453 10 1 [ [ X fcis-9453 10 2 2 2 NUM fcis-9453 10 3 ] ] PUNCT fcis-9453 10 4 . . PUNCT fcis-9453 11 1 traditional traditional ADJ fcis-9453 11 2 strip strip PROPN fcis-9453 11 3 inspection inspection NOUN fcis-9453 11 4 methods method NOUN fcis-9453 11 5 include include VERB fcis-9453 11 6 manual manual ADJ fcis-9453 11 7 sampling sampling NOUN fcis-9453 11 8 , , PUNCT fcis-9453 11 9 magnetic magnetic ADJ fcis-9453 11 10 fluxleakage fluxleakage NOUN fcis-9453 11 11 testing testing NOUN fcis-9453 11 12 , , PUNCT fcis-9453 11 13 eddy eddy PROPN fcis-9453 11 14 current current ADJ fcis-9453 11 15 testing testing NOUN fcis-9453 11 16 and and CCONJ fcis-9453 11 17 infrared infrared ADJ fcis-9453 11 18 detection detection NOUN fcis-9453 11 19 [ [ X fcis-9453 11 20 3 3 NUM fcis-9453 11 21 ] ] PUNCT fcis-9453 11 22 . . PUNCT fcis-9453 12 1 manual manual ADJ fcis-9453 12 2 sampling sampling NOUN fcis-9453 12 3 method method NOUN fcis-9453 12 4 refers refer VERB fcis-9453 12 5 to to ADP fcis-9453 12 6 the the DET fcis-9453 12 7 use use NOUN fcis-9453 12 8 of of ADP fcis-9453 12 9 the the DET fcis-9453 12 10 naked naked ADJ fcis-9453 12 11 eye eye NOUN fcis-9453 12 12 to to PART fcis-9453 12 13 distinguish distinguish VERB fcis-9453 12 14 defects defect NOUN fcis-9453 12 15 , , PUNCT fcis-9453 12 16 the the DET fcis-9453 12 17 method method NOUN fcis-9453 12 18 not not PART fcis-9453 12 19 only only ADV fcis-9453 12 20 wastes waste VERB fcis-9453 12 21 manpower manpower NOUN fcis-9453 12 22 but but CCONJ fcis-9453 12 23 also also ADV fcis-9453 12 24 has have VERB fcis-9453 12 25 the the DET fcis-9453 12 26 problems problem NOUN fcis-9453 12 27 of of ADP fcis-9453 12 28 leakage leakage NOUN fcis-9453 12 29 and and CCONJ fcis-9453 12 30 low low ADJ fcis-9453 12 31 accuracy accuracy NOUN fcis-9453 12 32 ; ; PUNCT fcis-9453 12 33 magnetic magnetic ADJ fcis-9453 12 34 fluxleakage fluxleakage NOUN fcis-9453 12 35 testing testing NOUN fcis-9453 12 36 [ [ X fcis-9453 12 37 4 4 X fcis-9453 12 38 ] ] PUNCT fcis-9453 12 39 uses use VERB fcis-9453 12 40 magnetic magnetic ADJ fcis-9453 12 41 sensors sensor NOUN fcis-9453 12 42 to to PART fcis-9453 12 43 detect detect VERB fcis-9453 12 44 surface surface NOUN fcis-9453 12 45 defects defect NOUN fcis-9453 12 46 , , PUNCT fcis-9453 12 47 but but CCONJ fcis-9453 12 48 can can AUX fcis-9453 12 49 not not PART fcis-9453 12 50 detect detect VERB fcis-9453 12 51 closed closed ADJ fcis-9453 12 52 cracks crack NOUN fcis-9453 12 53 and and CCONJ fcis-9453 12 54 limit limit VERB fcis-9453 12 55 the the DET fcis-9453 12 56 types type NOUN fcis-9453 12 57 of of ADP fcis-9453 12 58 defects defect NOUN fcis-9453 12 59 ; ; PUNCT fcis-9453 12 60 eddy eddy PROPN fcis-9453 12 61 current current ADJ fcis-9453 12 62 detection detection NOUN fcis-9453 12 63 testing testing NOUN fcis-9453 12 64 [ [ X fcis-9453 12 65 5 5 X fcis-9453 12 66 ] ] PUNCT fcis-9453 12 67 uses use VERB fcis-9453 12 68 the the DET fcis-9453 12 69 principle principle NOUN fcis-9453 12 70 of of ADP fcis-9453 12 71 electromagnetic electromagnetic ADJ fcis-9453 12 72 induction induction NOUN fcis-9453 12 73 to to PART fcis-9453 12 74 detect detect VERB fcis-9453 12 75 metal metal NOUN fcis-9453 12 76 surface surface NOUN fcis-9453 12 77 defects defect NOUN fcis-9453 12 78 , , PUNCT fcis-9453 12 79 the the DET fcis-9453 12 80 method method NOUN fcis-9453 12 81 requires require VERB fcis-9453 12 82 professional professional ADJ fcis-9453 12 83 analysis analysis NOUN fcis-9453 12 84 and and CCONJ fcis-9453 12 85 judgement judgement NOUN fcis-9453 12 86 , , PUNCT fcis-9453 12 87 customized customize VERB fcis-9453 12 88 solutions solution NOUN fcis-9453 12 89 and and CCONJ fcis-9453 12 90 high high ADJ fcis-9453 12 91 detection detection NOUN fcis-9453 12 92 costs cost NOUN fcis-9453 12 93 ; ; PUNCT fcis-9453 12 94 infrared infrared ADJ fcis-9453 12 95 detection detection NOUN fcis-9453 12 96 method method NOUN fcis-9453 12 97 [ [ X fcis-9453 12 98 6 6 NUM fcis-9453 12 99 ] ] PUNCT fcis-9453 12 100 uses use VERB fcis-9453 12 101 the the DET fcis-9453 12 102 surface surface NOUN fcis-9453 12 103 temperature temperature NOUN fcis-9453 12 104 of of ADP fcis-9453 12 105 defective defective ADJ fcis-9453 12 106 materials material NOUN fcis-9453 12 107 to to PART fcis-9453 12 108 detect detect VERB fcis-9453 12 109 defects defect NOUN fcis-9453 12 110 , , PUNCT fcis-9453 12 111 but but CCONJ fcis-9453 12 112 the the DET fcis-9453 12 113 detection detection NOUN fcis-9453 12 114 sensitivity sensitivity NOUN fcis-9453 12 115 is be AUX fcis-9453 12 116 related relate VERB fcis-9453 12 117 to to ADP fcis-9453 12 118 thermal thermal ADJ fcis-9453 12 119 emissivity emissivity NOUN fcis-9453 12 120 , , PUNCT fcis-9453 12 121 affected affect VERB fcis-9453 12 122 by by ADP fcis-9453 12 123 time time NOUN fcis-9453 12 124 , , PUNCT fcis-9453 12 125 temperature temperature NOUN fcis-9453 12 126 , , PUNCT fcis-9453 12 127 location location NOUN fcis-9453 12 128 and and CCONJ fcis-9453 12 129 size size NOUN fcis-9453 12 130 , , PUNCT fcis-9453 12 131 and and CCONJ fcis-9453 12 132 can can AUX fcis-9453 12 133 not not PART fcis-9453 12 134 accurately accurately ADV fcis-9453 12 135 distinguish distinguish VERB fcis-9453 12 136 the the DET fcis-9453 12 137 types type NOUN fcis-9453 12 138 of of ADP fcis-9453 12 139 defects defect NOUN fcis-9453 12 140 . . PUNCT fcis-9453 13 1 with with ADP fcis-9453 13 2 the the DET fcis-9453 13 3 rapid rapid ADJ fcis-9453 13 4 development development NOUN fcis-9453 13 5 of of ADP fcis-9453 13 6 computer computer NOUN fcis-9453 13 7 vision vision NOUN fcis-9453 13 8 and and CCONJ fcis-9453 13 9 deep deep ADJ fcis-9453 13 10 learning learning NOUN fcis-9453 13 11 [ [ X fcis-9453 13 12 7 7 NUM fcis-9453 13 13 ] ] PUNCT fcis-9453 13 14 , , PUNCT fcis-9453 13 15 target target NOUN fcis-9453 13 16 detection detection NOUN fcis-9453 13 17 algorithms algorithm NOUN fcis-9453 13 18 based base VERB fcis-9453 13 19 on on ADP fcis-9453 13 20 deep deep ADJ fcis-9453 13 21 neural neural ADJ fcis-9453 13 22 networks network NOUN fcis-9453 14 1 [ [ X fcis-9453 14 2 8 8 NUM fcis-9453 14 3 ] ] PUNCT fcis-9453 14 4 are be AUX fcis-9453 14 5 widely widely ADV fcis-9453 14 6 used use VERB fcis-9453 14 7 in in ADP fcis-9453 14 8 defect defect ADJ fcis-9453 14 9 detection detection NOUN fcis-9453 14 10 . . PUNCT fcis-9453 15 1 at at ADP fcis-9453 15 2 this this DET fcis-9453 15 3 stage stage NOUN fcis-9453 15 4 , , PUNCT fcis-9453 15 5 target target VERB fcis-9453 15 6 detection detection NOUN fcis-9453 15 7 algorithms algorithm NOUN fcis-9453 15 8 [ [ X fcis-9453 15 9 9 9 NUM fcis-9453 15 10 ] ] PUNCT fcis-9453 15 11 can can AUX fcis-9453 15 12 be be AUX fcis-9453 15 13 divided divide VERB fcis-9453 15 14 into into ADP fcis-9453 15 15 two two NUM fcis-9453 15 16 categories category NOUN fcis-9453 15 17 according accord VERB fcis-9453 15 18 to to ADP fcis-9453 15 19 the the DET fcis-9453 15 20 existence existence NOUN fcis-9453 15 21 of of ADP fcis-9453 15 22 candidate candidate NOUN fcis-9453 15 23 regions region NOUN fcis-9453 15 24 : : PUNCT fcis-9453 15 25 one one NUM fcis-9453 15 26 is be AUX fcis-9453 15 27 the the DET fcis-9453 15 28 two two NUM fcis-9453 15 29 - - PUNCT fcis-9453 15 30 stage stage NOUN fcis-9453 15 31 target target NOUN fcis-9453 15 32 detection detection NOUN fcis-9453 15 33 algorithm algorithm NOUN fcis-9453 15 34 represented represent VERB fcis-9453 15 35 by by ADP fcis-9453 15 36 rcnn rcnn PROPN fcis-9453 16 1 [ [ X fcis-9453 16 2 10 10 NUM fcis-9453 16 3 ] ] PUNCT fcis-9453 16 4 , , PUNCT fcis-9453 16 5 sppnet sppnet PROPN fcis-9453 16 6 [ [ X fcis-9453 16 7 11 11 NUM fcis-9453 16 8 ] ] PUNCT fcis-9453 16 9 , , PUNCT fcis-9453 16 10 fast fast ADJ fcis-9453 16 11 rcnn rcnn NOUN fcis-9453 17 1 [ [ X fcis-9453 17 2 12 12 NUM fcis-9453 17 3 ] ] X fcis-9453 17 4 , , PUNCT fcis-9453 17 5 faster fast ADJ fcis-9453 17 6 rcnn rcnn NOUN fcis-9453 18 1 [ [ X fcis-9453 18 2 13 13 NUM fcis-9453 18 3 ] ] PUNCT fcis-9453 18 4 ; ; PUNCT fcis-9453 18 5 the the DET fcis-9453 18 6 other other ADJ fcis-9453 18 7 is be AUX fcis-9453 18 8 the the DET fcis-9453 18 9 single single ADJ fcis-9453 18 10 - - PUNCT fcis-9453 18 11 stage stage NOUN fcis-9453 18 12 target target NOUN fcis-9453 18 13 detection detection NOUN fcis-9453 18 14 algorithm algorithm NOUN fcis-9453 19 1 [ [ X fcis-9453 19 2 14 14 NUM fcis-9453 19 3 ] ] PUNCT fcis-9453 19 4 represented represent VERB fcis-9453 19 5 by by ADP fcis-9453 19 6 ssd ssd NOUN fcis-9453 20 1 [ [ X fcis-9453 20 2 15 15 NUM fcis-9453 20 3 ] ] X fcis-9453 20 4 , , PUNCT fcis-9453 20 5 yolo yolo ADJ fcis-9453 20 6 series series NOUN fcis-9453 20 7 [ [ X fcis-9453 20 8 16 16 NUM fcis-9453 20 9 ] ] PUNCT fcis-9453 20 10 , , PUNCT fcis-9453 20 11 retinanet retinanet NOUN fcis-9453 21 1 [ [ X fcis-9453 21 2 17 17 NUM fcis-9453 21 3 ] ] PUNCT fcis-9453 21 4 . . PUNCT fcis-9453 22 1 at at ADP fcis-9453 22 2 present present ADJ fcis-9453 22 3 , , PUNCT fcis-9453 22 4 the the DET fcis-9453 22 5 development development NOUN fcis-9453 22 6 of of ADP fcis-9453 22 7 deep deep ADJ fcis-9453 22 8 learning learning NOUN fcis-9453 22 9 - - PUNCT fcis-9453 22 10 based base VERB fcis-9453 22 11 surface surface NOUN fcis-9453 22 12 defect defect NOUN fcis-9453 22 13 detection detection NOUN fcis-9453 22 14 technology technology NOUN fcis-9453 22 15 for for ADP fcis-9453 22 16 hot hot ADV fcis-9453 22 17 - - PUNCT fcis-9453 22 18 rolled roll VERB fcis-9453 22 19 strip strip NOUN fcis-9453 22 20 steel steel NOUN fcis-9453 22 21 has have AUX fcis-9453 22 22 advanced advance VERB fcis-9453 22 23 rapidly rapidly ADV fcis-9453 22 24 , , PUNCT fcis-9453 22 25 and and CCONJ fcis-9453 22 26 the the DET fcis-9453 22 27 method method NOUN fcis-9453 22 28 not not PART fcis-9453 22 29 only only ADV fcis-9453 22 30 improves improve VERB fcis-9453 22 31 the the DET fcis-9453 22 32 detection detection NOUN fcis-9453 22 33 accuracy accuracy NOUN fcis-9453 22 34 and and CCONJ fcis-9453 22 35 precision precision NOUN fcis-9453 22 36 , , PUNCT fcis-9453 22 37 but but CCONJ fcis-9453 22 38 also also ADV fcis-9453 22 39 saves save VERB fcis-9453 22 40 a a DET fcis-9453 22 41 lot lot NOUN fcis-9453 22 42 of of ADP fcis-9453 22 43 labor labor NOUN fcis-9453 22 44 costs cost NOUN fcis-9453 22 45 , , PUNCT fcis-9453 22 46 and and CCONJ fcis-9453 22 47 is be AUX fcis-9453 22 48 widely widely ADV fcis-9453 22 49 used use VERB fcis-9453 22 50 in in ADP fcis-9453 22 51 practical practical ADJ fcis-9453 22 52 production production NOUN fcis-9453 22 53 in in ADP fcis-9453 22 54 . . PUNCT fcis-9453 23 1 an an DET fcis-9453 23 2 improved improved ADJ fcis-9453 23 3 yolov3 yolov3 PROPN fcis-9453 23 4 algorithm algorithm PROPN fcis-9453 23 5 model model NOUN fcis-9453 23 6 is be AUX fcis-9453 23 7 proposed propose VERB fcis-9453 23 8 in in ADP fcis-9453 23 9 the the DET fcis-9453 23 10 literature literature NOUN fcis-9453 23 11 [ [ X fcis-9453 23 12 18 18 NUM fcis-9453 23 13 ] ] PUNCT fcis-9453 23 14 , , PUNCT fcis-9453 23 15 using use VERB fcis-9453 23 16 a a DET fcis-9453 23 17 weighted weighted ADJ fcis-9453 23 18 k k ADJ fcis-9453 23 19 - - PUNCT fcis-9453 23 20 means mean VERB fcis-9453 23 21 clustering cluster VERB fcis-9453 23 22 algorithm algorithm NOUN fcis-9453 23 23 to to PART fcis-9453 23 24 improve improve VERB fcis-9453 23 25 the the DET fcis-9453 23 26 matching matching NOUN fcis-9453 23 27 graph graph NOUN fcis-9453 23 28 of of ADP fcis-9453 23 29 the the DET fcis-9453 23 30 a a DET fcis-9453 23 31 priori priori ADJ fcis-9453 23 32 frame frame NOUN fcis-9453 23 33 and and CCONJ fcis-9453 23 34 the the DET fcis-9453 23 35 feature feature NOUN fcis-9453 23 36 layer layer NOUN fcis-9453 23 37 , , PUNCT fcis-9453 23 38 which which PRON fcis-9453 23 39 improves improve VERB fcis-9453 23 40 the the DET fcis-9453 23 41 inspection inspection NOUN fcis-9453 23 42 accuracy accuracy NOUN fcis-9453 23 43 of of ADP fcis-9453 23 44 the the DET fcis-9453 23 45 algorithm algorithm NOUN fcis-9453 23 46 . . PUNCT fcis-9453 24 1 wang wang PROPN fcis-9453 24 2 daolei daolei VERB fcis-9453 24 3 et et PROPN fcis-9453 24 4 al al PROPN fcis-9453 25 1 [ [ X fcis-9453 25 2 19 19 NUM fcis-9453 25 3 ] ] PUNCT fcis-9453 25 4 proposed propose VERB fcis-9453 25 5 an an DET fcis-9453 25 6 improved improved ADJ fcis-9453 25 7 algorithm algorithm NOUN fcis-9453 25 8 based base VERB fcis-9453 25 9 on on ADP fcis-9453 25 10 yolov4 yolov4 PROPN fcis-9453 25 11 - - PUNCT fcis-9453 25 12 tiny tiny ADJ fcis-9453 25 13 , , PUNCT fcis-9453 25 14 which which PRON fcis-9453 25 15 combined combine VERB fcis-9453 25 16 multi multi ADJ fcis-9453 25 17 - - ADJ fcis-9453 25 18 scale scale ADJ fcis-9453 25 19 detection detection NOUN fcis-9453 25 20 and and CCONJ fcis-9453 25 21 attention attention NOUN fcis-9453 25 22 mechanism mechanism NOUN fcis-9453 25 23 to to PART fcis-9453 25 24 improve improve VERB fcis-9453 25 25 lightweight lightweight ADJ fcis-9453 25 26 target target NOUN fcis-9453 25 27 detection detection NOUN fcis-9453 25 28 accuracy accuracy NOUN fcis-9453 25 29 . . PUNCT fcis-9453 26 1 for for ADP fcis-9453 26 2 the the DET fcis-9453 26 3 problems problem NOUN fcis-9453 26 4 of of ADP fcis-9453 26 5 small small ADJ fcis-9453 26 6 size size NOUN fcis-9453 26 7 of of ADP fcis-9453 26 8 strip strip PROPN fcis-9453 26 9 steel steel NOUN fcis-9453 26 10 surface surface NOUN fcis-9453 26 11 defects defect NOUN fcis-9453 26 12 , , PUNCT fcis-9453 26 13 fuzzy fuzzy ADJ fcis-9453 26 14 features feature NOUN fcis-9453 26 15 and and CCONJ fcis-9453 26 16 easy easy ADJ fcis-9453 26 17 to to PART fcis-9453 26 18 miss miss VERB fcis-9453 26 19 detection detection NOUN fcis-9453 26 20 . . PUNCT fcis-9453 27 1 zhou zhou PROPN fcis-9453 27 2 jinwei jinwei PROPN fcis-9453 27 3 et et PROPN fcis-9453 27 4 al al PROPN fcis-9453 28 1 [ [ X fcis-9453 28 2 20 20 NUM fcis-9453 28 3 ] ] PUNCT fcis-9453 28 4 proposed propose VERB fcis-9453 28 5 an an DET fcis-9453 28 6 improved improved ADJ fcis-9453 28 7 algorithm algorithm NOUN fcis-9453 28 8 based base VERB fcis-9453 28 9 on on ADP fcis-9453 28 10 yolov5 yolov5 NOUN fcis-9453 28 11 by by ADP fcis-9453 28 12 designing design VERB fcis-9453 28 13 a a DET fcis-9453 28 14 new new ADJ fcis-9453 28 15 feature feature NOUN fcis-9453 28 16 extraction extraction NOUN fcis-9453 28 17 module module NOUN fcis-9453 28 18 and and CCONJ fcis-9453 28 19 modifying modify VERB fcis-9453 28 20 the the DET fcis-9453 28 21 confidence confidence NOUN fcis-9453 28 22 loss loss NOUN fcis-9453 28 23 function function NOUN fcis-9453 28 24 to to PART fcis-9453 28 25 improve improve VERB fcis-9453 28 26 the the DET fcis-9453 28 27 stability stability NOUN fcis-9453 28 28 of of ADP fcis-9453 28 29 the the DET fcis-9453 28 30 algorithm algorithm NOUN fcis-9453 28 31 convergence convergence NOUN fcis-9453 28 32 . . PUNCT fcis-9453 29 1 liu liu PROPN fcis-9453 29 2 jinchuan jinchuan PROPN fcis-9453 29 3 et et PROPN fcis-9453 30 1 al al PROPN fcis-9453 31 1 [ [ X fcis-9453 31 2 21 21 NUM fcis-9453 31 3 ] ] PUNCT fcis-9453 31 4 added add VERB fcis-9453 31 5 a a DET fcis-9453 31 6 small small ADJ fcis-9453 31 7 target target NOUN fcis-9453 31 8 detection detection NOUN fcis-9453 31 9 layer layer NOUN fcis-9453 31 10 to to PART fcis-9453 31 11 address address VERB fcis-9453 31 12 the the DET fcis-9453 31 13 problem problem NOUN fcis-9453 31 14 of of ADP fcis-9453 31 15 small small ADJ fcis-9453 31 16 target target NOUN fcis-9453 31 17 miss miss NOUN fcis-9453 31 18 and and CCONJ fcis-9453 31 19 error error NOUN fcis-9453 31 20 detection detection NOUN fcis-9453 31 21 ; ; PUNCT fcis-9453 31 22 and and CCONJ fcis-9453 31 23 introduced introduce VERB fcis-9453 31 24 the the DET fcis-9453 31 25 transformer transformer NOUN fcis-9453 31 26 encoder encoder NOUN fcis-9453 31 27 block block NOUN fcis-9453 31 28 module module NOUN fcis-9453 31 29 and and CCONJ fcis-9453 31 30 the the DET fcis-9453 31 31 convolutional convolutional ADJ fcis-9453 31 32 block block NOUN fcis-9453 31 33 attention attention NOUN fcis-9453 31 34 model model NOUN fcis-9453 31 35 ( ( PUNCT fcis-9453 31 36 cbam cbam NOUN fcis-9453 31 37 ) ) PUNCT fcis-9453 31 38 attention attention NOUN fcis-9453 31 39 mechanism mechanism NOUN fcis-9453 31 40 module module NOUN fcis-9453 31 41 to to PART fcis-9453 31 42 address address VERB fcis-9453 31 43 the the DET fcis-9453 31 44 problems problem NOUN fcis-9453 31 45 of of ADP fcis-9453 31 46 image image NOUN fcis-9453 31 47 crossover crossover NOUN fcis-9453 31 48 and and CCONJ fcis-9453 31 49 overlap overlap VERB fcis-9453 31 50 , , PUNCT fcis-9453 31 51 improving improve VERB fcis-9453 31 52 the the DET fcis-9453 31 53 detection detection NOUN fcis-9453 31 54 capability capability NOUN fcis-9453 31 55 of of ADP fcis-9453 31 56 the the DET fcis-9453 31 57 algorithm algorithm NOUN fcis-9453 31 58 in in ADP fcis-9453 31 59 complex complex ADJ fcis-9453 31 60 backgrounds background NOUN fcis-9453 31 61 . . PUNCT fcis-9453 32 1 pan pan PROPN fcis-9453 32 2 meng meng PROPN fcis-9453 32 3 et et PROPN fcis-9453 32 4 al al PROPN fcis-9453 33 1 [ [ X fcis-9453 33 2 22 22 NUM fcis-9453 33 3 ] ] PUNCT fcis-9453 33 4 introduced introduce VERB fcis-9453 33 5 1x1 1x1 NUM fcis-9453 33 6 convolutional convolutional ADJ fcis-9453 33 7 side side NOUN fcis-9453 33 8 branches branch NOUN fcis-9453 33 9 by by ADP fcis-9453 33 10 reconstructing reconstruct VERB fcis-9453 33 11 convolution convolution NOUN fcis-9453 33 12 to to PART fcis-9453 33 13 improve improve VERB fcis-9453 33 14 the the DET fcis-9453 33 15 feature feature NOUN fcis-9453 33 16 extraction extraction NOUN fcis-9453 33 17 capability capability NOUN fcis-9453 33 18 of of ADP fcis-9453 33 19 the the DET fcis-9453 33 20 network network NOUN fcis-9453 33 21 ; ; PUNCT fcis-9453 33 22 added add VERB fcis-9453 33 23 an an DET fcis-9453 33 24 attention attention NOUN fcis-9453 33 25 mechanism mechanism NOUN fcis-9453 33 26 with with ADP fcis-9453 33 27 channels channel NOUN fcis-9453 33 28 to to PART fcis-9453 33 29 retain retain VERB fcis-9453 33 30 more more ADJ fcis-9453 33 31 spatial spatial ADJ fcis-9453 33 32 information information NOUN fcis-9453 33 33 ; ; PUNCT fcis-9453 33 34 and and CCONJ fcis-9453 33 35 switched switch VERB fcis-9453 33 36 to to ADP fcis-9453 33 37 a a DET fcis-9453 33 38 weighted weight VERB fcis-9453 33 39 bidirectional bidirectional ADJ fcis-9453 33 40 feature feature NOUN fcis-9453 33 41 pyramid pyramid NOUN fcis-9453 33 42 network network NOUN fcis-9453 33 43 to to PART fcis-9453 33 44 improve improve VERB fcis-9453 33 45 small small ADJ fcis-9453 33 46 target target NOUN fcis-9453 33 47 detection detection NOUN fcis-9453 33 48 . . PUNCT fcis-9453 34 1 wang wang PROPN fcis-9453 34 2 bo bo PROPN fcis-9453 34 3 et et PROPN fcis-9453 34 4 al al PROPN fcis-9453 35 1 [ [ X fcis-9453 35 2 23 23 NUM fcis-9453 35 3 ] ] PUNCT fcis-9453 35 4 enhanced enhance VERB fcis-9453 35 5 the the DET fcis-9453 35 6 fusion fusion NOUN fcis-9453 35 7 of of ADP fcis-9453 35 8 image image NOUN fcis-9453 35 9 information information NOUN fcis-9453 35 10 by by ADP fcis-9453 35 11 combining combine VERB fcis-9453 35 12 the the DET fcis-9453 35 13 transformer transformer NOUN fcis-9453 35 14 layer layer NOUN fcis-9453 35 15 with with ADP fcis-9453 35 16 the the DET fcis-9453 35 17 bifpn bifpn PROPN fcis-9453 35 18 network network NOUN fcis-9453 35 19 structure structure NOUN fcis-9453 35 20 ; ; PUNCT fcis-9453 35 21 replaced replace VERB fcis-9453 35 22 the the DET fcis-9453 35 23 convolutional convolutional ADJ fcis-9453 35 24 layer layer NOUN fcis-9453 35 25 in in ADP fcis-9453 35 26 the the DET fcis-9453 35 27 backbone backbone NOUN fcis-9453 35 28 network network NOUN fcis-9453 35 29 with with ADP fcis-9453 35 30 a a DET fcis-9453 35 31 lightweight lightweight ADJ fcis-9453 35 32 network network NOUN fcis-9453 35 33 , , PUNCT fcis-9453 35 34 repvgg repvgg NOUN fcis-9453 35 35 , , PUNCT fcis-9453 35 36 to to PART fcis-9453 35 37 enhance enhance VERB fcis-9453 35 38 the the DET fcis-9453 35 39 feature feature NOUN fcis-9453 35 40 extraction extraction NOUN fcis-9453 35 41 capability capability NOUN fcis-9453 35 42 of of ADP fcis-9453 35 43 the the DET fcis-9453 35 44 backbone backbone NOUN fcis-9453 35 45 network network NOUN fcis-9453 35 46 ; ; PUNCT fcis-9453 35 47 and and CCONJ fcis-9453 35 48 added add VERB fcis-9453 35 49 a a DET fcis-9453 35 50 prediction prediction NOUN fcis-9453 35 51 layer layer NOUN fcis-9453 35 52 to to PART fcis-9453 35 53 improve improve VERB fcis-9453 35 54 the the DET fcis-9453 35 55 multiscale multiscale ADJ fcis-9453 35 56 target target NOUN fcis-9453 35 57 detection detection NOUN fcis-9453 35 58 capability capability NOUN fcis-9453 35 59 . . PUNCT fcis-9453 36 1 to to PART fcis-9453 36 2 address address VERB fcis-9453 36 3 the the DET fcis-9453 36 4 problems problem NOUN fcis-9453 36 5 of of ADP fcis-9453 36 6 varying vary VERB fcis-9453 36 7 size size NOUN fcis-9453 36 8 and and CCONJ fcis-9453 36 9 uneven uneven ADJ fcis-9453 36 10 distribution distribution NOUN fcis-9453 36 11 of of ADP fcis-9453 36 12 strip strip NOUN fcis-9453 36 13 surface surface NOUN fcis-9453 36 14 defects defect NOUN fcis-9453 36 15 , , PUNCT fcis-9453 36 16 multiple multiple ADJ fcis-9453 36 17 types type NOUN fcis-9453 36 18 of of ADP fcis-9453 36 19 defects defect NOUN fcis-9453 36 20 and and CCONJ fcis-9453 36 21 complex complex ADJ fcis-9453 36 22 backgrounds background NOUN fcis-9453 36 23 , , PUNCT fcis-9453 36 24 this this DET fcis-9453 36 25 paper paper NOUN fcis-9453 36 26 proposes propose VERB fcis-9453 36 27 an an DET fcis-9453 36 28 improved improved ADJ fcis-9453 36 29 yolov5l yolov5l NOUN fcis-9453 36 30 - - PUNCT fcis-9453 36 31 based base VERB fcis-9453 36 32 algorithm algorithm NOUN fcis-9453 36 33 for for ADP fcis-9453 36 34 detecting detect VERB fcis-9453 36 35 strip strip NOUN fcis-9453 36 36 surface surface NOUN fcis-9453 36 37 defects defect NOUN fcis-9453 36 38 , , PUNCT fcis-9453 36 39 68 68 NUM fcis-9453 36 40 which which PRON fcis-9453 36 41 improves improve VERB fcis-9453 36 42 the the DET fcis-9453 36 43 detection detection NOUN fcis-9453 36 44 accuracy accuracy NOUN fcis-9453 36 45 of of ADP fcis-9453 36 46 the the DET fcis-9453 36 47 model model NOUN fcis-9453 36 48 while while SCONJ fcis-9453 36 49 satisfying satisfy VERB fcis-9453 36 50 high high ADJ fcis-9453 36 51 detection detection NOUN fcis-9453 36 52 accuracy accuracy NOUN fcis-9453 36 53 and and CCONJ fcis-9453 36 54 basically basically ADV fcis-9453 36 55 unchanged unchanged ADJ fcis-9453 36 56 number number NOUN fcis-9453 36 57 of of ADP fcis-9453 36 58 parameters parameter NOUN fcis-9453 36 59 and and CCONJ fcis-9453 36 60 computational computational ADJ fcis-9453 36 61 complexity complexity NOUN fcis-9453 36 62 , , PUNCT fcis-9453 36 63 adds add VERB fcis-9453 36 64 the the DET fcis-9453 36 65 simam simam ADJ fcis-9453 36 66 attention attention NOUN fcis-9453 36 67 mechanism mechanism NOUN fcis-9453 36 68 module module NOUN fcis-9453 36 69 at at ADP fcis-9453 36 70 the the DET fcis-9453 36 71 head head NOUN fcis-9453 36 72 end end NOUN fcis-9453 36 73 and and CCONJ fcis-9453 36 74 replaces replace VERB fcis-9453 36 75 the the DET fcis-9453 36 76 c3 c3 NOUN fcis-9453 36 77 module module NOUN fcis-9453 36 78 with with ADP fcis-9453 36 79 the the DET fcis-9453 36 80 c2f c2f NOUN fcis-9453 36 81 module module NOUN fcis-9453 36 82 to to PART fcis-9453 36 83 improve improve VERB fcis-9453 36 84 the the DET fcis-9453 36 85 detection detection NOUN fcis-9453 36 86 performance performance NOUN fcis-9453 36 87 of of ADP fcis-9453 36 88 the the DET fcis-9453 36 89 algorithm algorithm NOUN fcis-9453 36 90 and and CCONJ fcis-9453 36 91 meet meet VERB fcis-9453 36 92 the the DET fcis-9453 36 93 needs need NOUN fcis-9453 36 94 of of ADP fcis-9453 36 95 industrial industrial ADJ fcis-9453 36 96 deployment deployment NOUN fcis-9453 36 97 . . PUNCT fcis-9453 37 1 2 2 NUM fcis-9453 37 2 . . X fcis-9453 37 3 yolov5l yolov5l PROPN fcis-9453 37 4 algorithm algorithm NOUN fcis-9453 37 5 and and CCONJ fcis-9453 37 6 improvements improvement NOUN fcis-9453 37 7 2.1 2.1 NUM fcis-9453 37 8 . . PUNCT fcis-9453 38 1 introduction introduction NOUN fcis-9453 38 2 to to ADP fcis-9453 38 3 the the DET fcis-9453 38 4 yolov5l yolov5l PROPN fcis-9453 38 5 algorithm algorithm NOUN fcis-9453 38 6 the the DET fcis-9453 38 7 yolo yolo ADJ fcis-9453 38 8 series series NOUN fcis-9453 38 9 of of ADP fcis-9453 38 10 algorithms algorithm NOUN fcis-9453 38 11 are be AUX fcis-9453 38 12 deep deep ADJ fcis-9453 38 13 learning learning NOUN fcis-9453 38 14 - - PUNCT fcis-9453 38 15 based base VERB fcis-9453 38 16 regression regression NOUN fcis-9453 38 17 methods method NOUN fcis-9453 38 18 that that PRON fcis-9453 38 19 use use VERB fcis-9453 38 20 only only ADV fcis-9453 38 21 a a DET fcis-9453 38 22 single single ADJ fcis-9453 38 23 convolutional convolutional ADJ fcis-9453 38 24 neural neural ADJ fcis-9453 38 25 network network NOUN fcis-9453 38 26 ( ( PUNCT fcis-9453 38 27 cnn cnn PROPN fcis-9453 38 28 ) ) PUNCT fcis-9453 38 29 network network NOUN fcis-9453 38 30 to to PART fcis-9453 38 31 directly directly ADV fcis-9453 38 32 predict predict VERB fcis-9453 38 33 the the DET fcis-9453 38 34 class class NOUN fcis-9453 38 35 and and CCONJ fcis-9453 38 36 location location NOUN fcis-9453 38 37 of of ADP fcis-9453 38 38 different different ADJ fcis-9453 38 39 targets target NOUN fcis-9453 38 40 . . PUNCT fcis-9453 39 1 yolov5 yolov5 NOUN fcis-9453 39 2 is be AUX fcis-9453 39 3 an an DET fcis-9453 39 4 improvement improvement NOUN fcis-9453 39 5 on on ADP fcis-9453 39 6 the the DET fcis-9453 39 7 network network NOUN fcis-9453 39 8 structure structure NOUN fcis-9453 39 9 of of ADP fcis-9453 39 10 yolov4 yolov4 PROPN fcis-9453 39 11 , , PUNCT fcis-9453 39 12 with with SCONJ fcis-9453 39 13 four four NUM fcis-9453 39 14 models model NOUN fcis-9453 39 15 of of ADP fcis-9453 39 16 target target NOUN fcis-9453 39 17 detection detection NOUN fcis-9453 39 18 networks network NOUN fcis-9453 39 19 , , PUNCT fcis-9453 39 20 yolov5s yolov5s PROPN fcis-9453 39 21 , , PUNCT fcis-9453 39 22 yolov5 yolov5 PROPN fcis-9453 39 23 m m PROPN fcis-9453 39 24 , , PUNCT fcis-9453 39 25 yolov5l yolov5l PROPN fcis-9453 39 26 and and CCONJ fcis-9453 39 27 yolov5x yolov5x PROPN fcis-9453 39 28 , , PUNCT fcis-9453 39 29 based base VERB fcis-9453 39 30 on on ADP fcis-9453 39 31 the the DET fcis-9453 39 32 network network NOUN fcis-9453 39 33 structure structure NOUN fcis-9453 39 34 of of ADP fcis-9453 39 35 yolov5l yolov5l PROPN fcis-9453 39 36 version version PROPN fcis-9453 39 37 6.0 6.0 NUM fcis-9453 39 38 is be AUX fcis-9453 39 39 divided divide VERB fcis-9453 39 40 into into ADP fcis-9453 39 41 four four NUM fcis-9453 39 42 modules module NOUN fcis-9453 39 43 : : PUNCT fcis-9453 39 44 input input NOUN fcis-9453 39 45 , , PUNCT fcis-9453 39 46 backbone backbone NOUN fcis-9453 39 47 , , PUNCT fcis-9453 39 48 neck neck NOUN fcis-9453 39 49 and and CCONJ fcis-9453 39 50 head head NOUN fcis-9453 39 51 , , PUNCT fcis-9453 39 52 which which PRON fcis-9453 39 53 improves improve VERB fcis-9453 39 54 the the DET fcis-9453 39 55 detection detection NOUN fcis-9453 39 56 accuracy accuracy NOUN fcis-9453 39 57 and and CCONJ fcis-9453 39 58 learning learn VERB fcis-9453 39 59 speed speed NOUN fcis-9453 39 60 compared compare VERB fcis-9453 39 61 with with ADP fcis-9453 39 62 yolov4 yolov4 PROPN fcis-9453 40 1 [ [ X fcis-9453 40 2 24 24 NUM fcis-9453 40 3 ] ] PUNCT fcis-9453 40 4 . . PUNCT fcis-9453 41 1 the the DET fcis-9453 41 2 specific specific ADJ fcis-9453 41 3 structure structure NOUN fcis-9453 41 4 of of ADP fcis-9453 41 5 the the DET fcis-9453 41 6 network network NOUN fcis-9453 41 7 in in ADP fcis-9453 41 8 version version NOUN fcis-9453 41 9 yolov5l yolov5l PROPN fcis-9453 41 10 6.0 6.0 NUM fcis-9453 41 11 is be AUX fcis-9453 41 12 shown show VERB fcis-9453 41 13 in in ADP fcis-9453 41 14 figure figure NOUN fcis-9453 41 15 1 1 NUM fcis-9453 41 16 . . PUNCT fcis-9453 41 17 input input PROPN fcis-9453 41 18 c3 c3 PROPN fcis-9453 41 19 conv conv PROPN fcis-9453 41 20 c3 c3 PROPN fcis-9453 41 21 sppf sppf PROPN fcis-9453 41 22 conv conv PROPN fcis-9453 41 23 conv conv PROPN fcis-9453 41 24 c3 c3 PROPN fcis-9453 41 25 conv conv PROPN fcis-9453 41 26 contact contact PROPN fcis-9453 41 27 unsample unsample PROPN fcis-9453 41 28 conv conv PROPN fcis-9453 41 29 c3 c3 PROPN fcis-9453 41 30 contact contact PROPN fcis-9453 41 31 unsample unsample PROPN fcis-9453 41 32 conv conv PROPN fcis-9453 41 33 c3 c3 PROPN fcis-9453 41 34 contact contact PROPN fcis-9453 41 35 conv conv PROPN fcis-9453 41 36 c3 c3 PROPN fcis-9453 41 37 contact contact PROPN fcis-9453 41 38 conv conv PROPN fcis-9453 41 39 c3 c3 PROPN fcis-9453 41 40 conv conv PROPN fcis-9453 41 41 c3 c3 PROPN fcis-9453 41 42 headneck headneck PROPN fcis-9453 41 43 backbone backbone PROPN fcis-9453 41 44 80x80x255 80x80x255 NUM fcis-9453 41 45 40x40x255 40x40x255 NUM fcis-9453 41 46 20x20x255 20x20x255 NUM fcis-9453 41 47     SPACE fcis-9453 41 48 figure figure NOUN fcis-9453 41 49 1 1 NUM fcis-9453 41 50 . . PUNCT fcis-9453 41 51 yolov5l yolov5l PROPN fcis-9453 41 52 structure structure PROPN fcis-9453 41 53 diagram diagram PROPN fcis-9453 41 54 2.1.1 2.1.1 NUM fcis-9453 41 55 . . PUNCT fcis-9453 42 1 input input NOUN fcis-9453 42 2 compared compare VERB fcis-9453 42 3 with with ADP fcis-9453 42 4 yolov4 yolov4 PROPN fcis-9453 42 5 , , PUNCT fcis-9453 42 6 yolov5l yolov5l PROPN fcis-9453 42 7 uses use VERB fcis-9453 42 8 mosaic mosaic ADJ fcis-9453 42 9 data datum NOUN fcis-9453 42 10 augmentation augmentation NOUN fcis-9453 42 11 on on ADP fcis-9453 42 12 the the DET fcis-9453 42 13 input input NOUN fcis-9453 42 14 side side NOUN fcis-9453 42 15 of of ADP fcis-9453 42 16 its its PRON fcis-9453 42 17 network network NOUN fcis-9453 42 18 structure structure NOUN fcis-9453 42 19 to to PART fcis-9453 42 20 randomly randomly VERB fcis-9453 42 21 scale scale NOUN fcis-9453 42 22 , , PUNCT fcis-9453 42 23 crop crop NOUN fcis-9453 42 24 and and CCONJ fcis-9453 42 25 arrange arrange VERB fcis-9453 42 26 the the DET fcis-9453 42 27 dataset dataset NOUN fcis-9453 42 28 , , PUNCT fcis-9453 42 29 thus thus ADV fcis-9453 42 30 improving improve VERB fcis-9453 42 31 the the DET fcis-9453 42 32 small small ADJ fcis-9453 42 33 target target NOUN fcis-9453 42 34 detection detection NOUN fcis-9453 42 35 accuracy accuracy NOUN fcis-9453 42 36 ; ; PUNCT fcis-9453 42 37 secondly secondly ADV fcis-9453 42 38 , , PUNCT fcis-9453 42 39 it it PRON fcis-9453 42 40 adds add VERB fcis-9453 42 41 adaptive adaptive ADJ fcis-9453 42 42 anchor anchor NOUN fcis-9453 42 43 frames frame NOUN fcis-9453 42 44 to to PART fcis-9453 42 45 calculate calculate VERB fcis-9453 42 46 the the DET fcis-9453 42 47 best good ADJ fcis-9453 42 48 anchor anchor NOUN fcis-9453 42 49 frame frame NOUN fcis-9453 42 50 values value NOUN fcis-9453 42 51 for for ADP fcis-9453 42 52 different different ADJ fcis-9453 42 53 training training NOUN fcis-9453 42 54 sets set NOUN fcis-9453 42 55 ; ; PUNCT fcis-9453 42 56 finally finally ADV fcis-9453 42 57 , , PUNCT fcis-9453 42 58 it it PRON fcis-9453 42 59 uses use VERB fcis-9453 42 60 adaptive adaptive ADJ fcis-9453 42 61 image image NOUN fcis-9453 42 62 scaling scale VERB fcis-9453 42 63 to to PART fcis-9453 42 64 adaptively adaptively ADV fcis-9453 42 65 add add VERB fcis-9453 42 66 the the DET fcis-9453 42 67 least least ADJ fcis-9453 42 68 black black ADJ fcis-9453 42 69 edges edge NOUN fcis-9453 42 70 to to ADP fcis-9453 42 71 the the DET fcis-9453 42 72 original original ADJ fcis-9453 42 73 image image NOUN fcis-9453 42 74 to to PART fcis-9453 42 75 reduce reduce VERB fcis-9453 42 76 information information NOUN fcis-9453 42 77 redundancy redundancy NOUN fcis-9453 42 78 , , PUNCT fcis-9453 42 79 reduce reduce VERB fcis-9453 42 80 computational computational ADJ fcis-9453 42 81 effort effort NOUN fcis-9453 42 82 and and CCONJ fcis-9453 42 83 improve improve VERB fcis-9453 42 84 inference inference NOUN fcis-9453 42 85 speed speed NOUN fcis-9453 42 86 . . PUNCT fcis-9453 43 1 2.1.2 2.1.2 X fcis-9453 43 2 . . PUNCT fcis-9453 43 3 backbone backbone PROPN fcis-9453 43 4 yolov5l yolov5l PROPN fcis-9453 43 5 version version PROPN fcis-9453 43 6 6.0 6.0 NUM fcis-9453 43 7 of of ADP fcis-9453 43 8 backbone backbone NOUN fcis-9453 43 9 is be AUX fcis-9453 43 10 mainly mainly ADV fcis-9453 43 11 divided divide VERB fcis-9453 43 12 into into ADP fcis-9453 43 13 conv conv ADJ fcis-9453 43 14 module module NOUN fcis-9453 43 15 , , PUNCT fcis-9453 43 16 cspdarknet53 cspdarknet53 NOUN fcis-9453 43 17 and and CCONJ fcis-9453 43 18 sppf sppf ADJ fcis-9453 43 19 module module NOUN fcis-9453 43 20 . . PUNCT fcis-9453 44 1 among among ADP fcis-9453 44 2 them they PRON fcis-9453 44 3 , , PUNCT fcis-9453 44 4 the the DET fcis-9453 44 5 conv conv NOUN fcis-9453 44 6 module module NOUN fcis-9453 44 7 replaces replace VERB fcis-9453 44 8 focus focus NOUN fcis-9453 44 9 in in ADP fcis-9453 44 10 the the DET fcis-9453 44 11 old old ADJ fcis-9453 44 12 version version NOUN fcis-9453 44 13 to to PART fcis-9453 44 14 improve improve VERB fcis-9453 44 15 model model NOUN fcis-9453 44 16 efficiency efficiency NOUN fcis-9453 44 17 while while SCONJ fcis-9453 44 18 facilitating facilitate VERB fcis-9453 44 19 model model NOUN fcis-9453 44 20 export export NOUN fcis-9453 44 21 ; ; PUNCT fcis-9453 44 22 cspnet cspnet NOUN fcis-9453 44 23 reduces reduce VERB fcis-9453 44 24 computation computation NOUN fcis-9453 44 25 , , PUNCT fcis-9453 44 26 improves improve VERB fcis-9453 44 27 inference inference NOUN fcis-9453 44 28 speed speed NOUN fcis-9453 44 29 and and CCONJ fcis-9453 44 30 obtains obtain VERB fcis-9453 44 31 richer rich ADJ fcis-9453 44 32 gradient gradient ADJ fcis-9453 44 33 combination combination NOUN fcis-9453 44 34 information information NOUN fcis-9453 44 35 by by ADP fcis-9453 44 36 segmenting segment VERB fcis-9453 44 37 the the DET fcis-9453 44 38 gradient gradient NOUN fcis-9453 44 39 flow flow NOUN fcis-9453 44 40 , , PUNCT fcis-9453 44 41 while while SCONJ fcis-9453 44 42 ensuring ensure VERB fcis-9453 44 43 no no DET fcis-9453 44 44 degradation degradation NOUN fcis-9453 44 45 in in ADP fcis-9453 44 46 model model NOUN fcis-9453 44 47 detection detection NOUN fcis-9453 44 48 and and CCONJ fcis-9453 44 49 recognition recognition NOUN fcis-9453 44 50 accuracy accuracy NOUN fcis-9453 44 51 ; ; PUNCT fcis-9453 44 52 the the DET fcis-9453 44 53 spff spff PROPN fcis-9453 44 54 module module NOUN fcis-9453 44 55 uses use VERB fcis-9453 44 56 multiple multiple ADJ fcis-9453 44 57 small small ADJ fcis-9453 44 58 size size NOUN fcis-9453 44 59 pooling pool VERB fcis-9453 44 60 nucleus nucleus ADJ fcis-9453 44 61 cascade cascade NOUN fcis-9453 44 62 instead instead ADV fcis-9453 44 63 of of ADP fcis-9453 44 64 a a DET fcis-9453 44 65 single single ADJ fcis-9453 44 66 large large ADJ fcis-9453 44 67 size size NOUN fcis-9453 44 68 pooling pool VERB fcis-9453 44 69 nucleus nucleus NOUN fcis-9453 44 70 in in ADP fcis-9453 44 71 the the DET fcis-9453 44 72 spp spp NOUN fcis-9453 44 73 module module NOUN fcis-9453 44 74 , , PUNCT fcis-9453 44 75 which which PRON fcis-9453 44 76 improves improve VERB fcis-9453 44 77 the the DET fcis-9453 44 78 operation operation NOUN fcis-9453 44 79 speed speed NOUN fcis-9453 44 80 by by ADP fcis-9453 44 81 merging merge VERB fcis-9453 44 82 the the DET fcis-9453 44 83 feature feature NOUN fcis-9453 44 84 map map NOUN fcis-9453 44 85 of of ADP fcis-9453 44 86 different different ADJ fcis-9453 44 87 receptive receptive ADJ fcis-9453 44 88 fields field NOUN fcis-9453 44 89 and and CCONJ fcis-9453 44 90 enriching enrich VERB fcis-9453 44 91 the the DET fcis-9453 44 92 expression expression NOUN fcis-9453 44 93 ability ability NOUN fcis-9453 44 94 of of ADP fcis-9453 44 95 the the DET fcis-9453 44 96 feature feature NOUN fcis-9453 44 97 map map NOUN fcis-9453 44 98 . . PUNCT fcis-9453 45 1 2.1.3 2.1.3 NUM fcis-9453 45 2 . . PUNCT fcis-9453 45 3 neck neck PROPN fcis-9453 45 4 the the DET fcis-9453 45 5 yolov4 yolov4 PROPN fcis-9453 45 6 and and CCONJ fcis-9453 45 7 yolov5l yolov5l PROPN fcis-9453 45 8 neck neck NOUN fcis-9453 45 9 modules module NOUN fcis-9453 45 10 both both PRON fcis-9453 45 11 use use VERB fcis-9453 45 12 an an DET fcis-9453 45 13 fpn+pan fpn+pan NOUN fcis-9453 45 14 structure structure NOUN fcis-9453 45 15 to to PART fcis-9453 45 16 perform perform VERB fcis-9453 45 17 multi multi ADJ fcis-9453 45 18 - - ADJ fcis-9453 45 19 scale scale ADJ fcis-9453 45 20 feature feature NOUN fcis-9453 45 21 fusion fusion NOUN fcis-9453 45 22 of of ADP fcis-9453 45 23 strip strip NOUN fcis-9453 45 24 surface surface NOUN fcis-9453 45 25 defects defect NOUN fcis-9453 45 26 by by ADP fcis-9453 45 27 fpn fpn NOUN fcis-9453 45 28 and and CCONJ fcis-9453 45 29 pan pan PROPN fcis-9453 45 30 ( ( PUNCT fcis-9453 45 31 see see VERB fcis-9453 45 32 figure figure NOUN fcis-9453 45 33 2 2 NUM fcis-9453 45 34 ) ) PUNCT fcis-9453 45 35 to to PART fcis-9453 45 36 obtain obtain VERB fcis-9453 45 37 feature feature NOUN fcis-9453 45 38 maps map NOUN fcis-9453 45 39 at at ADP fcis-9453 45 40 three three NUM fcis-9453 45 41 different different ADJ fcis-9453 45 42 scales scale NOUN fcis-9453 45 43 ; ; PUNCT fcis-9453 45 44 fpn fpn VERB fcis-9453 45 45 conveys convey VERB fcis-9453 45 46 top top ADJ fcis-9453 45 47 - - PUNCT fcis-9453 45 48 down down ADP fcis-9453 45 49 semantic semantic ADJ fcis-9453 45 50 feature feature NOUN fcis-9453 45 51 enhancement enhancement NOUN fcis-9453 45 52 and and CCONJ fcis-9453 45 53 pan pan NOUN fcis-9453 45 54 conveys convey VERB fcis-9453 45 55 bottom bottom ADJ fcis-9453 45 56 - - PUNCT fcis-9453 45 57 up up ADP fcis-9453 45 58 localization localization NOUN fcis-9453 45 59 features feature NOUN fcis-9453 45 60 ; ; PUNCT fcis-9453 45 61 the the DET fcis-9453 45 62 difference difference NOUN fcis-9453 45 63 is be AUX fcis-9453 45 64 that that SCONJ fcis-9453 45 65 yolov4 yolov4 PROPN fcis-9453 45 66 uses use VERB fcis-9453 45 67 a a DET fcis-9453 45 68 normal normal ADJ fcis-9453 45 69 convolution convolution NOUN fcis-9453 45 70 operation operation NOUN fcis-9453 45 71 and and CCONJ fcis-9453 45 72 yolov5l yolov5l PROPN fcis-9453 45 73 uses use VERB fcis-9453 45 74 a a DET fcis-9453 45 75 c3 c3 NOUN fcis-9453 45 76 module module NOUN fcis-9453 45 77 to to PART fcis-9453 45 78 enhance enhance VERB fcis-9453 45 79 feature feature NOUN fcis-9453 45 80 fusion fusion NOUN fcis-9453 45 81 . . PUNCT fcis-9453 46 1 76 76 NUM fcis-9453 47 1 * * SYM fcis-9453 47 2 76 76 NUM fcis-9453 47 3 19 19 NUM fcis-9453 47 4 * * NUM fcis-9453 47 5 19 19 NUM fcis-9453 47 6 38 38 NUM fcis-9453 47 7 * * SYM fcis-9453 47 8 38 38 NUM fcis-9453 47 9 608 608 NUM fcis-9453 47 10 * * NUM fcis-9453 47 11 608 608 NUM fcis-9453 47 12 * * SYM fcis-9453 47 13 3 3 NUM fcis-9453 47 14 76 76 NUM fcis-9453 47 15 * * SYM fcis-9453 47 16 76 76 NUM fcis-9453 47 17 19 19 NUM fcis-9453 47 18 * * NUM fcis-9453 47 19 19 19 NUM fcis-9453 47 20 38 38 NUM fcis-9453 47 21 * * NUM fcis-9453 47 22 38 38 NUM fcis-9453 47 23 76 76 NUM fcis-9453 47 24 * * SYM fcis-9453 47 25 76 76 NUM fcis-9453 47 26 19 19 NUM fcis-9453 47 27 * * NUM fcis-9453 47 28 19 19 NUM fcis-9453 47 29 38 38 NUM fcis-9453 47 30 * * SYM fcis-9453 47 31 38 38 NUM fcis-9453 47 32 fpn fpn VERB fcis-9453 47 33 botton botton NOUN fcis-9453 47 34 - - PUNCT fcis-9453 47 35 up up ADP fcis-9453 47 36 下采样 下采样 NOUN fcis-9453 47 37 上采样 上采样 NOUN fcis-9453 47 38     SPACE fcis-9453 47 39 figure figure NOUN fcis-9453 47 40 2 2 NUM fcis-9453 47 41 . . PUNCT fcis-9453 47 42     SPACE fcis-9453 47 43 structural structural ADJ fcis-9453 47 44 diagram diagram NOUN fcis-9453 47 45 of of ADP fcis-9453 47 46 fpn fpn PROPN fcis-9453 47 47 + + CCONJ fcis-9453 47 48 pan pan PROPN fcis-9453 47 49     SPACE fcis-9453 47 50 2.1.4 2.1.4 PROPN fcis-9453 47 51 . . PUNCT fcis-9453 48 1 head head PROPN fcis-9453 48 2 yolov5l yolov5l PROPN fcis-9453 48 3 head head NOUN fcis-9453 48 4 contains contain VERB fcis-9453 48 5 three three NUM fcis-9453 48 6 detection detection NOUN fcis-9453 48 7 layers layer NOUN fcis-9453 48 8 , , PUNCT fcis-9453 48 9 corresponding correspond VERB fcis-9453 48 10 to to ADP fcis-9453 48 11 the the DET fcis-9453 48 12 three three NUM fcis-9453 48 13 different different ADJ fcis-9453 48 14 sizes size NOUN fcis-9453 48 15 of of ADP fcis-9453 48 16 feature feature NOUN fcis-9453 48 17 maps map NOUN fcis-9453 48 18 obtained obtain VERB fcis-9453 48 19 in in ADP fcis-9453 48 20 neck neck NOUN fcis-9453 48 21 , , PUNCT fcis-9453 48 22 and and CCONJ fcis-9453 48 23 three three NUM fcis-9453 48 24 anchors anchor NOUN fcis-9453 48 25 with with ADP fcis-9453 48 26 different different ADJ fcis-9453 48 27 aspect aspect NOUN fcis-9453 48 28 ratios ratio NOUN fcis-9453 48 29 are be AUX fcis-9453 48 30 preset preset ADJ fcis-9453 48 31 for for ADP fcis-9453 48 32 the the DET fcis-9453 48 33 grid grid NOUN fcis-9453 48 34 divided divide VERB fcis-9453 48 35 on on ADP fcis-9453 48 36 each each DET fcis-9453 48 37 feature feature NOUN fcis-9453 48 38 map map NOUN fcis-9453 48 39 to to PART fcis-9453 48 40 predict predict VERB fcis-9453 48 41 and and CCONJ fcis-9453 48 42 regress regress ADJ fcis-9453 48 43 targets target NOUN fcis-9453 48 44 ; ; PUNCT fcis-9453 48 45 ciou_loss ciou_loss NOUN fcis-9453 48 46 is be AUX fcis-9453 48 47 used use VERB fcis-9453 48 48 as as ADP fcis-9453 48 49 the the DET fcis-9453 48 50 loss loss NOUN fcis-9453 48 51 function function NOUN fcis-9453 48 52 of of ADP fcis-9453 48 53 bounding bounding NOUN fcis-9453 48 54 box box NOUN fcis-9453 48 55 , , PUNCT fcis-9453 48 56 and and CCONJ fcis-9453 48 57 for for ADP fcis-9453 48 58 the the DET fcis-9453 48 59 screening screening NOUN fcis-9453 48 60 of of ADP fcis-9453 48 61 multitarget multitarget PROPN fcis-9453 48 62 boxes box NOUN fcis-9453 48 63 , , PUNCT fcis-9453 48 64 a a DET fcis-9453 48 65 weighted weight VERB fcis-9453 48 66 nms nms NOUN fcis-9453 48 67 is be AUX fcis-9453 48 68 used use VERB fcis-9453 48 69 on on ADP fcis-9453 48 70 the the DET fcis-9453 48 71 basis basis NOUN fcis-9453 48 72 of of ADP fcis-9453 48 73 diou_loss diou_loss PROPN fcis-9453 48 74 to to PART fcis-9453 48 75 enhance enhance VERB fcis-9453 48 76 the the DET fcis-9453 48 77 detection detection NOUN fcis-9453 48 78 accuracy accuracy NOUN fcis-9453 48 79 of of ADP fcis-9453 48 80 occluded occluded ADJ fcis-9453 48 81 overlapping overlap VERB fcis-9453 48 82 targets target NOUN fcis-9453 48 83 . . PUNCT fcis-9453 49 1 2.2 2.2 NUM fcis-9453 49 2 . . PUNCT fcis-9453 49 3 improvements improvement NOUN fcis-9453 49 4 to to ADP fcis-9453 49 5 the the DET fcis-9453 49 6 yolov5l yolov5l PROPN fcis-9453 49 7 algorithm algorithm NOUN fcis-9453 49 8 2.2.1 2.2.1 NUM fcis-9453 49 9 . . PUNCT fcis-9453 50 1 introduction introduction NOUN fcis-9453 50 2 of of ADP fcis-9453 50 3 the the DET fcis-9453 50 4 simam simam ADJ fcis-9453 50 5 attention attention NOUN fcis-9453 50 6 mechanism mechanism NOUN fcis-9453 50 7 module module NOUN fcis-9453 50 8 adding add VERB fcis-9453 50 9 an an DET fcis-9453 50 10 attention attention NOUN fcis-9453 50 11 mechanism mechanism NOUN fcis-9453 50 12 can can AUX fcis-9453 50 13 effectively effectively ADV fcis-9453 50 14 enhance enhance VERB fcis-9453 50 15 the the DET fcis-9453 50 16 model model NOUN fcis-9453 50 17 's 's PART fcis-9453 50 18 ability ability NOUN fcis-9453 50 19 to to PART fcis-9453 50 20 extract extract VERB fcis-9453 50 21 features feature NOUN fcis-9453 50 22 from from ADP fcis-9453 50 23 images image NOUN fcis-9453 50 24 . . PUNCT fcis-9453 51 1 the the DET fcis-9453 51 2 simam simam ADJ fcis-9453 51 3 attention attention NOUN fcis-9453 51 4 mechanism mechanism NOUN fcis-9453 51 5 module module NOUN fcis-9453 51 6 [ [ X fcis-9453 51 7 25 25 NUM fcis-9453 51 8 ] ] PUNCT fcis-9453 51 9 is be AUX fcis-9453 51 10 a a DET fcis-9453 51 11 simple simple ADJ fcis-9453 51 12 and and CCONJ fcis-9453 51 13 very very ADV fcis-9453 51 14 effective effective ADJ fcis-9453 51 15 attention attention NOUN fcis-9453 51 16 module module NOUN fcis-9453 51 17 for for ADP fcis-9453 51 18 convolutional convolutional ADJ fcis-9453 51 19 neural neural ADJ fcis-9453 51 20 networks network NOUN fcis-9453 51 21 based base VERB fcis-9453 51 22 on on ADP fcis-9453 51 23 neuroscience neuroscience NOUN fcis-9453 51 24 theory theory NOUN fcis-9453 51 25 . . PUNCT fcis-9453 52 1 unlike unlike ADP fcis-9453 52 2 existing exist VERB fcis-9453 52 3 channel channel NOUN fcis-9453 52 4 or or CCONJ fcis-9453 52 5 null null ADJ fcis-9453 52 6 - - PUNCT fcis-9453 52 7 field field NOUN fcis-9453 52 8 attention attention NOUN fcis-9453 52 9 modules module NOUN fcis-9453 52 10 , , PUNCT fcis-9453 52 11 this this DET fcis-9453 52 12 module module NOUN fcis-9453 52 13 derives derive VERB fcis-9453 52 14 3d 3d NUM fcis-9453 52 15 attention attention NOUN fcis-9453 52 16 weights weight NOUN fcis-9453 52 17 in in ADP fcis-9453 52 18 the the DET fcis-9453 52 19 network network NOUN fcis-9453 52 20 layers layer NOUN fcis-9453 52 21 without without ADP fcis-9453 52 22 adding add VERB fcis-9453 52 23 any any DET fcis-9453 52 24 parameters parameter NOUN fcis-9453 52 25 , , PUNCT fcis-9453 52 26 and and CCONJ fcis-9453 52 27 a a DET fcis-9453 52 28 schematic schematic ADJ fcis-9453 52 29 of of ADP fcis-9453 52 30 3d 3d NUM fcis-9453 52 31 attention attention NOUN fcis-9453 52 32 weight weight NOUN fcis-9453 52 33 assignment assignment NOUN fcis-9453 52 34 is be AUX fcis-9453 52 35 shown show VERB fcis-9453 52 36 in in ADP fcis-9453 52 37 figure figure NOUN fcis-9453 52 38 3 3 NUM fcis-9453 52 39 . . PUNCT fcis-9453 53 1 the the DET fcis-9453 53 2 module module NOUN fcis-9453 53 3 aims aim VERB fcis-9453 53 4 to to PART fcis-9453 53 5 find find VERB fcis-9453 53 6 important important ADJ fcis-9453 53 7 neurons neuron NOUN fcis-9453 53 8 by by ADP fcis-9453 53 9 optimizing optimize VERB fcis-9453 53 10 the the DET fcis-9453 53 11 energy energy NOUN fcis-9453 53 12 function function NOUN fcis-9453 53 13 , , PUNCT fcis-9453 53 14 using use VERB fcis-9453 53 15 a a DET fcis-9453 53 16 linearly linearly ADV fcis-9453 53 17 branchable branchable ADJ fcis-9453 53 18 metric metric NOUN fcis-9453 53 19 between between ADP fcis-9453 53 20 neurons neuron NOUN fcis-9453 53 21 . . PUNCT fcis-9453 54 1 the the DET fcis-9453 54 2 energy energy NOUN fcis-9453 54 3 function function NOUN fcis-9453 54 4 as as SCONJ fcis-9453 54 5 defined define VERB fcis-9453 54 6 by by ADP fcis-9453 54 7 each each DET fcis-9453 54 8 neuron neuron NOUN fcis-9453 54 9 is be AUX fcis-9453 54 10 : : PUNCT fcis-9453 54 11 ( ( PUNCT fcis-9453 54 12 1 1 X fcis-9453 54 13 ) ) PUNCT fcis-9453 54 14 among among ADP fcis-9453 54 15 : : PUNCT fcis-9453 54 16 { { PUNCT fcis-9453 54 17 ˆ ˆ NOUN fcis-9453 54 18 ˆ ˆ ADP fcis-9453 54 19 tt tt PROPN fcis-9453 54 20 titi titi PROPN fcis-9453 54 21 btwt btwt PROPN fcis-9453 54 22 bxwx bxwx PROPN fcis-9453 54 23   PROPN fcis-9453 54 24   ADJ fcis-9453 54 25 ( ( PUNCT fcis-9453 54 26 2 2 NUM fcis-9453 54 27 ) ) PUNCT fcis-9453 54 28 where where SCONJ fcis-9453 54 29 t t PROPN fcis-9453 54 30 and and CCONJ fcis-9453 54 31 xi xi PROPN fcis-9453 54 32 are be AUX fcis-9453 54 33 the the DET fcis-9453 54 34 target target NOUN fcis-9453 54 35 neuron neuron NOUN fcis-9453 54 36 and and CCONJ fcis-9453 54 37 other other ADJ fcis-9453 54 38 neurons neuron NOUN fcis-9453 54 39 of of ADP fcis-9453 54 40 the the DET fcis-9453 54 41 input input NOUN fcis-9453 54 42 feature feature NOUN fcis-9453 54 43 tensor tensor NOUN fcis-9453 54 44 x x NOUN fcis-9453 54 45 , , PUNCT fcis-9453 55 1 x x PROPN fcis-9453 55 2 ∈ ∈ PROPN fcis-9453 55 3 rc rc PROPN fcis-9453 55 4 × × NOUN fcis-9453 55 5 h h NOUN fcis-9453 55 6 × × PROPN fcis-9453 55 7 w w PROPN fcis-9453 55 8 , , PUNCT fcis-9453 55 9 c c NOUN fcis-9453 55 10 , , PUNCT fcis-9453 55 11 h h NOUN fcis-9453 55 12 and and CCONJ fcis-9453 55 13 w w PROPN fcis-9453 55 14 are be AUX fcis-9453 55 15 the the DET fcis-9453 55 16 number number NOUN fcis-9453 55 17 of of ADP fcis-9453 55 18 channels channel NOUN fcis-9453 55 19 , , PUNCT fcis-9453 55 20 height height NOUN fcis-9453 55 21 and and CCONJ fcis-9453 55 22 width width NOUN fcis-9453 55 23 of of ADP fcis-9453 55 24 the the DET fcis-9453 55 25 feature feature NOUN fcis-9453 55 26 tensor tensor NOUN fcis-9453 55 27 , , PUNCT fcis-9453 55 28 respectively respectively ADV fcis-9453 55 29 ; ; PUNCT fcis-9453 55 30 i i PRON fcis-9453 55 31 is be AUX fcis-9453 55 32 the the DET fcis-9453 55 33 index index NOUN fcis-9453 55 34 of of ADP fcis-9453 55 35 neurons neuron NOUN fcis-9453 55 36 on on ADP fcis-9453 55 37 a a DET fcis-9453 55 38 channel channel NOUN fcis-9453 55 39 ; ; PUNCT fcis-9453 55 40 m m VERB fcis-9453 55 41 is be AUX fcis-9453 55 42 the the DET fcis-9453 55 43 number number NOUN fcis-9453 55 44 of of ADP fcis-9453 55 45 all all DET fcis-9453 55 46 neurons neuron NOUN fcis-9453 55 47 on on ADP fcis-9453 55 48 a a DET fcis-9453 55 49 channel channel NOUN fcis-9453 55 50 , , PUNCT fcis-9453 55 51 m m VERB fcis-9453 55 52 = = NOUN fcis-9453 55 53 h h NOUN fcis-9453 55 54 × × PROPN fcis-9453 55 55 w w NOUN fcis-9453 55 56 ; ; PUNCT fcis-9453 55 57 wt wt ADP fcis-9453 55 58 and and CCONJ fcis-9453 55 59 bt bt NOUN fcis-9453 55 60 are be AUX fcis-9453 55 61 the the DET fcis-9453 55 62 weights weight NOUN fcis-9453 55 63 and and CCONJ fcis-9453 55 64 biases bias NOUN fcis-9453 55 65 of of ADP fcis-9453 55 66 the the DET fcis-9453 55 67 target target NOUN fcis-9453 55 68 neuron neuron NOUN fcis-9453 55 69 when when SCONJ fcis-9453 55 70 transformed transform VERB fcis-9453 55 71 , , PUNCT fcis-9453 55 72 respectively respectively ADV fcis-9453 55 73 ; ; PUNCT fcis-9453 55 74 all all DET fcis-9453 55 75 values value NOUN fcis-9453 55 76 in in ADP fcis-9453 55 77 eq.(1 eq.(1 ADJ fcis-9453 55 78 ) ) PUNCT fcis-9453 55 79 are be AUX fcis-9453 55 80 scalars scalar NOUN fcis-9453 55 81 , , PUNCT fcis-9453 55 82 where where SCONJ fcis-9453 55 83 yt yt NOUN fcis-9453 55 84 and and CCONJ fcis-9453 55 85 y0 y0 NOUN fcis-9453 55 86 are be AUX fcis-9453 55 87 different different ADJ fcis-9453 55 88 values value NOUN fcis-9453 55 89 , , PUNCT fcis-9453 55 90 and and CCONJ fcis-9453 55 91 eq.(1 eq.(1 ADJ fcis-9453 55 92 ) ) PUNCT fcis-9453 55 93 is be AUX fcis-9453 55 94 minimized minimize VERB fcis-9453 55 95 when when SCONJ fcis-9453 55 96 t̂ t̂ PUNCT fcis-9453 56 1 = = NOUN fcis-9453 56 2 yt yt NOUN fcis-9453 56 3 and and CCONJ fcis-9453 56 4   VERB fcis-9453 56 5   VERB fcis-9453 56 6   PROPN fcis-9453 56 7   PROPN fcis-9453 56 8   NOUN fcis-9453 56 9   NUM fcis-9453 56 10 1 1 NUM fcis-9453 56 11 1 1 NUM fcis-9453 56 12 2 2 NUM fcis-9453 56 13 0 0 NUM fcis-9453 56 14 2 2 NUM fcis-9453 56 15 ) ) PUNCT fcis-9453 56 16 ˆ ˆ PROPN fcis-9453 56 17 ( ( PUNCT fcis-9453 56 18 1 1 NUM fcis-9453 56 19 1 1 NUM fcis-9453 56 20 ) ) PUNCT fcis-9453 56 21 ˆ ˆ NOUN fcis-9453 56 22 ( ( PUNCT fcis-9453 56 23 ) ) PUNCT fcis-9453 56 24 , , PUNCT fcis-9453 56 25 , , PUNCT fcis-9453 56 26 , , PUNCT fcis-9453 56 27 ( ( PUNCT fcis-9453 56 28 m m VERB fcis-9453 56 29 i i PRON fcis-9453 56 30 itittt itittt VERB fcis-9453 56 31 xy xy PROPN fcis-9453 56 32 m m PROPN fcis-9453 56 33 tyxybwe tyxybwe ADJ fcis-9453 56 34 69 69 NUM fcis-9453 56 35 has have VERB fcis-9453 56 36 ix̂ ix̂ VERB fcis-9453 56 37 = = NOUN fcis-9453 56 38 y0 y0 NOUN fcis-9453 56 39 for for ADP fcis-9453 56 40 all all DET fcis-9453 56 41 other other ADJ fcis-9453 56 42 neurons neuron NOUN fcis-9453 56 43 . . PUNCT fcis-9453 57 1 the the DET fcis-9453 57 2 minimization minimization NOUN fcis-9453 57 3 formula formula NOUN fcis-9453 57 4 is be AUX fcis-9453 57 5 equivalent equivalent ADJ fcis-9453 57 6 to to ADP fcis-9453 57 7 finding find VERB fcis-9453 57 8 the the DET fcis-9453 57 9 linear linear ADJ fcis-9453 57 10 differentiability differentiability NOUN fcis-9453 57 11 of of ADP fcis-9453 57 12 the the DET fcis-9453 57 13 target target NOUN fcis-9453 57 14 neuron neuron PROPN fcis-9453 57 15 t t PROPN fcis-9453 57 16 and and CCONJ fcis-9453 57 17 other other ADJ fcis-9453 57 18 neurons neuron NOUN fcis-9453 57 19 within within ADP fcis-9453 57 20 the the DET fcis-9453 57 21 same same ADJ fcis-9453 57 22 channel channel NOUN fcis-9453 57 23 . . PUNCT fcis-9453 58 1 using use VERB fcis-9453 58 2 binary binary ADJ fcis-9453 58 3 labels label NOUN fcis-9453 58 4 and and CCONJ fcis-9453 58 5 adding add VERB fcis-9453 58 6 regular regular ADJ fcis-9453 58 7 terms term NOUN fcis-9453 58 8 , , PUNCT fcis-9453 58 9 the the DET fcis-9453 58 10 final final ADJ fcis-9453 58 11 energy energy NOUN fcis-9453 58 12 function function NOUN fcis-9453 58 13 is be AUX fcis-9453 58 14 : : PUNCT fcis-9453 58 15     SPACE fcis-9453 58 16 2 2 NUM fcis-9453 58 17 t t NOUN fcis-9453 58 18 1 1 NUM fcis-9453 58 19 1 1 NUM fcis-9453 58 20 22 22 NUM fcis-9453 58 21 ) ) PUNCT fcis-9453 58 22 ] ] X fcis-9453 58 23 ( ( PUNCT fcis-9453 58 24 1 1 NUM fcis-9453 58 25 [ [ SYM fcis-9453 58 26 1 1 NUM fcis-9453 58 27 1 1 NUM fcis-9453 58 28 ) ) PUNCT fcis-9453 58 29 ] ] X fcis-9453 58 30 ( ( PUNCT fcis-9453 58 31 1 1 NUM fcis-9453 58 32 [ [ NOUN fcis-9453 58 33 ) ) PUNCT fcis-9453 58 34 , , PUNCT fcis-9453 58 35 , , PUNCT fcis-9453 58 36 , , PUNCT fcis-9453 58 37 ( ( PUNCT fcis-9453 58 38 wbxw wbxw NOUN fcis-9453 58 39 m m VERB fcis-9453 58 40 btw btw ADV fcis-9453 58 41 xybwe xybwe PROPN fcis-9453 58 42 m m VERB fcis-9453 59 1 i i PRON fcis-9453 59 2 tittt tittt VERB fcis-9453 59 3 ittt ittt PROPN fcis-9453 59 4   PROPN fcis-9453 59 5   PROPN fcis-9453 59 6   PROPN fcis-9453 59 7   PRON fcis-9453 59 8   PROPN fcis-9453 59 9   VERB fcis-9453 59 10   PRON fcis-9453 59 11 ( ( PUNCT fcis-9453 59 12 3 3 NUM fcis-9453 59 13 ) ) PUNCT fcis-9453 59 14 among among ADP fcis-9453 59 15 : : PUNCT fcis-9453 59 16     SPACE fcis-9453 59 17 { { PUNCT fcis-9453 59 18 22 22 NUM fcis-9453 59 19 ) ) PUNCT fcis-9453 60 1 ( ( PUNCT fcis-9453 60 2 ) ) PUNCT fcis-9453 60 3 ( ( PUNCT fcis-9453 60 4 2 2 NUM fcis-9453 60 5 ) ) PUNCT fcis-9453 60 6 ( ( PUNCT fcis-9453 60 7 2 2 NUM fcis-9453 60 8 1 1 NUM fcis-9453 60 9 22 22 NUM fcis-9453 60 10   NOUN fcis-9453 60 11   NOUN fcis-9453 60 12   NOUN fcis-9453 60 13   NOUN fcis-9453 60 14   NOUN fcis-9453 60 15   PROPN fcis-9453 60 16   NUM fcis-9453 60 17 tt tt PROPN fcis-9453 60 18 t t PROPN fcis-9453 60 19 t t PROPN fcis-9453 60 20 ttt ttt PROPN fcis-9453 60 21 t t PROPN fcis-9453 60 22 t t PROPN fcis-9453 60 23 w w PROPN fcis-9453 60 24 wtb wtb PROPN fcis-9453 60 25 ( ( PUNCT fcis-9453 60 26 4 4 NUM fcis-9453 60 27 ) ) PUNCT fcis-9453 60 28 { { PUNCT fcis-9453 60 29 1 1 NUM fcis-9453 60 30 1 1 NUM fcis-9453 60 31 2 2 NUM fcis-9453 60 32 t t NOUN fcis-9453 60 33 2 2 NUM fcis-9453 60 34 1 1 NUM fcis-9453 60 35 ) ) PUNCT fcis-9453 60 36 ˆ ˆ PROPN fcis-9453 60 37 ( ( PUNCT fcis-9453 60 38 1 1 NUM fcis-9453 60 39   ADV fcis-9453 60 40   X fcis-9453 60 41   PROPN fcis-9453 60 42   PROPN fcis-9453 61 1   NOUN fcis-9453 61 2 m m VERB fcis-9453 61 3 i i PRON fcis-9453 61 4 it it PRON fcis-9453 61 5 m m VERB fcis-9453 61 6 i i PRON fcis-9453 61 7 it it PRON fcis-9453 61 8 x x NOUN fcis-9453 61 9 m m VERB fcis-9453 61 10 x x VERB fcis-9453 61 11 m m VERB fcis-9453 61 12   NOUN fcis-9453 61 13   X fcis-9453 61 14 ( ( PUNCT fcis-9453 61 15 5 5 NUM fcis-9453 61 16 ) ) PUNCT fcis-9453 61 17 where where SCONJ fcis-9453 61 18 λ λ PROPN fcis-9453 61 19 is be AUX fcis-9453 61 20 the the DET fcis-9453 61 21 regularization regularization NOUN fcis-9453 61 22 factor factor NOUN fcis-9453 61 23 ; ; PUNCT fcis-9453 61 24 wi wi PROPN fcis-9453 61 25 is be AUX fcis-9453 61 26 the the DET fcis-9453 61 27 weight weight NOUN fcis-9453 61 28 of of ADP fcis-9453 61 29 the the DET fcis-9453 61 30 ith ith PROPN fcis-9453 61 31 neuron neuron NOUN fcis-9453 61 32 when when SCONJ fcis-9453 61 33 transformed transform VERB fcis-9453 61 34 ; ; PUNCT fcis-9453 61 35 from from ADP fcis-9453 61 36 eq.(4 eq.(4 ADV fcis-9453 61 37 ) ) PUNCT fcis-9453 61 38 it it PRON fcis-9453 61 39 can can AUX fcis-9453 61 40 be be AUX fcis-9453 61 41 inferred infer VERB fcis-9453 61 42 that that SCONJ fcis-9453 61 43 other other ADJ fcis-9453 61 44 neurons neuron NOUN fcis-9453 61 45 in in ADP fcis-9453 61 46 the the DET fcis-9453 61 47 same same ADJ fcis-9453 61 48 channel channel NOUN fcis-9453 61 49 satisfy satisfy VERB fcis-9453 61 50 the the DET fcis-9453 61 51 same same ADJ fcis-9453 61 52 distribution distribution NOUN fcis-9453 61 53 , , PUNCT fcis-9453 61 54 so so CCONJ fcis-9453 61 55 the the DET fcis-9453 61 56 mean mean NOUN fcis-9453 61 57 and and CCONJ fcis-9453 61 58 variance variance NOUN fcis-9453 61 59 of of ADP fcis-9453 61 60 all all DET fcis-9453 61 61 neurons neuron NOUN fcis-9453 61 62 can can AUX fcis-9453 61 63 be be AUX fcis-9453 61 64 calculated calculate VERB fcis-9453 61 65 , , PUNCT fcis-9453 61 66 replacing replace VERB fcis-9453 61 67 μt μt ADP fcis-9453 61 68 and and CCONJ fcis-9453 61 69 2 2 NUM fcis-9453 61 70 t t NOUN fcis-9453 61 71 is be AUX fcis-9453 61 72 the the DET fcis-9453 61 73 mean mean NOUN fcis-9453 61 74 and and CCONJ fcis-9453 61 75 variance variance NOUN fcis-9453 61 76 of of ADP fcis-9453 61 77 all all DET fcis-9453 61 78 neurons neuron NOUN fcis-9453 61 79 in in ADP fcis-9453 61 80 the the DET fcis-9453 61 81 corresponding correspond VERB fcis-9453 61 82 channel channel NOUN fcis-9453 61 83 after after ADP fcis-9453 61 84 removing remove VERB fcis-9453 61 85 neuron neuron PROPN fcis-9453 61 86 t t PROPN fcis-9453 61 87 , , PUNCT fcis-9453 61 88 and and CCONJ fcis-9453 61 89 all all DET fcis-9453 61 90 neurons neuron NOUN fcis-9453 61 91 on on ADP fcis-9453 61 92 the the DET fcis-9453 61 93 same same ADJ fcis-9453 61 94 channel channel NOUN fcis-9453 61 95 are be AUX fcis-9453 61 96 multiplexed multiplexe VERB fcis-9453 61 97 with with ADP fcis-9453 61 98 this this DET fcis-9453 61 99 mean mean NOUN fcis-9453 61 100 and and CCONJ fcis-9453 61 101 variance variance NOUN fcis-9453 61 102 , , PUNCT fcis-9453 61 103 reducing reduce VERB fcis-9453 61 104 the the DET fcis-9453 61 105 computational computational ADJ fcis-9453 61 106 complexity complexity NOUN fcis-9453 61 107 of of ADP fcis-9453 61 108 each each DET fcis-9453 61 109 location location NOUN fcis-9453 61 110 , , PUNCT fcis-9453 61 111 the the PRON fcis-9453 61 112 lower low ADJ fcis-9453 61 113 the the DET fcis-9453 61 114 energy energy NOUN fcis-9453 61 115 , , PUNCT fcis-9453 61 116 the the PRON fcis-9453 61 117 greater great ADJ fcis-9453 61 118 the the DET fcis-9453 61 119 difference difference NOUN fcis-9453 61 120 between between ADP fcis-9453 61 121 neuron neuron PROPN fcis-9453 61 122 t t PROPN fcis-9453 61 123 and and CCONJ fcis-9453 61 124 the the DET fcis-9453 61 125 surrounding surround VERB fcis-9453 61 126 neurons neuron NOUN fcis-9453 61 127 , , PUNCT fcis-9453 61 128 and and CCONJ fcis-9453 61 129 ultimately ultimately ADV fcis-9453 61 130 the the DET fcis-9453 61 131 minimum minimum ADJ fcis-9453 61 132 energy energy NOUN fcis-9453 61 133 et et NOUN fcis-9453 61 134 * * PUNCT fcis-9453 61 135 at at ADP fcis-9453 61 136 each each DET fcis-9453 61 137 location location NOUN fcis-9453 61 138 is be AUX fcis-9453 61 139 calculated calculate VERB fcis-9453 61 140 as as SCONJ fcis-9453 61 141 follows follow VERB fcis-9453 61 142 : : PUNCT fcis-9453 62 1   ADP fcis-9453 62 2   PROPN fcis-9453 62 3 2ˆ2)ˆ 2ˆ2)ˆ NUM fcis-9453 62 4 ( ( PUNCT fcis-9453 63 1 ) ) PUNCT fcis-9453 63 2 ˆ(4 ˆ(4 NOUN fcis-9453 63 3 22 22 NUM fcis-9453 63 4 2 2 NUM fcis-9453 63 5 * * PUNCT fcis-9453 63 6   PROPN fcis-9453 63 7   X fcis-9453 63 8   PROPN fcis-9453 63 9 t t X fcis-9453 63 10 et et NOUN fcis-9453 63 11 ( ( PUNCT fcis-9453 63 12 6 6 NUM fcis-9453 63 13 ) ) PUNCT fcis-9453 63 14 c c NOUN fcis-9453 63 15 w w PROPN fcis-9453 63 16 h h PROPN fcis-9453 63 17 ceneration ceneration NOUN fcis-9453 63 18 3d 3d NUM fcis-9453 63 19 - - PUNCT fcis-9453 63 20 weights weight NOUN fcis-9453 63 21 fusion fusion NOUN fcis-9453 63 22 expansion expansion NOUN fcis-9453 63 23 c c PROPN fcis-9453 63 24 w w NOUN fcis-9453 63 25 h h NOUN fcis-9453 63 26     SPACE fcis-9453 63 27 figure figure NOUN fcis-9453 63 28 3 3 NUM fcis-9453 63 29 . . PUNCT fcis-9453 63 30     SPACE fcis-9453 63 31 3d 3d ADJ fcis-9453 63 32 schematic schematic ADJ fcis-9453 63 33 representation representation NOUN fcis-9453 63 34 of of ADP fcis-9453 63 35 the the DET fcis-9453 63 36 attention attention NOUN fcis-9453 63 37 weight weight NOUN fcis-9453 63 38 allocation allocation NOUN fcis-9453 63 39     SPACE fcis-9453 63 40 2.2.2 2.2.2 NUM fcis-9453 63 41 . . PUNCT fcis-9453 64 1 introduction introduction NOUN fcis-9453 64 2 of of ADP fcis-9453 64 3 the the DET fcis-9453 64 4 c2f c2f NOUN fcis-9453 64 5 module module NOUN fcis-9453 64 6 a a DET fcis-9453 64 7 c3 c3 NOUN fcis-9453 64 8 module module NOUN fcis-9453 64 9 containing contain VERB fcis-9453 64 10 three three NUM fcis-9453 64 11 standard standard ADJ fcis-9453 64 12 convolutional convolutional ADJ fcis-9453 64 13 layers layer NOUN fcis-9453 64 14 ( ( PUNCT fcis-9453 64 15 conv+bn+silu conv+bn+silu NOUN fcis-9453 64 16 ) ) PUNCT fcis-9453 64 17 and and CCONJ fcis-9453 64 18 n n ADP fcis-9453 64 19 bottleneck bottleneck NOUN fcis-9453 64 20 modules module NOUN fcis-9453 64 21 was be AUX fcis-9453 64 22 designed design VERB fcis-9453 64 23 in in ADP fcis-9453 64 24 yolov5l yolov5l PROPN fcis-9453 64 25 with with ADP fcis-9453 64 26 the the DET fcis-9453 64 27 help help NOUN fcis-9453 64 28 of of ADP fcis-9453 64 29 the the DET fcis-9453 64 30 idea idea NOUN fcis-9453 64 31 of of ADP fcis-9453 64 32 cspnet cspnet NOUN fcis-9453 64 33 to to PART fcis-9453 64 34 extract extract VERB fcis-9453 64 35 the the DET fcis-9453 64 36 divergence divergence NOUN fcis-9453 64 37 and and CCONJ fcis-9453 64 38 residual residual ADJ fcis-9453 64 39 structure structure NOUN fcis-9453 64 40 , , PUNCT fcis-9453 64 41 which which PRON fcis-9453 64 42 is be AUX fcis-9453 64 43 the the DET fcis-9453 64 44 main main ADJ fcis-9453 64 45 module module NOUN fcis-9453 64 46 for for ADP fcis-9453 64 47 learning learn VERB fcis-9453 64 48 on on ADP fcis-9453 64 49 residual residual ADJ fcis-9453 64 50 features feature NOUN fcis-9453 64 51 , , PUNCT fcis-9453 64 52 with with ADP fcis-9453 64 53 two two NUM fcis-9453 64 54 types type NOUN fcis-9453 64 55 of of ADP fcis-9453 64 56 structure structure NOUN fcis-9453 64 57 , , PUNCT fcis-9453 64 58 one one NOUN fcis-9453 64 59 using use VERB fcis-9453 64 60 multiple multiple ADJ fcis-9453 64 61 bottleneck bottleneck NOUN fcis-9453 64 62 stacks stack NOUN fcis-9453 64 63 and and CCONJ fcis-9453 64 64 three three NUM fcis-9453 64 65 standard standard ADJ fcis-9453 64 66 convolutional convolutional ADJ fcis-9453 64 67 layers layer NOUN fcis-9453 64 68 ; ; PUNCT fcis-9453 64 69 the the DET fcis-9453 64 70 other other ADJ fcis-9453 64 71 class class NOUN fcis-9453 64 72 uses use VERB fcis-9453 64 73 only only ADV fcis-9453 64 74 one one NUM fcis-9453 64 75 basic basic ADJ fcis-9453 64 76 convolution convolution NOUN fcis-9453 64 77 module module NOUN fcis-9453 64 78 , , PUNCT fcis-9453 64 79 and and CCONJ fcis-9453 64 80 the the DET fcis-9453 64 81 two two NUM fcis-9453 64 82 classes class NOUN fcis-9453 64 83 are be AUX fcis-9453 64 84 combined combine VERB fcis-9453 64 85 for for ADP fcis-9453 64 86 concat concat NOUN fcis-9453 64 87 operations operation NOUN fcis-9453 64 88 . . PUNCT fcis-9453 65 1 in in ADP fcis-9453 65 2 this this DET fcis-9453 65 3 paper paper NOUN fcis-9453 65 4 , , PUNCT fcis-9453 65 5 the the DET fcis-9453 65 6 yolov5l yolov5l PROPN fcis-9453 65 7 model model NOUN fcis-9453 65 8 is be AUX fcis-9453 65 9 improved improve VERB fcis-9453 65 10 by by ADP fcis-9453 65 11 replacing replace VERB fcis-9453 65 12 the the DET fcis-9453 65 13 c3 c3 NOUN fcis-9453 65 14 module module NOUN fcis-9453 65 15 with with ADP fcis-9453 65 16 the the DET fcis-9453 65 17 c2f c2f NOUN fcis-9453 65 18 module module NOUN fcis-9453 65 19 , , PUNCT fcis-9453 65 20 so so SCONJ fcis-9453 65 21 that that SCONJ fcis-9453 65 22 the the DET fcis-9453 65 23 improved improved ADJ fcis-9453 65 24 model model NOUN fcis-9453 65 25 can can AUX fcis-9453 65 26 obtain obtain VERB fcis-9453 65 27 richer rich ADJ fcis-9453 65 28 gradient gradient NOUN fcis-9453 65 29 flow flow NOUN fcis-9453 65 30 information information NOUN fcis-9453 65 31 and and CCONJ fcis-9453 65 32 improve improve VERB fcis-9453 65 33 the the DET fcis-9453 65 34 detection detection NOUN fcis-9453 65 35 accuracy accuracy NOUN fcis-9453 65 36 of of ADP fcis-9453 65 37 the the DET fcis-9453 65 38 model model NOUN fcis-9453 65 39 while while SCONJ fcis-9453 65 40 ensuring ensure VERB fcis-9453 65 41 its its PRON fcis-9453 65 42 light light ADJ fcis-9453 65 43 weight weight NOUN fcis-9453 65 44 . . PUNCT fcis-9453 66 1 the the DET fcis-9453 66 2 c3 c3 PROPN fcis-9453 66 3 and and CCONJ fcis-9453 66 4 c2f c2f NOUN fcis-9453 66 5 module module NOUN fcis-9453 66 6 structure structure NOUN fcis-9453 66 7 pairs pair NOUN fcis-9453 66 8 are be AUX fcis-9453 66 9 shown show VERB fcis-9453 66 10 in in ADP fcis-9453 66 11 figure figure NOUN fcis-9453 66 12 4 4 NUM fcis-9453 66 13 . . PUNCT fcis-9453 66 14 convbnsilu convbnsilu NOUN fcis-9453 66 15 convbnsilu convbnsilu NOUN fcis-9453 66 16 bottleneck bottleneck NOUN fcis-9453 66 17 concat concat NOUN fcis-9453 66 18 convbnsilu convbnsilu NOUN fcis-9453 66 19 c3 c3 X fcis-9453 66 20 convbnsilu convbnsilu NOUN fcis-9453 66 21 convbnsilu convbnsilu NOUN fcis-9453 66 22 bottleneck bottleneck NOUN fcis-9453 66 23 bottleneck bottleneck NOUN fcis-9453 66 24 bottleneck bottleneck NOUN fcis-9453 66 25 split split VERB fcis-9453 66 26 c c PROPN fcis-9453 66 27 c2f c2f NOUN fcis-9453 66 28 } } PUNCT fcis-9453 66 29 n n PRON fcis-9453 66 30 figure figure VERB fcis-9453 66 31 4 4 NUM fcis-9453 66 32 . . PUNCT fcis-9453 66 33     SPACE fcis-9453 66 34 structural structural ADJ fcis-9453 66 35 comparison comparison NOUN fcis-9453 66 36 of of ADP fcis-9453 66 37 c c PROPN fcis-9453 66 38 3 3 NUM fcis-9453 66 39 and and CCONJ fcis-9453 66 40 c2f c2f NOUN fcis-9453 66 41 models model NOUN fcis-9453 66 42     SPACE fcis-9453 66 43 3 3 X fcis-9453 66 44 . . PUNCT fcis-9453 66 45 experimental experimental ADJ fcis-9453 66 46 results result NOUN fcis-9453 66 47 and and CCONJ fcis-9453 66 48 analysis analysis NOUN fcis-9453 66 49 3.1 3.1 NUM fcis-9453 66 50 . . PUNCT fcis-9453 67 1 experimental experimental ADJ fcis-9453 67 2 environment environment NOUN fcis-9453 67 3 setup setup VERB fcis-9453 67 4 the the DET fcis-9453 67 5 hardware hardware NOUN fcis-9453 67 6 environment environment NOUN fcis-9453 67 7 for for ADP fcis-9453 67 8 the the DET fcis-9453 67 9 experiments experiment NOUN fcis-9453 67 10 is be AUX fcis-9453 67 11 windows window NOUN fcis-9453 67 12 10 10 NUM fcis-9453 67 13 , , PUNCT fcis-9453 67 14 the the DET fcis-9453 67 15 cpu cpu NOUN fcis-9453 67 16 is be AUX fcis-9453 67 17 intel intel PROPN fcis-9453 67 18 core core PROPN fcis-9453 67 19 ( ( PUNCT fcis-9453 67 20 tm tm NOUN fcis-9453 67 21 ) ) PUNCT fcis-9453 67 22 i7 i7 NOUN fcis-9453 67 23 - - PUNCT fcis-9453 67 24 9700k 9700k NUM fcis-9453 67 25 , , PUNCT fcis-9453 67 26 the the DET fcis-9453 67 27 memory memory NOUN fcis-9453 67 28 is be AUX fcis-9453 67 29 32 32 NUM fcis-9453 67 30 gb gb NOUN fcis-9453 67 31 , , PUNCT fcis-9453 67 32 the the DET fcis-9453 67 33 gpu gpu NOUN fcis-9453 67 34 is be AUX fcis-9453 67 35 nvidia nvidia PROPN fcis-9453 67 36 geforce geforce NOUN fcis-9453 67 37 rtx2080 rtx2080 NOUN fcis-9453 67 38 ti ti NOUN fcis-9453 67 39 , , PUNCT fcis-9453 67 40 and and CCONJ fcis-9453 67 41 the the DET fcis-9453 67 42 software software NOUN fcis-9453 67 43 environment environment NOUN fcis-9453 67 44 is be AUX fcis-9453 67 45 pytorch pytorch NOUN fcis-9453 67 46 ( ( PUNCT fcis-9453 67 47 1.10.0 1.10.0 NUM fcis-9453 67 48 ) ) PUNCT fcis-9453 67 49 ; ; PUNCT fcis-9453 67 50 cuda cuda NOUN fcis-9453 67 51 ( ( PUNCT fcis-9453 67 52 12.0 12.0 NUM fcis-9453 67 53 ) ) PUNCT fcis-9453 67 54 ; ; PUNCT fcis-9453 67 55 numpy numpy NOUN fcis-9453 67 56 ( ( PUNCT fcis-9453 67 57 1.24.3 1.24.3 NUM fcis-9453 67 58 ) ) PUNCT fcis-9453 67 59 ; ; PUNCT fcis-9453 67 60 python python NOUN fcis-9453 67 61 3.8 3.8 NUM fcis-9453 67 62 ; ; PUNCT fcis-9453 67 63 pycharm pycharm VERB fcis-9453 67 64 2022.1 2022.1 NUM fcis-9453 67 65 . . PUNCT fcis-9453 67 66 3.2 3.2 NUM fcis-9453 67 67 . . PUNCT fcis-9453 68 1 dataset dataset VERB fcis-9453 68 2 this this DET fcis-9453 68 3 dataset dataset NOUN fcis-9453 68 4 was be AUX fcis-9453 68 5 obtained obtain VERB fcis-9453 68 6 from from ADP fcis-9453 68 7 the the DET fcis-9453 68 8 public public ADJ fcis-9453 68 9 tape tape NOUN fcis-9453 68 10 steel steel NOUN fcis-9453 68 11 neudet neudet NOUN fcis-9453 68 12 file file NOUN fcis-9453 68 13 of of ADP fcis-9453 68 14 northeastern northeastern ADJ fcis-9453 68 15 university university NOUN fcis-9453 68 16 , , PUNCT fcis-9453 68 17 which which PRON fcis-9453 68 18 contains contain VERB fcis-9453 68 19 6 6 NUM fcis-9453 68 20 different different ADJ fcis-9453 68 21 types type NOUN fcis-9453 68 22 of of ADP fcis-9453 68 23 crazing crazing NOUN fcis-9453 68 24 , , PUNCT fcis-9453 68 25 rolled roll VERB fcis-9453 68 26 - - PUNCT fcis-9453 68 27 in in ADP fcis-9453 68 28 scale scale NOUN fcis-9453 68 29 , , PUNCT fcis-9453 68 30 scratches scratch NOUN fcis-9453 68 31 , , PUNCT fcis-9453 68 32 inclusion inclusion NOUN fcis-9453 68 33 , , PUNCT fcis-9453 68 34 patches patch NOUN fcis-9453 68 35 and and CCONJ fcis-9453 68 36 pitted pit VERB fcis-9453 68 37 surface surface NOUN fcis-9453 68 38 defects defect NOUN fcis-9453 68 39 , , PUNCT fcis-9453 68 40 with with ADP fcis-9453 68 41 an an DET fcis-9453 68 42 image image NOUN fcis-9453 68 43 size size NOUN fcis-9453 68 44 of of ADP fcis-9453 68 45 200x200 200x200 NUM fcis-9453 68 46 and and CCONJ fcis-9453 68 47 a a DET fcis-9453 68 48 total total NOUN fcis-9453 68 49 of of ADP fcis-9453 68 50 1800 1800 NUM fcis-9453 68 51 grey grey ADJ fcis-9453 68 52 - - PUNCT fcis-9453 68 53 scale scale NOUN fcis-9453 68 54 images image NOUN fcis-9453 68 55 . . PUNCT fcis-9453 69 1 after after ADP fcis-9453 69 2 screening screen VERB fcis-9453 69 3 the the DET fcis-9453 69 4 data datum NOUN fcis-9453 69 5 set set VERB fcis-9453 69 6 in in ADP fcis-9453 69 7 this this DET fcis-9453 69 8 paper paper NOUN fcis-9453 69 9 , , PUNCT fcis-9453 69 10 1400 1400 NUM fcis-9453 69 11 images image NOUN fcis-9453 69 12 were be AUX fcis-9453 69 13 selected select VERB fcis-9453 69 14 randomly randomly ADV fcis-9453 69 15 according accord VERB fcis-9453 69 16 to to ADP fcis-9453 69 17 the the DET fcis-9453 69 18 ratio ratio NOUN fcis-9453 69 19 of of ADP fcis-9453 69 20 training training NOUN fcis-9453 69 21 set set NOUN fcis-9453 69 22 : : PUNCT fcis-9453 69 23 validation validation NOUN fcis-9453 69 24 set set NOUN fcis-9453 69 25 : : PUNCT fcis-9453 69 26 test test NOUN fcis-9453 69 27 set set VERB fcis-9453 69 28 6:2:2 6:2:2 NUM fcis-9453 69 29 . . PUNCT fcis-9453 70 1 an an DET fcis-9453 70 2 example example NOUN fcis-9453 70 3 of of ADP fcis-9453 70 4 the the DET fcis-9453 70 5 dataset dataset NOUN fcis-9453 70 6 is be AUX fcis-9453 70 7 shown show VERB fcis-9453 70 8 in in ADP fcis-9453 70 9 figure figure NOUN fcis-9453 70 10 5 5 NUM fcis-9453 70 11 . . PUNCT fcis-9453 70 12     SPACE fcis-9453 70 13 figure figure NOUN fcis-9453 70 14 5 5 NUM fcis-9453 70 15 . . PUNCT fcis-9453 70 16     SPACE fcis-9453 70 17 dataset dataset VERB fcis-9453 70 18 example example NOUN fcis-9453 70 19     SPACE fcis-9453 70 20 3.3 3.3 NUM fcis-9453 70 21 . . PUNCT fcis-9453 71 1 parameter parameter NOUN fcis-9453 71 2 setting setting NOUN fcis-9453 71 3 and and CCONJ fcis-9453 71 4 evaluation evaluation NOUN fcis-9453 71 5 index index NOUN fcis-9453 71 6 in in ADP fcis-9453 71 7 this this DET fcis-9453 71 8 paper paper NOUN fcis-9453 71 9 , , PUNCT fcis-9453 71 10 we we PRON fcis-9453 71 11 set set VERB fcis-9453 71 12 epoch=200 epoch=200 NOUN fcis-9453 71 13 , , PUNCT fcis-9453 71 14 batch batch NOUN fcis-9453 71 15 size=4 size=4 PROPN fcis-9453 71 16 , , PUNCT fcis-9453 71 17 conf_thres=0.5 conf_thres=0.5 VERB fcis-9453 71 18 , , PUNCT fcis-9453 71 19 initial initial ADJ fcis-9453 71 20 learning learning NOUN fcis-9453 71 21 rate rate NOUN fcis-9453 71 22 is be AUX fcis-9453 71 23 0.01 0.01 NUM fcis-9453 71 24 , , PUNCT fcis-9453 71 25 learning learn VERB fcis-9453 71 26 rate rate NOUN fcis-9453 71 27 momentum momentum NOUN fcis-9453 71 28 is be AUX fcis-9453 71 29 0.937 0.937 NUM fcis-9453 71 30 , , PUNCT fcis-9453 71 31 weight weight NOUN fcis-9453 71 32 decay decay NOUN fcis-9453 71 33 coefficient coefficient NOUN fcis-9453 71 34 is be AUX fcis-9453 71 35 0.005 0.005 NUM fcis-9453 71 36 , , PUNCT fcis-9453 71 37 sgd sgd PROPN fcis-9453 71 38 algorithm algorithm NOUN fcis-9453 71 39 is be AUX fcis-9453 71 40 used use VERB fcis-9453 71 41 for for ADP fcis-9453 71 42 training training NOUN fcis-9453 71 43 , , PUNCT fcis-9453 71 44 and and CCONJ fcis-9453 71 45 precision precision NOUN fcis-9453 71 46 rate rate NOUN fcis-9453 71 47 p p NOUN fcis-9453 71 48 ( ( PUNCT fcis-9453 71 49 precision precision NOUN fcis-9453 71 50 ) ) PUNCT fcis-9453 71 51 , , PUNCT fcis-9453 71 52 recall recall NOUN fcis-9453 71 53 rate rate NOUN fcis-9453 71 54 r r NOUN fcis-9453 71 55 ( ( PUNCT fcis-9453 71 56 recall recall NOUN fcis-9453 71 57 ) ) PUNCT fcis-9453 71 58 and and CCONJ fcis-9453 71 59 average average ADJ fcis-9453 71 60 precision precision NOUN fcis-9453 71 61 value value NOUN fcis-9453 71 62 map map NOUN fcis-9453 71 63 ( ( PUNCT fcis-9453 71 64 mean mean VERB fcis-9453 71 65 average average ADJ fcis-9453 71 66 precision precision NOUN fcis-9453 71 67 ) ) PUNCT fcis-9453 71 68 as as ADP fcis-9453 71 69 the the DET fcis-9453 71 70 model model NOUN fcis-9453 71 71 performance performance NOUN fcis-9453 71 72 index index NOUN fcis-9453 71 73 . . PUNCT fcis-9453 72 1 the the DET fcis-9453 72 2 formula formula NOUN fcis-9453 72 3 is be AUX fcis-9453 72 4 as as SCONJ fcis-9453 72 5 follows follow VERB fcis-9453 72 6 . . PUNCT fcis-9453 73 1 fptp fptp PROPN fcis-9453 73 2 tp tp ADP fcis-9453 74 1 p p X fcis-9453 74 2   VERB fcis-9453 74 3   PROPN fcis-9453 74 4                                                                 SPACE fcis-9453 74 5 ( ( PUNCT fcis-9453 74 6 7 7 NUM fcis-9453 74 7 ) ) PUNCT fcis-9453 74 8                         SPACE fcis-9453 74 9 fntp fntp NOUN fcis-9453 74 10 tp tp ADP fcis-9453 74 11 r r PROPN fcis-9453 74 12   PUNCT fcis-9453 74 13   PROPN fcis-9453 74 14                         SPACE fcis-9453 74 15 ( ( PUNCT fcis-9453 74 16 8) 8) NUM fcis-9453 74 17                         SPACE fcis-9453 75 1 i i PRON fcis-9453 75 2 c c VERB fcis-9453 76 1 i i PRON fcis-9453 76 2 ap ap PROPN fcis-9453 77 1 c c PROPN fcis-9453 77 2 ap ap PROPN fcis-9453 77 3   X fcis-9453 78 1   NUM fcis-9453 78 2   NOUN fcis-9453 78 3 1 1 NUM fcis-9453 78 4 1 1 NUM fcis-9453 78 5 m m VERB fcis-9453 78 6         SPACE fcis-9453 78 7 ( ( PUNCT fcis-9453 78 8 9 9 NUM fcis-9453 78 9 ) ) PUNCT fcis-9453 78 10                             SPACE fcis-9453 78 11 where where SCONJ fcis-9453 78 12 tp tp PART fcis-9453 78 13 ( ( PUNCT fcis-9453 78 14 true true ADJ fcis-9453 78 15 positive positive ADJ fcis-9453 78 16 ) ) PUNCT fcis-9453 78 17 indicates indicate VERB fcis-9453 78 18 a a DET fcis-9453 78 19 positive positive ADJ fcis-9453 78 20 sample sample NOUN fcis-9453 78 21 with with ADP fcis-9453 78 22 positive positive ADJ fcis-9453 78 23 prediction prediction NOUN fcis-9453 79 1 ; ; PUNCT fcis-9453 79 2 fp fp X fcis-9453 79 3 ( ( PUNCT fcis-9453 79 4 false false ADJ fcis-9453 79 5 positive positive ADJ fcis-9453 79 6 ) ) PUNCT fcis-9453 79 7 indicates indicate VERB fcis-9453 79 8 a a DET fcis-9453 79 9 negative negative ADJ fcis-9453 79 10 70 70 NUM fcis-9453 79 11 sample sample NOUN fcis-9453 79 12 with with ADP fcis-9453 79 13 positive positive ADJ fcis-9453 79 14 prediction prediction NOUN fcis-9453 79 15 ; ; PUNCT fcis-9453 79 16 fn fn PROPN fcis-9453 79 17 ( ( PUNCT fcis-9453 79 18 false false ADJ fcis-9453 79 19 negative negative NOUN fcis-9453 79 20 ) ) PUNCT fcis-9453 79 21 indicates indicate VERB fcis-9453 79 22 a a DET fcis-9453 79 23 positive positive ADJ fcis-9453 79 24 sample sample NOUN fcis-9453 79 25 with with ADP fcis-9453 79 26 negative negative ADJ fcis-9453 79 27 prediction prediction NOUN fcis-9453 79 28 ; ; PUNCT fcis-9453 79 29 ap ap PROPN fcis-9453 79 30 is be AUX fcis-9453 79 31 the the DET fcis-9453 79 32 average average ADJ fcis-9453 79 33 precision precision NOUN fcis-9453 79 34 of of ADP fcis-9453 79 35 a a DET fcis-9453 79 36 single single ADJ fcis-9453 79 37 target target NOUN fcis-9453 79 38 category category NOUN fcis-9453 79 39 ; ; PUNCT fcis-9453 79 40 map map NOUN fcis-9453 79 41 is be AUX fcis-9453 79 42 the the DET fcis-9453 79 43 average average ADJ fcis-9453 79 44 precision precision NOUN fcis-9453 79 45 value value NOUN fcis-9453 79 46 of of ADP fcis-9453 79 47 ap ap PROPN fcis-9453 79 48 for for ADP fcis-9453 79 49 all all DET fcis-9453 79 50 categories category NOUN fcis-9453 79 51 ; ; PUNCT fcis-9453 79 52 the the DET fcis-9453 79 53 p p PROPN fcis-9453 79 54 - - PUNCT fcis-9453 79 55 r r NOUN fcis-9453 79 56 curve curve NOUN fcis-9453 79 57 generated generate VERB fcis-9453 79 58 by by ADP fcis-9453 79 59 the the DET fcis-9453 79 60 improved improve VERB fcis-9453 79 61 yolov5l yolov5l NOUN fcis-9453 79 62 is be AUX fcis-9453 79 63 shown show VERB fcis-9453 79 64 in in ADP fcis-9453 79 65 figure figure NOUN fcis-9453 79 66 6 6 NUM fcis-9453 79 67 , , PUNCT fcis-9453 79 68 where where SCONJ fcis-9453 79 69 p p NOUN fcis-9453 79 70 is be AUX fcis-9453 79 71 the the DET fcis-9453 79 72 vertical vertical ADJ fcis-9453 79 73 coordinate coordinate NOUN fcis-9453 79 74 , , PUNCT fcis-9453 79 75 r r NOUN fcis-9453 79 76 is be AUX fcis-9453 79 77 the the DET fcis-9453 79 78 horizontal horizontal ADJ fcis-9453 79 79 coordinate coordinate NOUN fcis-9453 79 80 , , PUNCT fcis-9453 79 81 and and CCONJ fcis-9453 79 82 the the DET fcis-9453 79 83 area area NOUN fcis-9453 79 84 enclosed enclose VERB fcis-9453 79 85 by by ADP fcis-9453 79 86 the the DET fcis-9453 79 87 p p PROPN fcis-9453 79 88 - - PUNCT fcis-9453 79 89 r r NOUN fcis-9453 79 90 curve curve NOUN fcis-9453 79 91 and and CCONJ fcis-9453 79 92 the the DET fcis-9453 79 93 coordinate coordinate NOUN fcis-9453 79 94 axis axis NOUN fcis-9453 79 95 is be AUX fcis-9453 79 96 ap ap PROPN fcis-9453 79 97 . . PUNCT fcis-9453 79 98     SPACE fcis-9453 79 99 figure figure NOUN fcis-9453 79 100 6 6 NUM fcis-9453 79 101 . . PUNCT fcis-9453 79 102     SPACE fcis-9453 79 103 analysis analysis NOUN fcis-9453 79 104 of of ADP fcis-9453 79 105 the the DET fcis-9453 79 106 results result NOUN fcis-9453 79 107 of of ADP fcis-9453 79 108 the the DET fcis-9453 79 109 improved improved ADJ fcis-9453 79 110 p p NOUN fcis-9453 79 111 - - PUNCT fcis-9453 79 112 r r NOUN fcis-9453 79 113 curves curve NOUN fcis-9453 79 114     SPACE fcis-9453 79 115 3.4 3.4 NUM fcis-9453 79 116 . . PUNCT fcis-9453 80 1 analysis analysis NOUN fcis-9453 80 2 of of ADP fcis-9453 80 3 experimental experimental ADJ fcis-9453 80 4 results result NOUN fcis-9453 80 5 in in ADP fcis-9453 80 6 this this DET fcis-9453 80 7 paper paper NOUN fcis-9453 80 8 , , PUNCT fcis-9453 80 9 the the DET fcis-9453 80 10 improved improve VERB fcis-9453 80 11 yolov5l yolov5l PROPN fcis-9453 80 12 model model NOUN fcis-9453 80 13 is be AUX fcis-9453 80 14 compared compare VERB fcis-9453 80 15 with with ADP fcis-9453 80 16 the the DET fcis-9453 80 17 original original ADJ fcis-9453 80 18 yolov5l yolov5l PROPN fcis-9453 80 19 model model NOUN fcis-9453 80 20 , , PUNCT fcis-9453 80 21 and and CCONJ fcis-9453 80 22 the the PRON fcis-9453 80 23 improved improve VERB fcis-9453 80 24 before before ADV fcis-9453 80 25 and and CCONJ fcis-9453 80 26 after after ADP fcis-9453 80 27 models model NOUN fcis-9453 80 28 are be AUX fcis-9453 80 29 trained train VERB fcis-9453 80 30 on on ADP fcis-9453 80 31 the the DET fcis-9453 80 32 same same ADJ fcis-9453 80 33 dataset dataset NOUN fcis-9453 80 34 for for ADP fcis-9453 80 35 epoch epoch NOUN fcis-9453 80 36 times time NOUN fcis-9453 80 37 respectively respectively ADV fcis-9453 80 38 , , PUNCT fcis-9453 80 39 and and CCONJ fcis-9453 80 40 the the DET fcis-9453 80 41 comprehensive comprehensive ADJ fcis-9453 80 42 evaluation evaluation NOUN fcis-9453 80 43 results result NOUN fcis-9453 80 44 are be AUX fcis-9453 80 45 shown show VERB fcis-9453 80 46 in in ADP fcis-9453 80 47 table table NOUN fcis-9453 80 48 1 1 NUM fcis-9453 80 49 . . PUNCT fcis-9453 80 50 table table NOUN fcis-9453 80 51 1 1 NUM fcis-9453 80 52 . . PUNCT fcis-9453 80 53 comprehensive comprehensive ADJ fcis-9453 80 54 assessment assessment NOUN fcis-9453 80 55 comparison comparison NOUN fcis-9453 80 56 network network NOUN fcis-9453 80 57 p/% p/% PROPN fcis-9453 80 58 r/% r/% PROPN fcis-9453 80 59 map@0.5/% map@0.5/% PROPN fcis-9453 80 60 map@0.95/% map@0.95/% PROPN fcis-9453 80 61 yolov5l yolov5l PROPN fcis-9453 80 62 62.6 62.6 NUM fcis-9453 80 63 64.2 64.2 NUM fcis-9453 80 64 65.7 65.7 NUM fcis-9453 80 65 30.5 30.5 NUM fcis-9453 80 66 yolov5lsimam yolov5lsimam NOUN fcis-9453 80 67 59.8 59.8 NUM fcis-9453 80 68 69.7 69.7 NUM fcis-9453 80 69 68.1 68.1 NUM fcis-9453 80 70 29.5 29.5 NUM fcis-9453 80 71 yolov5lc2f yolov5lc2f NOUN fcis-9453 80 72 67.8 67.8 NUM fcis-9453 80 73 63.3 63.3 NUM fcis-9453 80 74 66.1 66.1 NUM fcis-9453 80 75 33.5 33.5 NUM fcis-9453 80 76 yolov5lsimamc2f yolov5lsimamc2f NOUN fcis-9453 80 77 70.9 70.9 NUM fcis-9453 80 78 64.3 64.3 NUM fcis-9453 80 79 70.9 70.9 NUM fcis-9453 80 80 36.4 36.4 NUM fcis-9453 80 81     SPACE fcis-9453 80 82 the the DET fcis-9453 80 83 experimental experimental ADJ fcis-9453 80 84 results result NOUN fcis-9453 80 85 show show VERB fcis-9453 80 86 that that SCONJ fcis-9453 80 87 the the DET fcis-9453 80 88 detection detection NOUN fcis-9453 80 89 accuracy accuracy NOUN fcis-9453 80 90 is be AUX fcis-9453 80 91 significantly significantly ADV fcis-9453 80 92 improved improve VERB fcis-9453 80 93 after after ADP fcis-9453 80 94 the the DET fcis-9453 80 95 introduction introduction NOUN fcis-9453 80 96 of of ADP fcis-9453 80 97 c2f c2f NOUN fcis-9453 80 98 and and CCONJ fcis-9453 80 99 simam simam ADJ fcis-9453 80 100 attention attention NOUN fcis-9453 80 101 mechanism mechanism NOUN fcis-9453 80 102 in in ADP fcis-9453 80 103 the the DET fcis-9453 80 104 original original ADJ fcis-9453 80 105 yolov5l yolov5l NOUN fcis-9453 80 106 network network NOUN fcis-9453 80 107 . . PUNCT fcis-9453 81 1 this this DET fcis-9453 81 2 paper paper NOUN fcis-9453 81 3 is be AUX fcis-9453 81 4 based base VERB fcis-9453 81 5 on on ADP fcis-9453 81 6 the the DET fcis-9453 81 7 improved improve VERB fcis-9453 81 8 yolov5l yolov5l PROPN fcis-9453 81 9 model model NOUN fcis-9453 81 10 can can AUX fcis-9453 81 11 detect detect VERB fcis-9453 81 12 strip strip NOUN fcis-9453 81 13 surface surface NOUN fcis-9453 81 14 defects defect NOUN fcis-9453 81 15 more more ADV fcis-9453 81 16 effectively effectively ADV fcis-9453 81 17 compared compare VERB fcis-9453 81 18 to to ADP fcis-9453 81 19 the the DET fcis-9453 81 20 original original ADJ fcis-9453 81 21 yolov5l yolov5l NOUN fcis-9453 81 22 , , PUNCT fcis-9453 81 23 and and CCONJ fcis-9453 81 24 the the DET fcis-9453 81 25 comparison comparison NOUN fcis-9453 81 26 graph graph NOUN fcis-9453 81 27 of of ADP fcis-9453 81 28 detection detection NOUN fcis-9453 81 29 effect effect NOUN fcis-9453 81 30 is be AUX fcis-9453 81 31 shown show VERB fcis-9453 81 32 in in ADP fcis-9453 81 33 figure figure NOUN fcis-9453 81 34 7 7 NUM fcis-9453 81 35 . . PUNCT fcis-9453 81 36     SPACE fcis-9453 81 37 figure figure NOUN fcis-9453 81 38 7 7 NUM fcis-9453 81 39 . . PUNCT fcis-9453 81 40     SPACE fcis-9453 81 41 comparison comparison NOUN fcis-9453 81 42 of of ADP fcis-9453 81 43 the the DET fcis-9453 81 44 detection detection NOUN fcis-9453 81 45 effects effect NOUN fcis-9453 81 46 before before ADV fcis-9453 81 47 and and CCONJ fcis-9453 81 48 after after ADP fcis-9453 81 49 the the DET fcis-9453 81 50 algorithm algorithm NOUN fcis-9453 81 51 improvement improvement NOUN fcis-9453 81 52     SPACE fcis-9453 81 53 4 4 NUM fcis-9453 81 54 . . PUNCT fcis-9453 81 55 summary summary NOUN fcis-9453 81 56 in in ADP fcis-9453 81 57 the the DET fcis-9453 81 58 process process NOUN fcis-9453 81 59 of of ADP fcis-9453 81 60 detecting detect VERB fcis-9453 81 61 defects defect NOUN fcis-9453 81 62 on on ADP fcis-9453 81 63 the the DET fcis-9453 81 64 surface surface NOUN fcis-9453 81 65 of of ADP fcis-9453 81 66 hot hot ADJ fcis-9453 81 67 rolled roll VERB fcis-9453 81 68 strip strip NOUN fcis-9453 81 69 steel steel NOUN fcis-9453 81 70 , , PUNCT fcis-9453 81 71 due due ADP fcis-9453 81 72 to to ADP fcis-9453 81 73 the the DET fcis-9453 81 74 problems problem NOUN fcis-9453 81 75 of of ADP fcis-9453 81 76 irregular irregular ADJ fcis-9453 81 77 target target NOUN fcis-9453 81 78 shape shape NOUN fcis-9453 81 79 , , PUNCT fcis-9453 81 80 different different ADJ fcis-9453 81 81 scales scale NOUN fcis-9453 81 82 , , PUNCT fcis-9453 81 83 complex complex ADJ fcis-9453 81 84 background background NOUN fcis-9453 81 85 and and CCONJ fcis-9453 81 86 easy easy ADJ fcis-9453 81 87 false false ADJ fcis-9453 81 88 detection detection NOUN fcis-9453 81 89 and and CCONJ fcis-9453 81 90 omission omission NOUN fcis-9453 81 91 , , PUNCT fcis-9453 81 92 this this DET fcis-9453 81 93 paper paper NOUN fcis-9453 81 94 designs design VERB fcis-9453 81 95 a a DET fcis-9453 81 96 new new ADJ fcis-9453 81 97 detection detection NOUN fcis-9453 81 98 algorithm algorithm NOUN fcis-9453 81 99 based base VERB fcis-9453 81 100 on on ADP fcis-9453 81 101 the the DET fcis-9453 81 102 yolov5l yolov5l PROPN fcis-9453 81 103 algorithm algorithm NOUN fcis-9453 81 104 . . PUNCT fcis-9453 82 1 firstly firstly ADV fcis-9453 82 2 , , PUNCT fcis-9453 82 3 the the DET fcis-9453 82 4 simam simam ADJ fcis-9453 82 5 attention attention NOUN fcis-9453 82 6 mechanism mechanism NOUN fcis-9453 82 7 module module NOUN fcis-9453 82 8 is be AUX fcis-9453 82 9 added add VERB fcis-9453 82 10 to to ADP fcis-9453 82 11 the the DET fcis-9453 82 12 head head NOUN fcis-9453 82 13 side side NOUN fcis-9453 82 14 to to PART fcis-9453 82 15 improve improve VERB fcis-9453 82 16 the the DET fcis-9453 82 17 recall recall NOUN fcis-9453 82 18 of of ADP fcis-9453 82 19 the the DET fcis-9453 82 20 original original ADJ fcis-9453 82 21 algorithm algorithm NOUN fcis-9453 82 22 without without ADP fcis-9453 82 23 affecting affect VERB fcis-9453 82 24 the the DET fcis-9453 82 25 model model NOUN fcis-9453 82 26 parameters parameter NOUN fcis-9453 82 27 and and CCONJ fcis-9453 82 28 computational computational ADJ fcis-9453 82 29 complexity complexity NOUN fcis-9453 82 30 ; ; PUNCT fcis-9453 82 31 secondly secondly ADV fcis-9453 82 32 , , PUNCT fcis-9453 82 33 all all DET fcis-9453 82 34 c3 c3 NOUN fcis-9453 82 35 modules module NOUN fcis-9453 82 36 in in ADP fcis-9453 82 37 backbone backbone NOUN fcis-9453 82 38 and and CCONJ fcis-9453 82 39 head head NOUN fcis-9453 82 40 are be AUX fcis-9453 82 41 replaced replace VERB fcis-9453 82 42 with with ADP fcis-9453 82 43 c2f c2f NOUN fcis-9453 82 44 modules module NOUN fcis-9453 82 45 in in ADP fcis-9453 82 46 the the DET fcis-9453 82 47 original original ADJ fcis-9453 82 48 yolov5l yolov5l PROPN fcis-9453 82 49 model model NOUN fcis-9453 82 50 , , PUNCT fcis-9453 82 51 on on ADP fcis-9453 82 52 the the DET fcis-9453 82 53 basis basis NOUN fcis-9453 82 54 of of ADP fcis-9453 82 55 ensuring ensure VERB fcis-9453 82 56 its its PRON fcis-9453 82 57 lightweight lightweight NOUN fcis-9453 82 58 , , PUNCT fcis-9453 82 59 the the DET fcis-9453 82 60 improved improved ADJ fcis-9453 82 61 model model NOUN fcis-9453 82 62 obtains obtain VERB fcis-9453 82 63 more more ADV fcis-9453 82 64 abundant abundant ADJ fcis-9453 82 65 gradient gradient ADJ fcis-9453 82 66 flow flow NOUN fcis-9453 82 67 information information NOUN fcis-9453 82 68 and and CCONJ fcis-9453 82 69 improves improve VERB fcis-9453 82 70 the the DET fcis-9453 82 71 identification identification NOUN fcis-9453 82 72 accuracy accuracy NOUN fcis-9453 82 73 of of ADP fcis-9453 82 74 the the DET fcis-9453 82 75 model model NOUN fcis-9453 82 76 on on ADP fcis-9453 82 77 the the DET fcis-9453 82 78 defects defect NOUN fcis-9453 82 79 ; ; PUNCT fcis-9453 82 80 finally finally ADV fcis-9453 82 81 , , PUNCT fcis-9453 82 82 the the DET fcis-9453 82 83 experimental experimental ADJ fcis-9453 82 84 results result NOUN fcis-9453 82 85 on on ADP fcis-9453 82 86 the the DET fcis-9453 82 87 fused fuse VERB fcis-9453 82 88 simam simam NOUN fcis-9453 82 89 and and CCONJ fcis-9453 82 90 c2f c2f NOUN fcis-9453 82 91 module module NOUN fcis-9453 82 92 on on ADP fcis-9453 82 93 the the DET fcis-9453 82 94 neu neu PROPN fcis-9453 82 95 - - PUNCT fcis-9453 82 96 det det PROPN fcis-9453 82 97 dataset dataset PROPN fcis-9453 82 98 show show NOUN fcis-9453 82 99 that that SCONJ fcis-9453 82 100 this this DET fcis-9453 82 101 model model NOUN fcis-9453 82 102 achieves achieve VERB fcis-9453 82 103 5.3 5.3 NUM fcis-9453 82 104 % % NOUN fcis-9453 82 105 improvement improvement NOUN fcis-9453 82 106 in in ADP fcis-9453 82 107 the the DET fcis-9453 82 108 accuracy accuracy NOUN fcis-9453 82 109 of of ADP fcis-9453 82 110 detecting detect VERB fcis-9453 82 111 strip strip NOUN fcis-9453 82 112 surface surface NOUN fcis-9453 82 113 defects defect NOUN fcis-9453 82 114 compared compare VERB fcis-9453 82 115 to to ADP fcis-9453 82 116 conventional conventional ADJ fcis-9453 82 117 neural neural ADJ fcis-9453 82 118 networks network NOUN fcis-9453 82 119 . . PUNCT fcis-9453 83 1 references reference NOUN fcis-9453 83 2 [ [ X fcis-9453 83 3 1 1 NUM fcis-9453 83 4 ] ] X fcis-9453 83 5 li li PROPN fcis-9453 83 6 , , PUNCT fcis-9453 83 7 y. y. PROPN fcis-9453 83 8 , , PUNCT fcis-9453 83 9 et et PROPN fcis-9453 83 10 al al PROPN fcis-9453 83 11 . . PUNCT fcis-9453 84 1 " " PUNCT fcis-9453 84 2 progress progress NOUN fcis-9453 84 3 in in ADP fcis-9453 84 4 surface surface NOUN fcis-9453 84 5 defect defect NOUN fcis-9453 84 6 detection detection NOUN fcis-9453 84 7 methods method NOUN fcis-9453 84 8 for for ADP fcis-9453 84 9 strip strip NOUN fcis-9453 84 10 steel steel NOUN fcis-9453 84 11 . . PUNCT fcis-9453 84 12 " " PUNCT fcis-9453 85 1 journal journal NOUN fcis-9453 85 2 of of ADP fcis-9453 85 3 iron iron NOUN fcis-9453 85 4 and and CCONJ fcis-9453 85 5 steel steel NOUN fcis-9453 85 6 research research NOUN fcis-9453 85 7 . . PUNCT fcis-9453 86 1 doi doi NOUN fcis-9453 86 2 : : PUNCT fcis-9453 86 3 10 10 NUM fcis-9453 86 4 . . X fcis-9453 86 5 13228/ 13228/ NUM fcis-9453 86 6 j. j. PROPN fcis-9453 86 7 boyuan boyuan PROPN fcis-9453 86 8 . . PUNCT fcis-9453 87 1 issn1001 issn1001 PROPN fcis-9453 87 2 - - PUNCT fcis-9453 87 3 0963.20220363 0963.20220363 PROPN fcis-9453 87 4 . . PUNCT fcis-9453 88 1 [ [ X fcis-9453 88 2 2 2 X fcis-9453 88 3 ] ] PUNCT fcis-9453 88 4 wang wang PROPN fcis-9453 88 5 meng meng PROPN fcis-9453 88 6 . . PUNCT fcis-9453 88 7 " " PUNCT fcis-9453 89 1 a a DET fcis-9453 89 2 multi multi ADJ fcis-9453 89 3 - - ADJ fcis-9453 89 4 scale scale ADJ fcis-9453 89 5 feature feature NOUN fcis-9453 89 6 map map NOUN fcis-9453 89 7 - - PUNCT fcis-9453 89 8 based base VERB fcis-9453 89 9 method method NOUN fcis-9453 89 10 for for ADP fcis-9453 89 11 detecting detect VERB fcis-9453 89 12 defects defect NOUN fcis-9453 89 13 in in ADP fcis-9453 89 14 strip strip NOUN fcis-9453 89 15 steel steel NOUN fcis-9453 89 16 . . PUNCT fcis-9453 89 17 " " PUNCT fcis-9453 90 1 digital digital ADJ fcis-9453 90 2 technology technology NOUN fcis-9453 90 3 and and CCONJ fcis-9453 90 4 applications application NOUN fcis-9453 90 5 40.04(2022):36 40.04(2022):36 NUM fcis-9453 90 6 - - SYM fcis-9453 90 7 39 39 NUM fcis-9453 90 8 . . PUNCT fcis-9453 91 1 doi:10.19695 doi:10.19695 PROPN fcis-9453 91 2 / / SYM fcis-9453 91 3 j.cnki.cn121369 j.cnki.cn121369 PROPN fcis-9453 91 4 . . PUNCT fcis-9453 91 5 2022.04.12 2022.04.12 NUM fcis-9453 91 6 . . PUNCT fcis-9453 92 1 [ [ X fcis-9453 92 2 3 3 X fcis-9453 92 3 ] ] X fcis-9453 92 4 zhang zhang PROPN fcis-9453 92 5 yan yan PROPN fcis-9453 92 6 , , PUNCT fcis-9453 92 7 and and CCONJ fcis-9453 92 8 feng feng PROPN fcis-9453 92 9 feng feng PROPN fcis-9453 92 10 . . PUNCT fcis-9453 92 11 " " PUNCT fcis-9453 93 1 exploration exploration NOUN fcis-9453 93 2 of of ADP fcis-9453 93 3 strip strip PROPN fcis-9453 93 4 steel steel NOUN fcis-9453 93 5 surface surface NOUN fcis-9453 93 6 defect defect NOUN fcis-9453 93 7 detection detection NOUN fcis-9453 93 8 technology technology NOUN fcis-9453 93 9 . . PUNCT fcis-9453 93 10 " " PUNCT fcis-9453 94 1 information information NOUN fcis-9453 94 2 and and CCONJ fcis-9453 94 3 computer computer NOUN fcis-9453 94 4 ( ( PUNCT fcis-9453 94 5 theoretical theoretical ADJ fcis-9453 94 6 edition edition NOUN fcis-9453 94 7 ) ) PUNCT fcis-9453 94 8 33.11(2021):19 33.11(2021):19 NUM fcis-9453 94 9 - - SYM fcis-9453 94 10 22 22 NUM fcis-9453 94 11 . . PUNCT fcis-9453 95 1 [ [ X fcis-9453 95 2 4 4 X fcis-9453 95 3 ] ] PUNCT fcis-9453 95 4 pan pan PROPN fcis-9453 95 5 meng meng PROPN fcis-9453 95 6 , , PUNCT fcis-9453 95 7 zhou zhou PROPN fcis-9453 95 8 deqiang deqiang PROPN fcis-9453 95 9 , , PUNCT fcis-9453 95 10 and and CCONJ fcis-9453 95 11 chang chang PROPN fcis-9453 95 12 xiang xiang PROPN fcis-9453 95 13 . . PUNCT fcis-9453 95 14 " " PUNCT fcis-9453 95 15 characterization characterization NOUN fcis-9453 95 16 of of ADP fcis-9453 95 17 surface surface NOUN fcis-9453 95 18 defect defect NOUN fcis-9453 95 19 detection detection NOUN fcis-9453 95 20 by by ADP fcis-9453 95 21 a a DET fcis-9453 95 22 novel novel ADJ fcis-9453 95 23 pulsed pulse VERB fcis-9453 95 24 leakage leakage NOUN fcis-9453 95 25 magnetic magnetic ADJ fcis-9453 95 26 detection detection NOUN fcis-9453 95 27 method method NOUN fcis-9453 95 28 . . PUNCT fcis-9453 95 29 " " PUNCT fcis-9453 96 1 sensors sensor NOUN fcis-9453 96 2 and and CCONJ fcis-9453 96 3 microsystems microsystem NOUN fcis-9453 96 4 36.12 36.12 NUM fcis-9453 96 5 ( ( PUNCT fcis-9453 96 6 2017 2017 NUM fcis-9453 96 7 ): ): PUNCT fcis-9453 96 8 32 32 NUM fcis-9453 96 9 - - SYM fcis-9453 96 10 35 35 NUM fcis-9453 96 11 . . PUNCT fcis-9453 96 12 doi:10.13873 doi:10.13873 PROPN fcis-9453 96 13 / / SYM fcis-9453 96 14 j.1000 j.1000 PROPN fcis-9453 96 15 - - PUNCT fcis-9453 96 16 9787(2017)12 9787(2017)12 PROPN fcis-9453 96 17 - - PUNCT fcis-9453 96 18 0032 0032 NUM fcis-9453 96 19 - - PUNCT fcis-9453 96 20 04 04 NUM fcis-9453 96 21 . . PUNCT fcis-9453 97 1 [ [ X fcis-9453 97 2 5 5 X fcis-9453 97 3 ] ] PUNCT fcis-9453 97 4 wang wang PROPN fcis-9453 97 5 b b PROPN fcis-9453 97 6 , , PUNCT fcis-9453 97 7 et et PROPN fcis-9453 97 8 al al PROPN fcis-9453 97 9 . . PUNCT fcis-9453 98 1 " " PUNCT fcis-9453 98 2 a a DET fcis-9453 98 3 new new ADJ fcis-9453 98 4 eddy eddy NOUN fcis-9453 98 5 current current ADJ fcis-9453 98 6 detection detection NOUN fcis-9453 98 7 method method NOUN fcis-9453 98 8 and and CCONJ fcis-9453 98 9 its its PRON fcis-9453 98 10 detection detection NOUN fcis-9453 98 11 effect effect NOUN fcis-9453 98 12 . . PUNCT fcis-9453 98 13 " " PUNCT fcis-9453 99 1 metallurgy metallurgy NOUN fcis-9453 99 2 of of ADP fcis-9453 99 3 china china PROPN fcis-9453 99 4 31.02(2021):50 31.02(2021):50 PROPN fcis-9453 99 5 - - SYM fcis-9453 99 6 54 54 NUM fcis-9453 99 7 . . PUNCT fcis-9453 100 1 doi:10.13228 doi:10.13228 PROPN fcis-9453 100 2 / / SYM fcis-9453 100 3 j.boyuan.issn1006 j.boyuan.issn1006 PROPN fcis-9453 100 4 - - PUNCT fcis-9453 100 5 9356.20200350 9356.20200350 PROPN fcis-9453 100 6 . . PUNCT fcis-9453 101 1 [ [ X fcis-9453 101 2 6 6 NUM fcis-9453 101 3 ] ] SYM fcis-9453 101 4 ma ma PROPN fcis-9453 101 5 , , PUNCT fcis-9453 101 6 k. k. PROPN fcis-9453 101 7 , , PUNCT fcis-9453 101 8 et et PROPN fcis-9453 101 9 al al PROPN fcis-9453 101 10 . . PUNCT fcis-9453 102 1 " " PUNCT fcis-9453 102 2 study study NOUN fcis-9453 102 3 on on ADP fcis-9453 102 4 sf_6 sf_6 PROPN fcis-9453 102 5 decomposition decomposition NOUN fcis-9453 102 6 under under ADP fcis-9453 102 7 pin pin NOUN fcis-9453 102 8 - - PUNCT fcis-9453 102 9 plate plate NOUN fcis-9453 102 10 defects defect NOUN fcis-9453 102 11 based base VERB fcis-9453 102 12 on on ADP fcis-9453 102 13 infrared infrared ADJ fcis-9453 102 14 detection detection NOUN fcis-9453 102 15 method method NOUN fcis-9453 102 16 . . PUNCT fcis-9453 102 17 " " PUNCT fcis-9453 103 1 high high ADJ fcis-9453 103 2 voltage voltage NOUN fcis-9453 103 3 electronics electronic NOUN fcis-9453 103 4 48.12 48.12 NUM fcis-9453 103 5 ( ( PUNCT fcis-9453 103 6 2012 2012 NUM fcis-9453 103 7 ): ): PUNCT fcis-9453 103 8 70 70 NUM fcis-9453 103 9 - - SYM fcis-9453 103 10 74 74 NUM fcis-9453 103 11 . . PUNCT fcis-9453 104 1 doi:10.13296 doi:10.13296 NOUN fcis-9453 104 2 / / SYM fcis-9453 104 3 j.1001 j.1001 NOUN fcis-9453 104 4 - - PUNCT fcis-9453 104 5 1609 1609 NUM fcis-9453 104 6 . . PUNCT fcis-9453 105 1 hva hva PROPN fcis-9453 105 2 . . PUNCT fcis-9453 106 1 2012.12.015 2012.12.015 X fcis-9453 106 2 . . PUNCT fcis-9453 107 1 [ [ X fcis-9453 107 2 7 7 NUM fcis-9453 107 3 ] ] X fcis-9453 107 4 tian tian PROPN fcis-9453 107 5 w. w. PROPN fcis-9453 107 6 research research PROPN fcis-9453 107 7 on on ADP fcis-9453 107 8 curriculum curriculum NOUN fcis-9453 107 9 design design NOUN fcis-9453 107 10 method method NOUN fcis-9453 107 11 of of ADP fcis-9453 107 12 teaching teach VERB fcis-9453 107 13 resource resource NOUN fcis-9453 107 14 library library NOUN fcis-9453 107 15 based base VERB fcis-9453 107 16 on on ADP fcis-9453 107 17 deep deep ADJ fcis-9453 107 18 learning learning NOUN fcis-9453 107 19 technology technology NOUN fcis-9453 107 20 [ [ X fcis-9453 107 21 c]// c]// PROPN fcis-9453 107 22 wuhan wuhan PROPN fcis-9453 107 23 zhicheng zhicheng PROPN fcis-9453 107 24 times times PROPN fcis-9453 107 25 cultural cultural ADJ fcis-9453 107 26 proceedings proceeding NOUN fcis-9453 107 27 of of ADP fcis-9453 107 28 6th 6th ADJ fcis-9453 107 29 international international ADJ fcis-9453 107 30 conference conference NOUN fcis-9453 107 31 on on ADP fcis-9453 107 32 computer computer NOUN fcis-9453 107 33 engineering engineering NOUN fcis-9453 107 34 , , PUNCT fcis-9453 107 35 information information NOUN fcis-9453 107 36 science science PROPN fcis-9453 107 37 & & CCONJ fcis-9453 107 38 application application NOUN fcis-9453 107 39 technology technology NOUN fcis-9453 107 40 ( ( PUNCT fcis-9453 107 41 iccia iccia NOUN fcis-9453 107 42 2023 2023 NUM fcis-9453 107 43 ) ) PUNCT fcis-9453 107 44 . . PUNCT fcis-9453 108 1 [ [ X fcis-9453 108 2 publisher publisher NOUN fcis-9453 108 3 unknown unknown ADJ fcis-9453 108 4 ] ] PUNCT fcis-9453 108 5 , , PUNCT fcis-9453 108 6 2023:148 2023:148 NUM fcis-9453 108 7 - - SYM fcis-9453 108 8 152 152 NUM fcis-9453 108 9 . . PUNCT fcis-9453 109 1 doi:10 doi:10 PROPN fcis-9453 109 2 . . PUNCT fcis-9453 110 1 26914/ 26914/ NUM fcis-9453 110 2 c. c. PROPN fcis-9453 110 3 cnkihy cnkihy PROPN fcis-9453 110 4 . . PUNCT fcis-9453 111 1 2023 2023 NUM fcis-9453 111 2 . . PUNCT fcis-9453 112 1 010734 010734 NUM fcis-9453 112 2 . . PUNCT fcis-9453 113 1 [ [ X fcis-9453 113 2 8 8 NUM fcis-9453 113 3 ] ] X fcis-9453 113 4 zhang zhang PROPN fcis-9453 113 5 zemiao zemiao PROPN fcis-9453 113 6 , , PUNCT fcis-9453 113 7 huo huo PROPN fcis-9453 113 8 huan huan PROPN fcis-9453 113 9 , , PUNCT fcis-9453 113 10 and and CCONJ fcis-9453 113 11 zhao zhao PROPN fcis-9453 113 12 fengyu fengyu NOUN fcis-9453 113 13 . . PUNCT fcis-9453 113 14 " " PUNCT fcis-9453 114 1 a a DET fcis-9453 114 2 review review NOUN fcis-9453 114 3 of of ADP fcis-9453 114 4 target target NOUN fcis-9453 114 5 detection detection NOUN fcis-9453 114 6 algorithms algorithm NOUN fcis-9453 114 7 for for ADP fcis-9453 114 8 deep deep ADJ fcis-9453 114 9 convolutional convolutional ADJ fcis-9453 114 10 neural neural ADJ fcis-9453 114 11 networks network NOUN fcis-9453 114 12 . . PUNCT fcis-9453 114 13 " " PUNCT fcis-9453 114 14 small small ADJ fcis-9453 114 15 microcomputer microcomputer NOUN fcis-9453 114 16 systems system NOUN fcis-9453 114 17 40.09(2019):18251831 40.09(2019):18251831 NOUN fcis-9453 114 18 . . PUNCT fcis-9453 115 1 [ [ X fcis-9453 115 2 9 9 NUM fcis-9453 115 3 ] ] SYM fcis-9453 115 4 ning ning NOUN fcis-9453 115 5 , , PUNCT fcis-9453 115 6 j. j. PROPN fcis-9453 115 7 , , PUNCT fcis-9453 115 8 et et PROPN fcis-9453 115 9 al al PROPN fcis-9453 115 10 . . PUNCT fcis-9453 116 1 " " PUNCT fcis-9453 116 2 a a DET fcis-9453 116 3 review review NOUN fcis-9453 116 4 of of ADP fcis-9453 116 5 target target NOUN fcis-9453 116 6 detection detection NOUN fcis-9453 116 7 algorithms algorithm NOUN fcis-9453 116 8 for for ADP fcis-9453 116 9 deep deep ADJ fcis-9453 116 10 learning learning NOUN fcis-9453 116 11 . . PUNCT fcis-9453 116 12 " " PUNCT fcis-9453 117 1 information information NOUN fcis-9453 117 2 recorded record VERB fcis-9453 117 3 materials material NOUN fcis-9453 117 4 23.10 23.10 NUM fcis-9453 117 5 ( ( PUNCT fcis-9453 117 6 2022 2022 NUM fcis-9453 117 7 ): ): PUNCT fcis-9453 117 8 1 1 NUM fcis-9453 117 9 - - SYM fcis-9453 117 10 4 4 NUM fcis-9453 117 11 . . PUNCT fcis-9453 117 12 doi:10.16009 doi:10.16009 PROPN fcis-9453 117 13 / / SYM fcis-9453 117 14 j.cnki.cn13 j.cnki.cn13 PROPN fcis-9453 117 15 - - PUNCT fcis-9453 117 16 1295 1295 NUM fcis-9453 117 17 / / SYM fcis-9453 117 18 tq.2022.10.057 tq.2022.10.057 NOUN fcis-9453 117 19 . . PUNCT fcis-9453 118 1 [ [ X fcis-9453 118 2 10 10 NUM fcis-9453 118 3 ] ] X fcis-9453 118 4 wen wen PROPN fcis-9453 118 5 , , PUNCT fcis-9453 118 6 xinglin xinglin PROPN fcis-9453 118 7 , , PUNCT fcis-9453 118 8 and and CCONJ fcis-9453 118 9 bai bai PROPN fcis-9453 118 10 , , PUNCT fcis-9453 118 11 tao tao PROPN fcis-9453 118 12 . . PUNCT fcis-9453 118 13 " " PUNCT fcis-9453 119 1 design design NOUN fcis-9453 119 2 of of ADP fcis-9453 119 3 audio audio ADJ fcis-9453 119 4 emotion emotion NOUN fcis-9453 119 5 recognition recognition NOUN fcis-9453 119 6 and and CCONJ fcis-9453 119 7 classification classification NOUN fcis-9453 119 8 model model NOUN fcis-9453 119 9 based base VERB fcis-9453 119 10 on on ADP fcis-9453 119 11 improved improved ADJ fcis-9453 119 12 multimodal multimodal NOUN fcis-9453 119 13 rcnn rcnn NOUN fcis-9453 119 14 . . PUNCT fcis-9453 119 15 " " PUNCT fcis-9453 120 1 modern modern ADJ fcis-9453 120 2 electronics electronic NOUN fcis-9453 120 3 46.11(2023):114118 46.11(2023):114118 NUM fcis-9453 120 4 . . PUNCT fcis-9453 120 5 doi:10.16652 doi:10.16652 NOUN fcis-9453 120 6 / / SYM fcis-9453 120 7 j.issn.1004 j.issn.1004 NOUN fcis-9453 120 8 - - PUNCT fcis-9453 120 9 373x.2023.11.021 373x.2023.11.021 NUM fcis-9453 120 10 . . PUNCT fcis-9453 121 1 [ [ X fcis-9453 121 2 11 11 NUM fcis-9453 121 3 ] ] SYM fcis-9453 121 4 yang yang PROPN fcis-9453 121 5 , , PUNCT fcis-9453 121 6 h. h. PROPN fcis-9453 121 7 zhou zhou PROPN fcis-9453 121 8 , , PUNCT fcis-9453 121 9 and and CCONJ fcis-9453 121 10 li li PROPN fcis-9453 121 11 , , PUNCT fcis-9453 121 12 d. d. PROPN fcis-9453 121 13 li li PROPN fcis-9453 121 14 . . PUNCT fcis-9453 121 15 " " PUNCT fcis-9453 122 1 yolov4 yolov4 PROPN fcis-9453 122 2 target target NOUN fcis-9453 122 3 detection detection NOUN fcis-9453 122 4 based base VERB fcis-9453 122 5 on on ADP fcis-9453 122 6 improved improved ADJ fcis-9453 122 7 sppnet sppnet NOUN fcis-9453 122 8 . . PUNCT fcis-9453 122 9 " " PUNCT fcis-9453 123 1 electronic electronic ADJ fcis-9453 123 2 fabrication fabrication NOUN fcis-9453 123 3 .22(2021):52 .22(2021):52 NOUN fcis-9453 123 4 - - PUNCT fcis-9453 123 5 54 54 NUM fcis-9453 123 6 . . PUNCT fcis-9453 124 1 doi:10.16589 doi:10.16589 NOUN fcis-9453 124 2 / / SYM fcis-9453 124 3 j.cnki.cn11 j.cnki.cn11 PROPN fcis-9453 124 4 - - PUNCT fcis-9453 124 5 3571 3571 NUM fcis-9453 124 6 / / SYM fcis-9453 124 7 tn.2021.22.018 tn.2021.22.018 NOUN fcis-9453 124 8 . . PUNCT fcis-9453 125 1 [ [ X fcis-9453 125 2 12 12 NUM fcis-9453 125 3 ] ] X fcis-9453 125 4 wang wang PROPN fcis-9453 125 5 , , PUNCT fcis-9453 125 6 yanqing yanqing PROPN fcis-9453 125 7 , , PUNCT fcis-9453 125 8 et et PROPN fcis-9453 125 9 al al PROPN fcis-9453 125 10 . . PUNCT fcis-9453 126 1 " " PUNCT fcis-9453 126 2 application application NOUN fcis-9453 126 3 of of ADP fcis-9453 126 4 fastrcnn fastrcnn PROPN fcis-9453 126 5 and and CCONJ fcis-9453 126 6 cnn cnn PROPN fcis-9453 126 7 techniques technique NOUN fcis-9453 126 8 in in ADP fcis-9453 126 9 the the DET fcis-9453 126 10 morphological morphological ADJ fcis-9453 126 11 identification identification NOUN fcis-9453 126 12 of of ADP fcis-9453 126 13 three three NUM fcis-9453 126 14 parasite parasite ADJ fcis-9453 126 15 eggs egg NOUN fcis-9453 126 16 . . PUNCT fcis-9453 126 17 " " PUNCT fcis-9453 127 1 journal journal NOUN fcis-9453 127 2 of of ADP fcis-9453 127 3 qiqihar qiqihar PROPN fcis-9453 127 4 medical medical PROPN fcis-9453 127 5 college college PROPN fcis-9453 127 6 43.03(2022):234237 43.03(2022):234237 PROPN fcis-9453 127 7 . . PUNCT fcis-9453 128 1 71 71 NUM fcis-9453 129 1 [ [ SYM fcis-9453 129 2 13 13 NUM fcis-9453 129 3 ] ] X fcis-9453 129 4 bai bai PROPN fcis-9453 129 5 chenshuai chenshuai PROPN fcis-9453 129 6 , , PUNCT fcis-9453 129 7 et et PROPN fcis-9453 129 8 al al PROPN fcis-9453 129 9 . . PUNCT fcis-9453 130 1 " " PUNCT fcis-9453 130 2 an an DET fcis-9453 130 3 improved improved ADJ fcis-9453 130 4 target target NOUN fcis-9453 130 5 detection detection NOUN fcis-9453 130 6 algorithm algorithm NOUN fcis-9453 130 7 based base VERB fcis-9453 130 8 on on ADP fcis-9453 130 9 faster fast ADJ fcis-9453 130 10 - - PUNCT fcis-9453 130 11 rcnn rcnn NOUN fcis-9453 130 12 ( ( PUNCT fcis-9453 130 13 in in ADP fcis-9453 130 14 english english PROPN fcis-9453 130 15 ) ) PUNCT fcis-9453 130 16 . . PUNCT fcis-9453 130 17 " " PUNCT fcis-9453 131 1 journal journal NOUN fcis-9453 131 2 of of ADP fcis-9453 131 3 measurement measurement NOUN fcis-9453 131 4 science science NOUN fcis-9453 131 5 and and CCONJ fcis-9453 131 6 instrumentation instrumentation NOUN fcis-9453 131 7 . . PUNCT fcis-9453 132 1 [ [ X fcis-9453 132 2 14 14 NUM fcis-9453 132 3 ] ] X fcis-9453 132 4 wang wang PROPN fcis-9453 132 5 , , PUNCT fcis-9453 132 6 dao dao NOUN fcis-9453 132 7 - - PUNCT fcis-9453 132 8 lei lei VERB fcis-9453 132 9 , , PUNCT fcis-9453 132 10 et et PROPN fcis-9453 132 11 al al PROPN fcis-9453 132 12 . . PUNCT fcis-9453 133 1 " " PUNCT fcis-9453 133 2 improved improve VERB fcis-9453 133 3 ssd ssd NOUN fcis-9453 133 4 method method NOUN fcis-9453 133 5 for for ADP fcis-9453 133 6 detection detection NOUN fcis-9453 133 7 of of ADP fcis-9453 133 8 hot hot ADJ fcis-9453 133 9 spot spot NOUN fcis-9453 133 10 defects defect NOUN fcis-9453 133 11 in in ADP fcis-9453 133 12 photovoltaic photovoltaic NOUN fcis-9453 133 13 modules module NOUN fcis-9453 133 14 . . PUNCT fcis-9453 133 15 " " PUNCT fcis-9453 134 1 journal journal NOUN fcis-9453 134 2 of of ADP fcis-9453 134 3 solar solar ADJ fcis-9453 134 4 energy energy NOUN fcis-9453 134 5 44 44 NUM fcis-9453 134 6 . . PUNCT fcis-9453 134 7 04 04 NUM fcis-9453 134 8 ( ( PUNCT fcis-9453 134 9 2023):420 2023):420 NUM fcis-9453 134 10 - - SYM fcis-9453 134 11 425 425 NUM fcis-9453 134 12 . . PUNCT fcis-9453 135 1 doi doi NOUN fcis-9453 135 2 : : PUNCT fcis-9453 135 3 10 10 NUM fcis-9453 135 4 . . NUM fcis-9453 135 5 19912/ 19912/ NUM fcis-9453 135 6 j.0254 j.0254 ADV fcis-9453 135 7 - - PUNCT fcis-9453 135 8 0096 0096 NUM fcis-9453 135 9 . . PUNCT fcis-9453 136 1 tynxb tynxb NOUN fcis-9453 136 2 . . PUNCT fcis-9453 137 1 2021 2021 NUM fcis-9453 137 2 - - SYM fcis-9453 137 3 1470 1470 NUM fcis-9453 137 4 . . PUNCT fcis-9453 138 1 [ [ X fcis-9453 138 2 15 15 NUM fcis-9453 138 3 ] ] X fcis-9453 138 4 zhou zhou PROPN fcis-9453 138 5 , , PUNCT fcis-9453 138 6 jinwei jinwei PROPN fcis-9453 138 7 , , PUNCT fcis-9453 138 8 and and CCONJ fcis-9453 138 9 wang wang PROPN fcis-9453 138 10 , , PUNCT fcis-9453 138 11 jianping jianping NOUN fcis-9453 138 12 . . PUNCT fcis-9453 138 13 " " PUNCT fcis-9453 139 1 a a DET fcis-9453 139 2 review review NOUN fcis-9453 139 3 of of ADP fcis-9453 139 4 yolo yolo ADJ fcis-9453 139 5 object object NOUN fcis-9453 139 6 detection detection NOUN fcis-9453 139 7 algorithm algorithm NOUN fcis-9453 139 8 research research NOUN fcis-9453 139 9 . . PUNCT fcis-9453 139 10 " " PUNCT fcis-9453 140 1 journal journal PROPN fcis-9453 140 2 of of ADP fcis-9453 140 3 changzhou changzhou PROPN fcis-9453 140 4 institute institute PROPN fcis-9453 140 5 of of ADP fcis-9453 140 6 technology technology PROPN fcis-9453 140 7 36.01(2023):18 36.01(2023):18 PROPN fcis-9453 140 8 - - PUNCT fcis-9453 140 9 23 23 NUM fcis-9453 140 10 + + NOUN fcis-9453 140 11 88 88 NUM fcis-9453 140 12 . . PUNCT fcis-9453 141 1 [ [ X fcis-9453 141 2 16 16 NUM fcis-9453 141 3 ] ] X fcis-9453 141 4 liu liu PROPN fcis-9453 141 5 , , PUNCT fcis-9453 141 6 j. j. PROPN fcis-9453 141 7 chuan chuan PROPN fcis-9453 141 8 , , PUNCT fcis-9453 141 9 et et PROPN fcis-9453 141 10 al al PROPN fcis-9453 141 11 . . PUNCT fcis-9453 142 1 " " PUNCT fcis-9453 142 2 improved improved ADJ fcis-9453 142 3 retinanet retinanet NOUN fcis-9453 142 4 for for ADP fcis-9453 142 5 uav uav PROPN fcis-9453 142 6 small small ADJ fcis-9453 142 7 target target NOUN fcis-9453 142 8 detection detection NOUN fcis-9453 142 9 . . PUNCT fcis-9453 142 10 " " PUNCT fcis-9453 143 1 science science NOUN fcis-9453 143 2 technology technology NOUN fcis-9453 143 3 and and CCONJ fcis-9453 143 4 engineering engineer VERB fcis-9453 143 5 23.01 23.01 NUM fcis-9453 143 6 ( ( PUNCT fcis-9453 143 7 2023 2023 NUM fcis-9453 143 8 ): ): PUNCT fcis-9453 143 9 274 274 NUM fcis-9453 143 10 - - SYM fcis-9453 143 11 282 282 NUM fcis-9453 143 12 . . PUNCT fcis-9453 144 1 [ [ X fcis-9453 144 2 17 17 NUM fcis-9453 144 3 ] ] X fcis-9453 144 4 zhang zhang PROPN fcis-9453 144 5 , , PUNCT fcis-9453 144 6 zhengchao zhengchao PROPN fcis-9453 144 7 . . PUNCT fcis-9453 144 8 " " PUNCT fcis-9453 144 9 improving improve VERB fcis-9453 144 10 yolov5 yolov5 NOUN fcis-9453 144 11 for for ADP fcis-9453 144 12 lightweight lightweight ADJ fcis-9453 144 13 strip strip PROPN fcis-9453 144 14 steel steel NOUN fcis-9453 144 15 surface surface NOUN fcis-9453 144 16 defect defect NOUN fcis-9453 144 17 detection detection NOUN fcis-9453 144 18 . . PUNCT fcis-9453 144 19 " " PUNCT fcis-9453 144 20 computer computer NOUN fcis-9453 144 21 systems system NOUN fcis-9453 144 22 applications application NOUN fcis-9453 144 23 . . PUNCT fcis-9453 145 1 doi:10.15888 doi:10.15888 PROPN fcis-9453 145 2 / / SYM fcis-9453 145 3 j.cnki.csa.009162 j.cnki.csa.009162 PROPN fcis-9453 145 4 . . PUNCT fcis-9453 146 1 [ [ X fcis-9453 146 2 18 18 NUM fcis-9453 146 3 ] ] SYM fcis-9453 146 4 li li PROPN fcis-9453 146 5 , , PUNCT fcis-9453 146 6 w.g w.g PROPN fcis-9453 146 7 . . PROPN fcis-9453 146 8 , , PUNCT fcis-9453 146 9 et et PROPN fcis-9453 146 10 al al PROPN fcis-9453 146 11 . . PUNCT fcis-9453 146 12 " " PUNCT fcis-9453 146 13 surface surface NOUN fcis-9453 146 14 defect defect NOUN fcis-9453 146 15 detection detection NOUN fcis-9453 146 16 of of ADP fcis-9453 146 17 strip strip NOUN fcis-9453 146 18 steel steel NOUN fcis-9453 146 19 based base VERB fcis-9453 146 20 on on ADP fcis-9453 146 21 improved improved ADJ fcis-9453 146 22 yolov3 yolov3 PROPN fcis-9453 146 23 algorithm algorithm PROPN fcis-9453 146 24 . . PUNCT fcis-9453 146 25 " " PUNCT fcis-9453 147 1 journal journal NOUN fcis-9453 147 2 of of ADP fcis-9453 147 3 electronics electronic NOUN fcis-9453 147 4 48.07 48.07 NUM fcis-9453 147 5 ( ( PUNCT fcis-9453 147 6 2020):1284 2020):1284 NOUN fcis-9453 147 7 - - PUNCT fcis-9453 147 8 1292 1292 NUM fcis-9453 147 9 . . PUNCT fcis-9453 148 1 [ [ X fcis-9453 148 2 19 19 NUM fcis-9453 148 3 ] ] SYM fcis-9453 148 4 zou zou PROPN fcis-9453 148 5 , , PUNCT fcis-9453 148 6 wang wang PROPN fcis-9453 148 7 , , PUNCT fcis-9453 148 8 and and CCONJ fcis-9453 148 9 ji ji PROPN fcis-9453 148 10 , , PUNCT fcis-9453 148 11 chang chang PROPN fcis-9453 148 12 . . PUNCT fcis-9453 148 13 " " PUNCT fcis-9453 148 14 an an DET fcis-9453 148 15 improved improved ADJ fcis-9453 148 16 yolov4 yolov4 NOUN fcis-9453 148 17 - - PUNCT fcis-9453 148 18 tiny tiny ADJ fcis-9453 148 19 method method NOUN fcis-9453 148 20 for for ADP fcis-9453 148 21 real real ADJ fcis-9453 148 22 - - PUNCT fcis-9453 148 23 time time NOUN fcis-9453 148 24 detection detection NOUN fcis-9453 148 25 of of ADP fcis-9453 148 26 strip strip PROPN fcis-9453 148 27 steel steel NOUN fcis-9453 148 28 surface surface NOUN fcis-9453 148 29 defects defect NOUN fcis-9453 148 30 . . PUNCT fcis-9453 148 31 " " PUNCT fcis-9453 149 1 mechanical mechanical ADJ fcis-9453 149 2 science science NOUN fcis-9453 149 3 and and CCONJ fcis-9453 149 4 technology technology NOUN fcis-9453 149 5 . . PUNCT fcis-9453 150 1 doi:10 doi:10 PROPN fcis-9453 150 2 . . PUNCT fcis-9453 151 1 13433/ 13433/ NUM fcis-9453 151 2 j. j. PROPN fcis-9453 151 3 cnki cnki PROPN fcis-9453 151 4 . . PROPN fcis-9453 151 5 1003 1003 NUM fcis-9453 151 6 - - PUNCT fcis-9453 151 7 8728.20230034 8728.20230034 NUM fcis-9453 151 8 . . PUNCT fcis-9453 152 1 [ [ X fcis-9453 152 2 20 20 NUM fcis-9453 152 3 ] ] SYM fcis-9453 152 4 li li PROPN fcis-9453 152 5 , , PUNCT fcis-9453 152 6 shun shun PROPN fcis-9453 152 7 , , PUNCT fcis-9453 152 8 and and CCONJ fcis-9453 152 9 yang yang PROPN fcis-9453 152 10 , , PUNCT fcis-9453 152 11 ying ying PROPN fcis-9453 152 12 . . PUNCT fcis-9453 152 13 " " PUNCT fcis-9453 152 14 improved improve VERB fcis-9453 152 15 yolov5 yolov5 NOUN fcis-9453 152 16 - - PUNCT fcis-9453 152 17 based base VERB fcis-9453 152 18 defect defect NOUN fcis-9453 152 19 detection detection NOUN fcis-9453 152 20 for for ADP fcis-9453 152 21 hot hot ADV fcis-9453 152 22 - - PUNCT fcis-9453 152 23 rolled roll VERB fcis-9453 152 24 strip strip NOUN fcis-9453 152 25 steel steel NOUN fcis-9453 152 26 . . PUNCT fcis-9453 152 27 " " PUNCT fcis-9453 153 1 computer computer NOUN fcis-9453 153 2 simulation simulation NOUN fcis-9453 153 3 . . PUNCT fcis-9453 154 1 [ [ X fcis-9453 154 2 21 21 NUM fcis-9453 154 3 ] ] X fcis-9453 154 4 wu wu PROPN fcis-9453 154 5 di di PROPN fcis-9453 154 6 , , PUNCT fcis-9453 154 7 et et PROPN fcis-9453 154 8 al al PROPN fcis-9453 154 9 . . PUNCT fcis-9453 155 1 " " PUNCT fcis-9453 155 2 improved improved ADJ fcis-9453 155 3 yolov5 yolov5 NOUN fcis-9453 155 4 - - PUNCT fcis-9453 155 5 based base VERB fcis-9453 155 6 surface surface NOUN fcis-9453 155 7 defect defect NOUN fcis-9453 155 8 detection detection NOUN fcis-9453 155 9 in in ADP fcis-9453 155 10 steel steel NOUN fcis-9453 155 11 . . PUNCT fcis-9453 155 12 " " PUNCT fcis-9453 155 13 journal journal PROPN fcis-9453 155 14 of of ADP fcis-9453 155 15 shaanxi shaanxi PROPN fcis-9453 155 16 university university PROPN fcis-9453 155 17 of of ADP fcis-9453 155 18 science science NOUN fcis-9453 155 19 and and CCONJ fcis-9453 155 20 technology technology NOUN fcis-9453 155 21 41.02(2023):162 41.02(2023):162 PROPN fcis-9453 155 22 - - PUNCT fcis-9453 155 23 169 169 NUM fcis-9453 155 24 . . PUNCT fcis-9453 156 1 doi doi NOUN fcis-9453 156 2 : : PUNCT fcis-9453 156 3 10 10 NUM fcis-9453 156 4 . . X fcis-9453 157 1 19481 19481 NUM fcis-9453 157 2 / / SYM fcis-9453 157 3 j. j. PROPN fcis-9453 157 4 cnki cnki PROPN fcis-9453 157 5 . . PUNCT fcis-9453 158 1 issn issn PROPN fcis-9453 158 2 2096 2096 NUM fcis-9453 158 3 - - SYM fcis-9453 158 4 398x.2023.02.009 398x.2023.02.009 NUM fcis-9453 158 5 . . PUNCT fcis-9453 159 1 [ [ X fcis-9453 159 2 22 22 NUM fcis-9453 159 3 ] ] X fcis-9453 159 4 zhang zhang PROPN fcis-9453 159 5 , , PUNCT fcis-9453 159 6 yang yang PROPN fcis-9453 159 7 , , PUNCT fcis-9453 159 8 liu liu PROPN fcis-9453 159 9 , , PUNCT fcis-9453 159 10 xiaofang xiaofang PROPN fcis-9453 159 11 , , PUNCT fcis-9453 159 12 and and CCONJ fcis-9453 159 13 li li PROPN fcis-9453 159 14 , , PUNCT fcis-9453 159 15 wenwei wenwei NOUN fcis-9453 159 16 . . PUNCT fcis-9453 159 17 " " PUNCT fcis-9453 160 1 an an DET fcis-9453 160 2 improved improved ADJ fcis-9453 160 3 yolov5n yolov5n NOUN fcis-9453 160 4 - - PUNCT fcis-9453 160 5 based base VERB fcis-9453 160 6 algorithm algorithm NOUN fcis-9453 160 7 for for ADP fcis-9453 160 8 strip strip PROPN fcis-9453 160 9 steel steel NOUN fcis-9453 160 10 surface surface NOUN fcis-9453 160 11 defect defect NOUN fcis-9453 160 12 detection detection NOUN fcis-9453 160 13 . . PUNCT fcis-9453 160 14 " " PUNCT fcis-9453 161 1 journal journal PROPN fcis-9453 161 2 of of ADP fcis-9453 161 3 sichuan sichuan PROPN fcis-9453 161 4 university university PROPN fcis-9453 161 5 of of ADP fcis-9453 161 6 light light NOUN fcis-9453 161 7 and and CCONJ fcis-9453 161 8 chemical chemical NOUN fcis-9453 161 9 technology technology NOUN fcis-9453 161 10 ( ( PUNCT fcis-9453 161 11 natural natural ADJ fcis-9453 161 12 science science NOUN fcis-9453 161 13 edition edition NOUN fcis-9453 161 14 ) ) PUNCT fcis-9453 161 15 35.05 35.05 NUM fcis-9453 161 16 ( ( PUNCT fcis-9453 161 17 2022 2022 NUM fcis-9453 161 18 ): ): PUNCT fcis-9453 161 19 60 60 NUM fcis-9453 161 20 - - SYM fcis-9453 161 21 67 67 NUM fcis-9453 161 22 . . PUNCT fcis-9453 162 1 [ [ X fcis-9453 162 2 23 23 NUM fcis-9453 162 3 ] ] X fcis-9453 162 4 ma ma PROPN fcis-9453 162 5 yanting yanting PROPN fcis-9453 162 6 , , PUNCT fcis-9453 162 7 et et PROPN fcis-9453 162 8 al al PROPN fcis-9453 162 9 . . PUNCT fcis-9453 163 1 " " PUNCT fcis-9453 163 2 improved improve VERB fcis-9453 163 3 surface surface NOUN fcis-9453 163 4 defect defect NOUN fcis-9453 163 5 detection detection NOUN fcis-9453 163 6 method method NOUN fcis-9453 163 7 for for ADP fcis-9453 163 8 strip strip NOUN fcis-9453 163 9 steel steel NOUN fcis-9453 163 10 with with ADP fcis-9453 163 11 yolov5 yolov5 NOUN fcis-9453 163 12 network network NOUN fcis-9453 163 13 . . PUNCT fcis-9453 163 14 " " PUNCT fcis-9453 164 1 journal journal NOUN fcis-9453 164 2 of of ADP fcis-9453 164 3 electronic electronic ADJ fcis-9453 164 4 measurement measurement NOUN fcis-9453 164 5 and and CCONJ fcis-9453 164 6 instrumentation instrumentation NOUN fcis-9453 164 7 36.08(2022):150 36.08(2022):150 PROPN fcis-9453 164 8 - - SYM fcis-9453 164 9 157 157 NUM fcis-9453 164 10 . . PUNCT fcis-9453 165 1 doi:10.13382 doi:10.13382 NOUN fcis-9453 165 2 / / SYM fcis-9453 165 3 j.jemi.b2205354 j.jemi.b2205354 PROPN fcis-9453 165 4 . . PUNCT fcis-9453 166 1 [ [ X fcis-9453 166 2 24 24 NUM fcis-9453 166 3 ] ] SYM fcis-9453 166 4 dong dong PROPN fcis-9453 166 5 , , PUNCT fcis-9453 166 6 yanhua yanhua PROPN fcis-9453 166 7 , , PUNCT fcis-9453 166 8 and and CCONJ fcis-9453 166 9 li li PROPN fcis-9453 166 10 , , PUNCT fcis-9453 166 11 jia'ao jia'ao PROPN fcis-9453 166 12 . . PUNCT fcis-9453 166 13 " " PUNCT fcis-9453 166 14 research research NOUN fcis-9453 166 15 on on ADP fcis-9453 166 16 improving improve VERB fcis-9453 166 17 yolov5s yolov5s PROPN fcis-9453 166 18 remote remote ADJ fcis-9453 166 19 sensing sense VERB fcis-9453 166 20 image image NOUN fcis-9453 166 21 recognition recognition NOUN fcis-9453 166 22 algorithm algorithm NOUN fcis-9453 166 23 . . PUNCT fcis-9453 166 24 " " PUNCT fcis-9453 167 1 journal journal PROPN fcis-9453 167 2 of of ADP fcis-9453 167 3 jilin jilin PROPN fcis-9453 167 4 normal normal PROPN fcis-9453 167 5 university university PROPN fcis-9453 167 6 ( ( PUNCT fcis-9453 167 7 natural natural ADJ fcis-9453 167 8 science science NOUN fcis-9453 167 9 edition edition NOUN fcis-9453 167 10 ) ) PUNCT fcis-9453 167 11 44.02 44.02 NUM fcis-9453 167 12 ( ( PUNCT fcis-9453 167 13 2023):117 2023):117 NUM fcis-9453 167 14 - - SYM fcis-9453 167 15 123 123 NUM fcis-9453 167 16 . . PUNCT fcis-9453 168 1 doi:10 doi:10 PROPN fcis-9453 168 2 . . PUNCT fcis-9453 169 1 16862/ 16862/ NUM fcis-9453 169 2 j. j. PROPN fcis-9453 169 3 cnki cnki PROPN fcis-9453 169 4 . . PUNCT fcis-9453 169 5 issn1674 issn1674 PROPN fcis-9453 169 6 - - PUNCT fcis-9453 169 7 3873 3873 NUM fcis-9453 169 8 . . PUNCT fcis-9453 169 9 2023 2023 NUM fcis-9453 169 10 . . PUNCT fcis-9453 170 1 02.017 02.017 NUM fcis-9453 170 2 . . PUNCT fcis-9453 171 1 [ [ X fcis-9453 171 2 25 25 NUM fcis-9453 171 3 ] ] PUNCT fcis-9453 171 4 chuanjun chuanjun X fcis-9453 171 5 zhu zhu PROPN fcis-9453 171 6 , , PUNCT fcis-9453 171 7 et et PROPN fcis-9453 171 8 al al PROPN fcis-9453 171 9 . . PUNCT fcis-9453 172 1 " " PUNCT fcis-9453 172 2 hybrid hybrid ADJ fcis-9453 172 3 defect defect NOUN fcis-9453 172 4 detection detection NOUN fcis-9453 172 5 model model NOUN fcis-9453 172 6 based base VERB fcis-9453 172 7 on on ADP fcis-9453 172 8 simam simam ADJ fcis-9453 172 9 module module NOUN fcis-9453 172 10 and and CCONJ fcis-9453 172 11 resnet34 resnet34 NOUN fcis-9453 172 12 network network NOUN fcis-9453 172 13 . . PUNCT fcis-9453 172 14 " " PUNCT fcis-9453 173 1 modern modern ADJ fcis-9453 173 2 manufacturing manufacturing NOUN fcis-9453 173 3 engineering engineering NOUN fcis-9453 173 4 .02(2023):1 .02(2023):1 NOUN fcis-9453 173 5 - - PUNCT fcis-9453 173 6 9 9 NUM fcis-9453 173 7 . . PUNCT fcis-9453 174 1 doi:10.16731/ doi:10.16731/ PROPN fcis-9453 174 2 j. j. PROPN fcis-9453 174 3 cnki cnki PROPN fcis-9453 174 4 . . PUNCT fcis-9453 175 1 1671 1671 NUM fcis-9453 175 2 - - SYM fcis-9453 175 3 3133.2023.02.001 3133.2023.02.001 NUM fcis-9453 175 4 . . PUNCT