id	sid	tid	token	lemma	pos
ajst-2549	1	1	academic	academic	ADJ
ajst-2549	1	2	journal	journal	NOUN
ajst-2549	1	3	of	of	ADP
ajst-2549	1	4	science	science	NOUN
ajst-2549	1	5	and	and	CCONJ
ajst-2549	1	6	technology	technology	NOUN
ajst-2549	1	7	issn	issn	NOUN
ajst-2549	1	8	:	:	PUNCT
ajst-2549	1	9	2771	2771	NUM
ajst-2549	1	10	-	-	SYM
ajst-2549	1	11	3032	3032	NUM
ajst-2549	1	12	|	|	NOUN
ajst-2549	1	13	vol	vol	NOUN
ajst-2549	1	14	.	.	PROPN
ajst-2549	2	1	3	3	NUM
ajst-2549	2	2	,	,	PUNCT
ajst-2549	2	3	no	no	INTJ
ajst-2549	2	4	.	.	NOUN
ajst-2549	2	5	3	3	NUM
ajst-2549	2	6	,	,	PUNCT
ajst-2549	2	7	2022	2022	NUM
ajst-2549	2	8	64	64	NUM
ajst-2549	2	9	improved	improved	ADJ
ajst-2549	2	10	angle	angle	NOUN
ajst-2549	2	11	steel	steel	NOUN
ajst-2549	2	12	character	character	NOUN
ajst-2549	2	13	detection	detection	NOUN
ajst-2549	2	14	algorithm	algorithm	NOUN
ajst-2549	2	15	based	base	VERB
ajst-2549	2	16	on	on	ADP
ajst-2549	2	17	dbnet	dbnet	PROPN
ajst-2549	2	18	xun	xun	PROPN
ajst-2549	2	19	he	he	PRON
ajst-2549	2	20	,	,	PUNCT
ajst-2549	2	21	xiaofang	xiaofang	PROPN
ajst-2549	2	22	liu	liu	PROPN
ajst-2549	2	23	school	school	PROPN
ajst-2549	2	24	of	of	ADP
ajst-2549	2	25	computer	computer	NOUN
ajst-2549	2	26	science	science	NOUN
ajst-2549	2	27	and	and	CCONJ
ajst-2549	2	28	engineering	engineering	NOUN
ajst-2549	2	29	,	,	PUNCT
ajst-2549	2	30	sichuan	sichuan	PROPN
ajst-2549	2	31	university	university	PROPN
ajst-2549	2	32	of	of	ADP
ajst-2549	2	33	science	science	PROPN
ajst-2549	2	34	&	&	CCONJ
ajst-2549	2	35	engineering	engineering	PROPN
ajst-2549	2	36	,	,	PUNCT
ajst-2549	2	37	yibin	yibin	PROPN
ajst-2549	2	38	644000	644000	NUM
ajst-2549	2	39	,	,	PUNCT
ajst-2549	2	40	china	china	PROPN
ajst-2549	2	41	abstract	abstract	NOUN
ajst-2549	2	42	:	:	PUNCT
ajst-2549	2	43	aiming	aim	VERB
ajst-2549	2	44	at	at	ADP
ajst-2549	2	45	the	the	DET
ajst-2549	2	46	problems	problem	NOUN
ajst-2549	2	47	of	of	ADP
ajst-2549	2	48	false	false	ADJ
ajst-2549	2	49	detection	detection	NOUN
ajst-2549	2	50	and	and	CCONJ
ajst-2549	2	51	missed	miss	VERB
ajst-2549	2	52	detection	detection	NOUN
ajst-2549	2	53	in	in	ADP
ajst-2549	2	54	manual	manual	ADJ
ajst-2549	2	55	recording	recording	NOUN
ajst-2549	2	56	of	of	ADP
ajst-2549	2	57	industrial	industrial	ADJ
ajst-2549	2	58	angle	angle	NOUN
ajst-2549	2	59	steel	steel	NOUN
ajst-2549	2	60	characters	character	NOUN
ajst-2549	2	61	,	,	PUNCT
ajst-2549	2	62	an	an	DET
ajst-2549	2	63	improved	improved	ADJ
ajst-2549	2	64	angle	angle	NOUN
ajst-2549	2	65	steel	steel	NOUN
ajst-2549	2	66	character	character	NOUN
ajst-2549	2	67	detection	detection	NOUN
ajst-2549	2	68	algorithm	algorithm	NOUN
ajst-2549	2	69	based	base	VERB
ajst-2549	2	70	on	on	ADP
ajst-2549	2	71	dbnet	dbnet	NOUN
ajst-2549	2	72	is	be	AUX
ajst-2549	2	73	proposed	propose	VERB
ajst-2549	2	74	.	.	PUNCT
ajst-2549	3	1	first	first	ADV
ajst-2549	3	2	,	,	PUNCT
ajst-2549	3	3	the	the	DET
ajst-2549	3	4	angle	angle	NOUN
ajst-2549	3	5	steel	steel	NOUN
ajst-2549	3	6	character	character	NOUN
ajst-2549	3	7	dataset	dataset	NOUN
ajst-2549	3	8	is	be	AUX
ajst-2549	3	9	collected	collect	VERB
ajst-2549	3	10	.	.	PUNCT
ajst-2549	4	1	the	the	DET
ajst-2549	4	2	algorithm	algorithm	NOUN
ajst-2549	4	3	uses	use	VERB
ajst-2549	4	4	mobilenetv3	mobilenetv3	PROPN
ajst-2549	4	5	as	as	ADP
ajst-2549	4	6	the	the	DET
ajst-2549	4	7	backbone	backbone	NOUN
ajst-2549	4	8	network	network	NOUN
ajst-2549	4	9	,	,	PUNCT
ajst-2549	4	10	and	and	CCONJ
ajst-2549	4	11	replaces	replace	VERB
ajst-2549	4	12	the	the	DET
ajst-2549	4	13	attention	attention	NOUN
ajst-2549	4	14	mechanism	mechanism	NOUN
ajst-2549	4	15	in	in	ADP
ajst-2549	4	16	the	the	DET
ajst-2549	4	17	backbone	backbone	NOUN
ajst-2549	4	18	network	network	NOUN
ajst-2549	4	19	with	with	ADP
ajst-2549	4	20	a	a	DET
ajst-2549	4	21	new	new	ADJ
ajst-2549	4	22	coordinate	coordinate	NOUN
ajst-2549	4	23	attention	attention	NOUN
ajst-2549	4	24	mechanism	mechanism	NOUN
ajst-2549	4	25	to	to	PART
ajst-2549	4	26	obtain	obtain	VERB
ajst-2549	4	27	larger	large	ADJ
ajst-2549	4	28	regional	regional	ADJ
ajst-2549	4	29	information	information	NOUN
ajst-2549	4	30	and	and	CCONJ
ajst-2549	4	31	enhance	enhance	VERB
ajst-2549	4	32	the	the	DET
ajst-2549	4	33	ability	ability	NOUN
ajst-2549	4	34	to	to	PART
ajst-2549	4	35	extract	extract	VERB
ajst-2549	4	36	complex	complex	ADJ
ajst-2549	4	37	features	feature	NOUN
ajst-2549	4	38	;	;	PUNCT
ajst-2549	4	39	secondly	secondly	ADV
ajst-2549	4	40	,	,	PUNCT
ajst-2549	4	41	weighted	weight	VERB
ajst-2549	4	42	bidirectional	bidirectional	ADJ
ajst-2549	4	43	the	the	DET
ajst-2549	4	44	feature	feature	NOUN
ajst-2549	4	45	pyramid	pyramid	NOUN
ajst-2549	4	46	network	network	NOUN
ajst-2549	4	47	improves	improve	VERB
ajst-2549	4	48	the	the	DET
ajst-2549	4	49	efficiency	efficiency	NOUN
ajst-2549	4	50	of	of	ADP
ajst-2549	4	51	multi	multi	ADJ
ajst-2549	4	52	-	-	ADJ
ajst-2549	4	53	scale	scale	ADJ
ajst-2549	4	54	fusion	fusion	NOUN
ajst-2549	4	55	and	and	CCONJ
ajst-2549	4	56	enhances	enhance	VERB
ajst-2549	4	57	the	the	DET
ajst-2549	4	58	detection	detection	NOUN
ajst-2549	4	59	ability	ability	NOUN
ajst-2549	4	60	of	of	ADP
ajst-2549	4	61	small	small	ADJ
ajst-2549	4	62	features	feature	NOUN
ajst-2549	4	63	;	;	PUNCT
ajst-2549	4	64	finally	finally	ADV
ajst-2549	4	65	,	,	PUNCT
ajst-2549	4	66	a	a	DET
ajst-2549	4	67	new	new	ADJ
ajst-2549	4	68	data	data	NOUN
ajst-2549	4	69	augmentation	augmentation	NOUN
ajst-2549	4	70	method	method	NOUN
ajst-2549	4	71	and	and	CCONJ
ajst-2549	4	72	learning	learn	VERB
ajst-2549	4	73	rate	rate	NOUN
ajst-2549	4	74	optimization	optimization	NOUN
ajst-2549	4	75	strategy	strategy	NOUN
ajst-2549	4	76	are	be	AUX
ajst-2549	4	77	used	use	VERB
ajst-2549	4	78	to	to	PART
ajst-2549	4	79	enhance	enhance	VERB
ajst-2549	4	80	the	the	DET
ajst-2549	4	81	robustness	robustness	NOUN
ajst-2549	4	82	and	and	CCONJ
ajst-2549	4	83	convergence	convergence	NOUN
ajst-2549	4	84	accuracy	accuracy	NOUN
ajst-2549	4	85	of	of	ADP
ajst-2549	4	86	the	the	DET
ajst-2549	4	87	model	model	NOUN
ajst-2549	4	88	.	.	PUNCT
ajst-2549	5	1	the	the	DET
ajst-2549	5	2	experimental	experimental	ADJ
ajst-2549	5	3	results	result	NOUN
ajst-2549	5	4	show	show	VERB
ajst-2549	5	5	that	that	SCONJ
ajst-2549	5	6	,	,	PUNCT
ajst-2549	5	7	compared	compare	VERB
ajst-2549	5	8	with	with	ADP
ajst-2549	5	9	the	the	DET
ajst-2549	5	10	original	original	ADJ
ajst-2549	5	11	dbnet	dbnet	NOUN
ajst-2549	5	12	algorithm	algorithm	NOUN
ajst-2549	5	13	,	,	PUNCT
ajst-2549	5	14	the	the	DET
ajst-2549	5	15	detection	detection	NOUN
ajst-2549	5	16	index	index	NOUN
ajst-2549	5	17	of	of	ADP
ajst-2549	5	18	the	the	DET
ajst-2549	5	19	improved	improved	ADJ
ajst-2549	5	20	dbnet	dbnet	NOUN
ajst-2549	5	21	algorithm	algorithm	NOUN
ajst-2549	5	22	is	be	AUX
ajst-2549	5	23	increased	increase	VERB
ajst-2549	5	24	from	from	ADP
ajst-2549	5	25	81.42	81.42	NUM
ajst-2549	5	26	%	%	NOUN
ajst-2549	5	27	to	to	ADP
ajst-2549	5	28	99.06	99.06	NUM
ajst-2549	5	29	%	%	NOUN
ajst-2549	5	30	on	on	ADP
ajst-2549	5	31	the	the	DET
ajst-2549	5	32	angle	angle	NOUN
ajst-2549	5	33	steel	steel	NOUN
ajst-2549	5	34	character	character	NOUN
ajst-2549	5	35	data	datum	NOUN
ajst-2549	5	36	set	set	VERB
ajst-2549	5	37	,	,	PUNCT
ajst-2549	5	38	which	which	PRON
ajst-2549	5	39	meets	meet	VERB
ajst-2549	5	40	the	the	DET
ajst-2549	5	41	needs	need	NOUN
ajst-2549	5	42	of	of	ADP
ajst-2549	5	43	angle	angle	NOUN
ajst-2549	5	44	steel	steel	NOUN
ajst-2549	5	45	character	character	NOUN
ajst-2549	5	46	detection	detection	NOUN
ajst-2549	5	47	in	in	ADP
ajst-2549	5	48	the	the	DET
ajst-2549	5	49	industrial	industrial	ADJ
ajst-2549	5	50	environment	environment	NOUN
ajst-2549	5	51	.	.	PUNCT
ajst-2549	6	1	keywords	keyword	NOUN
ajst-2549	6	2	:	:	PUNCT
ajst-2549	6	3	angle	angle	NOUN
ajst-2549	6	4	steel	steel	NOUN
ajst-2549	6	5	characters	character	NOUN
ajst-2549	6	6	,	,	PUNCT
ajst-2549	6	7	attention	attention	NOUN
ajst-2549	6	8	mechanisms	mechanism	NOUN
ajst-2549	6	9	,	,	PUNCT
ajst-2549	6	10	multiscale	multiscale	ADJ
ajst-2549	6	11	-	-	PUNCT
ajst-2549	6	12	scale	scale	NOUN
ajst-2549	6	13	feature	feature	NOUN
ajst-2549	6	14	fusion	fusion	NOUN
ajst-2549	6	15	,	,	PUNCT
ajst-2549	6	16	data	datum	NOUN
ajst-2549	6	17	augmentation	augmentation	NOUN
ajst-2549	6	18	.	.	PUNCT
ajst-2549	7	1	1	1	X
ajst-2549	7	2	.	.	X
ajst-2549	7	3	introduction	introduction	NOUN
ajst-2549	7	4	angle	angle	NOUN
ajst-2549	7	5	steel	steel	NOUN
ajst-2549	7	6	is	be	AUX
ajst-2549	7	7	one	one	NUM
ajst-2549	7	8	of	of	ADP
ajst-2549	7	9	the	the	DET
ajst-2549	7	10	very	very	ADV
ajst-2549	7	11	important	important	ADJ
ajst-2549	7	12	materials	material	NOUN
ajst-2549	7	13	in	in	ADP
ajst-2549	7	14	the	the	DET
ajst-2549	7	15	construction	construction	NOUN
ajst-2549	7	16	of	of	ADP
ajst-2549	7	17	the	the	DET
ajst-2549	7	18	national	national	ADJ
ajst-2549	7	19	power	power	NOUN
ajst-2549	7	20	grid	grid	NOUN
ajst-2549	7	21	,	,	PUNCT
ajst-2549	7	22	and	and	CCONJ
ajst-2549	7	23	is	be	AUX
ajst-2549	7	24	widely	widely	ADV
ajst-2549	7	25	used	use	VERB
ajst-2549	7	26	in	in	ADP
ajst-2549	7	27	the	the	DET
ajst-2549	7	28	structure	structure	NOUN
ajst-2549	7	29	of	of	ADP
ajst-2549	7	30	transmission	transmission	NOUN
ajst-2549	7	31	line	line	NOUN
ajst-2549	7	32	towers	tower	NOUN
ajst-2549	8	1	[	[	X
ajst-2549	8	2	1	1	NUM
ajst-2549	8	3	]	]	PUNCT
ajst-2549	8	4	.	.	PUNCT
ajst-2549	9	1	the	the	DET
ajst-2549	9	2	steel	steel	NOUN
ajst-2549	9	3	stamp	stamp	NOUN
ajst-2549	9	4	character	character	NOUN
ajst-2549	9	5	of	of	ADP
ajst-2549	9	6	the	the	DET
ajst-2549	9	7	angle	angle	NOUN
ajst-2549	9	8	steel	steel	NOUN
ajst-2549	9	9	is	be	AUX
ajst-2549	9	10	the	the	DET
ajst-2549	9	11	identity	identity	NOUN
ajst-2549	9	12	mark	mark	NOUN
ajst-2549	9	13	of	of	ADP
ajst-2549	9	14	the	the	DET
ajst-2549	9	15	angle	angle	NOUN
ajst-2549	9	16	steel	steel	NOUN
ajst-2549	9	17	,	,	PUNCT
ajst-2549	9	18	and	and	CCONJ
ajst-2549	9	19	its	its	PRON
ajst-2549	9	20	main	main	ADJ
ajst-2549	9	21	function	function	NOUN
ajst-2549	9	22	is	be	AUX
ajst-2549	9	23	to	to	PART
ajst-2549	9	24	distinguish	distinguish	VERB
ajst-2549	9	25	the	the	DET
ajst-2549	9	26	model	model	NOUN
ajst-2549	9	27	of	of	ADP
ajst-2549	9	28	the	the	DET
ajst-2549	9	29	angle	angle	NOUN
ajst-2549	9	30	steel	steel	NOUN
ajst-2549	9	31	material	material	NOUN
ajst-2549	9	32	.	.	PUNCT
ajst-2549	10	1	a	a	DET
ajst-2549	10	2	complete	complete	ADJ
ajst-2549	10	3	transmission	transmission	NOUN
ajst-2549	10	4	tower	tower	NOUN
ajst-2549	10	5	needs	need	VERB
ajst-2549	10	6	to	to	PART
ajst-2549	10	7	be	be	AUX
ajst-2549	10	8	composed	compose	VERB
ajst-2549	10	9	of	of	ADP
ajst-2549	10	10	different	different	ADJ
ajst-2549	10	11	types	type	NOUN
ajst-2549	10	12	of	of	ADP
ajst-2549	10	13	angle	angle	NOUN
ajst-2549	10	14	steel	steel	NOUN
ajst-2549	10	15	,	,	PUNCT
ajst-2549	10	16	and	and	CCONJ
ajst-2549	10	17	the	the	DET
ajst-2549	10	18	required	require	VERB
ajst-2549	10	19	quantity	quantity	NOUN
ajst-2549	10	20	of	of	ADP
ajst-2549	10	21	different	different	ADJ
ajst-2549	10	22	types	type	NOUN
ajst-2549	10	23	of	of	ADP
ajst-2549	10	24	angle	angle	NOUN
ajst-2549	10	25	steel	steel	NOUN
ajst-2549	10	26	is	be	AUX
ajst-2549	10	27	also	also	ADV
ajst-2549	10	28	different	different	ADJ
ajst-2549	10	29	,	,	PUNCT
ajst-2549	10	30	and	and	CCONJ
ajst-2549	10	31	in	in	ADP
ajst-2549	10	32	the	the	DET
ajst-2549	10	33	process	process	NOUN
ajst-2549	10	34	,	,	PUNCT
ajst-2549	10	35	the	the	DET
ajst-2549	10	36	operator	operator	NOUN
ajst-2549	10	37	needs	need	VERB
ajst-2549	10	38	to	to	PART
ajst-2549	10	39	manually	manually	ADV
ajst-2549	10	40	identify	identify	VERB
ajst-2549	10	41	and	and	CCONJ
ajst-2549	10	42	record	record	VERB
ajst-2549	10	43	and	and	CCONJ
ajst-2549	10	44	combine	combine	VERB
ajst-2549	10	45	the	the	DET
ajst-2549	10	46	characters	character	NOUN
ajst-2549	10	47	,	,	PUNCT
ajst-2549	10	48	which	which	PRON
ajst-2549	10	49	is	be	AUX
ajst-2549	10	50	easy	easy	ADJ
ajst-2549	10	51	to	to	PART
ajst-2549	10	52	cause	cause	VERB
ajst-2549	10	53	mistakes	mistake	NOUN
ajst-2549	10	54	.	.	PUNCT
ajst-2549	11	1	once	once	ADV
ajst-2549	11	2	a	a	DET
ajst-2549	11	3	mistake	mistake	NOUN
ajst-2549	11	4	is	be	AUX
ajst-2549	11	5	made	make	VERB
ajst-2549	11	6	,	,	PUNCT
ajst-2549	11	7	it	it	PRON
ajst-2549	11	8	will	will	AUX
ajst-2549	11	9	cost	cost	VERB
ajst-2549	11	10	a	a	DET
ajst-2549	11	11	lot	lot	NOUN
ajst-2549	11	12	of	of	ADP
ajst-2549	11	13	manpower	manpower	NOUN
ajst-2549	11	14	and	and	CCONJ
ajst-2549	11	15	material	material	NOUN
ajst-2549	11	16	resources	resource	NOUN
ajst-2549	11	17	to	to	PART
ajst-2549	11	18	re	re	VERB
ajst-2549	11	19	-	-	VERB
ajst-2549	11	20	match	match	VERB
ajst-2549	11	21	the	the	DET
ajst-2549	11	22	correct	correct	ADJ
ajst-2549	11	23	number	number	NOUN
ajst-2549	11	24	of	of	ADP
ajst-2549	11	25	models	model	NOUN
ajst-2549	11	26	.	.	PUNCT
ajst-2549	12	1	this	this	DET
ajst-2549	12	2	low	low	ADJ
ajst-2549	12	3	-	-	PUNCT
ajst-2549	12	4	efficiency	efficiency	NOUN
ajst-2549	12	5	and	and	CCONJ
ajst-2549	12	6	high	high	ADJ
ajst-2549	12	7	-	-	PUNCT
ajst-2549	12	8	cost	cost	NOUN
ajst-2549	12	9	method	method	NOUN
ajst-2549	12	10	is	be	AUX
ajst-2549	12	11	very	very	ADV
ajst-2549	12	12	unfavorable	unfavorable	ADJ
ajst-2549	12	13	for	for	ADP
ajst-2549	12	14	the	the	DET
ajst-2549	12	15	development	development	NOUN
ajst-2549	12	16	of	of	ADP
ajst-2549	12	17	current	current	ADJ
ajst-2549	12	18	production	production	NOUN
ajst-2549	12	19	enterprises	enterprise	NOUN
ajst-2549	12	20	.	.	PUNCT
ajst-2549	13	1	with	with	ADP
ajst-2549	13	2	the	the	DET
ajst-2549	13	3	rapid	rapid	ADJ
ajst-2549	13	4	development	development	NOUN
ajst-2549	13	5	of	of	ADP
ajst-2549	13	6	image	image	NOUN
ajst-2549	13	7	recognition	recognition	NOUN
ajst-2549	13	8	technology	technology	NOUN
ajst-2549	13	9	,	,	PUNCT
ajst-2549	13	10	text	text	NOUN
ajst-2549	13	11	detection	detection	NOUN
ajst-2549	13	12	algorithms	algorithm	NOUN
ajst-2549	13	13	based	base	VERB
ajst-2549	13	14	on	on	ADP
ajst-2549	13	15	deep	deep	ADJ
ajst-2549	13	16	learning	learning	NOUN
ajst-2549	13	17	have	have	AUX
ajst-2549	13	18	been	be	AUX
ajst-2549	13	19	widely	widely	ADV
ajst-2549	13	20	proposed	propose	VERB
ajst-2549	13	21	.	.	PUNCT
ajst-2549	14	1	the	the	DET
ajst-2549	14	2	realization	realization	NOUN
ajst-2549	14	3	principle	principle	NOUN
ajst-2549	14	4	of	of	ADP
ajst-2549	14	5	the	the	DET
ajst-2549	14	6	text	text	NOUN
ajst-2549	14	7	detection	detection	NOUN
ajst-2549	14	8	algorithm	algorithm	NOUN
ajst-2549	14	9	is	be	AUX
ajst-2549	14	10	to	to	PART
ajst-2549	14	11	first	first	ADV
ajst-2549	14	12	use	use	VERB
ajst-2549	14	13	the	the	DET
ajst-2549	14	14	target	target	NOUN
ajst-2549	14	15	detection	detection	NOUN
ajst-2549	14	16	framework	framework	NOUN
ajst-2549	14	17	to	to	PART
ajst-2549	14	18	detect	detect	VERB
ajst-2549	14	19	part	part	NOUN
ajst-2549	14	20	of	of	ADP
ajst-2549	14	21	the	the	DET
ajst-2549	14	22	text	text	NOUN
ajst-2549	14	23	area	area	NOUN
ajst-2549	14	24	,	,	PUNCT
ajst-2549	14	25	and	and	CCONJ
ajst-2549	14	26	then	then	ADV
ajst-2549	14	27	use	use	VERB
ajst-2549	14	28	the	the	DET
ajst-2549	14	29	post	post	ADJ
ajst-2549	14	30	-	-	ADJ
ajst-2549	14	31	processing	processing	ADJ
ajst-2549	14	32	method	method	NOUN
ajst-2549	14	33	to	to	PART
ajst-2549	14	34	obtain	obtain	VERB
ajst-2549	14	35	the	the	DET
ajst-2549	14	36	complete	complete	ADJ
ajst-2549	14	37	text	text	NOUN
ajst-2549	14	38	area	area	NOUN
ajst-2549	14	39	[	[	X
ajst-2549	14	40	2	2	NUM
ajst-2549	14	41	]	]	PUNCT
ajst-2549	14	42	.	.	PUNCT
ajst-2549	15	1	text	text	NOUN
ajst-2549	15	2	detection	detection	NOUN
ajst-2549	15	3	algorithms	algorithm	NOUN
ajst-2549	15	4	have	have	AUX
ajst-2549	15	5	been	be	AUX
ajst-2549	15	6	widely	widely	ADV
ajst-2549	15	7	used	use	VERB
ajst-2549	15	8	in	in	ADP
ajst-2549	15	9	finance	finance	NOUN
ajst-2549	15	10	,	,	PUNCT
ajst-2549	15	11	transportation	transportation	NOUN
ajst-2549	15	12	,	,	PUNCT
ajst-2549	15	13	logistics	logistic	NOUN
ajst-2549	15	14	and	and	CCONJ
ajst-2549	15	15	many	many	ADJ
ajst-2549	15	16	other	other	ADJ
ajst-2549	15	17	fields	field	NOUN
ajst-2549	15	18	,	,	PUNCT
ajst-2549	15	19	but	but	CCONJ
ajst-2549	15	20	in	in	ADP
ajst-2549	15	21	the	the	DET
ajst-2549	15	22	industrial	industrial	ADJ
ajst-2549	15	23	field	field	NOUN
ajst-2549	15	24	,	,	PUNCT
ajst-2549	15	25	they	they	PRON
ajst-2549	15	26	have	have	AUX
ajst-2549	15	27	not	not	PART
ajst-2549	15	28	been	be	AUX
ajst-2549	15	29	fully	fully	ADV
ajst-2549	15	30	applied	apply	VERB
ajst-2549	15	31	.	.	PUNCT
ajst-2549	16	1	the	the	DET
ajst-2549	16	2	main	main	ADJ
ajst-2549	16	3	reason	reason	NOUN
ajst-2549	16	4	is	be	AUX
ajst-2549	16	5	that	that	SCONJ
ajst-2549	16	6	there	there	PRON
ajst-2549	16	7	is	be	VERB
ajst-2549	16	8	no	no	DET
ajst-2549	16	9	public	public	ADJ
ajst-2549	16	10	dataset	dataset	NOUN
ajst-2549	16	11	of	of	ADP
ajst-2549	16	12	industrial	industrial	ADJ
ajst-2549	16	13	characters	character	NOUN
ajst-2549	16	14	,	,	PUNCT
ajst-2549	16	15	the	the	DET
ajst-2549	16	16	characters	character	NOUN
ajst-2549	16	17	are	be	AUX
ajst-2549	16	18	diverse	diverse	ADJ
ajst-2549	16	19	in	in	ADP
ajst-2549	16	20	shape	shape	NOUN
ajst-2549	16	21	,	,	PUNCT
ajst-2549	16	22	and	and	CCONJ
ajst-2549	16	23	the	the	DET
ajst-2549	16	24	background	background	NOUN
ajst-2549	16	25	is	be	AUX
ajst-2549	16	26	complex	complex	ADJ
ajst-2549	17	1	[	[	X
ajst-2549	17	2	3	3	NUM
ajst-2549	17	3	]	]	PUNCT
ajst-2549	17	4	.	.	PUNCT
ajst-2549	18	1	based	base	VERB
ajst-2549	18	2	on	on	ADP
ajst-2549	18	3	the	the	DET
ajst-2549	18	4	above	above	ADJ
ajst-2549	18	5	problems	problem	NOUN
ajst-2549	18	6	,	,	PUNCT
ajst-2549	18	7	this	this	DET
ajst-2549	18	8	paper	paper	NOUN
ajst-2549	18	9	collects	collect	VERB
ajst-2549	18	10	the	the	DET
ajst-2549	18	11	angle	angle	NOUN
ajst-2549	18	12	steel	steel	NOUN
ajst-2549	18	13	character	character	NOUN
ajst-2549	18	14	data	datum	NOUN
ajst-2549	18	15	set	set	VERB
ajst-2549	18	16	,	,	PUNCT
ajst-2549	18	17	combines	combine	VERB
ajst-2549	18	18	the	the	DET
ajst-2549	18	19	currently	currently	ADV
ajst-2549	18	20	widely	widely	ADV
ajst-2549	18	21	used	use	VERB
ajst-2549	18	22	dbnet	dbnet	NOUN
ajst-2549	18	23	to	to	PART
ajst-2549	18	24	study	study	VERB
ajst-2549	18	25	the	the	DET
ajst-2549	18	26	angle	angle	NOUN
ajst-2549	18	27	steel	steel	NOUN
ajst-2549	18	28	characters	character	NOUN
ajst-2549	18	29	,	,	PUNCT
ajst-2549	18	30	uses	use	VERB
ajst-2549	18	31	mobilenetv3[4	mobilenetv3[4	PROPN
ajst-2549	18	32	]	]	PUNCT
ajst-2549	18	33	as	as	ADP
ajst-2549	18	34	the	the	DET
ajst-2549	18	35	feature	feature	NOUN
ajst-2549	18	36	extraction	extraction	NOUN
ajst-2549	18	37	network	network	NOUN
ajst-2549	18	38	,	,	PUNCT
ajst-2549	18	39	and	and	CCONJ
ajst-2549	18	40	introduces	introduce	NOUN
ajst-2549	18	41	coordinate	coordinate	VERB
ajst-2549	18	42	attention	attention	NOUN
ajst-2549	18	43	(	(	PUNCT
ajst-2549	18	44	ca	ca	NOUN
ajst-2549	18	45	)	)	PUNCT
ajst-2549	19	1	[	[	X
ajst-2549	19	2	5	5	NUM
ajst-2549	19	3	]	]	PUNCT
ajst-2549	19	4	to	to	PART
ajst-2549	19	5	replace	replace	VERB
ajst-2549	19	6	the	the	DET
ajst-2549	19	7	original	original	ADJ
ajst-2549	19	8	attention	attention	NOUN
ajst-2549	19	9	mechanism	mechanism	NOUN
ajst-2549	19	10	,	,	PUNCT
ajst-2549	19	11	improve	improve	VERB
ajst-2549	19	12	multi	multi	ADJ
ajst-2549	19	13	-	-	ADJ
ajst-2549	19	14	scale	scale	ADJ
ajst-2549	19	15	feature	feature	NOUN
ajst-2549	19	16	fusion	fusion	NOUN
ajst-2549	19	17	,	,	PUNCT
ajst-2549	19	18	and	and	CCONJ
ajst-2549	19	19	finally	finally	ADV
ajst-2549	19	20	use	use	VERB
ajst-2549	19	21	data	datum	NOUN
ajst-2549	19	22	augmentation	augmentation	NOUN
ajst-2549	19	23	and	and	CCONJ
ajst-2549	19	24	learning	learn	VERB
ajst-2549	19	25	rate	rate	NOUN
ajst-2549	19	26	strategy	strategy	NOUN
ajst-2549	19	27	to	to	PART
ajst-2549	19	28	optimize	optimize	VERB
ajst-2549	19	29	the	the	DET
ajst-2549	19	30	algorithm	algorithm	NOUN
ajst-2549	19	31	.	.	PUNCT
ajst-2549	20	1	in	in	ADP
ajst-2549	20	2	order	order	NOUN
ajst-2549	20	3	to	to	PART
ajst-2549	20	4	prove	prove	VERB
ajst-2549	20	5	the	the	DET
ajst-2549	20	6	effectiveness	effectiveness	NOUN
ajst-2549	20	7	of	of	ADP
ajst-2549	20	8	the	the	DET
ajst-2549	20	9	above	above	ADJ
ajst-2549	20	10	improvements	improvement	NOUN
ajst-2549	20	11	,	,	PUNCT
ajst-2549	20	12	the	the	DET
ajst-2549	20	13	experiment	experiment	NOUN
ajst-2549	20	14	compares	compare	VERB
ajst-2549	20	15	the	the	DET
ajst-2549	20	16	improvement	improvement	NOUN
ajst-2549	20	17	points	point	NOUN
ajst-2549	20	18	through	through	ADP
ajst-2549	20	19	the	the	DET
ajst-2549	20	20	ablation	ablation	NOUN
ajst-2549	20	21	method	method	NOUN
ajst-2549	20	22	,	,	PUNCT
ajst-2549	20	23	and	and	CCONJ
ajst-2549	20	24	compares	compare	VERB
ajst-2549	20	25	the	the	DET
ajst-2549	20	26	common	common	ADJ
ajst-2549	20	27	detection	detection	NOUN
ajst-2549	20	28	algorithms	algorithm	NOUN
ajst-2549	20	29	also	also	ADV
ajst-2549	20	30	based	base	VERB
ajst-2549	20	31	on	on	ADP
ajst-2549	20	32	mobilenetv3	mobilenetv3	PROPN
ajst-2549	20	33	.	.	PUNCT
ajst-2549	21	1	the	the	DET
ajst-2549	21	2	final	final	ADJ
ajst-2549	21	3	index	index	NOUN
ajst-2549	21	4	and	and	CCONJ
ajst-2549	21	5	effect	effect	NOUN
ajst-2549	21	6	observation	observation	NOUN
ajst-2549	21	7	show	show	VERB
ajst-2549	21	8	that	that	SCONJ
ajst-2549	21	9	the	the	DET
ajst-2549	21	10	proposed	propose	VERB
ajst-2549	21	11	improved	improve	VERB
ajst-2549	21	12	method	method	NOUN
ajst-2549	21	13	is	be	AUX
ajst-2549	21	14	significantly	significantly	ADV
ajst-2549	21	15	improved	improve	VERB
ajst-2549	21	16	compared	compare	VERB
ajst-2549	21	17	with	with	ADP
ajst-2549	21	18	the	the	DET
ajst-2549	21	19	original	original	ADJ
ajst-2549	21	20	algorithm	algorithm	NOUN
ajst-2549	21	21	,	,	PUNCT
ajst-2549	21	22	which	which	PRON
ajst-2549	21	23	meets	meet	VERB
ajst-2549	21	24	the	the	DET
ajst-2549	21	25	requirements	requirement	NOUN
ajst-2549	21	26	of	of	ADP
ajst-2549	21	27	industrial	industrial	ADJ
ajst-2549	21	28	angle	angle	NOUN
ajst-2549	21	29	steel	steel	NOUN
ajst-2549	21	30	character	character	NOUN
ajst-2549	21	31	detection	detection	NOUN
ajst-2549	21	32	accuracy	accuracy	NOUN
ajst-2549	21	33	and	and	CCONJ
ajst-2549	21	34	provides	provide	VERB
ajst-2549	21	35	a	a	DET
ajst-2549	21	36	reference	reference	NOUN
ajst-2549	21	37	for	for	ADP
ajst-2549	21	38	industrial	industrial	ADJ
ajst-2549	21	39	angle	angle	NOUN
ajst-2549	21	40	steel	steel	NOUN
ajst-2549	21	41	character	character	NOUN
ajst-2549	21	42	detection	detection	NOUN
ajst-2549	21	43	methods	method	NOUN
ajst-2549	21	44	.	.	PUNCT
ajst-2549	22	1	2	2	X
ajst-2549	22	2	.	.	X
ajst-2549	22	3	dbnet	dbnet	NOUN
ajst-2549	22	4	network	network	NOUN
ajst-2549	22	5	architecture	architecture	NOUN
ajst-2549	22	6	at	at	ADP
ajst-2549	22	7	present	present	ADJ
ajst-2549	22	8	,	,	PUNCT
ajst-2549	22	9	there	there	PRON
ajst-2549	22	10	are	be	VERB
ajst-2549	22	11	three	three	NUM
ajst-2549	22	12	main	main	ADJ
ajst-2549	22	13	versions	version	NOUN
ajst-2549	22	14	of	of	ADP
ajst-2549	22	15	dbnet	dbnet	NOUN
ajst-2549	22	16	algorithm	algorithm	NOUN
ajst-2549	22	17	,	,	PUNCT
ajst-2549	22	18	dbnet_res50	dbnet_res50	ADJ
ajst-2549	22	19	,	,	PUNCT
ajst-2549	22	20	dbnet_res18	dbnet_res18	ADJ
ajst-2549	22	21	and	and	CCONJ
ajst-2549	22	22	dbnet_mov3	dbnet_mov3	ADJ
ajst-2549	22	23	.	.	PUNCT
ajst-2549	23	1	the	the	DET
ajst-2549	23	2	main	main	ADJ
ajst-2549	23	3	difference	difference	NOUN
ajst-2549	23	4	between	between	ADP
ajst-2549	23	5	the	the	DET
ajst-2549	23	6	three	three	NUM
ajst-2549	23	7	versions	version	NOUN
ajst-2549	23	8	is	be	AUX
ajst-2549	23	9	that	that	SCONJ
ajst-2549	23	10	the	the	DET
ajst-2549	23	11	backbone	backbone	NOUN
ajst-2549	23	12	network	network	NOUN
ajst-2549	23	13	is	be	AUX
ajst-2549	23	14	different	different	ADJ
ajst-2549	23	15	,	,	PUNCT
ajst-2549	23	16	and	and	CCONJ
ajst-2549	23	17	resnet50	resnet50	NOUN
ajst-2549	23	18	,	,	PUNCT
ajst-2549	23	19	resnet18	resnet18	NOUN
ajst-2549	23	20	and	and	CCONJ
ajst-2549	23	21	mobilenetv3	mobilenetv3	PROPN
ajst-2549	23	22	are	be	AUX
ajst-2549	23	23	used	use	VERB
ajst-2549	23	24	as	as	ADP
ajst-2549	23	25	the	the	DET
ajst-2549	23	26	backbone	backbone	NOUN
ajst-2549	23	27	network	network	NOUN
ajst-2549	23	28	respectively	respectively	ADV
ajst-2549	23	29	.	.	PUNCT
ajst-2549	24	1	the	the	DET
ajst-2549	24	2	algorithm	algorithm	NOUN
ajst-2549	24	3	in	in	ADP
ajst-2549	24	4	this	this	DET
ajst-2549	24	5	paper	paper	NOUN
ajst-2549	24	6	is	be	AUX
ajst-2549	24	7	improved	improve	VERB
ajst-2549	24	8	based	base	VERB
ajst-2549	24	9	on	on	ADP
ajst-2549	24	10	mobilenetv3	mobilenetv3	PROPN
ajst-2549	24	11	.	.	PUNCT
ajst-2549	25	1	the	the	DET
ajst-2549	25	2	network	network	NOUN
ajst-2549	25	3	structure	structure	NOUN
ajst-2549	25	4	of	of	ADP
ajst-2549	25	5	dbnet	dbnet	NOUN
ajst-2549	25	6	algorithm	algorithm	NOUN
ajst-2549	25	7	consists	consist	VERB
ajst-2549	25	8	of	of	ADP
ajst-2549	25	9	three	three	NUM
ajst-2549	25	10	parts	part	NOUN
ajst-2549	25	11	:	:	PUNCT
ajst-2549	25	12	backbone	backbone	NOUN
ajst-2549	25	13	,	,	PUNCT
ajst-2549	25	14	neck	neck	NOUN
ajst-2549	25	15	and	and	CCONJ
ajst-2549	25	16	head	head	NOUN
ajst-2549	25	17	.	.	PUNCT
ajst-2549	26	1	the	the	DET
ajst-2549	26	2	dbnet	dbnet	PROPN
ajst-2549	26	3	network	network	NOUN
ajst-2549	26	4	structure	structure	NOUN
ajst-2549	26	5	is	be	AUX
ajst-2549	26	6	shown	show	VERB
ajst-2549	26	7	in	in	ADP
ajst-2549	26	8	fig	fig	NOUN
ajst-2549	26	9	.	.	PUNCT
ajst-2549	27	1	1	1	X
ajst-2549	27	2	.	.	PUNCT
ajst-2549	27	3	the	the	DET
ajst-2549	27	4	backbone	backbone	NOUN
ajst-2549	27	5	part	part	NOUN
ajst-2549	27	6	adopts	adopt	VERB
ajst-2549	27	7	the	the	DET
ajst-2549	27	8	image	image	NOUN
ajst-2549	27	9	classification	classification	NOUN
ajst-2549	27	10	network	network	NOUN
ajst-2549	27	11	,	,	PUNCT
ajst-2549	27	12	which	which	PRON
ajst-2549	27	13	is	be	AUX
ajst-2549	27	14	responsible	responsible	ADJ
ajst-2549	27	15	for	for	ADP
ajst-2549	27	16	extracting	extract	VERB
ajst-2549	27	17	the	the	DET
ajst-2549	27	18	multi	multi	ADJ
ajst-2549	27	19	-	-	ADJ
ajst-2549	27	20	scale	scale	ADJ
ajst-2549	27	21	features	feature	NOUN
ajst-2549	27	22	of	of	ADP
ajst-2549	27	23	the	the	DET
ajst-2549	27	24	image	image	NOUN
ajst-2549	27	25	,	,	PUNCT
ajst-2549	27	26	mainly	mainly	ADV
ajst-2549	27	27	including	include	VERB
ajst-2549	27	28	the	the	DET
ajst-2549	27	29	cbh	cbh	NOUN
ajst-2549	27	30	convolution	convolution	NOUN
ajst-2549	27	31	module	module	NOUN
ajst-2549	27	32	(	(	PUNCT
ajst-2549	27	33	convolution	convolution	NOUN
ajst-2549	27	34	layer	layer	NOUN
ajst-2549	27	35	conv	conv	NOUN
ajst-2549	27	36	,	,	PUNCT
ajst-2549	27	37	normalization	normalization	NOUN
ajst-2549	27	38	layer	layer	NOUN
ajst-2549	27	39	bn	bn	NOUN
ajst-2549	27	40	and	and	CCONJ
ajst-2549	27	41	activation	activation	NOUN
ajst-2549	27	42	function	function	VERB
ajst-2549	27	43	hard	hard	ADV
ajst-2549	27	44	-	-	PUNCT
ajst-2549	27	45	swish	swish	NOUN
ajst-2549	27	46	)	)	PUNCT
ajst-2549	27	47	,	,	PUNCT
ajst-2549	27	48	brc	brc	NOUN
ajst-2549	27	49	deconvolution	deconvolution	NOUN
ajst-2549	27	50	module	module	NOUN
ajst-2549	27	51	(	(	PUNCT
ajst-2549	27	52	normalization	normalization	NOUN
ajst-2549	27	53	layer	layer	NOUN
ajst-2549	27	54	bn	bn	NOUN
ajst-2549	27	55	,	,	PUNCT
ajst-2549	27	56	activation	activation	NOUN
ajst-2549	27	57	function	function	NOUN
ajst-2549	27	58	relu	relu	NOUN
ajst-2549	27	59	and	and	CCONJ
ajst-2549	27	60	deconvolution	deconvolution	NOUN
ajst-2549	27	61	layer	layer	NOUN
ajst-2549	27	62	convtranspose	convtranspose	ADJ
ajst-2549	27	63	)	)	PUNCT
ajst-2549	27	64	,	,	PUNCT
ajst-2549	27	65	res	re	VERB
ajst-2549	27	66	residual	residual	ADJ
ajst-2549	27	67	module	module	NOUN
ajst-2549	27	68	and	and	CCONJ
ajst-2549	27	69	res_se	res_se	DET
ajst-2549	27	70	residual	residual	ADJ
ajst-2549	27	71	module	module	NOUN
ajst-2549	27	72	with	with	ADP
ajst-2549	27	73	se	se	NOUN
ajst-2549	27	74	attention	attention	NOUN
ajst-2549	27	75	mechanism	mechanism	NOUN
ajst-2549	27	76	(	(	PUNCT
ajst-2549	27	77	squeeze	squeeze	NOUN
ajst-2549	27	78	-	-	PUNCT
ajst-2549	27	79	and	and	CCONJ
ajst-2549	27	80	-	-	PUNCT
ajst-2549	27	81	excitation	excitation	NOUN
ajst-2549	27	82	)	)	PUNCT
ajst-2549	28	1	[	[	X
ajst-2549	28	2	9	9	NUM
ajst-2549	28	3	]	]	PUNCT
ajst-2549	28	4	.	.	PUNCT
ajst-2549	29	1	the	the	DET
ajst-2549	29	2	res_se	res_se	PRON
ajst-2549	29	3	module	module	NOUN
ajst-2549	29	4	mainly	mainly	ADV
ajst-2549	29	5	includes	include	VERB
ajst-2549	29	6	cbh	cbh	NOUN
ajst-2549	29	7	convolution	convolution	NOUN
ajst-2549	29	8	module	module	NOUN
ajst-2549	29	9	,	,	PUNCT
ajst-2549	29	10	dbh	dbh	PROPN
ajst-2549	29	11	convolution	convolution	NOUN
ajst-2549	29	12	module	module	NOUN
ajst-2549	29	13	(	(	PUNCT
ajst-2549	29	14	depth	depth	NOUN
ajst-2549	29	15	separable	separable	ADJ
ajst-2549	29	16	convolution	convolution	NOUN
ajst-2549	29	17	layer	layer	NOUN
ajst-2549	29	18	dw	dw	PROPN
ajst-2549	29	19	,	,	PUNCT
ajst-2549	29	20	normalization	normalization	NOUN
ajst-2549	29	21	layer	layer	NOUN
ajst-2549	29	22	bn	bn	NOUN
ajst-2549	29	23	and	and	CCONJ
ajst-2549	29	24	activation	activation	NOUN
ajst-2549	29	25	function	function	VERB
ajst-2549	29	26	hard	hard	ADV
ajst-2549	29	27	-	-	PUNCT
ajst-2549	29	28	swish	swish	NOUN
ajst-2549	29	29	)	)	PUNCT
ajst-2549	29	30	,	,	PUNCT
ajst-2549	29	31	attention	attention	NOUN
ajst-2549	29	32	mechanism	mechanism	NOUN
ajst-2549	29	33	module	module	NOUN
ajst-2549	29	34	se	se	PROPN
ajst-2549	29	35	and	and	CCONJ
ajst-2549	29	36	fully	fully	ADV
ajst-2549	29	37	connected	connected	ADJ
ajst-2549	29	38	convolution	convolution	NOUN
ajst-2549	29	39	module	module	NOUN
ajst-2549	29	40	cb	cb	PROPN
ajst-2549	29	41	.	.	PUNCT
ajst-2549	30	1	the	the	DET
ajst-2549	30	2	neck	neck	NOUN
ajst-2549	30	3	part	part	NOUN
ajst-2549	30	4	is	be	AUX
ajst-2549	30	5	located	locate	VERB
ajst-2549	30	6	between	between	ADP
ajst-2549	30	7	backbone	backbone	NOUN
ajst-2549	30	8	and	and	CCONJ
ajst-2549	30	9	head	head	NOUN
ajst-2549	30	10	.	.	PUNCT
ajst-2549	31	1	the	the	DET
ajst-2549	31	2	feature	feature	NOUN
ajst-2549	31	3	pyramid	pyramid	NOUN
ajst-2549	31	4	structure	structure	NOUN
ajst-2549	31	5	fpn	fpn	VERB
ajst-2549	31	6	is	be	AUX
ajst-2549	31	7	used	use	VERB
ajst-2549	31	8	to	to	PART
ajst-2549	31	9	enhance	enhance	VERB
ajst-2549	31	10	the	the	DET
ajst-2549	31	11	image	image	NOUN
ajst-2549	31	12	features	feature	VERB
ajst-2549	31	13	through	through	ADP
ajst-2549	31	14	multi	multi	ADJ
ajst-2549	31	15	-	-	ADJ
ajst-2549	31	16	scale	scale	ADJ
ajst-2549	31	17	feature	feature	NOUN
ajst-2549	31	18	fusion	fusion	NOUN
ajst-2549	31	19	.	.	PUNCT
ajst-2549	32	1	the	the	DET
ajst-2549	32	2	four	four	NUM
ajst-2549	32	3	feature	feature	NOUN
ajst-2549	32	4	maps	map	NOUN
ajst-2549	32	5	output	output	NOUN
ajst-2549	32	6	by	by	ADP
ajst-2549	32	7	backbone	backbone	NOUN
ajst-2549	32	8	are	be	AUX
ajst-2549	32	9	enhanced	enhance	VERB
ajst-2549	32	10	and	and	CCONJ
ajst-2549	32	11	then	then	ADV
ajst-2549	32	12	spliced	splice	VERB
ajst-2549	32	13	(	(	PUNCT
ajst-2549	32	14	concat	concat	NOUN
ajst-2549	32	15	)	)	PUNCT
ajst-2549	32	16	,	,	PUNCT
ajst-2549	32	17	and	and	CCONJ
ajst-2549	32	18	finally	finally	ADV
ajst-2549	32	19	the	the	DET
ajst-2549	32	20	features	feature	NOUN
ajst-2549	32	21	output	output	NOUN
ajst-2549	32	22	by	by	ADP
ajst-2549	32	23	neck	neck	NOUN
ajst-2549	32	24	the	the	DET
ajst-2549	32	25	height	height	NOUN
ajst-2549	32	26	and	and	CCONJ
ajst-2549	32	27	width	width	NOUN
ajst-2549	32	28	of	of	ADP
ajst-2549	32	29	the	the	DET
ajst-2549	32	30	image	image	NOUN
ajst-2549	32	31	are	be	AUX
ajst-2549	32	32	one	one	NUM
ajst-2549	32	33	-	-	PUNCT
ajst-2549	32	34	fourth	fourth	NOUN
ajst-2549	32	35	of	of	ADP
ajst-2549	32	36	the	the	DET
ajst-2549	32	37	original	original	ADJ
ajst-2549	32	38	image	image	NOUN
ajst-2549	32	39	.	.	PUNCT
ajst-2549	33	1	the	the	DET
ajst-2549	33	2	head	head	NOUN
ajst-2549	33	3	part	part	NOUN
ajst-2549	33	4	first	first	ADV
ajst-2549	33	5	goes	go	VERB
ajst-2549	33	6	through	through	ADP
ajst-2549	33	7	a	a	DET
ajst-2549	33	8	convolutional	convolutional	ADJ
ajst-2549	33	9	layer	layer	NOUN
ajst-2549	33	10	,	,	PUNCT
ajst-2549	33	11	then	then	ADV
ajst-2549	33	12	deconvolutions	deconvolution	NOUN
ajst-2549	33	13	twice	twice	ADV
ajst-2549	33	14	,	,	PUNCT
ajst-2549	33	15	maps	map	VERB
ajst-2549	33	16	the	the	DET
ajst-2549	33	17	fpn	fpn	ADJ
ajst-2549	33	18	feature	feature	NOUN
ajst-2549	33	19	output	output	NOUN
ajst-2549	33	20	by	by	ADP
ajst-2549	33	21	neck	neck	NOUN
ajst-2549	33	22	from	from	ADP
ajst-2549	33	23	the	the	DET
ajst-2549	33	24	quarter	quarter	NOUN
ajst-2549	33	25	size	size	NOUN
ajst-2549	33	26	of	of	ADP
ajst-2549	33	27	the	the	DET
ajst-2549	33	28	original	original	ADJ
ajst-2549	33	29	image	image	NOUN
ajst-2549	33	30	to	to	ADP
ajst-2549	33	31	the	the	DET
ajst-2549	33	32	original	original	ADJ
ajst-2549	33	33	65	65	NUM
ajst-2549	33	34	image	image	NOUN
ajst-2549	33	35	size	size	NOUN
ajst-2549	33	36	,	,	PUNCT
ajst-2549	33	37	and	and	CCONJ
ajst-2549	33	38	finally	finally	ADV
ajst-2549	33	39	activates	activate	VERB
ajst-2549	33	40	the	the	DET
ajst-2549	33	41	output	output	NOUN
ajst-2549	33	42	through	through	ADP
ajst-2549	33	43	sigmoid	sigmoid	NOUN
ajst-2549	33	44	figure	figure	NOUN
ajst-2549	33	45	1	1	NUM
ajst-2549	33	46	.	.	PUNCT
ajst-2549	33	47	dbnet	dbnet	NOUN
ajst-2549	33	48	network	network	NOUN
ajst-2549	33	49	structure	structure	NOUN
ajst-2549	33	50	3	3	NUM
ajst-2549	33	51	.	.	PUNCT
ajst-2549	33	52	improved	improve	VERB
ajst-2549	33	53	dbnet	dbnet	NOUN
ajst-2549	33	54	network	network	NOUN
ajst-2549	33	55	this	this	DET
ajst-2549	33	56	paper	paper	NOUN
ajst-2549	33	57	makes	make	VERB
ajst-2549	33	58	three	three	NUM
ajst-2549	33	59	improvements	improvement	NOUN
ajst-2549	33	60	to	to	ADP
ajst-2549	33	61	the	the	DET
ajst-2549	33	62	dbnet	dbnet	NOUN
ajst-2549	33	63	algorithm	algorithm	NOUN
ajst-2549	33	64	:	:	PUNCT
ajst-2549	33	65	the	the	DET
ajst-2549	33	66	backbone	backbone	NOUN
ajst-2549	33	67	part	part	NOUN
ajst-2549	33	68	replaces	replace	VERB
ajst-2549	33	69	the	the	DET
ajst-2549	33	70	se	se	X
ajst-2549	33	71	attention	attention	NOUN
ajst-2549	33	72	mechanism	mechanism	NOUN
ajst-2549	33	73	in	in	ADP
ajst-2549	33	74	the	the	DET
ajst-2549	33	75	residual	residual	ADJ
ajst-2549	33	76	module	module	NOUN
ajst-2549	33	77	with	with	ADP
ajst-2549	33	78	the	the	DET
ajst-2549	33	79	new	new	ADJ
ajst-2549	33	80	ca	ca	NOUN
ajst-2549	33	81	attention	attention	NOUN
ajst-2549	33	82	mechanism	mechanism	NOUN
ajst-2549	33	83	;	;	PUNCT
ajst-2549	33	84	the	the	DET
ajst-2549	33	85	neck	neck	NOUN
ajst-2549	33	86	part	part	NOUN
ajst-2549	33	87	uses	use	VERB
ajst-2549	33	88	the	the	DET
ajst-2549	33	89	weighted	weighted	ADJ
ajst-2549	33	90	bidirectional	bidirectional	ADJ
ajst-2549	33	91	feature	feature	NOUN
ajst-2549	33	92	pyramid	pyramid	NOUN
ajst-2549	33	93	(	(	PUNCT
ajst-2549	33	94	bifpn	bifpn	PROPN
ajst-2549	33	95	)	)	PUNCT
ajst-2549	33	96	to	to	PART
ajst-2549	33	97	improve	improve	VERB
ajst-2549	33	98	the	the	DET
ajst-2549	33	99	fpn	fpn	NOUN
ajst-2549	33	100	;	;	PUNCT
ajst-2549	33	101	except	except	SCONJ
ajst-2549	33	102	for	for	ADP
ajst-2549	33	103	the	the	DET
ajst-2549	33	104	network	network	NOUN
ajst-2549	33	105	structure	structure	NOUN
ajst-2549	33	106	part	part	NOUN
ajst-2549	33	107	,	,	PUNCT
ajst-2549	33	108	other	other	ADJ
ajst-2549	33	109	nonthe	nonthe	ADJ
ajst-2549	33	110	network	network	NOUN
ajst-2549	33	111	structure	structure	NOUN
ajst-2549	33	112	has	have	AUX
ajst-2549	33	113	also	also	ADV
ajst-2549	33	114	been	be	AUX
ajst-2549	33	115	optimized	optimize	VERB
ajst-2549	33	116	.	.	PUNCT
ajst-2549	34	1	the	the	DET
ajst-2549	34	2	data	data	NOUN
ajst-2549	34	3	part	part	NOUN
ajst-2549	34	4	is	be	AUX
ajst-2549	34	5	introduced	introduce	VERB
ajst-2549	34	6	into	into	ADP
ajst-2549	34	7	copypaste	copypaste	NOUN
ajst-2549	34	8	[	[	X
ajst-2549	34	9	10	10	NUM
ajst-2549	34	10	]	]	PUNCT
ajst-2549	34	11	for	for	SCONJ
ajst-2549	34	12	data	datum	NOUN
ajst-2549	34	13	augmentation	augmentation	NOUN
ajst-2549	34	14	to	to	PART
ajst-2549	34	15	improve	improve	VERB
ajst-2549	34	16	the	the	DET
ajst-2549	34	17	generalization	generalization	NOUN
ajst-2549	34	18	of	of	ADP
ajst-2549	34	19	the	the	DET
ajst-2549	34	20	model	model	NOUN
ajst-2549	34	21	.	.	PUNCT
ajst-2549	35	1	the	the	DET
ajst-2549	35	2	hyperparameter	hyperparameter	NOUN
ajst-2549	35	3	part	part	NOUN
ajst-2549	35	4	adopts	adopt	VERB
ajst-2549	35	5	the	the	DET
ajst-2549	35	6	cosine	cosine	NOUN
ajst-2549	35	7	learning	learn	VERB
ajst-2549	35	8	rate	rate	NOUN
ajst-2549	35	9	reduction	reduction	NOUN
ajst-2549	35	10	strategy	strategy	NOUN
ajst-2549	35	11	and	and	CCONJ
ajst-2549	35	12	the	the	DET
ajst-2549	35	13	learning	learning	NOUN
ajst-2549	35	14	rate	rate	NOUN
ajst-2549	35	15	warm	warm	ADJ
ajst-2549	35	16	-	-	PUNCT
ajst-2549	35	17	up	up	ADP
ajst-2549	35	18	strategy	strategy	NOUN
ajst-2549	35	19	to	to	PART
ajst-2549	35	20	ensure	ensure	VERB
ajst-2549	35	21	training	training	NOUN
ajst-2549	35	22	.	.	PUNCT
ajst-2549	36	1	stability	stability	NOUN
ajst-2549	36	2	in	in	ADP
ajst-2549	36	3	the	the	DET
ajst-2549	36	4	initial	initial	ADJ
ajst-2549	36	5	stages	stage	NOUN
ajst-2549	36	6	and	and	CCONJ
ajst-2549	36	7	when	when	SCONJ
ajst-2549	36	8	the	the	DET
ajst-2549	36	9	model	model	NOUN
ajst-2549	36	10	converges	converge	VERB
ajst-2549	36	11	.	.	PUNCT
ajst-2549	37	1	the	the	DET
ajst-2549	37	2	improved	improved	ADJ
ajst-2549	37	3	dbnet	dbnet	NOUN
ajst-2549	37	4	network	network	NOUN
ajst-2549	37	5	structure	structure	NOUN
ajst-2549	37	6	is	be	AUX
ajst-2549	37	7	shown	show	VERB
ajst-2549	37	8	in	in	ADP
ajst-2549	37	9	fig.2	fig.2	PROPN
ajst-2549	37	10	.	.	PUNCT
ajst-2549	38	1	figure	figure	NOUN
ajst-2549	38	2	2	2	NUM
ajst-2549	38	3	.	.	PUNCT
ajst-2549	38	4	improved	improve	VERB
ajst-2549	38	5	dbnet	dbnet	NOUN
ajst-2549	38	6	network	network	NOUN
ajst-2549	38	7	structure	structure	NOUN
ajst-2549	38	8	66	66	NUM
ajst-2549	38	9	3.1	3.1	NUM
ajst-2549	38	10	.	.	PUNCT
ajst-2549	39	1	residual	residual	ADJ
ajst-2549	39	2	module	module	NOUN
ajst-2549	39	3	introduces	introduce	VERB
ajst-2549	39	4	a	a	DET
ajst-2549	39	5	new	new	ADJ
ajst-2549	39	6	attention	attention	NOUN
ajst-2549	39	7	mechanism	mechanism	NOUN
ajst-2549	39	8	in	in	ADP
ajst-2549	39	9	this	this	DET
ajst-2549	39	10	paper	paper	NOUN
ajst-2549	39	11	,	,	PUNCT
ajst-2549	39	12	the	the	DET
ajst-2549	39	13	se	se	PROPN
ajst-2549	39	14	attention	attention	NOUN
ajst-2549	39	15	mechanism	mechanism	NOUN
ajst-2549	39	16	of	of	ADP
ajst-2549	39	17	the	the	DET
ajst-2549	39	18	residual	residual	ADJ
ajst-2549	39	19	module	module	NOUN
ajst-2549	39	20	in	in	ADP
ajst-2549	39	21	the	the	DET
ajst-2549	39	22	backbone	backbone	NOUN
ajst-2549	39	23	network	network	NOUN
ajst-2549	39	24	is	be	AUX
ajst-2549	39	25	changed	change	VERB
ajst-2549	39	26	to	to	ADP
ajst-2549	39	27	a	a	DET
ajst-2549	39	28	novel	novel	NOUN
ajst-2549	39	29	ca	ca	NOUN
ajst-2549	39	30	attention	attention	NOUN
ajst-2549	39	31	mechanism	mechanism	NOUN
ajst-2549	39	32	.	.	PUNCT
ajst-2549	40	1	the	the	DET
ajst-2549	40	2	ca	ca	NOUN
ajst-2549	40	3	attention	attention	NOUN
ajst-2549	40	4	mechanism	mechanism	NOUN
ajst-2549	40	5	can	can	AUX
ajst-2549	40	6	make	make	VERB
ajst-2549	40	7	the	the	DET
ajst-2549	40	8	lightweight	lightweight	ADJ
ajst-2549	40	9	network	network	NOUN
ajst-2549	40	10	mobilenetv3	mobilenetv3	PROPN
ajst-2549	40	11	obtain	obtain	VERB
ajst-2549	40	12	information	information	NOUN
ajst-2549	40	13	of	of	ADP
ajst-2549	40	14	a	a	DET
ajst-2549	40	15	larger	large	ADJ
ajst-2549	40	16	area	area	NOUN
ajst-2549	40	17	without	without	ADP
ajst-2549	40	18	introducing	introduce	VERB
ajst-2549	40	19	large	large	ADJ
ajst-2549	40	20	overhead	overhead	NOUN
ajst-2549	40	21	by	by	ADP
ajst-2549	40	22	embedding	embed	VERB
ajst-2549	40	23	the	the	DET
ajst-2549	40	24	position	position	NOUN
ajst-2549	40	25	information	information	NOUN
ajst-2549	40	26	into	into	ADP
ajst-2549	40	27	the	the	DET
ajst-2549	40	28	channel	channel	NOUN
ajst-2549	40	29	attention	attention	NOUN
ajst-2549	40	30	.	.	PUNCT
ajst-2549	41	1	the	the	DET
ajst-2549	41	2	structure	structure	NOUN
ajst-2549	41	3	comparison	comparison	NOUN
ajst-2549	41	4	of	of	ADP
ajst-2549	41	5	the	the	DET
ajst-2549	41	6	attention	attention	NOUN
ajst-2549	41	7	mechanism	mechanism	NOUN
ajst-2549	41	8	is	be	AUX
ajst-2549	41	9	shown	show	VERB
ajst-2549	41	10	in	in	ADP
ajst-2549	41	11	fig	fig	NOUN
ajst-2549	41	12	.	.	PUNCT
ajst-2549	42	1	3	3	X
ajst-2549	42	2	.	.	X
ajst-2549	42	3	the	the	DET
ajst-2549	42	4	se	se	PROPN
ajst-2549	42	5	module	module	NOUN
ajst-2549	42	6	in	in	ADP
ajst-2549	42	7	fig	fig	NOUN
ajst-2549	42	8	.	.	PUNCT
ajst-2549	43	1	3	3	NUM
ajst-2549	43	2	is	be	AUX
ajst-2549	43	3	divided	divide	VERB
ajst-2549	43	4	into	into	ADP
ajst-2549	43	5	two	two	NUM
ajst-2549	43	6	steps	step	NOUN
ajst-2549	43	7	in	in	ADP
ajst-2549	43	8	structure	structure	NOUN
ajst-2549	43	9	,	,	PUNCT
ajst-2549	43	10	compression	compression	NOUN
ajst-2549	43	11	and	and	CCONJ
ajst-2549	43	12	excitation	excitation	NOUN
ajst-2549	43	13	,	,	PUNCT
ajst-2549	43	14	which	which	PRON
ajst-2549	43	15	are	be	AUX
ajst-2549	43	16	respectively	respectively	ADV
ajst-2549	43	17	designed	design	VERB
ajst-2549	43	18	for	for	ADP
ajst-2549	43	19	global	global	ADJ
ajst-2549	43	20	information	information	NOUN
ajst-2549	43	21	embedding	embed	VERB
ajst-2549	43	22	and	and	CCONJ
ajst-2549	43	23	adaptive	adaptive	ADJ
ajst-2549	43	24	recalibration	recalibration	NOUN
ajst-2549	43	25	of	of	ADP
ajst-2549	43	26	channel	channel	NOUN
ajst-2549	43	27	relationships	relationship	NOUN
ajst-2549	43	28	.	.	PUNCT
ajst-2549	44	1	the	the	DET
ajst-2549	44	2	specific	specific	ADJ
ajst-2549	44	3	structure	structure	NOUN
ajst-2549	44	4	is	be	AUX
ajst-2549	44	5	shown	show	VERB
ajst-2549	44	6	in	in	ADP
ajst-2549	44	7	fig	fig	NOUN
ajst-2549	44	8	.	.	PUNCT
ajst-2549	45	1	3	3	NUM
ajst-2549	45	2	(	(	PUNCT
ajst-2549	45	3	a	a	NOUN
ajst-2549	45	4	)	)	PUNCT
ajst-2549	45	5	,	,	PUNCT
ajst-2549	45	6	using	use	VERB
ajst-2549	45	7	global	global	ADJ
ajst-2549	45	8	average	average	ADJ
ajst-2549	45	9	pooling	pooling	NOUN
ajst-2549	45	10	(	(	PUNCT
ajst-2549	45	11	global	global	ADJ
ajst-2549	45	12	avg	avg	NOUN
ajst-2549	45	13	pool	pool	PROPN
ajst-2549	45	14	)	)	PUNCT
ajst-2549	45	15	as	as	ADP
ajst-2549	45	16	a	a	DET
ajst-2549	45	17	compression	compression	NOUN
ajst-2549	45	18	operation	operation	NOUN
ajst-2549	45	19	,	,	PUNCT
ajst-2549	45	20	followed	follow	VERB
ajst-2549	45	21	by	by	ADP
ajst-2549	45	22	a	a	DET
ajst-2549	45	23	fully	fully	ADV
ajst-2549	45	24	connected	connect	VERB
ajst-2549	45	25	layer	layer	NOUN
ajst-2549	45	26	(	(	PUNCT
ajst-2549	45	27	fc	fc	X
ajst-2549	45	28	)	)	PUNCT
ajst-2549	45	29	layer	layer	NOUN
ajst-2549	45	30	to	to	PART
ajst-2549	45	31	reduce	reduce	VERB
ajst-2549	45	32	the	the	DET
ajst-2549	45	33	feature	feature	NOUN
ajst-2549	45	34	dimension	dimension	NOUN
ajst-2549	45	35	by	by	ADP
ajst-2549	45	36	r	r	NOUN
ajst-2549	45	37	times	time	NOUN
ajst-2549	45	38	,	,	PUNCT
ajst-2549	45	39	and	and	CCONJ
ajst-2549	45	40	then	then	ADV
ajst-2549	45	41	activated	activate	VERB
ajst-2549	45	42	by	by	ADP
ajst-2549	45	43	the	the	DET
ajst-2549	45	44	relu	relu	NOUN
ajst-2549	45	45	[	[	X
ajst-2549	45	46	14	14	NUM
ajst-2549	45	47	]	]	PUNCT
ajst-2549	45	48	function	function	NOUN
ajst-2549	45	49	and	and	CCONJ
ajst-2549	45	50	then	then	ADV
ajst-2549	45	51	raised	raise	VERB
ajst-2549	45	52	back	back	ADV
ajst-2549	45	53	to	to	ADP
ajst-2549	45	54	the	the	DET
ajst-2549	45	55	original	original	ADJ
ajst-2549	45	56	feature	feature	NOUN
ajst-2549	45	57	dimension	dimension	NOUN
ajst-2549	45	58	through	through	ADP
ajst-2549	45	59	an	an	DET
ajst-2549	45	60	fc	fc	NOUN
ajst-2549	45	61	layer	layer	NOUN
ajst-2549	45	62	,	,	PUNCT
ajst-2549	45	63	then	then	ADV
ajst-2549	45	64	it	it	PRON
ajst-2549	45	65	is	be	AUX
ajst-2549	45	66	converted	convert	VERB
ajst-2549	45	67	into	into	ADP
ajst-2549	45	68	a	a	DET
ajst-2549	45	69	normalized	normalize	VERB
ajst-2549	45	70	weight	weight	NOUN
ajst-2549	45	71	of	of	ADP
ajst-2549	45	72	0~1	0~1	NUM
ajst-2549	45	73	through	through	ADP
ajst-2549	45	74	the	the	DET
ajst-2549	45	75	hard	hard	ADJ
ajst-2549	45	76	-	-	PUNCT
ajst-2549	45	77	sigmoid	sigmoid	NOUN
ajst-2549	45	78	function	function	NOUN
ajst-2549	45	79	,	,	PUNCT
ajst-2549	45	80	which	which	PRON
ajst-2549	45	81	is	be	AUX
ajst-2549	45	82	an	an	DET
ajst-2549	45	83	excitation	excitation	NOUN
ajst-2549	45	84	operation	operation	NOUN
ajst-2549	45	85	.	.	PUNCT
ajst-2549	46	1	finally	finally	ADV
ajst-2549	46	2	,	,	PUNCT
ajst-2549	46	3	the	the	DET
ajst-2549	46	4	newly	newly	ADV
ajst-2549	46	5	generated	generate	VERB
ajst-2549	46	6	feature	feature	NOUN
ajst-2549	46	7	map	map	NOUN
ajst-2549	46	8	and	and	CCONJ
ajst-2549	46	9	the	the	DET
ajst-2549	46	10	original	original	ADJ
ajst-2549	46	11	feature	feature	NOUN
ajst-2549	46	12	map	map	NOUN
ajst-2549	46	13	are	be	AUX
ajst-2549	46	14	multiplied	multiply	VERB
ajst-2549	46	15	and	and	CCONJ
ajst-2549	46	16	calibrated	calibrate	VERB
ajst-2549	46	17	,	,	PUNCT
ajst-2549	46	18	that	that	ADV
ajst-2549	46	19	is	is	ADV
ajst-2549	46	20	,	,	PUNCT
ajst-2549	46	21	the	the	DET
ajst-2549	46	22	re	re	ADJ
ajst-2549	46	23	-	-	NOUN
ajst-2549	46	24	weight	weight	ADJ
ajst-2549	46	25	operation	operation	NOUN
ajst-2549	46	26	;	;	PUNCT
ajst-2549	46	27	the	the	DET
ajst-2549	46	28	ca	ca	NOUN
ajst-2549	46	29	module	module	NOUN
ajst-2549	46	30	encodes	encode	VERB
ajst-2549	46	31	the	the	DET
ajst-2549	46	32	channel	channel	NOUN
ajst-2549	46	33	relationship	relationship	NOUN
ajst-2549	46	34	and	and	CCONJ
ajst-2549	46	35	long	long	ADJ
ajst-2549	46	36	-	-	PUNCT
ajst-2549	46	37	range	range	NOUN
ajst-2549	46	38	dependency	dependency	NOUN
ajst-2549	46	39	through	through	ADP
ajst-2549	46	40	the	the	DET
ajst-2549	46	41	spatial	spatial	ADJ
ajst-2549	46	42	domain	domain	NOUN
ajst-2549	46	43	.	.	PUNCT
ajst-2549	47	1	the	the	DET
ajst-2549	47	2	specific	specific	ADJ
ajst-2549	47	3	structure	structure	NOUN
ajst-2549	47	4	is	be	AUX
ajst-2549	47	5	shown	show	VERB
ajst-2549	47	6	in	in	ADP
ajst-2549	47	7	fig	fig	NOUN
ajst-2549	47	8	.	.	PUNCT
ajst-2549	48	1	3	3	NUM
ajst-2549	48	2	(	(	PUNCT
ajst-2549	48	3	b	b	NOUN
ajst-2549	48	4	)	)	PUNCT
ajst-2549	48	5	.	.	PUNCT
ajst-2549	49	1	first	first	ADV
ajst-2549	49	2	,	,	PUNCT
ajst-2549	49	3	along	along	ADP
ajst-2549	49	4	the	the	PRON
ajst-2549	49	5	x	x	NOUN
ajst-2549	49	6	the	the	DET
ajst-2549	49	7	average	average	ADJ
ajst-2549	49	8	pooling	pooling	NOUN
ajst-2549	49	9	(	(	PUNCT
ajst-2549	49	10	avg	avg	NOUN
ajst-2549	49	11	pool	pool	NOUN
ajst-2549	49	12	)	)	PUNCT
ajst-2549	49	13	operation	operation	NOUN
ajst-2549	49	14	is	be	AUX
ajst-2549	49	15	performed	perform	VERB
ajst-2549	49	16	in	in	ADP
ajst-2549	49	17	the	the	DET
ajst-2549	49	18	direction	direction	NOUN
ajst-2549	49	19	and	and	CCONJ
ajst-2549	49	20	the	the	DET
ajst-2549	49	21	y	y	PROPN
ajst-2549	49	22	direction	direction	NOUN
ajst-2549	49	23	respectively	respectively	ADV
ajst-2549	49	24	,	,	PUNCT
ajst-2549	49	25	and	and	CCONJ
ajst-2549	49	26	two	two	NUM
ajst-2549	49	27	feature	feature	NOUN
ajst-2549	49	28	maps	map	NOUN
ajst-2549	49	29	based	base	VERB
ajst-2549	49	30	on	on	ADP
ajst-2549	49	31	x	x	PUNCT
ajst-2549	49	32	and	and	CCONJ
ajst-2549	49	33	y	y	PROPN
ajst-2549	49	34	are	be	AUX
ajst-2549	49	35	obtained	obtain	VERB
ajst-2549	49	36	,	,	PUNCT
ajst-2549	49	37	and	and	CCONJ
ajst-2549	49	38	then	then	ADV
ajst-2549	49	39	the	the	DET
ajst-2549	49	40	concatenation	concatenation	NOUN
ajst-2549	49	41	(	(	PUNCT
ajst-2549	49	42	concat	concat	NOUN
ajst-2549	49	43	)	)	PUNCT
ajst-2549	49	44	and	and	CCONJ
ajst-2549	49	45	convolution	convolution	NOUN
ajst-2549	49	46	operations	operation	NOUN
ajst-2549	49	47	are	be	AUX
ajst-2549	49	48	performed	perform	VERB
ajst-2549	49	49	to	to	PART
ajst-2549	49	50	establish	establish	VERB
ajst-2549	49	51	a	a	DET
ajst-2549	49	52	longrange	longrange	NOUN
ajst-2549	49	53	dependency	dependency	NOUN
ajst-2549	49	54	,	,	PUNCT
ajst-2549	49	55	and	and	CCONJ
ajst-2549	49	56	then	then	ADV
ajst-2549	49	57	the	the	DET
ajst-2549	49	58	bn	bn	PROPN
ajst-2549	49	59	operation	operation	NOUN
ajst-2549	49	60	is	be	AUX
ajst-2549	49	61	performed	perform	VERB
ajst-2549	49	62	,	,	PUNCT
ajst-2549	49	63	and	and	CCONJ
ajst-2549	49	64	then	then	ADV
ajst-2549	49	65	the	the	DET
ajst-2549	49	66	feature	feature	NOUN
ajst-2549	49	67	map	map	NOUN
ajst-2549	49	68	with	with	ADP
ajst-2549	49	69	global	global	ADJ
ajst-2549	49	70	information	information	NOUN
ajst-2549	49	71	is	be	AUX
ajst-2549	49	72	redivided	redivide	VERB
ajst-2549	49	73	into	into	ADP
ajst-2549	49	74	feature	feature	NOUN
ajst-2549	49	75	maps	map	NOUN
ajst-2549	49	76	based	base	VERB
ajst-2549	49	77	on	on	ADP
ajst-2549	49	78	x	x	PUNCT
ajst-2549	49	79	and	and	CCONJ
ajst-2549	49	80	y	y	PROPN
ajst-2549	49	81	,	,	PUNCT
ajst-2549	49	82	and	and	CCONJ
ajst-2549	49	83	convolution	convolution	NOUN
ajst-2549	49	84	and	and	CCONJ
ajst-2549	49	85	sigmoid	sigmoid	NOUN
ajst-2549	49	86	normalization	normalization	NOUN
ajst-2549	49	87	are	be	AUX
ajst-2549	49	88	performed	perform	VERB
ajst-2549	49	89	respectively	respectively	ADV
ajst-2549	49	90	,	,	PUNCT
ajst-2549	49	91	and	and	CCONJ
ajst-2549	49	92	finally	finally	ADV
ajst-2549	49	93	the	the	DET
ajst-2549	49	94	position	position	NOUN
ajst-2549	49	95	information	information	NOUN
ajst-2549	49	96	along	along	ADP
ajst-2549	49	97	x	x	PUNCT
ajst-2549	49	98	and	and	CCONJ
ajst-2549	49	99	y	y	PROPN
ajst-2549	49	100	is	be	AUX
ajst-2549	49	101	weighted	weight	VERB
ajst-2549	49	102	on	on	ADP
ajst-2549	49	103	the	the	DET
ajst-2549	49	104	original	original	ADJ
ajst-2549	49	105	feature	feature	NOUN
ajst-2549	49	106	map	map	NOUN
ajst-2549	49	107	.	.	PUNCT
ajst-2549	50	1	in	in	ADP
ajst-2549	50	2	fig	fig	NOUN
ajst-2549	50	3	.	.	PUNCT
ajst-2549	51	1	3	3	NUM
ajst-2549	51	2	,	,	PUNCT
ajst-2549	51	3	"	"	PUNCT
ajst-2549	51	4	c	c	X
ajst-2549	51	5	"	"	PUNCT
ajst-2549	51	6	represents	represent	VERB
ajst-2549	51	7	the	the	DET
ajst-2549	51	8	channel	channel	NOUN
ajst-2549	51	9	,	,	PUNCT
ajst-2549	51	10	"	"	PUNCT
ajst-2549	51	11	r	r	NOUN
ajst-2549	51	12	"	"	PUNCT
ajst-2549	51	13	represents	represent	VERB
ajst-2549	51	14	the	the	DET
ajst-2549	51	15	multiple	multiple	NOUN
ajst-2549	51	16	of	of	ADP
ajst-2549	51	17	channel	channel	NOUN
ajst-2549	51	18	reduction	reduction	NOUN
ajst-2549	51	19	,	,	PUNCT
ajst-2549	51	20	and	and	CCONJ
ajst-2549	51	21	"	"	PUNCT
ajst-2549	51	22	h	h	NOUN
ajst-2549	51	23	"	"	PUNCT
ajst-2549	51	24	and	and	CCONJ
ajst-2549	51	25	"	"	PUNCT
ajst-2549	51	26	w	w	NOUN
ajst-2549	51	27	"	"	PUNCT
ajst-2549	51	28	represent	represent	VERB
ajst-2549	51	29	height	height	NOUN
ajst-2549	51	30	and	and	CCONJ
ajst-2549	51	31	width	width	ADJ
ajst-2549	51	32	,	,	PUNCT
ajst-2549	51	33	respectively	respectively	ADV
ajst-2549	51	34	.	.	PUNCT
ajst-2549	52	1	(	(	PUNCT
ajst-2549	52	2	a	a	X
ajst-2549	52	3	)	)	PUNCT
ajst-2549	52	4	se	se	NOUN
ajst-2549	52	5	attention	attention	NOUN
ajst-2549	52	6	mechanism	mechanism	NOUN
ajst-2549	52	7	structure	structure	NOUN
ajst-2549	52	8	(	(	PUNCT
ajst-2549	52	9	b	b	NOUN
ajst-2549	52	10	)	)	PUNCT
ajst-2549	52	11	ca	can	AUX
ajst-2549	52	12	attention	attention	NOUN
ajst-2549	52	13	mechanism	mechanism	NOUN
ajst-2549	52	14	structure	structure	NOUN
ajst-2549	52	15	figure	figure	NOUN
ajst-2549	52	16	3	3	NUM
ajst-2549	52	17	.	.	PUNCT
ajst-2549	52	18	attention	attention	NOUN
ajst-2549	52	19	mechanism	mechanism	NOUN
ajst-2549	52	20	structure	structure	NOUN
ajst-2549	52	21	comparison	comparison	NOUN
ajst-2549	52	22	3.2	3.2	NUM
ajst-2549	52	23	.	.	PUNCT
ajst-2549	53	1	introducing	introduce	VERB
ajst-2549	53	2	bifpn	bifpn	PROPN
ajst-2549	53	3	structure	structure	NOUN
ajst-2549	53	4	to	to	PART
ajst-2549	53	5	improve	improve	VERB
ajst-2549	53	6	multi	multi	ADJ
ajst-2549	53	7	-	-	ADJ
ajst-2549	53	8	scale	scale	ADJ
ajst-2549	53	9	feature	feature	NOUN
ajst-2549	53	10	fusion	fusion	NOUN
ajst-2549	53	11	the	the	DET
ajst-2549	53	12	bifpn	bifpn	NOUN
ajst-2549	53	13	introduced	introduce	VERB
ajst-2549	53	14	in	in	ADP
ajst-2549	53	15	this	this	DET
ajst-2549	53	16	paper	paper	NOUN
ajst-2549	53	17	is	be	AUX
ajst-2549	53	18	borrowed	borrow	VERB
ajst-2549	53	19	from	from	ADP
ajst-2549	53	20	the	the	DET
ajst-2549	53	21	feature	feature	NOUN
ajst-2549	53	22	enhancement	enhancement	NOUN
ajst-2549	53	23	network	network	NOUN
ajst-2549	53	24	in	in	ADP
ajst-2549	53	25	efficientdet	efficientdet	NOUN
ajst-2549	54	1	[	[	X
ajst-2549	54	2	15	15	NUM
ajst-2549	54	3	]	]	PUNCT
ajst-2549	54	4	.	.	PUNCT
ajst-2549	55	1	compared	compare	VERB
ajst-2549	55	2	with	with	ADP
ajst-2549	55	3	the	the	DET
ajst-2549	55	4	fpn	fpn	PROPN
ajst-2549	55	5	structure	structure	NOUN
ajst-2549	55	6	,	,	PUNCT
ajst-2549	55	7	in	in	ADP
ajst-2549	55	8	addition	addition	NOUN
ajst-2549	55	9	to	to	ADP
ajst-2549	55	10	the	the	DET
ajst-2549	55	11	bottom	bottom	ADJ
ajst-2549	55	12	-	-	PUNCT
ajst-2549	55	13	up	up	ADP
ajst-2549	55	14	feature	feature	NOUN
ajst-2549	55	15	fusion	fusion	NOUN
ajst-2549	55	16	,	,	PUNCT
ajst-2549	55	17	the	the	DET
ajst-2549	55	18	top	top	ADJ
ajst-2549	55	19	-	-	PUNCT
ajst-2549	55	20	down	down	ADP
ajst-2549	55	21	feature	feature	NOUN
ajst-2549	55	22	fusion	fusion	NOUN
ajst-2549	55	23	is	be	AUX
ajst-2549	55	24	added	add	VERB
ajst-2549	55	25	,	,	PUNCT
ajst-2549	55	26	and	and	CCONJ
ajst-2549	55	27	the	the	DET
ajst-2549	55	28	introduction	introduction	NOUN
ajst-2549	55	29	between	between	ADP
ajst-2549	55	30	the	the	DET
ajst-2549	55	31	input	input	NOUN
ajst-2549	55	32	node	node	NOUN
ajst-2549	55	33	and	and	CCONJ
ajst-2549	55	34	the	the	DET
ajst-2549	55	35	output	output	NOUN
ajst-2549	55	36	node	node	NOUN
ajst-2549	55	37	of	of	ADP
ajst-2549	55	38	the	the	DET
ajst-2549	55	39	same	same	ADJ
ajst-2549	55	40	scale	scale	NOUN
ajst-2549	55	41	is	be	AUX
ajst-2549	55	42	introduced	introduce	VERB
ajst-2549	55	43	.	.	PUNCT
ajst-2549	56	1	a	a	DET
ajst-2549	56	2	skip	skip	ADJ
ajst-2549	56	3	connection	connection	NOUN
ajst-2549	56	4	can	can	AUX
ajst-2549	56	5	fuse	fuse	VERB
ajst-2549	56	6	more	more	ADJ
ajst-2549	56	7	angle	angle	NOUN
ajst-2549	56	8	steel	steel	NOUN
ajst-2549	56	9	character	character	NOUN
ajst-2549	56	10	features	feature	VERB
ajst-2549	56	11	and	and	CCONJ
ajst-2549	56	12	improve	improve	VERB
ajst-2549	56	13	the	the	DET
ajst-2549	56	14	ability	ability	NOUN
ajst-2549	56	15	of	of	ADP
ajst-2549	56	16	network	network	NOUN
ajst-2549	56	17	feature	feature	NOUN
ajst-2549	56	18	enhancement	enhancement	NOUN
ajst-2549	56	19	.	.	PUNCT
ajst-2549	57	1	the	the	DET
ajst-2549	57	2	bifpn	bifpn	PROPN
ajst-2549	57	3	network	network	NOUN
ajst-2549	57	4	structure	structure	NOUN
ajst-2549	57	5	is	be	AUX
ajst-2549	57	6	shown	show	VERB
ajst-2549	57	7	in	in	ADP
ajst-2549	57	8	fig	fig	NOUN
ajst-2549	57	9	.	.	PUNCT
ajst-2549	58	1	4	4	X
ajst-2549	58	2	.	.	X
ajst-2549	58	3	this	this	DET
ajst-2549	58	4	structure	structure	NOUN
ajst-2549	58	5	is	be	AUX
ajst-2549	58	6	different	different	ADJ
ajst-2549	58	7	from	from	ADP
ajst-2549	58	8	the	the	DET
ajst-2549	58	9	original	original	ADJ
ajst-2549	58	10	bifpn	bifpn	NOUN
ajst-2549	58	11	structure	structure	NOUN
ajst-2549	58	12	,	,	PUNCT
ajst-2549	58	13	and	and	CCONJ
ajst-2549	58	14	is	be	AUX
ajst-2549	58	15	changed	change	VERB
ajst-2549	58	16	to	to	ADP
ajst-2549	58	17	a	a	DET
ajst-2549	58	18	four	four	NUM
ajst-2549	58	19	-	-	PUNCT
ajst-2549	58	20	layer	layer	NOUN
ajst-2549	58	21	structure	structure	NOUN
ajst-2549	58	22	in	in	ADP
ajst-2549	58	23	order	order	NOUN
ajst-2549	58	24	to	to	PART
ajst-2549	58	25	adapt	adapt	VERB
ajst-2549	58	26	to	to	ADP
ajst-2549	58	27	the	the	DET
ajst-2549	58	28	output	output	NOUN
ajst-2549	58	29	of	of	ADP
ajst-2549	58	30	the	the	DET
ajst-2549	58	31	backbone	backbone	NOUN
ajst-2549	58	32	network	network	NOUN
ajst-2549	58	33	.	.	PUNCT
ajst-2549	59	1	figure	figure	NOUN
ajst-2549	59	2	4	4	NUM
ajst-2549	59	3	.	.	PUNCT
ajst-2549	60	1	bifpn	bifpn	PROPN
ajst-2549	60	2	network	network	NOUN
ajst-2549	60	3	structure	structure	NOUN
ajst-2549	60	4	3.3	3.3	NUM
ajst-2549	60	5	.	.	PUNCT
ajst-2549	61	1	non	non	ADJ
ajst-2549	61	2	-	-	ADJ
ajst-2549	61	3	network	network	ADJ
ajst-2549	61	4	structure	structure	NOUN
ajst-2549	61	5	part	part	NOUN
ajst-2549	61	6	algorithm	algorithm	NOUN
ajst-2549	61	7	optimization	optimization	NOUN
ajst-2549	61	8	3.3.1	3.3.1	PROPN
ajst-2549	61	9	.	.	PUNCT
ajst-2549	61	10	copy	copy	NOUN
ajst-2549	61	11	-	-	PUNCT
ajst-2549	61	12	paste	paste	NOUN
ajst-2549	61	13	data	data	NOUN
ajst-2549	61	14	enhancement	enhancement	NOUN
ajst-2549	61	15	due	due	ADP
ajst-2549	61	16	to	to	ADP
ajst-2549	61	17	the	the	DET
ajst-2549	61	18	single	single	ADJ
ajst-2549	61	19	position	position	NOUN
ajst-2549	61	20	of	of	ADP
ajst-2549	61	21	the	the	DET
ajst-2549	61	22	text	text	NOUN
ajst-2549	61	23	area	area	NOUN
ajst-2549	61	24	in	in	ADP
ajst-2549	61	25	the	the	DET
ajst-2549	61	26	collected	collect	VERB
ajst-2549	61	27	angle	angle	NOUN
ajst-2549	61	28	steel	steel	NOUN
ajst-2549	61	29	data	datum	NOUN
ajst-2549	61	30	set	set	VERB
ajst-2549	61	31	images	image	NOUN
ajst-2549	61	32	,	,	PUNCT
ajst-2549	61	33	this	this	DET
ajst-2549	61	34	paper	paper	NOUN
ajst-2549	61	35	introduces	introduce	VERB
ajst-2549	61	36	the	the	DET
ajst-2549	61	37	copypaste	copypaste	NOUN
ajst-2549	61	38	method	method	NOUN
ajst-2549	61	39	to	to	PART
ajst-2549	61	40	optimize	optimize	VERB
ajst-2549	61	41	the	the	DET
ajst-2549	61	42	model	model	NOUN
ajst-2549	61	43	for	for	ADP
ajst-2549	61	44	data	data	NOUN
ajst-2549	61	45	enhancement	enhancement	NOUN
ajst-2549	61	46	,	,	PUNCT
ajst-2549	61	47	by	by	ADP
ajst-2549	61	48	randomly	randomly	ADV
ajst-2549	61	49	selecting	select	VERB
ajst-2549	61	50	two	two	NUM
ajst-2549	61	51	training	training	NOUN
ajst-2549	61	52	images	image	NOUN
ajst-2549	61	53	,	,	PUNCT
ajst-2549	61	54	such	such	ADJ
ajst-2549	61	55	as	as	ADP
ajst-2549	61	56	(	(	PUNCT
ajst-2549	61	57	a	a	NOUN
ajst-2549	61	58	)	)	PUNCT
ajst-2549	61	59	and	and	CCONJ
ajst-2549	61	60	(	(	PUNCT
ajst-2549	61	61	b	b	NOUN
ajst-2549	61	62	)	)	PUNCT
ajst-2549	61	63	in	in	ADP
ajst-2549	61	64	fig	fig	NOUN
ajst-2549	61	65	.	.	PUNCT
ajst-2549	62	1	5	5	NUM
ajst-2549	62	2	,	,	PUNCT
ajst-2549	62	3	to	to	PART
ajst-2549	62	4	perform	perform	VERB
ajst-2549	62	5	random	random	ADJ
ajst-2549	62	6	jitter	jitter	NOUN
ajst-2549	62	7	scaling	scaling	NOUN
ajst-2549	62	8	and	and	CCONJ
ajst-2549	62	9	random	random	ADJ
ajst-2549	62	10	flip	flip	NOUN
ajst-2549	62	11	horizontally	horizontally	ADV
ajst-2549	62	12	,	,	PUNCT
ajst-2549	62	13	select	select	VERB
ajst-2549	62	14	one	one	NUM
ajst-2549	62	15	of	of	ADP
ajst-2549	62	16	the	the	DET
ajst-2549	62	17	images	image	NOUN
ajst-2549	62	18	at	at	ADP
ajst-2549	62	19	random	random	ADJ
ajst-2549	62	20	,	,	PUNCT
ajst-2549	62	21	and	and	CCONJ
ajst-2549	62	22	paste	paste	VERB
ajst-2549	62	23	the	the	DET
ajst-2549	62	24	text	text	NOUN
ajst-2549	62	25	area	area	NOUN
ajst-2549	62	26	in	in	ADP
ajst-2549	62	27	the	the	DET
ajst-2549	62	28	image	image	NOUN
ajst-2549	62	29	at	at	ADP
ajst-2549	62	30	a	a	DET
ajst-2549	62	31	random	random	ADJ
ajst-2549	62	32	location	location	NOUN
ajst-2549	62	33	in	in	ADP
ajst-2549	62	34	the	the	DET
ajst-2549	62	35	other	other	ADJ
ajst-2549	62	36	image	image	NOUN
ajst-2549	62	37	.	.	PUNCT
ajst-2549	63	1	it	it	PRON
ajst-2549	63	2	can	can	AUX
ajst-2549	63	3	better	well	ADV
ajst-2549	63	4	improve	improve	VERB
ajst-2549	63	5	the	the	DET
ajst-2549	63	6	richness	richness	NOUN
ajst-2549	63	7	of	of	ADP
ajst-2549	63	8	the	the	DET
ajst-2549	63	9	training	training	NOUN
ajst-2549	63	10	set	set	NOUN
ajst-2549	63	11	and	and	CCONJ
ajst-2549	63	12	67	67	NUM
ajst-2549	63	13	enhance	enhance	VERB
ajst-2549	63	14	the	the	DET
ajst-2549	63	15	robustness	robustness	NOUN
ajst-2549	63	16	of	of	ADP
ajst-2549	63	17	the	the	DET
ajst-2549	63	18	model	model	NOUN
ajst-2549	63	19	to	to	ADP
ajst-2549	63	20	the	the	DET
ajst-2549	63	21	background	background	NOUN
ajst-2549	63	22	.	.	PUNCT
ajst-2549	64	1	the	the	DET
ajst-2549	64	2	copy	copy	NOUN
ajst-2549	64	3	-	-	PUNCT
ajst-2549	64	4	paste	paste	NOUN
ajst-2549	64	5	method	method	NOUN
ajst-2549	64	6	demonstration	demonstration	NOUN
ajst-2549	64	7	is	be	AUX
ajst-2549	64	8	shown	show	VERB
ajst-2549	64	9	in	in	ADP
ajst-2549	64	10	fig	fig	NOUN
ajst-2549	64	11	.	.	PUNCT
ajst-2549	65	1	5	5	X
ajst-2549	65	2	.	.	X
ajst-2549	65	3	(	(	PUNCT
ajst-2549	65	4	a)copy	a)copy	NOUN
ajst-2549	65	5	-	-	PUNCT
ajst-2549	65	6	paste	paste	NOUN
ajst-2549	65	7	(	(	PUNCT
ajst-2549	65	8	b	b	NOUN
ajst-2549	65	9	)	)	PUNCT
ajst-2549	65	10	img2	img2	NOUN
ajst-2549	65	11	(	(	PUNCT
ajst-2549	65	12	c)img1	c)img1	NOUN
ajst-2549	65	13	figure	figure	NOUN
ajst-2549	65	14	5	5	NUM
ajst-2549	65	15	.	.	PUNCT
ajst-2549	65	16	copy	copy	NOUN
ajst-2549	65	17	-	-	PUNCT
ajst-2549	65	18	paste	paste	NOUN
ajst-2549	65	19	method	method	NOUN
ajst-2549	65	20	demonstration	demonstration	NOUN
ajst-2549	65	21	3.3.2	3.3.2	NUM
ajst-2549	65	22	.	.	PUNCT
ajst-2549	66	1	cosine	cosine	NOUN
ajst-2549	66	2	learning	learning	NOUN
ajst-2549	66	3	rate	rate	NOUN
ajst-2549	66	4	descent	descent	NOUN
ajst-2549	66	5	strategy	strategy	NOUN
ajst-2549	66	6	in	in	ADP
ajst-2549	66	7	this	this	DET
ajst-2549	66	8	paper	paper	NOUN
ajst-2549	67	1	,	,	PUNCT
ajst-2549	67	2	the	the	DET
ajst-2549	67	3	cosine	cosine	NOUN
ajst-2549	67	4	learning	learn	VERB
ajst-2549	67	5	rate	rate	NOUN
ajst-2549	67	6	drop	drop	NOUN
ajst-2549	67	7	strategy	strategy	NOUN
ajst-2549	67	8	and	and	CCONJ
ajst-2549	67	9	the	the	DET
ajst-2549	67	10	learning	learning	NOUN
ajst-2549	67	11	rate	rate	NOUN
ajst-2549	67	12	warm	warm	ADJ
ajst-2549	67	13	-	-	PUNCT
ajst-2549	67	14	up	up	ADP
ajst-2549	67	15	strategy	strategy	NOUN
ajst-2549	67	16	are	be	AUX
ajst-2549	67	17	used	use	VERB
ajst-2549	67	18	to	to	PART
ajst-2549	67	19	control	control	VERB
ajst-2549	67	20	the	the	DET
ajst-2549	67	21	learning	learning	NOUN
ajst-2549	67	22	rate	rate	NOUN
ajst-2549	67	23	parameters	parameter	NOUN
ajst-2549	67	24	in	in	ADP
ajst-2549	67	25	training	training	NOUN
ajst-2549	67	26	.	.	PUNCT
ajst-2549	68	1	in	in	ADP
ajst-2549	68	2	the	the	DET
ajst-2549	68	3	early	early	ADJ
ajst-2549	68	4	stage	stage	NOUN
ajst-2549	68	5	of	of	ADP
ajst-2549	68	6	training	training	NOUN
ajst-2549	68	7	,	,	PUNCT
ajst-2549	68	8	a	a	DET
ajst-2549	68	9	smaller	small	ADJ
ajst-2549	68	10	learning	learning	NOUN
ajst-2549	68	11	rate	rate	NOUN
ajst-2549	68	12	is	be	AUX
ajst-2549	68	13	used	use	VERB
ajst-2549	68	14	to	to	PART
ajst-2549	68	15	train	train	VERB
ajst-2549	68	16	the	the	DET
ajst-2549	68	17	model	model	NOUN
ajst-2549	68	18	through	through	ADP
ajst-2549	68	19	the	the	DET
ajst-2549	68	20	learning	learning	NOUN
ajst-2549	68	21	rate	rate	NOUN
ajst-2549	68	22	warm	warm	ADJ
ajst-2549	68	23	-	-	PUNCT
ajst-2549	68	24	up	up	ADP
ajst-2549	68	25	strategy	strategy	NOUN
ajst-2549	68	26	,	,	PUNCT
ajst-2549	68	27	and	and	CCONJ
ajst-2549	68	28	the	the	DET
ajst-2549	68	29	learning	learning	NOUN
ajst-2549	68	30	rate	rate	NOUN
ajst-2549	68	31	is	be	AUX
ajst-2549	68	32	gradually	gradually	ADV
ajst-2549	68	33	increased	increase	VERB
ajst-2549	68	34	to	to	ADP
ajst-2549	68	35	the	the	DET
ajst-2549	68	36	specified	specified	ADJ
ajst-2549	68	37	epoch	epoch	NOUN
ajst-2549	68	38	;	;	PUNCT
ajst-2549	68	39	as	as	SCONJ
ajst-2549	68	40	the	the	DET
ajst-2549	68	41	model	model	NOUN
ajst-2549	68	42	gradually	gradually	ADV
ajst-2549	68	43	converges	converge	VERB
ajst-2549	68	44	,	,	PUNCT
ajst-2549	68	45	the	the	DET
ajst-2549	68	46	cosine	cosine	NOUN
ajst-2549	68	47	learning	learn	VERB
ajst-2549	68	48	rate	rate	NOUN
ajst-2549	68	49	reduction	reduction	NOUN
ajst-2549	68	50	strategy	strategy	NOUN
ajst-2549	68	51	is	be	AUX
ajst-2549	68	52	used	use	VERB
ajst-2549	68	53	in	in	ADP
ajst-2549	68	54	the	the	DET
ajst-2549	68	55	later	later	ADJ
ajst-2549	68	56	stage	stage	NOUN
ajst-2549	68	57	of	of	ADP
ajst-2549	68	58	training	training	NOUN
ajst-2549	68	59	to	to	PART
ajst-2549	68	60	gradually	gradually	ADV
ajst-2549	68	61	increase	increase	VERB
ajst-2549	68	62	the	the	DET
ajst-2549	68	63	learning	learning	NOUN
ajst-2549	68	64	rate	rate	NOUN
ajst-2549	68	65	.	.	PUNCT
ajst-2549	69	1	falling	fall	VERB
ajst-2549	69	2	to	to	ADP
ajst-2549	69	3	0	0	NUM
ajst-2549	69	4	,	,	PUNCT
ajst-2549	69	5	the	the	DET
ajst-2549	69	6	falling	fall	VERB
ajst-2549	69	7	state	state	NOUN
ajst-2549	69	8	is	be	AUX
ajst-2549	69	9	a	a	DET
ajst-2549	69	10	cosine	cosine	NOUN
ajst-2549	69	11	curve	curve	NOUN
ajst-2549	69	12	.	.	PUNCT
ajst-2549	70	1	this	this	DET
ajst-2549	70	2	strategy	strategy	NOUN
ajst-2549	70	3	can	can	AUX
ajst-2549	70	4	better	well	ADV
ajst-2549	70	5	improve	improve	VERB
ajst-2549	70	6	the	the	DET
ajst-2549	70	7	stability	stability	NOUN
ajst-2549	70	8	in	in	ADP
ajst-2549	70	9	the	the	DET
ajst-2549	70	10	early	early	ADJ
ajst-2549	70	11	and	and	CCONJ
ajst-2549	70	12	late	late	ADJ
ajst-2549	70	13	stages	stage	NOUN
ajst-2549	70	14	of	of	ADP
ajst-2549	70	15	training	training	NOUN
ajst-2549	70	16	,	,	PUNCT
ajst-2549	70	17	and	and	CCONJ
ajst-2549	70	18	effectively	effectively	ADV
ajst-2549	70	19	improve	improve	VERB
ajst-2549	70	20	the	the	DET
ajst-2549	70	21	convergence	convergence	NOUN
ajst-2549	70	22	accuracy	accuracy	NOUN
ajst-2549	70	23	of	of	ADP
ajst-2549	70	24	the	the	DET
ajst-2549	70	25	final	final	ADJ
ajst-2549	70	26	model	model	NOUN
ajst-2549	70	27	.	.	PUNCT
ajst-2549	71	1	4	4	X
ajst-2549	71	2	.	.	X
ajst-2549	71	3	experiment	experiment	NOUN
ajst-2549	71	4	and	and	CCONJ
ajst-2549	71	5	result	result	VERB
ajst-2549	71	6	analysis	analysis	NOUN
ajst-2549	71	7	4.1	4.1	NUM
ajst-2549	71	8	.	.	PUNCT
ajst-2549	72	1	data	datum	NOUN
ajst-2549	72	2	set	set	VERB
ajst-2549	72	3	creation	creation	NOUN
ajst-2549	72	4	to	to	PART
ajst-2549	72	5	realize	realize	VERB
ajst-2549	72	6	the	the	DET
ajst-2549	72	7	improved	improved	ADJ
ajst-2549	72	8	angle	angle	NOUN
ajst-2549	72	9	steel	steel	NOUN
ajst-2549	72	10	character	character	NOUN
ajst-2549	72	11	detection	detection	NOUN
ajst-2549	72	12	algorithm	algorithm	NOUN
ajst-2549	72	13	based	base	VERB
ajst-2549	72	14	on	on	ADP
ajst-2549	72	15	dbnet	dbnet	NOUN
ajst-2549	72	16	,	,	PUNCT
ajst-2549	72	17	the	the	DET
ajst-2549	72	18	real	real	ADJ
ajst-2549	72	19	industrial	industrial	ADJ
ajst-2549	72	20	angle	angle	NOUN
ajst-2549	72	21	steel	steel	NOUN
ajst-2549	72	22	image	image	NOUN
ajst-2549	72	23	is	be	AUX
ajst-2549	72	24	required	require	VERB
ajst-2549	72	25	to	to	PART
ajst-2549	72	26	be	be	AUX
ajst-2549	72	27	used	use	VERB
ajst-2549	72	28	as	as	ADP
ajst-2549	72	29	a	a	DET
ajst-2549	72	30	data	datum	NOUN
ajst-2549	72	31	set	set	VERB
ajst-2549	72	32	by	by	ADP
ajst-2549	72	33	marking	mark	VERB
ajst-2549	72	34	the	the	DET
ajst-2549	72	35	character	character	NOUN
ajst-2549	72	36	area	area	NOUN
ajst-2549	72	37	for	for	ADP
ajst-2549	72	38	algorithm	algorithm	NOUN
ajst-2549	72	39	training	training	NOUN
ajst-2549	72	40	and	and	CCONJ
ajst-2549	72	41	verification	verification	NOUN
ajst-2549	72	42	.	.	PUNCT
ajst-2549	73	1	the	the	DET
ajst-2549	73	2	data	datum	NOUN
ajst-2549	73	3	set	set	VERB
ajst-2549	73	4	in	in	ADP
ajst-2549	73	5	this	this	DET
ajst-2549	73	6	paper	paper	NOUN
ajst-2549	73	7	is	be	AUX
ajst-2549	73	8	collected	collect	VERB
ajst-2549	73	9	by	by	ADP
ajst-2549	73	10	using	use	VERB
ajst-2549	73	11	industrial	industrial	ADJ
ajst-2549	73	12	cameras	camera	NOUN
ajst-2549	73	13	to	to	PART
ajst-2549	73	14	take	take	VERB
ajst-2549	73	15	pictures	picture	NOUN
ajst-2549	73	16	at	at	ADP
ajst-2549	73	17	fixed	fix	VERB
ajst-2549	73	18	points	point	NOUN
ajst-2549	73	19	in	in	ADP
ajst-2549	73	20	an	an	DET
ajst-2549	73	21	industrial	industrial	ADJ
ajst-2549	73	22	environment	environment	NOUN
ajst-2549	73	23	,	,	PUNCT
ajst-2549	73	24	and	and	CCONJ
ajst-2549	73	25	it	it	PRON
ajst-2549	73	26	is	be	AUX
ajst-2549	73	27	marked	mark	VERB
ajst-2549	73	28	by	by	ADP
ajst-2549	73	29	baidu	baidu	PROPN
ajst-2549	73	30	's	's	PART
ajst-2549	73	31	open	open	ADJ
ajst-2549	73	32	source	source	NOUN
ajst-2549	73	33	tool	tool	NOUN
ajst-2549	73	34	ppocrlabel	ppocrlabel	NOUN
ajst-2549	73	35	.	.	PUNCT
ajst-2549	74	1	the	the	DET
ajst-2549	74	2	angle	angle	NOUN
ajst-2549	74	3	steel	steel	NOUN
ajst-2549	74	4	dataset	dataset	NOUN
ajst-2549	74	5	consists	consist	VERB
ajst-2549	74	6	of	of	ADP
ajst-2549	74	7	500	500	NUM
ajst-2549	74	8	images	image	NOUN
ajst-2549	74	9	in	in	ADP
ajst-2549	74	10	total	total	NOUN
ajst-2549	74	11	,	,	PUNCT
ajst-2549	74	12	of	of	ADP
ajst-2549	74	13	which	which	PRON
ajst-2549	74	14	400	400	NUM
ajst-2549	74	15	are	be	AUX
ajst-2549	74	16	used	use	VERB
ajst-2549	74	17	for	for	ADP
ajst-2549	74	18	training	training	NOUN
ajst-2549	74	19	and	and	CCONJ
ajst-2549	74	20	100	100	NUM
ajst-2549	74	21	are	be	AUX
ajst-2549	74	22	used	use	VERB
ajst-2549	74	23	for	for	ADP
ajst-2549	74	24	algorithm	algorithm	NOUN
ajst-2549	74	25	performance	performance	NOUN
ajst-2549	74	26	evaluation	evaluation	NOUN
ajst-2549	74	27	.	.	PUNCT
ajst-2549	75	1	each	each	DET
ajst-2549	75	2	angle	angle	NOUN
ajst-2549	75	3	steel	steel	NOUN
ajst-2549	75	4	character	character	NOUN
ajst-2549	75	5	area	area	NOUN
ajst-2549	75	6	frame	frame	NOUN
ajst-2549	75	7	is	be	AUX
ajst-2549	75	8	manually	manually	ADV
ajst-2549	75	9	marked	mark	VERB
ajst-2549	75	10	,	,	PUNCT
ajst-2549	75	11	and	and	CCONJ
ajst-2549	75	12	there	there	PRON
ajst-2549	75	13	is	be	VERB
ajst-2549	75	14	an	an	DET
ajst-2549	75	15	average	average	NOUN
ajst-2549	75	16	of	of	ADP
ajst-2549	75	17	about	about	ADV
ajst-2549	75	18	1	1	NUM
ajst-2549	75	19	character	character	NOUN
ajst-2549	75	20	area	area	NOUN
ajst-2549	75	21	frame	frame	NOUN
ajst-2549	75	22	in	in	ADP
ajst-2549	75	23	each	each	DET
ajst-2549	75	24	angle	angle	NOUN
ajst-2549	75	25	steel	steel	NOUN
ajst-2549	75	26	image	image	NOUN
ajst-2549	75	27	.	.	PUNCT
ajst-2549	76	1	the	the	DET
ajst-2549	76	2	information	information	NOUN
ajst-2549	76	3	statistics	statistic	NOUN
ajst-2549	76	4	of	of	ADP
ajst-2549	76	5	angle	angle	NOUN
ajst-2549	76	6	steel	steel	NOUN
ajst-2549	76	7	character	character	NOUN
ajst-2549	76	8	detection	detection	NOUN
ajst-2549	76	9	dataset	dataset	VERB
ajst-2549	76	10	are	be	AUX
ajst-2549	76	11	shown	show	VERB
ajst-2549	76	12	in	in	ADP
ajst-2549	76	13	table	table	NOUN
ajst-2549	76	14	1	1	NUM
ajst-2549	76	15	.	.	PUNCT
ajst-2549	76	16	table	table	NOUN
ajst-2549	76	17	1	1	NUM
ajst-2549	76	18	.	.	PUNCT
ajst-2549	76	19	angle	angle	NOUN
ajst-2549	76	20	steel	steel	NOUN
ajst-2549	76	21	character	character	NOUN
ajst-2549	76	22	detection	detection	NOUN
ajst-2549	76	23	data	datum	NOUN
ajst-2549	76	24	set	set	VERB
ajst-2549	76	25	information	information	NOUN
ajst-2549	76	26	statistics	statistic	NOUN
ajst-2549	76	27	dataset	dataset	VERB
ajst-2549	76	28	classification	classification	NOUN
ajst-2549	76	29	number	number	NOUN
ajst-2549	76	30	of	of	ADP
ajst-2549	76	31	images	image	NOUN
ajst-2549	76	32	number	number	NOUN
ajst-2549	76	33	of	of	ADP
ajst-2549	76	34	character	character	NOUN
ajst-2549	76	35	area	area	NOUN
ajst-2549	76	36	boxes	box	VERB
ajst-2549	76	37	training	training	NOUN
ajst-2549	76	38	set	set	VERB
ajst-2549	76	39	400	400	NUM
ajst-2549	76	40	471	471	NUM
ajst-2549	76	41	validation	validation	NOUN
ajst-2549	76	42	set	set	VERB
ajst-2549	76	43	100	100	NUM
ajst-2549	76	44	105	105	NUM
ajst-2549	76	45	total	total	NOUN
ajst-2549	76	46	500	500	NUM
ajst-2549	76	47	576	576	NUM
ajst-2549	76	48	4.2	4.2	NUM
ajst-2549	76	49	.	.	PUNCT
ajst-2549	77	1	experimental	experimental	ADJ
ajst-2549	77	2	platform	platform	NOUN
ajst-2549	77	3	and	and	CCONJ
ajst-2549	77	4	training	training	NOUN
ajst-2549	77	5	details	detail	NOUN
ajst-2549	77	6	the	the	DET
ajst-2549	77	7	experimental	experimental	ADJ
ajst-2549	77	8	environment	environment	NOUN
ajst-2549	77	9	is	be	AUX
ajst-2549	77	10	ubuntu	ubuntu	ADJ
ajst-2549	77	11	smp	smp	PROPN
ajst-2549	77	12	16.04.1	16.04.1	NUM
ajst-2549	77	13	operating	operating	NOUN
ajst-2549	77	14	system	system	NOUN
ajst-2549	77	15	,	,	PUNCT
ajst-2549	77	16	1	1	NUM
ajst-2549	77	17	4	4	NUM
ajst-2549	77	18	-	-	PUNCT
ajst-2549	77	19	core	core	NOUN
ajst-2549	77	20	intel(r	intel(r	PROPN
ajst-2549	77	21	)	)	PUNCT
ajst-2549	77	22	xeon(r	xeon(r	PROPN
ajst-2549	77	23	)	)	PUNCT
ajst-2549	77	24	gold	gold	NOUN
ajst-2549	77	25	6148	6148	NUM
ajst-2549	77	26	@	@	ADP
ajst-2549	77	27	2.40ghz	2.40ghz	NUM
ajst-2549	77	28	processor	processor	NOUN
ajst-2549	77	29	,	,	PUNCT
ajst-2549	77	30	1	1	NUM
ajst-2549	77	31	32	32	NUM
ajst-2549	77	32	g	g	PROPN
ajst-2549	77	33	nvidia	nvidia	PROPN
ajst-2549	77	34	v100	v100	PROPN
ajst-2549	77	35	tensor	tensor	NOUN
ajst-2549	77	36	core	core	NOUN
ajst-2549	77	37	graphics	graphic	NOUN
ajst-2549	77	38	card	card	NOUN
ajst-2549	77	39	,	,	PUNCT
ajst-2549	77	40	deep	deep	ADJ
ajst-2549	77	41	learning	learning	NOUN
ajst-2549	77	42	framework	framework	NOUN
ajst-2549	77	43	paddlepaddle	paddlepaddle	VERB
ajst-2549	77	44	2.2.2	2.2.2	NUM
ajst-2549	77	45	version	version	NOUN
ajst-2549	77	46	,	,	PUNCT
ajst-2549	77	47	the	the	DET
ajst-2549	77	48	code	code	NOUN
ajst-2549	77	49	running	running	NOUN
ajst-2549	77	50	environment	environment	NOUN
ajst-2549	77	51	is	be	AUX
ajst-2549	77	52	python	python	PROPN
ajst-2549	77	53	3.7.the	3.7.the	DET
ajst-2549	77	54	angle	angle	NOUN
ajst-2549	77	55	steel	steel	NOUN
ajst-2549	77	56	data	datum	NOUN
ajst-2549	77	57	set	set	VERB
ajst-2549	77	58	collected	collect	VERB
ajst-2549	77	59	in	in	ADP
ajst-2549	77	60	this	this	DET
ajst-2549	77	61	experiment	experiment	NOUN
ajst-2549	77	62	is	be	AUX
ajst-2549	77	63	relatively	relatively	ADV
ajst-2549	77	64	small	small	ADJ
ajst-2549	77	65	,	,	PUNCT
ajst-2549	77	66	so	so	SCONJ
ajst-2549	77	67	this	this	DET
ajst-2549	77	68	paper	paper	NOUN
ajst-2549	77	69	uses	use	VERB
ajst-2549	77	70	the	the	DET
ajst-2549	77	71	transfer	transfer	NOUN
ajst-2549	77	72	learning	learning	NOUN
ajst-2549	77	73	method	method	NOUN
ajst-2549	77	74	[	[	X
ajst-2549	77	75	17	17	NUM
ajst-2549	77	76	]	]	PUNCT
ajst-2549	77	77	,	,	PUNCT
ajst-2549	77	78	and	and	CCONJ
ajst-2549	77	79	uses	use	VERB
ajst-2549	77	80	the	the	DET
ajst-2549	77	81	corresponding	correspond	VERB
ajst-2549	77	82	classification	classification	NOUN
ajst-2549	77	83	network	network	NOUN
ajst-2549	77	84	trained	train	VERB
ajst-2549	77	85	on	on	ADP
ajst-2549	77	86	imagenet	imagenet	NOUN
ajst-2549	77	87	as	as	ADP
ajst-2549	77	88	a	a	DET
ajst-2549	77	89	pre	pre	ADJ
ajst-2549	77	90	-	-	ADJ
ajst-2549	77	91	training	training	ADJ
ajst-2549	77	92	model	model	NOUN
ajst-2549	77	93	to	to	PART
ajst-2549	77	94	speed	speed	VERB
ajst-2549	77	95	up	up	ADP
ajst-2549	77	96	the	the	DET
ajst-2549	77	97	training	training	NOUN
ajst-2549	77	98	.	.	PUNCT
ajst-2549	78	1	this	this	DET
ajst-2549	78	2	experiment	experiment	NOUN
ajst-2549	78	3	is	be	AUX
ajst-2549	78	4	trained	train	VERB
ajst-2549	78	5	for	for	ADP
ajst-2549	78	6	24,000	24,000	NUM
ajst-2549	78	7	iterations	iteration	NOUN
ajst-2549	78	8	,	,	PUNCT
ajst-2549	78	9	using	use	VERB
ajst-2549	78	10	the	the	DET
ajst-2549	78	11	adam	adam	PROPN
ajst-2549	78	12	optimizer	optimizer	NOUN
ajst-2549	78	13	,	,	PUNCT
ajst-2549	78	14	the	the	DET
ajst-2549	78	15	parameters	parameter	NOUN
ajst-2549	78	16	bata1	bata1	PROPN
ajst-2549	78	17	is	be	AUX
ajst-2549	78	18	0.9	0.9	NUM
ajst-2549	78	19	,	,	PUNCT
ajst-2549	78	20	bata2	bata2	NOUN
ajst-2549	78	21	is	be	AUX
ajst-2549	78	22	0.999	0.999	NUM
ajst-2549	78	23	,	,	PUNCT
ajst-2549	78	24	and	and	CCONJ
ajst-2549	78	25	the	the	DET
ajst-2549	78	26	learning	learning	NOUN
ajst-2549	78	27	rate	rate	NOUN
ajst-2549	78	28	is	be	AUX
ajst-2549	78	29	set	set	VERB
ajst-2549	78	30	to	to	ADP
ajst-2549	78	31	0.001	0.001	NUM
ajst-2549	78	32	.	.	PUNCT
ajst-2549	79	1	the	the	DET
ajst-2549	79	2	training	training	NOUN
ajst-2549	79	3	image	image	NOUN
ajst-2549	79	4	input	input	NOUN
ajst-2549	79	5	is	be	AUX
ajst-2549	79	6	adjusted	adjust	VERB
ajst-2549	79	7	to	to	ADP
ajst-2549	79	8	a	a	DET
ajst-2549	79	9	size	size	NOUN
ajst-2549	79	10	of	of	ADP
ajst-2549	79	11	960	960	NUM
ajst-2549	79	12	×	×	PROPN
ajst-2549	79	13	960	960	NUM
ajst-2549	79	14	pixel	pixel	NOUN
ajst-2549	79	15	.	.	PUNCT
ajst-2549	80	1	4.3	4.3	NUM
ajst-2549	80	2	.	.	PUNCT
ajst-2549	80	3	backbone	backbone	PROPN
ajst-2549	80	4	network	network	PROPN
ajst-2549	80	5	comparison	comparison	NOUN
ajst-2549	80	6	experiment	experiment	NOUN
ajst-2549	80	7	for	for	ADP
ajst-2549	80	8	this	this	DET
ajst-2549	80	9	angle	angle	NOUN
ajst-2549	80	10	steel	steel	NOUN
ajst-2549	80	11	data	datum	NOUN
ajst-2549	80	12	set	set	VERB
ajst-2549	80	13	,	,	PUNCT
ajst-2549	80	14	in	in	ADP
ajst-2549	80	15	order	order	NOUN
ajst-2549	80	16	to	to	PART
ajst-2549	80	17	select	select	VERB
ajst-2549	80	18	a	a	DET
ajst-2549	80	19	suitable	suitable	ADJ
ajst-2549	80	20	classification	classification	NOUN
ajst-2549	80	21	network	network	NOUN
ajst-2549	80	22	as	as	ADP
ajst-2549	80	23	the	the	DET
ajst-2549	80	24	backbone	backbone	NOUN
ajst-2549	80	25	network	network	NOUN
ajst-2549	80	26	,	,	PUNCT
ajst-2549	80	27	this	this	DET
ajst-2549	80	28	experiment	experiment	NOUN
ajst-2549	80	29	is	be	AUX
ajst-2549	80	30	based	base	VERB
ajst-2549	80	31	on	on	ADP
ajst-2549	80	32	the	the	DET
ajst-2549	80	33	dbnet	dbnet	NOUN
ajst-2549	80	34	model	model	NOUN
ajst-2549	80	35	to	to	PART
ajst-2549	80	36	train	train	VERB
ajst-2549	80	37	three	three	NUM
ajst-2549	80	38	classification	classification	NOUN
ajst-2549	80	39	networks	network	NOUN
ajst-2549	80	40	,	,	PUNCT
ajst-2549	80	41	namely	namely	ADV
ajst-2549	80	42	mobilenetv3	mobilenetv3	PROPN
ajst-2549	80	43	,	,	PUNCT
ajst-2549	80	44	resnet18	resnet18	NOUN
ajst-2549	80	45	and	and	CCONJ
ajst-2549	80	46	resnet50	resnet50	NOUN
ajst-2549	80	47	.	.	PUNCT
ajst-2549	81	1	the	the	DET
ajst-2549	81	2	comparison	comparison	NOUN
ajst-2549	81	3	of	of	ADP
ajst-2549	81	4	the	the	DET
ajst-2549	81	5	training	training	NOUN
ajst-2549	81	6	loss	loss	NOUN
ajst-2549	81	7	of	of	ADP
ajst-2549	81	8	different	different	ADJ
ajst-2549	81	9	backbone	backbone	NOUN
ajst-2549	81	10	networks	network	NOUN
ajst-2549	81	11	is	be	AUX
ajst-2549	81	12	shown	show	VERB
ajst-2549	81	13	in	in	ADP
ajst-2549	81	14	fig	fig	NOUN
ajst-2549	81	15	.	.	PUNCT
ajst-2549	82	1	6	6	NUM
ajst-2549	82	2	.	.	X
ajst-2549	82	3	it	it	PRON
ajst-2549	82	4	can	can	AUX
ajst-2549	82	5	be	be	AUX
ajst-2549	82	6	seen	see	VERB
ajst-2549	82	7	that	that	SCONJ
ajst-2549	82	8	mobilenetv3	mobilenetv3	PROPN
ajst-2549	82	9	and	and	CCONJ
ajst-2549	82	10	resnet18	resnet18	NOUN
ajst-2549	82	11	with	with	ADP
ajst-2549	82	12	shallow	shallow	ADJ
ajst-2549	82	13	networks	network	NOUN
ajst-2549	82	14	converge	converge	VERB
ajst-2549	82	15	faster	fast	ADV
ajst-2549	82	16	in	in	ADP
ajst-2549	82	17	dbnet	dbnet	NOUN
ajst-2549	82	18	,	,	PUNCT
ajst-2549	82	19	and	and	CCONJ
ajst-2549	82	20	the	the	DET
ajst-2549	82	21	convergence	convergence	NOUN
ajst-2549	82	22	is	be	AUX
ajst-2549	82	23	not	not	PART
ajst-2549	82	24	much	much	ADV
ajst-2549	82	25	different	different	ADJ
ajst-2549	82	26	.	.	PUNCT
ajst-2549	83	1	figure	figure	VERB
ajst-2549	83	2	6	6	NUM
ajst-2549	83	3	.	.	PUNCT
ajst-2549	83	4	comparison	comparison	NOUN
ajst-2549	83	5	of	of	ADP
ajst-2549	83	6	training	training	NOUN
ajst-2549	83	7	losses	loss	NOUN
ajst-2549	83	8	of	of	ADP
ajst-2549	83	9	different	different	ADJ
ajst-2549	83	10	backbone	backbone	NOUN
ajst-2549	83	11	networks	network	NOUN
ajst-2549	83	12	in	in	ADP
ajst-2549	83	13	addition	addition	NOUN
ajst-2549	83	14	to	to	ADP
ajst-2549	83	15	the	the	DET
ajst-2549	83	16	analysis	analysis	NOUN
ajst-2549	83	17	of	of	ADP
ajst-2549	83	18	the	the	DET
ajst-2549	83	19	loss	loss	NOUN
ajst-2549	83	20	situation	situation	NOUN
ajst-2549	83	21	,	,	PUNCT
ajst-2549	83	22	the	the	DET
ajst-2549	83	23	most	most	ADV
ajst-2549	83	24	important	important	ADJ
ajst-2549	83	25	thing	thing	NOUN
ajst-2549	83	26	is	be	AUX
ajst-2549	83	27	the	the	DET
ajst-2549	83	28	performance	performance	NOUN
ajst-2549	83	29	indicators	indicator	NOUN
ajst-2549	83	30	of	of	ADP
ajst-2549	83	31	different	different	ADJ
ajst-2549	83	32	classification	classification	NOUN
ajst-2549	83	33	networks	network	NOUN
ajst-2549	83	34	in	in	ADP
ajst-2549	83	35	the	the	DET
ajst-2549	83	36	dbnet	dbnet	NOUN
ajst-2549	83	37	algorithm	algorithm	NOUN
ajst-2549	83	38	,	,	PUNCT
ajst-2549	83	39	which	which	PRON
ajst-2549	83	40	are	be	AUX
ajst-2549	83	41	evaluated	evaluate	VERB
ajst-2549	83	42	using	use	VERB
ajst-2549	83	43	the	the	DET
ajst-2549	83	44	angle	angle	NOUN
ajst-2549	83	45	steel	steel	NOUN
ajst-2549	83	46	data	datum	NOUN
ajst-2549	83	47	set	set	VERB
ajst-2549	83	48	.	.	PUNCT
ajst-2549	84	1	the	the	DET
ajst-2549	84	2	experimental	experimental	ADJ
ajst-2549	84	3	comparison	comparison	NOUN
ajst-2549	84	4	results	result	NOUN
ajst-2549	84	5	of	of	ADP
ajst-2549	84	6	different	different	ADJ
ajst-2549	84	7	backbone	backbone	NOUN
ajst-2549	84	8	networks	network	NOUN
ajst-2549	84	9	are	be	AUX
ajst-2549	84	10	shown	show	VERB
ajst-2549	84	11	in	in	ADP
ajst-2549	84	12	table	table	NOUN
ajst-2549	84	13	2	2	NUM
ajst-2549	84	14	.	.	PUNCT
ajst-2549	85	1	it	it	PRON
ajst-2549	85	2	can	can	AUX
ajst-2549	85	3	be	be	AUX
ajst-2549	85	4	seen	see	VERB
ajst-2549	85	5	from	from	ADP
ajst-2549	85	6	table	table	NOUN
ajst-2549	85	7	2	2	NUM
ajst-2549	85	8	that	that	PRON
ajst-2549	85	9	mobilenetv3	mobilenetv3	PROPN
ajst-2549	85	10	is	be	AUX
ajst-2549	85	11	lighter	light	ADJ
ajst-2549	85	12	than	than	ADP
ajst-2549	85	13	resnet18	resnet18	NOUN
ajst-2549	85	14	under	under	ADP
ajst-2549	85	15	the	the	DET
ajst-2549	85	16	condition	condition	NOUN
ajst-2549	85	17	of	of	ADP
ajst-2549	85	18	similar	similar	ADJ
ajst-2549	85	19	convergence	convergence	NOUN
ajst-2549	85	20	speed	speed	NOUN
ajst-2549	85	21	,	,	PUNCT
ajst-2549	85	22	and	and	CCONJ
ajst-2549	85	23	the	the	DET
ajst-2549	85	24	evaluation	evaluation	NOUN
ajst-2549	85	25	index	index	NOUN
ajst-2549	85	26	is	be	AUX
ajst-2549	85	27	higher	high	ADJ
ajst-2549	85	28	,	,	PUNCT
ajst-2549	85	29	so	so	CCONJ
ajst-2549	85	30	mobilenetv3	mobilenetv3	PROPN
ajst-2549	85	31	is	be	AUX
ajst-2549	85	32	used	use	VERB
ajst-2549	85	33	as	as	ADP
ajst-2549	85	34	the	the	DET
ajst-2549	85	35	backbone	backbone	NOUN
ajst-2549	85	36	network	network	NOUN
ajst-2549	85	37	in	in	ADP
ajst-2549	85	38	the	the	DET
ajst-2549	85	39	experiments	experiment	NOUN
ajst-2549	85	40	later	later	ADV
ajst-2549	85	41	in	in	ADP
ajst-2549	85	42	this	this	DET
ajst-2549	85	43	paper	paper	NOUN
ajst-2549	85	44	.	.	PUNCT
ajst-2549	86	1	68	68	NUM
ajst-2549	86	2	table	table	NOUN
ajst-2549	86	3	2	2	NUM
ajst-2549	86	4	.	.	PUNCT
ajst-2549	86	5	comparison	comparison	NOUN
ajst-2549	86	6	experiment	experiment	NOUN
ajst-2549	86	7	of	of	ADP
ajst-2549	86	8	different	different	ADJ
ajst-2549	86	9	backbone	backbone	NOUN
ajst-2549	86	10	networks	network	NOUN
ajst-2549	86	11	backbone	backbone	NOUN
ajst-2549	86	12	precision	precision	NOUN
ajst-2549	86	13	rcall	rcall	NOUN
ajst-2549	86	14	f1	f1	NOUN
ajst-2549	86	15	-	-	PUNCT
ajst-2549	86	16	score	score	NOUN
ajst-2549	86	17	params(m	params(m	PROPN
ajst-2549	86	18	)	)	PUNCT
ajst-2549	86	19	mobilenetv3	mobilenetv3	PROPN
ajst-2549	86	20	76.03	76.03	NUM
ajst-2549	86	21	87.62	87.62	NUM
ajst-2549	86	22	81.42	81.42	NUM
ajst-2549	86	23	5.70	5.70	NUM
ajst-2549	86	24	resnet18	resnet18	NOUN
ajst-2549	86	25	71.85	71.85	NUM
ajst-2549	86	26	92.38	92.38	NUM
ajst-2549	86	27	80.83	80.83	NUM
ajst-2549	86	28	47.20	47.20	NUM
ajst-2549	86	29	resnet50	resnet50	NOUN
ajst-2549	86	30	77.57	77.57	NUM
ajst-2549	86	31	79.05	79.05	NUM
ajst-2549	86	32	78.30	78.30	NUM
ajst-2549	86	33	97.20	97.20	NUM
ajst-2549	86	34	4.4	4.4	NUM
ajst-2549	86	35	.	.	PUNCT
ajst-2549	87	1	ablation	ablation	NOUN
ajst-2549	87	2	experiment	experiment	NOUN
ajst-2549	87	3	in	in	ADP
ajst-2549	87	4	order	order	NOUN
ajst-2549	87	5	to	to	PART
ajst-2549	87	6	verify	verify	VERB
ajst-2549	87	7	the	the	DET
ajst-2549	87	8	effectiveness	effectiveness	NOUN
ajst-2549	87	9	of	of	ADP
ajst-2549	87	10	the	the	DET
ajst-2549	87	11	algorithm	algorithm	NOUN
ajst-2549	87	12	improvement	improvement	NOUN
ajst-2549	87	13	and	and	CCONJ
ajst-2549	87	14	the	the	DET
ajst-2549	87	15	importance	importance	NOUN
ajst-2549	87	16	of	of	ADP
ajst-2549	87	17	each	each	DET
ajst-2549	87	18	improvement	improvement	NOUN
ajst-2549	87	19	point	point	NOUN
ajst-2549	87	20	to	to	ADP
ajst-2549	87	21	the	the	DET
ajst-2549	87	22	algorithm	algorithm	NOUN
ajst-2549	87	23	,	,	PUNCT
ajst-2549	87	24	this	this	DET
ajst-2549	87	25	paper	paper	NOUN
ajst-2549	87	26	conducts	conduct	VERB
ajst-2549	87	27	ablation	ablation	NOUN
ajst-2549	87	28	experiments	experiment	NOUN
ajst-2549	87	29	on	on	ADP
ajst-2549	87	30	the	the	DET
ajst-2549	87	31	improvement	improvement	NOUN
ajst-2549	87	32	points	point	NOUN
ajst-2549	87	33	.	.	PUNCT
ajst-2549	88	1	the	the	DET
ajst-2549	88	2	results	result	NOUN
ajst-2549	88	3	of	of	ADP
ajst-2549	88	4	the	the	DET
ajst-2549	88	5	algorithm	algorithm	NOUN
ajst-2549	88	6	improvement	improvement	NOUN
ajst-2549	88	7	ablation	ablation	NOUN
ajst-2549	88	8	experiments	experiment	NOUN
ajst-2549	88	9	are	be	AUX
ajst-2549	88	10	shown	show	VERB
ajst-2549	88	11	in	in	ADP
ajst-2549	88	12	table	table	NOUN
ajst-2549	88	13	3	3	NUM
ajst-2549	88	14	.	.	PUNCT
ajst-2549	89	1	among	among	ADP
ajst-2549	89	2	them	they	PRON
ajst-2549	89	3	,	,	PUNCT
ajst-2549	89	4	strategy	strategy	NOUN
ajst-2549	89	5	a	a	PRON
ajst-2549	89	6	is	be	AUX
ajst-2549	89	7	to	to	PART
ajst-2549	89	8	replace	replace	VERB
ajst-2549	89	9	the	the	DET
ajst-2549	89	10	se	se	PROPN
ajst-2549	89	11	attention	attention	NOUN
ajst-2549	89	12	mechanism	mechanism	NOUN
ajst-2549	89	13	in	in	ADP
ajst-2549	89	14	the	the	DET
ajst-2549	89	15	backbone	backbone	NOUN
ajst-2549	89	16	network	network	NOUN
ajst-2549	89	17	with	with	ADP
ajst-2549	89	18	the	the	DET
ajst-2549	89	19	ca	ca	NOUN
ajst-2549	89	20	attention	attention	NOUN
ajst-2549	89	21	mechanism	mechanism	NOUN
ajst-2549	89	22	;	;	PUNCT
ajst-2549	89	23	strategy	strategy	NOUN
ajst-2549	89	24	b	b	PROPN
ajst-2549	89	25	is	be	AUX
ajst-2549	89	26	to	to	PART
ajst-2549	89	27	introduce	introduce	VERB
ajst-2549	89	28	the	the	DET
ajst-2549	89	29	bifpn	bifpn	PROPN
ajst-2549	89	30	structure	structure	NOUN
ajst-2549	89	31	to	to	PART
ajst-2549	89	32	improve	improve	VERB
ajst-2549	89	33	the	the	DET
ajst-2549	89	34	multi	multi	ADJ
ajst-2549	89	35	-	-	ADJ
ajst-2549	89	36	scale	scale	ADJ
ajst-2549	89	37	feature	feature	NOUN
ajst-2549	89	38	network	network	NOUN
ajst-2549	89	39	structure	structure	NOUN
ajst-2549	89	40	;	;	PUNCT
ajst-2549	89	41	strategy	strategy	NOUN
ajst-2549	89	42	c	c	PROPN
ajst-2549	89	43	is	be	AUX
ajst-2549	89	44	to	to	PART
ajst-2549	89	45	improve	improve	VERB
ajst-2549	89	46	the	the	DET
ajst-2549	89	47	non	non	ADJ
ajst-2549	89	48	-	-	ADJ
ajst-2549	89	49	network	network	ADJ
ajst-2549	89	50	structure	structure	NOUN
ajst-2549	89	51	part	part	NOUN
ajst-2549	89	52	,	,	PUNCT
ajst-2549	89	53	and	and	CCONJ
ajst-2549	89	54	perform	perform	VERB
ajst-2549	89	55	copy	copy	NOUN
ajst-2549	89	56	-	-	PUNCT
ajst-2549	89	57	paste	paste	NOUN
ajst-2549	89	58	data	datum	NOUN
ajst-2549	89	59	enhancement	enhancement	NOUN
ajst-2549	89	60	and	and	CCONJ
ajst-2549	89	61	learning	learn	VERB
ajst-2549	89	62	rate	rate	NOUN
ajst-2549	89	63	optimization	optimization	NOUN
ajst-2549	89	64	the	the	DET
ajst-2549	89	65	strategy	strategy	NOUN
ajst-2549	89	66	optimizes	optimize	VERB
ajst-2549	89	67	the	the	DET
ajst-2549	89	68	algorithm	algorithm	NOUN
ajst-2549	89	69	.	.	PUNCT
ajst-2549	90	1	table	table	NOUN
ajst-2549	90	2	3	3	NUM
ajst-2549	90	3	.	.	PUNCT
ajst-2549	90	4	ablation	ablation	NOUN
ajst-2549	90	5	experiment	experiment	NOUN
ajst-2549	90	6	strategy	strategy	NOUN
ajst-2549	90	7	precision	precision	NOUN
ajst-2549	90	8	rcall	rcall	NOUN
ajst-2549	90	9	f1	f1	NOUN
ajst-2549	90	10	-	-	PUNCT
ajst-2549	90	11	score	score	NOUN
ajst-2549	90	12	params(m	params(m	NOUN
ajst-2549	90	13	)	)	PUNCT
ajst-2549	90	14	dbnet	dbnet	NOUN
ajst-2549	90	15	76.03	76.03	NUM
ajst-2549	90	16	87.62	87.62	NUM
ajst-2549	90	17	81.42	81.42	NUM
ajst-2549	90	18	5.70	5.70	NUM
ajst-2549	90	19	dbnet+a+b	dbnet+a+b	NOUN
ajst-2549	90	20	80.67	80.67	NUM
ajst-2549	90	21	91.43	91.43	NUM
ajst-2549	90	22	85.71	85.71	NUM
ajst-2549	90	23	6.90	6.90	NUM
ajst-2549	90	24	dbnet+a+c	dbnet+a+c	ADP
ajst-2549	90	25	92.92	92.92	NUM
ajst-2549	90	26	100.00	100.00	NUM
ajst-2549	90	27	96.33	96.33	NUM
ajst-2549	90	28	6.00	6.00	NUM
ajst-2549	90	29	dbnet+b+c	dbnet+b+c	PROPN
ajst-2549	90	30	79.34	79.34	NUM
ajst-2549	90	31	91.43	91.43	NUM
ajst-2549	90	32	84.96	84.96	NUM
ajst-2549	90	33	6.60	6.60	NUM
ajst-2549	90	34	dbnet+a+b+c(ours	dbnet+a+b+c(our	NOUN
ajst-2549	90	35	)	)	PUNCT
ajst-2549	90	36	98.13	98.13	NUM
ajst-2549	90	37	100.00	100.00	NUM
ajst-2549	90	38	99.06	99.06	NUM
ajst-2549	90	39	6.90	6.90	NUM
ajst-2549	90	40	the	the	DET
ajst-2549	90	41	first	first	ADJ
ajst-2549	90	42	row	row	NOUN
ajst-2549	90	43	of	of	ADP
ajst-2549	90	44	table	table	NOUN
ajst-2549	90	45	3	3	NUM
ajst-2549	90	46	is	be	AUX
ajst-2549	90	47	the	the	DET
ajst-2549	90	48	reproduction	reproduction	NOUN
ajst-2549	90	49	result	result	NOUN
ajst-2549	90	50	of	of	ADP
ajst-2549	90	51	the	the	DET
ajst-2549	90	52	dbnet	dbnet	NOUN
ajst-2549	90	53	algorithm	algorithm	NOUN
ajst-2549	90	54	on	on	ADP
ajst-2549	90	55	the	the	DET
ajst-2549	90	56	angle	angle	NOUN
ajst-2549	90	57	steel	steel	NOUN
ajst-2549	90	58	character	character	NOUN
ajst-2549	90	59	data	datum	NOUN
ajst-2549	90	60	set	set	VERB
ajst-2549	90	61	in	in	ADP
ajst-2549	90	62	this	this	DET
ajst-2549	90	63	experiment	experiment	NOUN
ajst-2549	90	64	.	.	PUNCT
ajst-2549	91	1	as	as	ADP
ajst-2549	91	2	a	a	DET
ajst-2549	91	3	comparison	comparison	NOUN
ajst-2549	91	4	of	of	ADP
ajst-2549	91	5	the	the	DET
ajst-2549	91	6	results	result	NOUN
ajst-2549	91	7	of	of	ADP
ajst-2549	91	8	the	the	DET
ajst-2549	91	9	improved	improved	ADJ
ajst-2549	91	10	algorithm	algorithm	NOUN
ajst-2549	91	11	in	in	ADP
ajst-2549	91	12	this	this	DET
ajst-2549	91	13	paper	paper	NOUN
ajst-2549	91	14	,	,	PUNCT
ajst-2549	91	15	the	the	DET
ajst-2549	91	16	last	last	ADJ
ajst-2549	91	17	row	row	NOUN
ajst-2549	91	18	of	of	ADP
ajst-2549	91	19	table	table	NOUN
ajst-2549	91	20	3	3	NUM
ajst-2549	91	21	is	be	AUX
ajst-2549	91	22	the	the	DET
ajst-2549	91	23	result	result	NOUN
ajst-2549	91	24	of	of	ADP
ajst-2549	91	25	the	the	DET
ajst-2549	91	26	improved	improved	ADJ
ajst-2549	91	27	algorithm	algorithm	NOUN
ajst-2549	91	28	,	,	PUNCT
ajst-2549	91	29	which	which	PRON
ajst-2549	91	30	is	be	AUX
ajst-2549	91	31	used	use	VERB
ajst-2549	91	32	as	as	ADP
ajst-2549	91	33	the	the	DET
ajst-2549	91	34	benchmark	benchmark	NOUN
ajst-2549	91	35	for	for	ADP
ajst-2549	91	36	this	this	DET
ajst-2549	91	37	ablation	ablation	NOUN
ajst-2549	91	38	experiment	experiment	NOUN
ajst-2549	91	39	.	.	PUNCT
ajst-2549	92	1	①a+b+c	①a+b+c	PROPN
ajst-2549	92	2	:	:	PUNCT
ajst-2549	92	3	using	use	VERB
ajst-2549	92	4	all	all	DET
ajst-2549	92	5	improvement	improvement	NOUN
ajst-2549	92	6	strategies	strategy	NOUN
ajst-2549	92	7	,	,	PUNCT
ajst-2549	92	8	the	the	DET
ajst-2549	92	9	indicators	indicator	NOUN
ajst-2549	92	10	have	have	AUX
ajst-2549	92	11	improved	improve	VERB
ajst-2549	92	12	significantly	significantly	ADV
ajst-2549	92	13	,	,	PUNCT
ajst-2549	92	14	reaching	reach	VERB
ajst-2549	92	15	99.06	99.06	NUM
ajst-2549	92	16	%	%	NOUN
ajst-2549	92	17	,	,	PUNCT
ajst-2549	92	18	and	and	CCONJ
ajst-2549	92	19	the	the	DET
ajst-2549	92	20	number	number	NOUN
ajst-2549	92	21	of	of	ADP
ajst-2549	92	22	parameters	parameter	NOUN
ajst-2549	92	23	has	have	AUX
ajst-2549	92	24	reached	reach	VERB
ajst-2549	92	25	6.9	6.9	NUM
ajst-2549	92	26	m.	m.	NOUN
ajst-2549	92	27	②a+b	②a+b	ADP
ajst-2549	92	28	:	:	PUNCT
ajst-2549	92	29	without	without	ADP
ajst-2549	92	30	strategy	strategy	NOUN
ajst-2549	92	31	c	c	PROPN
ajst-2549	92	32	,	,	PUNCT
ajst-2549	92	33	the	the	DET
ajst-2549	92	34	f1	f1	NOUN
ajst-2549	92	35	-	-	PUNCT
ajst-2549	92	36	score	score	NOUN
ajst-2549	92	37	decreased	decrease	VERB
ajst-2549	92	38	by	by	ADP
ajst-2549	92	39	13.35	13.35	NUM
ajst-2549	92	40	%	%	NOUN
ajst-2549	92	41	,	,	PUNCT
ajst-2549	92	42	and	and	CCONJ
ajst-2549	92	43	the	the	DET
ajst-2549	92	44	parameters	parameter	NOUN
ajst-2549	92	45	remained	remain	VERB
ajst-2549	92	46	unchanged	unchanged	ADJ
ajst-2549	92	47	.	.	PUNCT
ajst-2549	93	1	③a+c	③a+c	NOUN
ajst-2549	93	2	:	:	PUNCT
ajst-2549	93	3	without	without	ADP
ajst-2549	93	4	strategy	strategy	NOUN
ajst-2549	93	5	b	b	PROPN
ajst-2549	93	6	,	,	PUNCT
ajst-2549	93	7	the	the	DET
ajst-2549	93	8	f1	f1	NOUN
ajst-2549	93	9	-	-	PUNCT
ajst-2549	93	10	score	score	NOUN
ajst-2549	93	11	decreased	decrease	VERB
ajst-2549	93	12	by	by	ADP
ajst-2549	93	13	2.73	2.73	NUM
ajst-2549	93	14	%	%	NOUN
ajst-2549	93	15	,	,	PUNCT
ajst-2549	93	16	and	and	CCONJ
ajst-2549	93	17	④	④	NUM
ajst-2549	93	18	the	the	DET
ajst-2549	93	19	parameter	parameter	NOUN
ajst-2549	93	20	amount	amount	NOUN
ajst-2549	93	21	decreased	decrease	VERB
ajst-2549	93	22	by	by	ADP
ajst-2549	93	23	0.9	0.9	NUM
ajst-2549	93	24	m.	m.	NOUN
ajst-2549	93	25	b+c	b+c	PROPN
ajst-2549	93	26	:	:	PUNCT
ajst-2549	93	27	without	without	ADP
ajst-2549	93	28	strategy	strategy	NOUN
ajst-2549	93	29	a	a	X
ajst-2549	93	30	,	,	PUNCT
ajst-2549	93	31	the	the	DET
ajst-2549	93	32	f1	f1	NOUN
ajst-2549	93	33	-	-	PUNCT
ajst-2549	93	34	score	score	NOUN
ajst-2549	93	35	decreased	decrease	VERB
ajst-2549	93	36	by	by	ADP
ajst-2549	93	37	14.1	14.1	NUM
ajst-2549	93	38	%	%	NOUN
ajst-2549	93	39	,	,	PUNCT
ajst-2549	93	40	and	and	CCONJ
ajst-2549	93	41	the	the	DET
ajst-2549	93	42	number	number	NOUN
ajst-2549	93	43	of	of	ADP
ajst-2549	93	44	parameters	parameter	NOUN
ajst-2549	93	45	decreased	decrease	VERB
ajst-2549	93	46	by	by	ADP
ajst-2549	93	47	0.3	0.3	NUM
ajst-2549	93	48	m.	m.	NOUN
ajst-2549	93	49	to	to	PART
ajst-2549	93	50	sum	sum	VERB
ajst-2549	93	51	up	up	ADP
ajst-2549	93	52	,	,	PUNCT
ajst-2549	93	53	it	it	PRON
ajst-2549	93	54	can	can	AUX
ajst-2549	93	55	be	be	AUX
ajst-2549	93	56	seen	see	VERB
ajst-2549	93	57	that	that	SCONJ
ajst-2549	93	58	without	without	ADP
ajst-2549	93	59	strategy	strategy	NOUN
ajst-2549	93	60	a	a	X
ajst-2549	93	61	,	,	PUNCT
ajst-2549	93	62	the	the	DET
ajst-2549	93	63	performance	performance	NOUN
ajst-2549	93	64	index	index	NOUN
ajst-2549	93	65	declines	decline	VERB
ajst-2549	93	66	most	most	ADV
ajst-2549	93	67	obviously	obviously	ADV
ajst-2549	93	68	,	,	PUNCT
ajst-2549	93	69	indicating	indicate	VERB
ajst-2549	93	70	that	that	SCONJ
ajst-2549	93	71	strategy	strategy	NOUN
ajst-2549	93	72	a	a	PRON
ajst-2549	93	73	is	be	AUX
ajst-2549	93	74	the	the	DET
ajst-2549	93	75	most	most	ADV
ajst-2549	93	76	important	important	ADJ
ajst-2549	93	77	in	in	ADP
ajst-2549	93	78	algorithm	algorithm	NOUN
ajst-2549	93	79	improvement	improvement	NOUN
ajst-2549	93	80	;	;	PUNCT
ajst-2549	93	81	without	without	ADP
ajst-2549	93	82	strategy	strategy	NOUN
ajst-2549	93	83	b	b	PROPN
ajst-2549	93	84	,	,	PUNCT
ajst-2549	93	85	the	the	DET
ajst-2549	93	86	performance	performance	NOUN
ajst-2549	93	87	declines	decline	VERB
ajst-2549	93	88	the	the	DET
ajst-2549	93	89	least	least	ADJ
ajst-2549	93	90	,	,	PUNCT
ajst-2549	93	91	indicating	indicate	VERB
ajst-2549	93	92	that	that	SCONJ
ajst-2549	93	93	strategy	strategy	NOUN
ajst-2549	93	94	b	b	NOUN
ajst-2549	93	95	has	have	VERB
ajst-2549	93	96	relatively	relatively	ADV
ajst-2549	93	97	little	little	ADJ
ajst-2549	93	98	impact	impact	NOUN
ajst-2549	93	99	on	on	ADP
ajst-2549	93	100	the	the	DET
ajst-2549	93	101	model	model	NOUN
ajst-2549	93	102	;	;	PUNCT
ajst-2549	93	103	strategy	strategy	NOUN
ajst-2549	93	104	c	c	PROPN
ajst-2549	93	105	,	,	PUNCT
ajst-2549	93	106	the	the	DET
ajst-2549	93	107	performance	performance	NOUN
ajst-2549	93	108	drops	drop	VERB
ajst-2549	93	109	more	more	ADV
ajst-2549	93	110	and	and	CCONJ
ajst-2549	93	111	the	the	DET
ajst-2549	93	112	parameter	parameter	NOUN
ajst-2549	93	113	quantity	quantity	NOUN
ajst-2549	93	114	does	do	AUX
ajst-2549	93	115	not	not	PART
ajst-2549	93	116	change	change	VERB
ajst-2549	93	117	,	,	PUNCT
ajst-2549	93	118	indicating	indicate	VERB
ajst-2549	93	119	that	that	SCONJ
ajst-2549	93	120	strategy	strategy	NOUN
ajst-2549	93	121	c	c	PROPN
ajst-2549	93	122	has	have	VERB
ajst-2549	93	123	a	a	DET
ajst-2549	93	124	greater	great	ADJ
ajst-2549	93	125	impact	impact	NOUN
ajst-2549	93	126	on	on	ADP
ajst-2549	93	127	the	the	DET
ajst-2549	93	128	model	model	NOUN
ajst-2549	93	129	.	.	PUNCT
ajst-2549	94	1	4.5	4.5	NUM
ajst-2549	94	2	.	.	PUNCT
ajst-2549	95	1	algorithm	algorithm	PROPN
ajst-2549	95	2	comparison	comparison	NOUN
ajst-2549	95	3	experiment	experiment	NOUN
ajst-2549	95	4	in	in	ADP
ajst-2549	95	5	order	order	NOUN
ajst-2549	95	6	to	to	PART
ajst-2549	95	7	verify	verify	VERB
ajst-2549	95	8	the	the	DET
ajst-2549	95	9	excellence	excellence	NOUN
ajst-2549	95	10	of	of	ADP
ajst-2549	95	11	the	the	DET
ajst-2549	95	12	improved	improved	ADJ
ajst-2549	95	13	algorithm	algorithm	NOUN
ajst-2549	95	14	,	,	PUNCT
ajst-2549	95	15	this	this	DET
ajst-2549	95	16	paper	paper	NOUN
ajst-2549	95	17	selects	select	VERB
ajst-2549	95	18	the	the	DET
ajst-2549	95	19	commonly	commonly	ADV
ajst-2549	95	20	used	use	VERB
ajst-2549	95	21	text	text	NOUN
ajst-2549	95	22	detection	detection	NOUN
ajst-2549	95	23	algorithms	algorithm	NOUN
ajst-2549	95	24	based	base	VERB
ajst-2549	95	25	on	on	ADP
ajst-2549	95	26	mobilenetv3	mobilenetv3	PROPN
ajst-2549	95	27	for	for	ADP
ajst-2549	95	28	comparison	comparison	NOUN
ajst-2549	95	29	.	.	PUNCT
ajst-2549	96	1	the	the	DET
ajst-2549	96	2	experimental	experimental	ADJ
ajst-2549	96	3	results	result	NOUN
ajst-2549	96	4	of	of	ADP
ajst-2549	96	5	different	different	ADJ
ajst-2549	96	6	algorithms	algorithm	NOUN
ajst-2549	96	7	are	be	AUX
ajst-2549	96	8	shown	show	VERB
ajst-2549	96	9	in	in	ADP
ajst-2549	96	10	table	table	NOUN
ajst-2549	96	11	4	4	NUM
ajst-2549	96	12	.	.	PUNCT
ajst-2549	97	1	as	as	SCONJ
ajst-2549	97	2	can	can	AUX
ajst-2549	97	3	be	be	AUX
ajst-2549	97	4	seen	see	VERB
ajst-2549	97	5	from	from	ADP
ajst-2549	97	6	table	table	NOUN
ajst-2549	97	7	4	4	NUM
ajst-2549	97	8	,	,	PUNCT
ajst-2549	97	9	for	for	ADP
ajst-2549	97	10	the	the	DET
ajst-2549	97	11	character	character	NOUN
ajst-2549	97	12	detection	detection	NOUN
ajst-2549	97	13	of	of	ADP
ajst-2549	97	14	angle	angle	NOUN
ajst-2549	97	15	steel	steel	NOUN
ajst-2549	97	16	data	datum	NOUN
ajst-2549	97	17	set	set	VERB
ajst-2549	97	18	,	,	PUNCT
ajst-2549	97	19	the	the	DET
ajst-2549	97	20	performance	performance	NOUN
ajst-2549	97	21	of	of	ADP
ajst-2549	97	22	segmentation	segmentation	NOUN
ajst-2549	97	23	algorithm	algorithm	NOUN
ajst-2549	97	24	is	be	AUX
ajst-2549	97	25	better	well	ADJ
ajst-2549	97	26	than	than	ADP
ajst-2549	97	27	that	that	PRON
ajst-2549	97	28	of	of	ADP
ajst-2549	97	29	regression	regression	NOUN
ajst-2549	97	30	algorithm	algorithm	NOUN
ajst-2549	97	31	.	.	PUNCT
ajst-2549	98	1	compared	compare	VERB
ajst-2549	98	2	with	with	ADP
ajst-2549	98	3	the	the	DET
ajst-2549	98	4	original	original	ADJ
ajst-2549	98	5	dbnet	dbnet	NOUN
ajst-2549	98	6	algorithm	algorithm	NOUN
ajst-2549	98	7	and	and	CCONJ
ajst-2549	98	8	other	other	ADJ
ajst-2549	98	9	algorithms	algorithm	NOUN
ajst-2549	98	10	based	base	VERB
ajst-2549	98	11	on	on	ADP
ajst-2549	98	12	mobilenetv3	mobilenetv3	PROPN
ajst-2549	98	13	as	as	ADP
ajst-2549	98	14	the	the	DET
ajst-2549	98	15	backbone	backbone	NOUN
ajst-2549	98	16	network	network	NOUN
ajst-2549	98	17	,	,	PUNCT
ajst-2549	98	18	the	the	DET
ajst-2549	98	19	improved	improved	ADJ
ajst-2549	98	20	dbnet	dbnet	NOUN
ajst-2549	98	21	algorithm	algorithm	NOUN
ajst-2549	98	22	based	base	VERB
ajst-2549	98	23	on	on	ADP
ajst-2549	98	24	the	the	DET
ajst-2549	98	25	coordinate	coordinate	ADJ
ajst-2549	98	26	attention	attention	NOUN
ajst-2549	98	27	mechanism	mechanism	NOUN
ajst-2549	98	28	has	have	VERB
ajst-2549	98	29	the	the	DET
ajst-2549	98	30	best	good	ADJ
ajst-2549	98	31	performance	performance	NOUN
ajst-2549	98	32	indicators	indicator	NOUN
ajst-2549	98	33	,	,	PUNCT
ajst-2549	98	34	and	and	CCONJ
ajst-2549	98	35	the	the	DET
ajst-2549	98	36	increase	increase	NOUN
ajst-2549	98	37	in	in	ADP
ajst-2549	98	38	the	the	DET
ajst-2549	98	39	amount	amount	NOUN
ajst-2549	98	40	of	of	ADP
ajst-2549	98	41	parameters	parameter	NOUN
ajst-2549	98	42	is	be	AUX
ajst-2549	98	43	also	also	ADV
ajst-2549	98	44	within	within	ADP
ajst-2549	98	45	the	the	DET
ajst-2549	98	46	acceptable	acceptable	ADJ
ajst-2549	98	47	range	range	NOUN
ajst-2549	98	48	.	.	PUNCT
ajst-2549	99	1	table	table	NOUN
ajst-2549	99	2	4	4	NUM
ajst-2549	99	3	.	.	PUNCT
ajst-2549	99	4	comparison	comparison	NOUN
ajst-2549	99	5	experiment	experiment	NOUN
ajst-2549	99	6	of	of	ADP
ajst-2549	99	7	different	different	ADJ
ajst-2549	99	8	algorithms	algorithm	NOUN
ajst-2549	99	9	algorithm	algorithm	NOUN
ajst-2549	99	10	precision	precision	NOUN
ajst-2549	99	11	rcall	rcall	NOUN
ajst-2549	99	12	f1	f1	NOUN
ajst-2549	99	13	-	-	PUNCT
ajst-2549	99	14	score	score	NOUN
ajst-2549	99	15	params(m	params(m	NOUN
ajst-2549	99	16	)	)	PUNCT
ajst-2549	99	17	dbnet	dbnet	NOUN
ajst-2549	99	18	76.03	76.03	NUM
ajst-2549	99	19	87.62	87.62	NUM
ajst-2549	99	20	81.42	81.42	NUM
ajst-2549	99	21	5.70	5.70	NUM
ajst-2549	99	22	psenet	psenet	NOUN
ajst-2549	99	23	78.91	78.91	NUM
ajst-2549	99	24	96.19	96.19	NUM
ajst-2549	99	25	86.70	86.70	NUM
ajst-2549	99	26	5.50	5.50	NUM
ajst-2549	99	27	east	east	NOUN
ajst-2549	99	28	51.91	51.91	NUM
ajst-2549	99	29	64.76	64.76	NUM
ajst-2549	99	30	57.63	57.63	NUM
ajst-2549	99	31	6.40	6.40	NUM
ajst-2549	99	32	ours	ours	PRON
ajst-2549	99	33	98.13	98.13	NUM
ajst-2549	99	34	100.00	100.00	NUM
ajst-2549	99	35	99.06	99.06	NUM
ajst-2549	99	36	6.90	6.90	NUM
ajst-2549	99	37	algorithm	algorithm	NOUN
ajst-2549	99	38	precision	precision	NOUN
ajst-2549	99	39	rcall	rcall	NOUN
ajst-2549	99	40	f1	f1	NOUN
ajst-2549	99	41	-	-	PUNCT
ajst-2549	99	42	score	score	NOUN
ajst-2549	99	43	params(m	params(m	NOUN
ajst-2549	99	44	)	)	PUNCT
ajst-2549	99	45	in	in	ADP
ajst-2549	99	46	order	order	NOUN
ajst-2549	99	47	to	to	PART
ajst-2549	99	48	more	more	ADV
ajst-2549	99	49	clearly	clearly	ADV
ajst-2549	99	50	show	show	VERB
ajst-2549	99	51	the	the	DET
ajst-2549	99	52	improved	improved	ADJ
ajst-2549	99	53	effect	effect	NOUN
ajst-2549	99	54	of	of	ADP
ajst-2549	99	55	the	the	DET
ajst-2549	99	56	algorithm	algorithm	NOUN
ajst-2549	99	57	in	in	ADP
ajst-2549	99	58	this	this	DET
ajst-2549	99	59	paper	paper	NOUN
ajst-2549	99	60	,	,	PUNCT
ajst-2549	99	61	4	4	NUM
ajst-2549	99	62	images	image	NOUN
ajst-2549	99	63	are	be	AUX
ajst-2549	99	64	randomly	randomly	ADV
ajst-2549	99	65	selected	select	VERB
ajst-2549	99	66	to	to	PART
ajst-2549	99	67	test	test	VERB
ajst-2549	99	68	the	the	DET
ajst-2549	99	69	performance	performance	NOUN
ajst-2549	99	70	of	of	ADP
ajst-2549	99	71	different	different	ADJ
ajst-2549	99	72	algorithms	algorithm	NOUN
ajst-2549	99	73	,	,	PUNCT
ajst-2549	99	74	and	and	CCONJ
ajst-2549	99	75	the	the	DET
ajst-2549	99	76	test	test	NOUN
ajst-2549	99	77	comparison	comparison	NOUN
ajst-2549	99	78	results	result	NOUN
ajst-2549	99	79	are	be	AUX
ajst-2549	99	80	shown	show	VERB
ajst-2549	99	81	in	in	ADP
ajst-2549	99	82	fig	fig	NOUN
ajst-2549	99	83	.	.	PUNCT
ajst-2549	100	1	7	7	X
ajst-2549	100	2	.	.	X
ajst-2549	100	3	it	it	PRON
ajst-2549	100	4	can	can	AUX
ajst-2549	100	5	be	be	AUX
ajst-2549	100	6	seen	see	VERB
ajst-2549	100	7	that	that	SCONJ
ajst-2549	100	8	in	in	ADP
ajst-2549	100	9	fig	fig	NOUN
ajst-2549	100	10	.	.	PUNCT
ajst-2549	101	1	7	7	NUM
ajst-2549	101	2	(	(	PUNCT
ajst-2549	101	3	a	a	X
ajst-2549	101	4	)	)	PUNCT
ajst-2549	101	5	the	the	DET
ajst-2549	101	6	algorithm	algorithm	NOUN
ajst-2549	101	7	has	have	VERB
ajst-2549	101	8	detection	detection	NOUN
ajst-2549	101	9	errors	error	NOUN
ajst-2549	101	10	and	and	CCONJ
ajst-2549	101	11	incomplete	incomplete	ADJ
ajst-2549	101	12	detection	detection	NOUN
ajst-2549	101	13	;	;	PUNCT
ajst-2549	101	14	(	(	PUNCT
ajst-2549	101	15	b	b	X
ajst-2549	101	16	)	)	PUNCT
ajst-2549	101	17	the	the	DET
ajst-2549	101	18	algorithm	algorithm	NOUN
ajst-2549	101	19	has	have	VERB
ajst-2549	101	20	too	too	ADV
ajst-2549	101	21	large	large	ADJ
ajst-2549	101	22	detection	detection	NOUN
ajst-2549	101	23	frames	frame	NOUN
ajst-2549	101	24	and	and	CCONJ
ajst-2549	101	25	too	too	ADV
ajst-2549	101	26	many	many	ADJ
ajst-2549	101	27	detection	detection	NOUN
ajst-2549	101	28	frames	frame	NOUN
ajst-2549	101	29	;	;	PUNCT
ajst-2549	101	30	(	(	PUNCT
ajst-2549	101	31	c	c	X
ajst-2549	101	32	)	)	PUNCT
ajst-2549	101	33	the	the	DET
ajst-2549	101	34	algorithm	algorithm	NOUN
ajst-2549	101	35	has	have	VERB
ajst-2549	101	36	incomplete	incomplete	ADJ
ajst-2549	101	37	detection	detection	NOUN
ajst-2549	101	38	and	and	CCONJ
ajst-2549	101	39	detection	detection	NOUN
ajst-2549	101	40	errors	error	NOUN
ajst-2549	101	41	and	and	CCONJ
ajst-2549	101	42	overlap	overlap	NOUN
ajst-2549	101	43	.	.	PUNCT
ajst-2549	102	1	;	;	PUNCT
ajst-2549	102	2	(	(	PUNCT
ajst-2549	102	3	d	d	X
ajst-2549	102	4	)	)	PUNCT
ajst-2549	102	5	the	the	DET
ajst-2549	102	6	algorithm	algorithm	NOUN
ajst-2549	102	7	is	be	AUX
ajst-2549	102	8	improved	improve	VERB
ajst-2549	102	9	for	for	ADP
ajst-2549	102	10	this	this	DET
ajst-2549	102	11	paper	paper	NOUN
ajst-2549	102	12	,	,	PUNCT
ajst-2549	102	13	the	the	DET
ajst-2549	102	14	detection	detection	NOUN
ajst-2549	102	15	frame	frame	NOUN
ajst-2549	102	16	is	be	AUX
ajst-2549	102	17	positioned	position	VERB
ajst-2549	102	18	accurately	accurately	ADV
ajst-2549	102	19	,	,	PUNCT
ajst-2549	102	20	and	and	CCONJ
ajst-2549	102	21	there	there	PRON
ajst-2549	102	22	is	be	VERB
ajst-2549	102	23	no	no	DET
ajst-2549	102	24	false	false	ADJ
ajst-2549	102	25	detection	detection	NOUN
ajst-2549	102	26	and	and	CCONJ
ajst-2549	102	27	missed	miss	VERB
ajst-2549	102	28	detection	detection	NOUN
ajst-2549	102	29	.	.	PUNCT
ajst-2549	103	1	69	69	NUM
ajst-2549	103	2	(	(	PUNCT
ajst-2549	103	3	a	a	NOUN
ajst-2549	103	4	)	)	PUNCT
ajst-2549	103	5	dbnet	dbnet	NOUN
ajst-2549	103	6	(	(	PUNCT
ajst-2549	103	7	b	b	NOUN
ajst-2549	103	8	)	)	PUNCT
ajst-2549	103	9	psenet	psenet	NOUN
ajst-2549	103	10	(	(	PUNCT
ajst-2549	103	11	c	c	NOUN
ajst-2549	103	12	)	)	PUNCT
ajst-2549	103	13	eastt	eastt	NOUN
ajst-2549	103	14	(	(	PUNCT
ajst-2549	103	15	d	d	NOUN
ajst-2549	103	16	)	)	PUNCT
ajst-2549	103	17	ours	ours	PRON
ajst-2549	103	18	figure	figure	VERB
ajst-2549	103	19	7	7	NUM
ajst-2549	103	20	.	.	PUNCT
ajst-2549	103	21	test	test	NOUN
ajst-2549	103	22	comparison	comparison	NOUN
ajst-2549	103	23	result	result	NOUN
ajst-2549	103	24	display	display	NOUN
ajst-2549	103	25	5	5	NUM
ajst-2549	103	26	.	.	PUNCT
ajst-2549	103	27	conclusion	conclusion	NOUN
ajst-2549	103	28	this	this	DET
ajst-2549	103	29	paper	paper	NOUN
ajst-2549	103	30	proposes	propose	VERB
ajst-2549	103	31	an	an	DET
ajst-2549	103	32	improved	improved	ADJ
ajst-2549	103	33	detection	detection	NOUN
ajst-2549	103	34	algorithm	algorithm	NOUN
ajst-2549	103	35	for	for	ADP
ajst-2549	103	36	angle	angle	NOUN
ajst-2549	103	37	steel	steel	NOUN
ajst-2549	103	38	characters	character	NOUN
ajst-2549	103	39	based	base	VERB
ajst-2549	103	40	on	on	ADP
ajst-2549	103	41	dbnet	dbnet	NOUN
ajst-2549	103	42	,	,	PUNCT
ajst-2549	103	43	aiming	aim	VERB
ajst-2549	103	44	at	at	ADP
ajst-2549	103	45	the	the	DET
ajst-2549	103	46	problems	problem	NOUN
ajst-2549	103	47	of	of	ADP
ajst-2549	103	48	false	false	ADJ
ajst-2549	103	49	detection	detection	NOUN
ajst-2549	103	50	and	and	CCONJ
ajst-2549	103	51	missed	miss	VERB
ajst-2549	103	52	detection	detection	NOUN
ajst-2549	103	53	in	in	ADP
ajst-2549	103	54	manual	manual	ADJ
ajst-2549	103	55	recording	recording	NOUN
ajst-2549	103	56	of	of	ADP
ajst-2549	103	57	industrial	industrial	ADJ
ajst-2549	103	58	angle	angle	NOUN
ajst-2549	103	59	steel	steel	NOUN
ajst-2549	103	60	characters	character	NOUN
ajst-2549	103	61	.	.	PUNCT
ajst-2549	104	1	among	among	ADP
ajst-2549	104	2	them	they	PRON
ajst-2549	104	3	,	,	PUNCT
ajst-2549	104	4	ca	ca	PRON
ajst-2549	104	5	and	and	CCONJ
ajst-2549	104	6	bifpn	bifpn	PROPN
ajst-2549	104	7	improve	improve	VERB
ajst-2549	104	8	the	the	DET
ajst-2549	104	9	feature	feature	NOUN
ajst-2549	104	10	perception	perception	NOUN
ajst-2549	104	11	ability	ability	NOUN
ajst-2549	104	12	of	of	ADP
ajst-2549	104	13	the	the	DET
ajst-2549	104	14	model	model	NOUN
ajst-2549	104	15	for	for	ADP
ajst-2549	104	16	the	the	DET
ajst-2549	104	17	angle	angle	NOUN
ajst-2549	104	18	steel	steel	NOUN
ajst-2549	104	19	character	character	NOUN
ajst-2549	104	20	area	area	NOUN
ajst-2549	104	21	,	,	PUNCT
ajst-2549	104	22	and	and	CCONJ
ajst-2549	104	23	the	the	DET
ajst-2549	104	24	copy	copy	NOUN
ajst-2549	104	25	-	-	PUNCT
ajst-2549	104	26	paste	paste	NOUN
ajst-2549	104	27	data	datum	NOUN
ajst-2549	104	28	enhancement	enhancement	NOUN
ajst-2549	104	29	and	and	CCONJ
ajst-2549	104	30	cosine	cosine	NOUN
ajst-2549	104	31	learning	learning	NOUN
ajst-2549	104	32	rate	rate	NOUN
ajst-2549	104	33	optimization	optimization	NOUN
ajst-2549	104	34	strategy	strategy	NOUN
ajst-2549	104	35	enhance	enhance	VERB
ajst-2549	104	36	the	the	DET
ajst-2549	104	37	robustness	robustness	NOUN
ajst-2549	104	38	of	of	ADP
ajst-2549	104	39	the	the	DET
ajst-2549	104	40	model	model	NOUN
ajst-2549	104	41	.	.	PUNCT
ajst-2549	105	1	the	the	DET
ajst-2549	105	2	experimental	experimental	ADJ
ajst-2549	105	3	results	result	NOUN
ajst-2549	105	4	on	on	ADP
ajst-2549	105	5	the	the	DET
ajst-2549	105	6	angle	angle	NOUN
ajst-2549	105	7	steel	steel	NOUN
ajst-2549	105	8	character	character	NOUN
ajst-2549	105	9	data	datum	NOUN
ajst-2549	105	10	set	set	VERB
ajst-2549	105	11	show	show	VERB
ajst-2549	105	12	that	that	SCONJ
ajst-2549	105	13	the	the	DET
ajst-2549	105	14	method	method	NOUN
ajst-2549	105	15	in	in	ADP
ajst-2549	105	16	this	this	DET
ajst-2549	105	17	paper	paper	NOUN
ajst-2549	105	18	has	have	VERB
ajst-2549	105	19	the	the	DET
ajst-2549	105	20	highest	high	ADJ
ajst-2549	105	21	index	index	NOUN
ajst-2549	105	22	and	and	CCONJ
ajst-2549	105	23	the	the	DET
ajst-2549	105	24	best	good	ADJ
ajst-2549	105	25	effect	effect	NOUN
ajst-2549	105	26	among	among	ADP
ajst-2549	105	27	the	the	DET
ajst-2549	105	28	common	common	ADJ
ajst-2549	105	29	text	text	NOUN
ajst-2549	105	30	detection	detection	NOUN
ajst-2549	105	31	algorithms	algorithm	NOUN
ajst-2549	105	32	based	base	VERB
ajst-2549	105	33	on	on	ADP
ajst-2549	105	34	mobilenetv3	mobilenetv3	PROPN
ajst-2549	105	35	.	.	PUNCT
ajst-2549	106	1	compared	compare	VERB
ajst-2549	106	2	with	with	ADP
ajst-2549	106	3	the	the	DET
ajst-2549	106	4	traditional	traditional	ADJ
ajst-2549	106	5	dbnet	dbnet	NOUN
ajst-2549	106	6	algorithm	algorithm	NOUN
ajst-2549	106	7	,	,	PUNCT
ajst-2549	106	8	the	the	DET
ajst-2549	106	9	f1	f1	ADJ
ajst-2549	106	10	-	-	PUNCT
ajst-2549	106	11	score	score	NOUN
ajst-2549	106	12	index	index	NOUN
ajst-2549	106	13	has	have	AUX
ajst-2549	106	14	increased	increase	VERB
ajst-2549	106	15	from	from	ADP
ajst-2549	106	16	81.42	81.42	NUM
ajst-2549	106	17	%	%	NOUN
ajst-2549	106	18	to	to	ADP
ajst-2549	106	19	99.06	99.06	NUM
ajst-2549	106	20	%	%	NOUN
ajst-2549	106	21	,	,	PUNCT
ajst-2549	106	22	which	which	PRON
ajst-2549	106	23	proves	prove	VERB
ajst-2549	106	24	that	that	SCONJ
ajst-2549	106	25	this	this	DET
ajst-2549	106	26	paper	paper	NOUN
ajst-2549	106	27	effectiveness	effectiveness	NOUN
ajst-2549	106	28	of	of	ADP
ajst-2549	106	29	the	the	DET
ajst-2549	106	30	method	method	NOUN
ajst-2549	106	31	.	.	PUNCT
ajst-2549	107	1	subsequent	subsequent	ADJ
ajst-2549	107	2	work	work	NOUN
ajst-2549	107	3	will	will	AUX
ajst-2549	107	4	collect	collect	VERB
ajst-2549	107	5	more	more	ADJ
ajst-2549	107	6	angle	angle	NOUN
ajst-2549	107	7	steel	steel	NOUN
ajst-2549	107	8	character	character	NOUN
ajst-2549	107	9	images	image	NOUN
ajst-2549	107	10	as	as	ADP
ajst-2549	107	11	datasets	dataset	NOUN
ajst-2549	107	12	to	to	PART
ajst-2549	107	13	further	far	ADV
ajst-2549	107	14	strengthen	strengthen	VERB
ajst-2549	107	15	the	the	DET
ajst-2549	107	16	generalization	generalization	NOUN
ajst-2549	107	17	of	of	ADP
ajst-2549	107	18	the	the	DET
ajst-2549	107	19	model	model	NOUN
ajst-2549	107	20	,	,	PUNCT
ajst-2549	107	21	thereby	thereby	ADV
ajst-2549	107	22	broadening	broaden	VERB
ajst-2549	107	23	the	the	DET
ajst-2549	107	24	detection	detection	NOUN
ajst-2549	107	25	scenarios	scenario	NOUN
ajst-2549	107	26	of	of	ADP
ajst-2549	107	27	the	the	DET
ajst-2549	107	28	model	model	NOUN
ajst-2549	107	29	.	.	PUNCT
ajst-2549	108	1	references	reference	NOUN
ajst-2549	108	2	[	[	X
ajst-2549	108	3	1	1	NUM
ajst-2549	108	4	]	]	X
ajst-2549	108	5	h.y	h.y	PROPN
ajst-2549	108	6	.	.	PROPN
ajst-2549	108	7	liu	liu	PROPN
ajst-2549	108	8	,	,	PUNCT
ajst-2549	108	9	z.l	z.l	PROPN
ajst-2549	108	10	.	.	PROPN
ajst-2549	108	11	li	li	PROPN
ajst-2549	108	12	,	,	PUNCT
ajst-2549	108	13	z.l	z.l	PROPN
ajst-2549	108	14	.	.	PROPN
ajst-2549	108	15	huang	huang	PROPN
ajst-2549	108	16	:	:	PUNCT
ajst-2549	108	17	progress	progress	NOUN
ajst-2549	108	18	in	in	ADP
ajst-2549	108	19	building	build	VERB
ajst-2549	108	20	steel	steel	NOUN
ajst-2549	108	21	structures	structure	NOUN
ajst-2549	108	22	,	,	PUNCT
ajst-2549	108	23	vol	vol	NOUN
ajst-2549	108	24	.	.	PROPN
ajst-2549	108	25	23	23	NUM
ajst-2549	108	26	(	(	PUNCT
ajst-2549	108	27	2021	2021	NUM
ajst-2549	108	28	)	)	PUNCT
ajst-2549	109	1	no	no	INTJ
ajst-2549	109	2	.	.	NOUN
ajst-2549	109	3	12	12	NUM
ajst-2549	109	4	,	,	PUNCT
ajst-2549	109	5	p.	p.	NOUN
ajst-2549	109	6	47	47	NUM
ajst-2549	110	1	-	-	SYM
ajst-2549	110	2	55	55	NUM
ajst-2549	110	3	.	.	PUNCT
ajst-2549	111	1	[	[	X
ajst-2549	111	2	2	2	NUM
ajst-2549	111	3	]	]	X
ajst-2549	111	4	y.h	y.h	PROPN
ajst-2549	111	5	.	.	PROPN
ajst-2549	111	6	li	li	PROPN
ajst-2549	111	7	,	,	PUNCT
ajst-2549	111	8	y.y	y.y	PROPN
ajst-2549	111	9	chen	chen	PROPN
ajst-2549	111	10	:	:	PUNCT
ajst-2549	111	11	computer	computer	NOUN
ajst-2549	111	12	engineering	engineering	NOUN
ajst-2549	111	13	and	and	CCONJ
ajst-2549	111	14	applications	application	NOUN
ajst-2549	111	15	,	,	PUNCT
ajst-2549	111	16	vol	vol	NOUN
ajst-2549	111	17	.	.	PROPN
ajst-2549	111	18	57	57	NUM
ajst-2549	111	19	(	(	PUNCT
ajst-2549	111	20	2021	2021	NUM
ajst-2549	111	21	)	)	PUNCT
ajst-2549	111	22	no	no	INTJ
ajst-2549	111	23	.	.	NOUN
ajst-2549	111	24	06	06	NUM
ajst-2549	111	25	,	,	PUNCT
ajst-2549	111	26	p.	p.	NOUN
ajst-2549	111	27	42	42	NUM
ajst-2549	111	28	-	-	SYM
ajst-2549	111	29	48	48	NUM
ajst-2549	111	30	.	.	PUNCT
ajst-2549	112	1	[	[	X
ajst-2549	112	2	3	3	X
ajst-2549	112	3	]	]	PUNCT
ajst-2549	112	4	k.	k.	PROPN
ajst-2549	112	5	wang	wang	PROPN
ajst-2549	112	6	,	,	PUNCT
ajst-2549	112	7	f.	f.	PROPN
ajst-2549	112	8	yang	yang	PROPN
ajst-2549	112	9	,	,	PUNCT
ajst-2549	112	10	s.	s.	PROPN
ajst-2549	112	11	jiang	jiang	PROPN
ajst-2549	112	12	:	:	PUNCT
ajst-2549	112	13	computer	computer	NOUN
ajst-2549	112	14	application	application	NOUN
ajst-2549	112	15	research	research	NOUN
ajst-2549	112	16	,	,	PUNCT
ajst-2549	112	17	vol	vol	NOUN
ajst-2549	112	18	.	.	PROPN
ajst-2549	112	19	37	37	NUM
ajst-2549	112	20	(	(	PUNCT
ajst-2549	112	21	2020	2020	NUM
ajst-2549	112	22	)	)	PUNCT
ajst-2549	112	23	no	no	NOUN
ajst-2549	112	24	.	.	PUNCT
ajst-2549	112	25	s2	s2	PROPN
ajst-2549	112	26	,	,	PUNCT
ajst-2549	112	27	p.	p.	NOUN
ajst-2549	112	28	22	22	NUM
ajst-2549	112	29	-	-	SYM
ajst-2549	112	30	24	24	NUM
ajst-2549	112	31	.	.	PUNCT
ajst-2549	113	1	[	[	X
ajst-2549	113	2	4	4	X
ajst-2549	113	3	]	]	X
ajst-2549	113	4	q.l	q.l	PROPN
ajst-2549	113	5	.	.	PROPN
ajst-2549	113	6	zhou	zhou	PROPN
ajst-2549	113	7	,	,	PUNCT
ajst-2549	113	8	l.	l.	PROPN
ajst-2549	113	9	ma	ma	PROPN
ajst-2549	113	10	,	,	PUNCT
ajst-2549	113	11	l.	l.	PROPN
ajst-2549	113	12	cao	cao	PROPN
ajst-2549	113	13	,	,	PUNCT
ajst-2549	113	14	et	et	PROPN
ajst-2549	113	15	al	al	PROPN
ajst-2549	113	16	:	:	PUNCT
ajst-2549	113	17	smart	smart	ADJ
ajst-2549	113	18	agriculture	agriculture	NOUN
ajst-2549	113	19	,	,	PUNCT
ajst-2549	113	20	vol	vol	NOUN
ajst-2549	113	21	.	.	PROPN
ajst-2549	113	22	4	4	NUM
ajst-2549	113	23	(	(	PUNCT
ajst-2549	113	24	2022	2022	NUM
ajst-2549	113	25	)	)	PUNCT
ajst-2549	113	26	no	no	INTJ
ajst-2549	113	27	.	.	NOUN
ajst-2549	113	28	01	01	NUM
ajst-2549	113	29	,	,	PUNCT
ajst-2549	113	30	p.	p.	NOUN
ajst-2549	113	31	47	47	NUM
ajst-2549	113	32	-	-	SYM
ajst-2549	113	33	56	56	NUM
ajst-2549	113	34	.	.	PUNCT
ajst-2549	114	1	[	[	X
ajst-2549	114	2	5	5	NUM
ajst-2549	114	3	]	]	PUNCT
ajst-2549	114	4	c.	c.	PROPN
ajst-2549	114	5	xie	xie	PROPN
ajst-2549	114	6	,	,	PUNCT
ajst-2549	114	7	h.y	h.y	PROPN
ajst-2549	114	8	.	.	PROPN
ajst-2549	114	9	zhu	zhu	PROPN
ajst-2549	114	10	,	,	PUNCT
ajst-2549	114	11	q.	q.	PROPN
ajst-2549	114	12	fei	fei	PROPN
ajst-2549	114	13	:	:	PUNCT
ajst-2549	114	14	iet	iet	PROPN
ajst-2549	114	15	image	image	PROPN
ajst-2549	114	16	processing	processing	NOUN
ajst-2549	114	17	,	,	PUNCT
ajst-2549	114	18	vol	vol	NOUN
ajst-2549	114	19	.	.	PROPN
ajst-2549	115	1	16	16	NUM
ajst-2549	115	2	(	(	PUNCT
ajst-2549	115	3	2022	2022	NUM
ajst-2549	115	4	)	)	PUNCT
ajst-2549	116	1	no	no	INTJ
ajst-2549	116	2	.	.	NOUN
ajst-2549	116	3	01	01	NUM
ajst-2549	116	4	,	,	PUNCT
ajst-2549	116	5	p.	p.	NOUN
ajst-2549	116	6	273	273	NUM
ajst-2549	116	7	-	-	SYM
ajst-2549	116	8	284	284	NUM
ajst-2549	116	9	.	.	PUNCT
ajst-2549	117	1	[	[	X
ajst-2549	117	2	6	6	NUM
ajst-2549	117	3	]	]	PUNCT
ajst-2549	117	4	j.	j.	PROPN
ajst-2549	117	5	hu	hu	PROPN
ajst-2549	117	6	,	,	PUNCT
ajst-2549	117	7	l.	l.	PROPN
ajst-2549	117	8	shen	shen	PROPN
ajst-2549	117	9	,	,	PUNCT
ajst-2549	117	10	s.	s.	PROPN
ajst-2549	117	11	albanie	albanie	PROPN
ajst-2549	117	12	,	,	PUNCT
ajst-2549	117	13	et	et	PROPN
ajst-2549	117	14	al	al	PROPN
ajst-2549	117	15	:	:	PUNCT
ajst-2549	117	16	ieee	ieee	NOUN
ajst-2549	117	17	trans	tran	NOUN
ajst-2549	117	18	on	on	ADP
ajst-2549	117	19	pattern	pattern	NOUN
ajst-2549	117	20	analysis	analysis	NOUN
ajst-2549	117	21	and	and	CCONJ
ajst-2549	117	22	machine	machine	NOUN
ajst-2549	117	23	intelligence	intelligence	NOUN
ajst-2549	117	24	,	,	PUNCT
ajst-2549	117	25	vol	vol	NOUN
ajst-2549	117	26	.	.	PROPN
ajst-2549	118	1	42	42	NUM
ajst-2549	118	2	(	(	PUNCT
ajst-2549	118	3	2020	2020	NUM
ajst-2549	118	4	)	)	PUNCT
ajst-2549	119	1	no	no	INTJ
ajst-2549	119	2	.	.	NOUN
ajst-2549	119	3	08	08	NUM
ajst-2549	119	4	,	,	PUNCT
ajst-2549	119	5	p.	p.	NOUN
ajst-2549	119	6	2011	2011	NUM
ajst-2549	119	7	-	-	SYM
ajst-2549	119	8	2023	2023	NUM
ajst-2549	119	9	.	.	PUNCT
ajst-2549	120	1	[	[	X
ajst-2549	120	2	7	7	NUM
ajst-2549	120	3	]	]	X
ajst-2549	120	4	q.f	q.f	NOUN
ajst-2549	120	5	.	.	PUNCT
ajst-2549	120	6	guo	guo	PROPN
ajst-2549	120	7	,	,	PUNCT
ajst-2549	120	8	l.	l.	PROPN
ajst-2549	120	9	liu	liu	PROPN
ajst-2549	120	10	,	,	PUNCT
ajst-2549	120	11	x.	x.	PROPN
ajst-2549	120	12	zhang	zhang	PROPN
ajst-2549	120	13	,	,	PUNCT
ajst-2549	120	14	et	et	PROPN
ajst-2549	120	15	al	al	PROPN
ajst-2549	120	16	:	:	PUNCT
ajst-2549	120	17	journal	journal	NOUN
ajst-2549	120	18	of	of	ADP
ajst-2549	120	19	engineering	engineering	NOUN
ajst-2549	120	20	mathematics	mathematic	NOUN
ajst-2549	120	21	,	,	PUNCT
ajst-2549	120	22	vol	vol	NOUN
ajst-2549	120	23	.	.	PROPN
ajst-2549	120	24	37	37	NUM
ajst-2549	120	25	(	(	PUNCT
ajst-2549	120	26	2020	2020	NUM
ajst-2549	120	27	)	)	PUNCT
ajst-2549	120	28	no	no	INTJ
ajst-2549	120	29	.	.	NOUN
ajst-2549	120	30	05	05	NUM
ajst-2549	120	31	,	,	PUNCT
ajst-2549	120	32	p.	p.	NOUN
ajst-2549	120	33	521	521	NUM
ajst-2549	120	34	-	-	SYM
ajst-2549	120	35	530	530	NUM
ajst-2549	120	36	.	.	PUNCT
ajst-2549	121	1	[	[	X
ajst-2549	121	2	8	8	NUM
ajst-2549	121	3	]	]	X
ajst-2549	121	4	b.m	b.m	PROPN
ajst-2549	121	5	.	.	PROPN
ajst-2549	121	6	li	li	PROPN
ajst-2549	121	7	,	,	PUNCT
ajst-2549	121	8	r.l	r.l	PROPN
ajst-2549	121	9	.	.	PROPN
ajst-2549	121	10	jin	jin	PROPN
ajst-2549	121	11	,	,	PUNCT
ajst-2549	121	12	z.f	z.f	PROPN
ajst-2549	121	13	.	.	PROPN
ajst-2549	121	14	xu	xu	PROPN
ajst-2549	121	15	,	,	PUNCT
ajst-2549	121	16	et	et	PROPN
ajst-2549	121	17	al	al	PROPN
ajst-2549	121	18	:	:	PUNCT
ajst-2549	121	19	journal	journal	PROPN
ajst-2549	121	20	of	of	ADP
ajst-2549	121	21	zhengzhou	zhengzhou	PROPN
ajst-2549	121	22	university	university	PROPN
ajst-2549	121	23	(	(	PUNCT
ajst-2549	121	24	engineering	engineering	NOUN
ajst-2549	121	25	edition	edition	NOUN
ajst-2549	121	26	)	)	PUNCT
ajst-2549	121	27	,	,	PUNCT
ajst-2549	121	28	vol	vol	NOUN
ajst-2549	121	29	.	.	PUNCT
ajst-2549	121	30	(	(	PUNCT
ajst-2549	121	31	2022	2022	NUM
ajst-2549	121	32	)	)	PUNCT
ajst-2549	121	33	no	no	INTJ
ajst-2549	121	34	.	.	NOUN
ajst-2549	121	35	01	01	NUM
ajst-2549	121	36	,	,	PUNCT
ajst-2549	121	37	p.	p.	NOUN
ajst-2549	121	38	2026	2026	NUM
ajst-2549	121	39	.	.	PUNCT
ajst-2549	122	1	[	[	X
ajst-2549	122	2	9	9	NUM
ajst-2549	122	3	]	]	X
ajst-2549	122	4	y.x	y.x	PROPN
ajst-2549	122	5	.	.	PUNCT
ajst-2549	122	6	feng	feng	PROPN
ajst-2549	122	7	,	,	PUNCT
ajst-2549	122	8	y.m	y.m	PROPN
ajst-2549	122	9	.	.	PROPN
ajst-2549	122	10	li	li	PROPN
ajst-2549	122	11	:	:	PUNCT
ajst-2549	122	12	data	data	PROPN
ajst-2549	122	13	mining	mining	NOUN
ajst-2549	122	14	,	,	PUNCT
ajst-2549	122	15	vol	vol	NOUN
ajst-2549	122	16	.	.	NOUN
ajst-2549	122	17	8	8	NUM
ajst-2549	122	18	(	(	PUNCT
ajst-2549	122	19	2018	2018	NUM
ajst-2549	122	20	)	)	PUNCT
ajst-2549	123	1	no	no	INTJ
ajst-2549	123	2	.	.	NOUN
ajst-2549	123	3	04	04	NUM
ajst-2549	123	4	,	,	PUNCT
ajst-2549	123	5	p.	p.	NOUN
ajst-2549	123	6	186	186	NUM
ajst-2549	123	7	-	-	SYM
ajst-2549	123	8	200	200	NUM
ajst-2549	123	9	.	.	PUNCT
ajst-2549	124	1	[	[	X
ajst-2549	124	2	10	10	NUM
ajst-2549	124	3	]	]	PUNCT
ajst-2549	124	4	j.	j.	PROPN
ajst-2549	124	5	schmidt	schmidt	PROPN
ajst-2549	124	6	-	-	PUNCT
ajst-2549	124	7	hieber	hieber	PROPN
ajst-2549	124	8	:	:	PUNCT
ajst-2549	124	9	annals	annal	NOUN
ajst-2549	124	10	of	of	ADP
ajst-2549	124	11	statistics	statistic	NOUN
ajst-2549	124	12	,	,	PUNCT
ajst-2549	124	13	vol	vol	NOUN
ajst-2549	124	14	.	.	PROPN
ajst-2549	125	1	48	48	NUM
ajst-2549	125	2	(	(	PUNCT
ajst-2549	125	3	2020	2020	NUM
ajst-2549	125	4	)	)	PUNCT
ajst-2549	126	1	no	no	INTJ
ajst-2549	126	2	.	.	NOUN
ajst-2549	126	3	04	04	NUM
ajst-2549	126	4	,	,	PUNCT
ajst-2549	126	5	p.	p.	NOUN
ajst-2549	126	6	1875	1875	NUM
ajst-2549	126	7	-	-	SYM
ajst-2549	126	8	1897	1897	NUM
ajst-2549	126	9	.	.	PUNCT