id	sid	tid	token	lemma	pos
ajst-23543	1	1	academic	academic	ADJ
ajst-23543	1	2	journal	journal	NOUN
ajst-23543	1	3	of	of	ADP
ajst-23543	1	4	science	science	NOUN
ajst-23543	1	5	and	and	CCONJ
ajst-23543	1	6	technology	technology	NOUN
ajst-23543	1	7	issn	issn	NOUN
ajst-23543	1	8	:	:	PUNCT
ajst-23543	1	9	2771	2771	NUM
ajst-23543	1	10	-	-	SYM
ajst-23543	1	11	3032	3032	NUM
ajst-23543	1	12	|	|	NOUN
ajst-23543	1	13	vol	vol	NOUN
ajst-23543	1	14	.	.	PROPN
ajst-23543	2	1	11	11	NUM
ajst-23543	2	2	,	,	PUNCT
ajst-23543	2	3	no	no	INTJ
ajst-23543	2	4	.	.	NOUN
ajst-23543	2	5	3	3	NUM
ajst-23543	2	6	,	,	PUNCT
ajst-23543	2	7	2024	2024	NUM
ajst-23543	2	8	85	85	NUM
ajst-23543	2	9	image	image	NOUN
ajst-23543	2	10	feature	feature	NOUN
ajst-23543	2	11	selection	selection	NOUN
ajst-23543	2	12	based	base	VERB
ajst-23543	2	13	on	on	ADP
ajst-23543	2	14	attention	attention	NOUN
ajst-23543	2	15	mechanism	mechanism	NOUN
ajst-23543	2	16	zuyong	zuyong	NOUN
ajst-23543	2	17	lu	lu	PROPN
ajst-23543	2	18	*	*	PUNCT
ajst-23543	2	19	college	college	PROPN
ajst-23543	2	20	of	of	ADP
ajst-23543	2	21	artificial	artificial	ADJ
ajst-23543	2	22	intelligence	intelligence	NOUN
ajst-23543	2	23	,	,	PUNCT
ajst-23543	2	24	automation	automation	NOUN
ajst-23543	2	25	,	,	PUNCT
ajst-23543	2	26	nanjing	nanjing	PROPN
ajst-23543	2	27	agricultural	agricultural	PROPN
ajst-23543	2	28	university	university	PROPN
ajst-23543	2	29	,	,	PUNCT
ajst-23543	2	30	nanjing	nanjing	PROPN
ajst-23543	2	31	,	,	PUNCT
ajst-23543	2	32	jiangsu	jiangsu	PROPN
ajst-23543	2	33	,	,	PUNCT
ajst-23543	2	34	210031	210031	NUM
ajst-23543	2	35	,	,	PUNCT
ajst-23543	2	36	china	china	PROPN
ajst-23543	2	37	*	*	PUNCT
ajst-23543	2	38	corresponding	correspond	VERB
ajst-23543	2	39	author	author	NOUN
ajst-23543	2	40	email	email	NOUN
ajst-23543	2	41	:	:	PUNCT
ajst-23543	2	42	9213020510@stu.njau.edu.cn	9213020510@stu.njau.edu.cn	NUM
ajst-23543	2	43	abstract	abstract	NOUN
ajst-23543	2	44	:	:	PUNCT
ajst-23543	2	45	in	in	ADP
ajst-23543	2	46	the	the	DET
ajst-23543	2	47	field	field	NOUN
ajst-23543	2	48	of	of	ADP
ajst-23543	2	49	deep	deep	ADJ
ajst-23543	2	50	learning	learning	NOUN
ajst-23543	2	51	,	,	PUNCT
ajst-23543	2	52	the	the	DET
ajst-23543	2	53	selection	selection	NOUN
ajst-23543	2	54	and	and	CCONJ
ajst-23543	2	55	extraction	extraction	NOUN
ajst-23543	2	56	of	of	ADP
ajst-23543	2	57	image	image	NOUN
ajst-23543	2	58	features	feature	NOUN
ajst-23543	2	59	are	be	AUX
ajst-23543	2	60	the	the	DET
ajst-23543	2	61	key	key	ADJ
ajst-23543	2	62	factors	factor	NOUN
ajst-23543	2	63	affecting	affect	VERB
ajst-23543	2	64	model	model	NOUN
ajst-23543	2	65	performance	performance	NOUN
ajst-23543	2	66	.	.	PUNCT
ajst-23543	3	1	traditional	traditional	ADJ
ajst-23543	3	2	image	image	NOUN
ajst-23543	3	3	feature	feature	NOUN
ajst-23543	3	4	selection	selection	NOUN
ajst-23543	3	5	methods	method	NOUN
ajst-23543	3	6	often	often	ADV
ajst-23543	3	7	rely	rely	VERB
ajst-23543	3	8	on	on	ADP
ajst-23543	3	9	artificially	artificially	ADV
ajst-23543	3	10	designed	design	VERB
ajst-23543	3	11	features	feature	NOUN
ajst-23543	3	12	,	,	PUNCT
ajst-23543	3	13	which	which	PRON
ajst-23543	3	14	is	be	AUX
ajst-23543	3	15	not	not	PART
ajst-23543	3	16	only	only	ADV
ajst-23543	3	17	timeconsuming	timeconsuming	ADJ
ajst-23543	3	18	but	but	CCONJ
ajst-23543	3	19	also	also	ADV
ajst-23543	3	20	difficult	difficult	ADJ
ajst-23543	3	21	to	to	PART
ajst-23543	3	22	capture	capture	VERB
ajst-23543	3	23	complex	complex	ADJ
ajst-23543	3	24	patterns	pattern	NOUN
ajst-23543	3	25	in	in	ADP
ajst-23543	3	26	the	the	DET
ajst-23543	3	27	image	image	NOUN
ajst-23543	3	28	.	.	PUNCT
ajst-23543	4	1	in	in	ADP
ajst-23543	4	2	recent	recent	ADJ
ajst-23543	4	3	years	year	NOUN
ajst-23543	4	4	,	,	PUNCT
ajst-23543	4	5	attention	attention	NOUN
ajst-23543	4	6	mechanism	mechanism	NOUN
ajst-23543	4	7	,	,	PUNCT
ajst-23543	4	8	as	as	ADP
ajst-23543	4	9	a	a	DET
ajst-23543	4	10	technique	technique	NOUN
ajst-23543	4	11	that	that	PRON
ajst-23543	4	12	enables	enable	VERB
ajst-23543	4	13	models	model	NOUN
ajst-23543	4	14	to	to	PART
ajst-23543	4	15	automatically	automatically	ADV
ajst-23543	4	16	focus	focus	VERB
ajst-23543	4	17	on	on	ADP
ajst-23543	4	18	key	key	ADJ
ajst-23543	4	19	parts	part	NOUN
ajst-23543	4	20	of	of	ADP
ajst-23543	4	21	input	input	NOUN
ajst-23543	4	22	data	datum	NOUN
ajst-23543	4	23	,	,	PUNCT
ajst-23543	4	24	has	have	AUX
ajst-23543	4	25	shown	show	VERB
ajst-23543	4	26	significant	significant	ADJ
ajst-23543	4	27	advantages	advantage	NOUN
ajst-23543	4	28	in	in	ADP
ajst-23543	4	29	many	many	ADJ
ajst-23543	4	30	fields	field	NOUN
ajst-23543	4	31	such	such	ADJ
ajst-23543	4	32	as	as	ADP
ajst-23543	4	33	natural	natural	ADJ
ajst-23543	4	34	language	language	NOUN
ajst-23543	4	35	processing	processing	NOUN
ajst-23543	4	36	and	and	CCONJ
ajst-23543	4	37	image	image	NOUN
ajst-23543	4	38	recognition	recognition	NOUN
ajst-23543	4	39	.	.	PUNCT
ajst-23543	5	1	in	in	ADP
ajst-23543	5	2	this	this	DET
ajst-23543	5	3	paper	paper	NOUN
ajst-23543	5	4	,	,	PUNCT
ajst-23543	5	5	an	an	DET
ajst-23543	5	6	attention	attention	NOUN
ajst-23543	5	7	-	-	PUNCT
ajst-23543	5	8	mechanism	mechanism	NOUN
ajst-23543	5	9	-	-	PUNCT
ajst-23543	5	10	based	base	VERB
ajst-23543	5	11	image	image	NOUN
ajst-23543	5	12	feature	feature	NOUN
ajst-23543	5	13	selection	selection	NOUN
ajst-23543	5	14	method	method	NOUN
ajst-23543	5	15	is	be	AUX
ajst-23543	5	16	proposed	propose	VERB
ajst-23543	5	17	to	to	PART
ajst-23543	5	18	improve	improve	VERB
ajst-23543	5	19	the	the	DET
ajst-23543	5	20	accuracy	accuracy	NOUN
ajst-23543	5	21	and	and	CCONJ
ajst-23543	5	22	efficiency	efficiency	NOUN
ajst-23543	5	23	of	of	ADP
ajst-23543	5	24	image	image	NOUN
ajst-23543	5	25	classification	classification	NOUN
ajst-23543	5	26	and	and	CCONJ
ajst-23543	5	27	object	object	NOUN
ajst-23543	5	28	detection	detection	NOUN
ajst-23543	5	29	tasks	task	NOUN
ajst-23543	5	30	.	.	PUNCT
ajst-23543	6	1	first	first	ADV
ajst-23543	6	2	,	,	PUNCT
ajst-23543	6	3	we	we	PRON
ajst-23543	6	4	introduce	introduce	VERB
ajst-23543	6	5	the	the	DET
ajst-23543	6	6	basic	basic	ADJ
ajst-23543	6	7	principles	principle	NOUN
ajst-23543	6	8	of	of	ADP
ajst-23543	6	9	the	the	DET
ajst-23543	6	10	attention	attention	NOUN
ajst-23543	6	11	mechanism	mechanism	NOUN
ajst-23543	6	12	,	,	PUNCT
ajst-23543	6	13	and	and	CCONJ
ajst-23543	6	14	then	then	ADV
ajst-23543	6	15	we	we	PRON
ajst-23543	6	16	design	design	VERB
ajst-23543	6	17	a	a	DET
ajst-23543	6	18	convolutional	convolutional	ADJ
ajst-23543	6	19	neural	neural	ADJ
ajst-23543	6	20	network	network	NOUN
ajst-23543	6	21	(	(	PUNCT
ajst-23543	6	22	cnn	cnn	PROPN
ajst-23543	6	23	)	)	PUNCT
ajst-23543	6	24	framework	framework	NOUN
ajst-23543	6	25	with	with	ADP
ajst-23543	6	26	integrated	integrate	VERB
ajst-23543	6	27	attention	attention	NOUN
ajst-23543	6	28	modules	module	NOUN
ajst-23543	6	29	that	that	PRON
ajst-23543	6	30	can	can	AUX
ajst-23543	6	31	adaptively	adaptively	ADV
ajst-23543	6	32	adjust	adjust	VERB
ajst-23543	6	33	the	the	DET
ajst-23543	6	34	weights	weight	NOUN
ajst-23543	6	35	during	during	ADP
ajst-23543	6	36	training	training	NOUN
ajst-23543	6	37	to	to	PART
ajst-23543	6	38	highlight	highlight	VERB
ajst-23543	6	39	important	important	ADJ
ajst-23543	6	40	areas	area	NOUN
ajst-23543	6	41	in	in	ADP
ajst-23543	6	42	the	the	DET
ajst-23543	6	43	image	image	NOUN
ajst-23543	6	44	and	and	CCONJ
ajst-23543	6	45	ignore	ignore	VERB
ajst-23543	6	46	irrelevant	irrelevant	ADJ
ajst-23543	6	47	backgrounds	background	NOUN
ajst-23543	6	48	.	.	PUNCT
ajst-23543	7	1	by	by	ADP
ajst-23543	7	2	introducing	introduce	VERB
ajst-23543	7	3	the	the	DET
ajst-23543	7	4	attention	attention	NOUN
ajst-23543	7	5	mechanism	mechanism	NOUN
ajst-23543	7	6	,	,	PUNCT
ajst-23543	7	7	our	our	PRON
ajst-23543	7	8	model	model	NOUN
ajst-23543	7	9	can	can	AUX
ajst-23543	7	10	learn	learn	VERB
ajst-23543	7	11	the	the	DET
ajst-23543	7	12	key	key	ADJ
ajst-23543	7	13	features	feature	NOUN
ajst-23543	7	14	in	in	ADP
ajst-23543	7	15	the	the	DET
ajst-23543	7	16	image	image	NOUN
ajst-23543	7	17	more	more	ADV
ajst-23543	7	18	effectively	effectively	ADV
ajst-23543	7	19	,	,	PUNCT
ajst-23543	7	20	reduce	reduce	VERB
ajst-23543	7	21	the	the	DET
ajst-23543	7	22	waste	waste	NOUN
ajst-23543	7	23	of	of	ADP
ajst-23543	7	24	computing	compute	VERB
ajst-23543	7	25	resources	resource	NOUN
ajst-23543	7	26	,	,	PUNCT
ajst-23543	7	27	and	and	CCONJ
ajst-23543	7	28	improve	improve	VERB
ajst-23543	7	29	the	the	DET
ajst-23543	7	30	generalization	generalization	NOUN
ajst-23543	7	31	ability	ability	NOUN
ajst-23543	7	32	of	of	ADP
ajst-23543	7	33	the	the	DET
ajst-23543	7	34	model	model	NOUN
ajst-23543	7	35	.	.	PUNCT
ajst-23543	8	1	finally	finally	ADV
ajst-23543	8	2	,	,	PUNCT
ajst-23543	8	3	we	we	PRON
ajst-23543	8	4	verify	verify	VERB
ajst-23543	8	5	the	the	DET
ajst-23543	8	6	reliability	reliability	NOUN
ajst-23543	8	7	of	of	ADP
ajst-23543	8	8	the	the	DET
ajst-23543	8	9	model	model	NOUN
ajst-23543	8	10	on	on	ADP
ajst-23543	8	11	several	several	ADJ
ajst-23543	8	12	data	datum	NOUN
ajst-23543	8	13	sets	set	NOUN
ajst-23543	8	14	.	.	PUNCT
ajst-23543	9	1	keywords	keyword	NOUN
ajst-23543	9	2	:	:	PUNCT
ajst-23543	9	3	attention	attention	NOUN
ajst-23543	9	4	;	;	PUNCT
ajst-23543	9	5	neural	neural	ADJ
ajst-23543	9	6	network	network	NOUN
ajst-23543	9	7	;	;	PUNCT
ajst-23543	9	8	feature	feature	NOUN
ajst-23543	9	9	extraction	extraction	NOUN
ajst-23543	9	10	.	.	PUNCT
ajst-23543	10	1	1	1	X
ajst-23543	10	2	.	.	X
ajst-23543	10	3	introduction	introduction	NOUN
ajst-23543	10	4	in	in	ADP
ajst-23543	10	5	the	the	DET
ajst-23543	10	6	wide	wide	ADJ
ajst-23543	10	7	field	field	NOUN
ajst-23543	10	8	of	of	ADP
ajst-23543	10	9	deep	deep	ADJ
ajst-23543	10	10	learning	learning	NOUN
ajst-23543	10	11	,	,	PUNCT
ajst-23543	10	12	the	the	DET
ajst-23543	10	13	selection	selection	NOUN
ajst-23543	10	14	and	and	CCONJ
ajst-23543	10	15	extraction	extraction	NOUN
ajst-23543	10	16	of	of	ADP
ajst-23543	10	17	image	image	NOUN
ajst-23543	10	18	features	feature	NOUN
ajst-23543	10	19	is	be	AUX
ajst-23543	10	20	one	one	NUM
ajst-23543	10	21	of	of	ADP
ajst-23543	10	22	the	the	DET
ajst-23543	10	23	core	core	NOUN
ajst-23543	10	24	factors	factor	NOUN
ajst-23543	10	25	affecting	affect	VERB
ajst-23543	10	26	model	model	NOUN
ajst-23543	10	27	performance	performance	NOUN
ajst-23543	10	28	.	.	PUNCT
ajst-23543	11	1	with	with	ADP
ajst-23543	11	2	the	the	DET
ajst-23543	11	3	development	development	NOUN
ajst-23543	11	4	of	of	ADP
ajst-23543	11	5	technology	technology	NOUN
ajst-23543	11	6	,	,	PUNCT
ajst-23543	11	7	the	the	DET
ajst-23543	11	8	field	field	NOUN
ajst-23543	11	9	has	have	AUX
ajst-23543	11	10	gradually	gradually	ADV
ajst-23543	11	11	shifted	shift	VERB
ajst-23543	11	12	from	from	ADP
ajst-23543	11	13	traditional	traditional	ADJ
ajst-23543	11	14	manual	manual	ADJ
ajst-23543	11	15	feature	feature	NOUN
ajst-23543	11	16	design	design	NOUN
ajst-23543	11	17	to	to	ADP
ajst-23543	11	18	automated	automate	VERB
ajst-23543	11	19	feature	feature	NOUN
ajst-23543	11	20	learning	learning	NOUN
ajst-23543	11	21	.	.	PUNCT
ajst-23543	12	1	however	however	ADV
ajst-23543	12	2	,	,	PUNCT
ajst-23543	12	3	while	while	SCONJ
ajst-23543	12	4	automated	automated	ADJ
ajst-23543	12	5	feature	feature	NOUN
ajst-23543	12	6	learning	learning	NOUN
ajst-23543	12	7	methods	method	NOUN
ajst-23543	12	8	,	,	PUNCT
ajst-23543	12	9	particularly	particularly	ADV
ajst-23543	12	10	convolutional	convolutional	ADJ
ajst-23543	12	11	neural	neural	ADJ
ajst-23543	12	12	networks	network	NOUN
ajst-23543	12	13	,	,	PUNCT
ajst-23543	12	14	have	have	AUX
ajst-23543	12	15	made	make	VERB
ajst-23543	12	16	significant	significant	ADJ
ajst-23543	12	17	progress	progress	NOUN
ajst-23543	12	18	in	in	ADP
ajst-23543	12	19	tasks	task	NOUN
ajst-23543	12	20	such	such	ADJ
ajst-23543	12	21	as	as	ADP
ajst-23543	12	22	image	image	NOUN
ajst-23543	12	23	recognition	recognition	NOUN
ajst-23543	12	24	,	,	PUNCT
ajst-23543	12	25	classification	classification	NOUN
ajst-23543	12	26	,	,	PUNCT
ajst-23543	12	27	and	and	CCONJ
ajst-23543	12	28	detection	detection	NOUN
ajst-23543	12	29	,	,	PUNCT
ajst-23543	12	30	these	these	DET
ajst-23543	12	31	methods	method	NOUN
ajst-23543	12	32	often	often	ADV
ajst-23543	12	33	rely	rely	VERB
ajst-23543	12	34	on	on	ADP
ajst-23543	12	35	large	large	ADJ
ajst-23543	12	36	amounts	amount	NOUN
ajst-23543	12	37	of	of	ADP
ajst-23543	12	38	data	datum	NOUN
ajst-23543	12	39	and	and	CCONJ
ajst-23543	12	40	computational	computational	ADJ
ajst-23543	12	41	resources	resource	NOUN
ajst-23543	12	42	,	,	PUNCT
ajst-23543	12	43	and	and	CCONJ
ajst-23543	12	44	in	in	ADP
ajst-23543	12	45	some	some	DET
ajst-23543	12	46	cases	case	NOUN
ajst-23543	12	47	still	still	ADV
ajst-23543	12	48	struggle	struggle	VERB
ajst-23543	12	49	to	to	PART
ajst-23543	12	50	capture	capture	VERB
ajst-23543	12	51	complex	complex	ADJ
ajst-23543	12	52	patterns	pattern	NOUN
ajst-23543	12	53	and	and	CCONJ
ajst-23543	12	54	subtle	subtle	ADJ
ajst-23543	12	55	differences	difference	NOUN
ajst-23543	12	56	in	in	ADP
ajst-23543	12	57	images	image	NOUN
ajst-23543	12	58	.	.	PUNCT
ajst-23543	13	1	therefore	therefore	ADV
ajst-23543	13	2	,	,	PUNCT
ajst-23543	13	3	how	how	SCONJ
ajst-23543	13	4	to	to	PART
ajst-23543	13	5	effectively	effectively	ADV
ajst-23543	13	6	select	select	VERB
ajst-23543	13	7	and	and	CCONJ
ajst-23543	13	8	extract	extract	VERB
ajst-23543	13	9	the	the	DET
ajst-23543	13	10	key	key	ADJ
ajst-23543	13	11	image	image	NOUN
ajst-23543	13	12	features	feature	VERB
ajst-23543	13	13	helpful	helpful	ADJ
ajst-23543	13	14	to	to	ADP
ajst-23543	13	15	the	the	DET
ajst-23543	13	16	task	task	NOUN
ajst-23543	13	17	has	have	AUX
ajst-23543	13	18	become	become	VERB
ajst-23543	13	19	the	the	DET
ajst-23543	13	20	key	key	NOUN
ajst-23543	13	21	to	to	PART
ajst-23543	13	22	improve	improve	VERB
ajst-23543	13	23	the	the	DET
ajst-23543	13	24	performance	performance	NOUN
ajst-23543	13	25	of	of	ADP
ajst-23543	13	26	the	the	DET
ajst-23543	13	27	model	model	NOUN
ajst-23543	13	28	.	.	PUNCT
ajst-23543	14	1	in	in	ADP
ajst-23543	14	2	this	this	DET
ajst-23543	14	3	research	research	NOUN
ajst-23543	14	4	context	context	NOUN
ajst-23543	14	5	,	,	PUNCT
ajst-23543	14	6	attention	attention	NOUN
ajst-23543	14	7	mechanism	mechanism	NOUN
ajst-23543	14	8	as	as	ADP
ajst-23543	14	9	a	a	DET
ajst-23543	14	10	new	new	ADJ
ajst-23543	14	11	technology	technology	NOUN
ajst-23543	14	12	,	,	PUNCT
ajst-23543	14	13	the	the	DET
ajst-23543	14	14	core	core	NOUN
ajst-23543	14	15	idea	idea	NOUN
ajst-23543	14	16	is	be	AUX
ajst-23543	14	17	derived	derive	VERB
ajst-23543	14	18	from	from	ADP
ajst-23543	14	19	the	the	DET
ajst-23543	14	20	working	work	VERB
ajst-23543	14	21	principle	principle	NOUN
ajst-23543	14	22	of	of	ADP
ajst-23543	14	23	the	the	DET
ajst-23543	14	24	human	human	ADJ
ajst-23543	14	25	visual	visual	ADJ
ajst-23543	14	26	system	system	NOUN
ajst-23543	14	27	,	,	PUNCT
ajst-23543	14	28	that	that	ADV
ajst-23543	14	29	is	is	ADV
ajst-23543	14	30	,	,	PUNCT
ajst-23543	14	31	when	when	SCONJ
ajst-23543	14	32	processing	process	VERB
ajst-23543	14	33	a	a	DET
ajst-23543	14	34	large	large	ADJ
ajst-23543	14	35	amount	amount	NOUN
ajst-23543	14	36	of	of	ADP
ajst-23543	14	37	information	information	NOUN
ajst-23543	14	38	,	,	PUNCT
ajst-23543	14	39	it	it	PRON
ajst-23543	14	40	can	can	AUX
ajst-23543	14	41	automatically	automatically	ADV
ajst-23543	14	42	filter	filter	VERB
ajst-23543	14	43	and	and	CCONJ
ajst-23543	14	44	focus	focus	VERB
ajst-23543	14	45	on	on	ADP
ajst-23543	14	46	the	the	DET
ajst-23543	14	47	key	key	ADJ
ajst-23543	14	48	information	information	NOUN
ajst-23543	14	49	.	.	PUNCT
ajst-23543	15	1	this	this	DET
ajst-23543	15	2	mechanism	mechanism	NOUN
ajst-23543	15	3	allows	allow	VERB
ajst-23543	15	4	the	the	DET
ajst-23543	15	5	model	model	NOUN
ajst-23543	15	6	to	to	PART
ajst-23543	15	7	dynamically	dynamically	ADV
ajst-23543	15	8	adjust	adjust	VERB
ajst-23543	15	9	the	the	DET
ajst-23543	15	10	attention	attention	NOUN
ajst-23543	15	11	of	of	ADP
ajst-23543	15	12	different	different	ADJ
ajst-23543	15	13	regions	region	NOUN
ajst-23543	15	14	or	or	CCONJ
ajst-23543	15	15	features	feature	NOUN
ajst-23543	15	16	when	when	SCONJ
ajst-23543	15	17	processing	processing	NOUN
ajst-23543	15	18	data	datum	NOUN
ajst-23543	15	19	,	,	PUNCT
ajst-23543	15	20	thus	thus	ADV
ajst-23543	15	21	achieving	achieve	VERB
ajst-23543	15	22	remarkable	remarkable	ADJ
ajst-23543	15	23	results	result	NOUN
ajst-23543	15	24	in	in	ADP
ajst-23543	15	25	many	many	ADJ
ajst-23543	15	26	fields	field	NOUN
ajst-23543	15	27	such	such	ADJ
ajst-23543	15	28	as	as	ADP
ajst-23543	15	29	natural	natural	ADJ
ajst-23543	15	30	language	language	NOUN
ajst-23543	15	31	processing	processing	NOUN
ajst-23543	15	32	,	,	PUNCT
ajst-23543	15	33	speech	speech	NOUN
ajst-23543	15	34	recognition	recognition	NOUN
ajst-23543	15	35	,	,	PUNCT
ajst-23543	15	36	and	and	CCONJ
ajst-23543	15	37	image	image	NOUN
ajst-23543	15	38	recognition	recognition	NOUN
ajst-23543	15	39	.	.	PUNCT
ajst-23543	16	1	the	the	DET
ajst-23543	16	2	attention	attention	NOUN
ajst-23543	16	3	mechanism	mechanism	NOUN
ajst-23543	16	4	not	not	PART
ajst-23543	16	5	only	only	ADV
ajst-23543	16	6	improves	improve	VERB
ajst-23543	16	7	the	the	DET
ajst-23543	16	8	efficiency	efficiency	NOUN
ajst-23543	16	9	of	of	ADP
ajst-23543	16	10	the	the	DET
ajst-23543	16	11	model	model	NOUN
ajst-23543	16	12	in	in	ADP
ajst-23543	16	13	processing	processing	NOUN
ajst-23543	16	14	information	information	NOUN
ajst-23543	16	15	,	,	PUNCT
ajst-23543	16	16	but	but	CCONJ
ajst-23543	16	17	also	also	ADV
ajst-23543	16	18	enhances	enhance	VERB
ajst-23543	16	19	the	the	DET
ajst-23543	16	20	ability	ability	NOUN
ajst-23543	16	21	of	of	ADP
ajst-23543	16	22	the	the	DET
ajst-23543	16	23	model	model	NOUN
ajst-23543	16	24	to	to	PART
ajst-23543	16	25	capture	capture	VERB
ajst-23543	16	26	important	important	ADJ
ajst-23543	16	27	features	feature	NOUN
ajst-23543	16	28	in	in	ADP
ajst-23543	16	29	the	the	DET
ajst-23543	16	30	data	datum	NOUN
ajst-23543	16	31	,	,	PUNCT
ajst-23543	16	32	enabling	enable	VERB
ajst-23543	16	33	the	the	DET
ajst-23543	16	34	model	model	NOUN
ajst-23543	16	35	to	to	PART
ajst-23543	16	36	make	make	VERB
ajst-23543	16	37	more	more	ADV
ajst-23543	16	38	accurate	accurate	ADJ
ajst-23543	16	39	and	and	CCONJ
ajst-23543	16	40	robust	robust	ADJ
ajst-23543	16	41	predictions	prediction	NOUN
ajst-23543	16	42	in	in	ADP
ajst-23543	16	43	the	the	DET
ajst-23543	16	44	complex	complex	ADJ
ajst-23543	16	45	data	datum	NOUN
ajst-23543	16	46	environment	environment	NOUN
ajst-23543	17	1	[	[	X
ajst-23543	17	2	1	1	NUM
ajst-23543	17	3	]	]	PUNCT
ajst-23543	17	4	.	.	PUNCT
ajst-23543	18	1	in	in	ADP
ajst-23543	18	2	view	view	NOUN
ajst-23543	18	3	of	of	ADP
ajst-23543	18	4	the	the	DET
ajst-23543	18	5	great	great	ADJ
ajst-23543	18	6	potential	potential	NOUN
ajst-23543	18	7	of	of	ADP
ajst-23543	18	8	attention	attention	NOUN
ajst-23543	18	9	mechanism	mechanism	NOUN
ajst-23543	18	10	in	in	ADP
ajst-23543	18	11	improving	improve	VERB
ajst-23543	18	12	model	model	NOUN
ajst-23543	18	13	performance	performance	NOUN
ajst-23543	18	14	,	,	PUNCT
ajst-23543	18	15	a	a	DET
ajst-23543	18	16	novel	novel	ADJ
ajst-23543	18	17	image	image	NOUN
ajst-23543	18	18	feature	feature	NOUN
ajst-23543	18	19	selection	selection	NOUN
ajst-23543	18	20	method	method	NOUN
ajst-23543	18	21	based	base	VERB
ajst-23543	18	22	on	on	ADP
ajst-23543	18	23	attention	attention	NOUN
ajst-23543	18	24	mechanism	mechanism	NOUN
ajst-23543	18	25	is	be	AUX
ajst-23543	18	26	proposed	propose	VERB
ajst-23543	18	27	in	in	ADP
ajst-23543	18	28	this	this	DET
ajst-23543	18	29	paper	paper	NOUN
ajst-23543	18	30	.	.	PUNCT
ajst-23543	19	1	this	this	DET
ajst-23543	19	2	method	method	NOUN
ajst-23543	19	3	aims	aim	VERB
ajst-23543	19	4	to	to	PART
ajst-23543	19	5	optimize	optimize	VERB
ajst-23543	19	6	the	the	DET
ajst-23543	19	7	feature	feature	NOUN
ajst-23543	19	8	learning	learn	VERB
ajst-23543	19	9	ability	ability	NOUN
ajst-23543	19	10	of	of	ADP
ajst-23543	19	11	convolutional	convolutional	ADJ
ajst-23543	19	12	neural	neural	ADJ
ajst-23543	19	13	networks	network	NOUN
ajst-23543	19	14	by	by	ADP
ajst-23543	19	15	introducing	introduce	VERB
ajst-23543	19	16	attention	attention	NOUN
ajst-23543	19	17	mechanisms	mechanism	NOUN
ajst-23543	19	18	,	,	PUNCT
ajst-23543	19	19	so	so	SCONJ
ajst-23543	19	20	as	as	SCONJ
ajst-23543	19	21	to	to	PART
ajst-23543	19	22	improve	improve	VERB
ajst-23543	19	23	the	the	DET
ajst-23543	19	24	accuracy	accuracy	NOUN
ajst-23543	19	25	and	and	CCONJ
ajst-23543	19	26	efficiency	efficiency	NOUN
ajst-23543	19	27	of	of	ADP
ajst-23543	19	28	image	image	NOUN
ajst-23543	19	29	classification	classification	NOUN
ajst-23543	19	30	and	and	CCONJ
ajst-23543	19	31	object	object	NOUN
ajst-23543	19	32	detection	detection	NOUN
ajst-23543	19	33	tasks	task	NOUN
ajst-23543	19	34	[	[	X
ajst-23543	19	35	2	2	NUM
ajst-23543	19	36	]	]	PUNCT
ajst-23543	19	37	.	.	PUNCT
ajst-23543	20	1	we	we	PRON
ajst-23543	20	2	first	first	ADV
ajst-23543	20	3	introduce	introduce	VERB
ajst-23543	20	4	the	the	DET
ajst-23543	20	5	fundamentals	fundamental	NOUN
ajst-23543	20	6	of	of	ADP
ajst-23543	20	7	the	the	DET
ajst-23543	20	8	attention	attention	NOUN
ajst-23543	20	9	mechanism	mechanism	NOUN
ajst-23543	20	10	and	and	CCONJ
ajst-23543	20	11	detail	detail	NOUN
ajst-23543	20	12	how	how	SCONJ
ajst-23543	20	13	it	it	PRON
ajst-23543	20	14	helps	help	VERB
ajst-23543	20	15	the	the	DET
ajst-23543	20	16	model	model	NOUN
ajst-23543	20	17	focus	focus	VERB
ajst-23543	20	18	on	on	ADP
ajst-23543	20	19	key	key	ADJ
ajst-23543	20	20	areas	area	NOUN
ajst-23543	20	21	in	in	ADP
ajst-23543	20	22	the	the	DET
ajst-23543	20	23	image	image	NOUN
ajst-23543	20	24	.	.	PUNCT
ajst-23543	21	1	next	next	ADV
ajst-23543	21	2	,	,	PUNCT
ajst-23543	21	3	we	we	PRON
ajst-23543	21	4	design	design	VERB
ajst-23543	21	5	and	and	CCONJ
ajst-23543	21	6	implement	implement	VERB
ajst-23543	21	7	a	a	DET
ajst-23543	21	8	cnn	cnn	NOUN
ajst-23543	21	9	framework	framework	NOUN
ajst-23543	21	10	with	with	ADP
ajst-23543	21	11	an	an	DET
ajst-23543	21	12	integrated	integrate	VERB
ajst-23543	21	13	attention	attention	NOUN
ajst-23543	21	14	module	module	NOUN
ajst-23543	21	15	,	,	PUNCT
ajst-23543	21	16	which	which	PRON
ajst-23543	21	17	adaptively	adaptively	ADV
ajst-23543	21	18	adjusts	adjust	VERB
ajst-23543	21	19	the	the	DET
ajst-23543	21	20	weights	weight	NOUN
ajst-23543	21	21	to	to	PART
ajst-23543	21	22	highlight	highlight	VERB
ajst-23543	21	23	important	important	ADJ
ajst-23543	21	24	features	feature	NOUN
ajst-23543	21	25	and	and	CCONJ
ajst-23543	21	26	suppress	suppress	VERB
ajst-23543	21	27	irrelevant	irrelevant	ADJ
ajst-23543	21	28	background	background	NOUN
ajst-23543	21	29	,	,	PUNCT
ajst-23543	21	30	thus	thus	ADV
ajst-23543	21	31	making	make	VERB
ajst-23543	21	32	the	the	DET
ajst-23543	21	33	model	model	NOUN
ajst-23543	21	34	more	more	ADV
ajst-23543	21	35	focused	focused	ADJ
ajst-23543	21	36	on	on	ADP
ajst-23543	21	37	learning	learn	VERB
ajst-23543	21	38	meaningful	meaningful	ADJ
ajst-23543	21	39	image	image	NOUN
ajst-23543	21	40	information	information	NOUN
ajst-23543	21	41	.	.	PUNCT
ajst-23543	22	1	through	through	ADP
ajst-23543	22	2	this	this	DET
ajst-23543	22	3	innovative	innovative	ADJ
ajst-23543	22	4	approach	approach	NOUN
ajst-23543	22	5	,	,	PUNCT
ajst-23543	22	6	our	our	PRON
ajst-23543	22	7	model	model	NOUN
ajst-23543	22	8	can	can	AUX
ajst-23543	22	9	not	not	PART
ajst-23543	22	10	only	only	ADV
ajst-23543	22	11	learn	learn	VERB
ajst-23543	22	12	the	the	DET
ajst-23543	22	13	key	key	ADJ
ajst-23543	22	14	features	feature	NOUN
ajst-23543	22	15	in	in	ADP
ajst-23543	22	16	the	the	DET
ajst-23543	22	17	image	image	NOUN
ajst-23543	22	18	more	more	ADV
ajst-23543	22	19	effectively	effectively	ADV
ajst-23543	22	20	,	,	PUNCT
ajst-23543	22	21	but	but	CCONJ
ajst-23543	22	22	also	also	ADV
ajst-23543	22	23	reduce	reduce	VERB
ajst-23543	22	24	the	the	DET
ajst-23543	22	25	waste	waste	NOUN
ajst-23543	22	26	of	of	ADP
ajst-23543	22	27	computing	compute	VERB
ajst-23543	22	28	resources	resource	NOUN
ajst-23543	22	29	and	and	CCONJ
ajst-23543	22	30	improve	improve	VERB
ajst-23543	22	31	the	the	DET
ajst-23543	22	32	generalization	generalization	NOUN
ajst-23543	22	33	ability	ability	NOUN
ajst-23543	22	34	of	of	ADP
ajst-23543	22	35	the	the	DET
ajst-23543	22	36	model	model	NOUN
ajst-23543	22	37	.	.	PUNCT
ajst-23543	23	1	to	to	PART
ajst-23543	23	2	verify	verify	VERB
ajst-23543	23	3	the	the	DET
ajst-23543	23	4	validity	validity	NOUN
ajst-23543	23	5	of	of	ADP
ajst-23543	23	6	the	the	DET
ajst-23543	23	7	proposed	propose	VERB
ajst-23543	23	8	method	method	NOUN
ajst-23543	23	9	,	,	PUNCT
ajst-23543	23	10	we	we	PRON
ajst-23543	23	11	conducted	conduct	VERB
ajst-23543	23	12	extensive	extensive	ADJ
ajst-23543	23	13	experiments	experiment	NOUN
ajst-23543	23	14	on	on	ADP
ajst-23543	23	15	multiple	multiple	ADJ
ajst-23543	23	16	public	public	ADJ
ajst-23543	23	17	datasets	dataset	NOUN
ajst-23543	23	18	and	and	CCONJ
ajst-23543	23	19	compared	compare	VERB
ajst-23543	23	20	them	they	PRON
ajst-23543	23	21	with	with	ADP
ajst-23543	23	22	traditional	traditional	ADJ
ajst-23543	23	23	methods	method	NOUN
ajst-23543	23	24	[	[	X
ajst-23543	23	25	3,4	3,4	NUM
ajst-23543	23	26	]	]	PUNCT
ajst-23543	23	27	.	.	PUNCT
ajst-23543	24	1	the	the	DET
ajst-23543	24	2	experimental	experimental	ADJ
ajst-23543	24	3	results	result	NOUN
ajst-23543	24	4	demonstrate	demonstrate	VERB
ajst-23543	24	5	the	the	DET
ajst-23543	24	6	superiority	superiority	NOUN
ajst-23543	24	7	of	of	ADP
ajst-23543	24	8	our	our	PRON
ajst-23543	24	9	method	method	NOUN
ajst-23543	24	10	in	in	ADP
ajst-23543	24	11	improving	improve	VERB
ajst-23543	24	12	the	the	DET
ajst-23543	24	13	performance	performance	NOUN
ajst-23543	24	14	of	of	ADP
ajst-23543	24	15	image	image	NOUN
ajst-23543	24	16	processing	processing	NOUN
ajst-23543	24	17	tasks	task	NOUN
ajst-23543	24	18	.	.	PUNCT
ajst-23543	25	1	in	in	ADP
ajst-23543	25	2	addition	addition	NOUN
ajst-23543	25	3	,	,	PUNCT
ajst-23543	25	4	we	we	PRON
ajst-23543	25	5	further	far	ADV
ajst-23543	25	6	explore	explore	VERB
ajst-23543	25	7	the	the	DET
ajst-23543	25	8	effect	effect	NOUN
ajst-23543	25	9	of	of	ADP
ajst-23543	25	10	attention	attention	NOUN
ajst-23543	25	11	mechanism	mechanism	NOUN
ajst-23543	25	12	on	on	ADP
ajst-23543	25	13	different	different	ADJ
ajst-23543	25	14	types	type	NOUN
ajst-23543	25	15	of	of	ADP
ajst-23543	25	16	image	image	NOUN
ajst-23543	25	17	data	datum	NOUN
ajst-23543	25	18	,	,	PUNCT
ajst-23543	25	19	including	include	VERB
ajst-23543	25	20	remote	remote	ADJ
ajst-23543	25	21	sensing	sense	VERB
ajst-23543	25	22	image	image	NOUN
ajst-23543	25	23	,	,	PUNCT
ajst-23543	25	24	medical	medical	ADJ
ajst-23543	25	25	image	image	NOUN
ajst-23543	25	26	and	and	CCONJ
ajst-23543	25	27	natural	natural	ADJ
ajst-23543	25	28	scene	scene	NOUN
ajst-23543	25	29	image	image	NOUN
ajst-23543	25	30	.	.	PUNCT
ajst-23543	26	1	these	these	DET
ajst-23543	26	2	experimental	experimental	ADJ
ajst-23543	26	3	results	result	NOUN
ajst-23543	26	4	show	show	VERB
ajst-23543	26	5	that	that	SCONJ
ajst-23543	26	6	our	our	PRON
ajst-23543	26	7	approach	approach	NOUN
ajst-23543	26	8	can	can	AUX
ajst-23543	26	9	effectively	effectively	ADV
ajst-23543	26	10	improve	improve	VERB
ajst-23543	26	11	the	the	DET
ajst-23543	26	12	performance	performance	NOUN
ajst-23543	26	13	of	of	ADP
ajst-23543	26	14	the	the	DET
ajst-23543	26	15	model	model	NOUN
ajst-23543	26	16	in	in	ADP
ajst-23543	26	17	both	both	CCONJ
ajst-23543	26	18	high	high	ADJ
ajst-23543	26	19	resolution	resolution	NOUN
ajst-23543	26	20	and	and	CCONJ
ajst-23543	26	21	low	low	ADJ
ajst-23543	26	22	light	light	ADJ
ajst-23543	26	23	conditions	condition	NOUN
ajst-23543	26	24	,	,	PUNCT
ajst-23543	26	25	and	and	CCONJ
ajst-23543	26	26	in	in	ADP
ajst-23543	26	27	both	both	CCONJ
ajst-23543	26	28	simple	simple	ADJ
ajst-23543	26	29	and	and	CCONJ
ajst-23543	26	30	complex	complex	ADJ
ajst-23543	26	31	scenes	scene	NOUN
ajst-23543	26	32	.	.	PUNCT
ajst-23543	27	1	this	this	PRON
ajst-23543	27	2	is	be	AUX
ajst-23543	27	3	particularly	particularly	ADV
ajst-23543	27	4	important	important	ADJ
ajst-23543	27	5	because	because	SCONJ
ajst-23543	27	6	it	it	PRON
ajst-23543	27	7	shows	show	VERB
ajst-23543	27	8	that	that	SCONJ
ajst-23543	27	9	our	our	PRON
ajst-23543	27	10	approach	approach	NOUN
ajst-23543	27	11	has	have	VERB
ajst-23543	27	12	strong	strong	ADJ
ajst-23543	27	13	generalization	generalization	NOUN
ajst-23543	27	14	and	and	CCONJ
ajst-23543	27	15	adaptability	adaptability	NOUN
ajst-23543	27	16	,	,	PUNCT
ajst-23543	27	17	and	and	CCONJ
ajst-23543	27	18	can	can	AUX
ajst-23543	27	19	meet	meet	VERB
ajst-23543	27	20	various	various	ADJ
ajst-23543	27	21	challenges	challenge	NOUN
ajst-23543	27	22	in	in	ADP
ajst-23543	27	23	practical	practical	ADJ
ajst-23543	27	24	application	application	NOUN
ajst-23543	27	25	scenarios	scenario	NOUN
ajst-23543	27	26	.	.	PUNCT
ajst-23543	28	1	to	to	PART
ajst-23543	28	2	sum	sum	VERB
ajst-23543	28	3	up	up	ADP
ajst-23543	28	4	,	,	PUNCT
ajst-23543	28	5	the	the	DET
ajst-23543	28	6	research	research	NOUN
ajst-23543	28	7	in	in	ADP
ajst-23543	28	8	this	this	DET
ajst-23543	28	9	paper	paper	NOUN
ajst-23543	28	10	not	not	PART
ajst-23543	28	11	only	only	ADV
ajst-23543	28	12	provides	provide	VERB
ajst-23543	28	13	a	a	DET
ajst-23543	28	14	new	new	ADJ
ajst-23543	28	15	solution	solution	NOUN
ajst-23543	28	16	for	for	ADP
ajst-23543	28	17	image	image	NOUN
ajst-23543	28	18	feature	feature	NOUN
ajst-23543	28	19	selection	selection	NOUN
ajst-23543	28	20	and	and	CCONJ
ajst-23543	28	21	extraction	extraction	NOUN
ajst-23543	28	22	,	,	PUNCT
ajst-23543	28	23	but	but	CCONJ
ajst-23543	28	24	also	also	ADV
ajst-23543	28	25	opens	open	VERB
ajst-23543	28	26	up	up	ADP
ajst-23543	28	27	new	new	ADJ
ajst-23543	28	28	possibilities	possibility	NOUN
ajst-23543	28	29	for	for	ADP
ajst-23543	28	30	the	the	DET
ajst-23543	28	31	research	research	NOUN
ajst-23543	28	32	and	and	CCONJ
ajst-23543	28	33	application	application	NOUN
ajst-23543	28	34	of	of	ADP
ajst-23543	28	35	deep	deep	ADJ
ajst-23543	28	36	learning	learning	NOUN
ajst-23543	28	37	.	.	PUNCT
ajst-23543	29	1	we	we	PRON
ajst-23543	29	2	believe	believe	VERB
ajst-23543	29	3	that	that	SCONJ
ajst-23543	29	4	as	as	SCONJ
ajst-23543	29	5	technology	technology	NOUN
ajst-23543	29	6	continues	continue	VERB
ajst-23543	29	7	to	to	PART
ajst-23543	29	8	advance	advance	VERB
ajst-23543	29	9	and	and	CCONJ
ajst-23543	29	10	improve	improve	VERB
ajst-23543	29	11	,	,	PUNCT
ajst-23543	29	12	attention	attention	NOUN
ajst-23543	29	13	mechanisms	mechanism	NOUN
ajst-23543	29	14	will	will	AUX
ajst-23543	29	15	play	play	VERB
ajst-23543	29	16	an	an	DET
ajst-23543	29	17	even	even	ADV
ajst-23543	29	18	more	more	ADV
ajst-23543	29	19	important	important	ADJ
ajst-23543	29	20	role	role	NOUN
ajst-23543	29	21	in	in	ADP
ajst-23543	29	22	image	image	NOUN
ajst-23543	29	23	processing	processing	NOUN
ajst-23543	29	24	and	and	CCONJ
ajst-23543	29	25	computer	computer	NOUN
ajst-23543	29	26	vision	vision	NOUN
ajst-23543	29	27	tasks	task	NOUN
ajst-23543	29	28	in	in	ADP
ajst-23543	29	29	the	the	DET
ajst-23543	29	30	future	future	NOUN
ajst-23543	29	31	.	.	PUNCT
ajst-23543	30	1	2	2	X
ajst-23543	30	2	.	.	X
ajst-23543	30	3	convolutional	convolutional	ADJ
ajst-23543	30	4	neural	neural	ADJ
ajst-23543	30	5	network	network	NOUN
ajst-23543	30	6	in	in	ADP
ajst-23543	30	7	the	the	DET
ajst-23543	30	8	fully	fully	ADV
ajst-23543	30	9	connected	connected	ADJ
ajst-23543	30	10	neural	neural	ADJ
ajst-23543	30	11	network	network	NOUN
ajst-23543	30	12	(	(	PUNCT
ajst-23543	30	13	fully	fully	ADV
ajst-23543	30	14	connected	connected	ADJ
ajst-23543	30	15	neural	neural	ADJ
ajst-23543	30	16	network	network	NOUN
ajst-23543	30	17	,	,	PUNCT
ajst-23543	30	18	fc	fc	PROPN
ajst-23543	30	19	)	)	PUNCT
ajst-23543	30	20	,	,	PUNCT
ajst-23543	30	21	it	it	PRON
ajst-23543	30	22	is	be	AUX
ajst-23543	30	23	assumed	assume	VERB
ajst-23543	30	24	that	that	SCONJ
ajst-23543	30	25	the	the	DET
ajst-23543	30	26	number	number	NOUN
ajst-23543	30	27	of	of	ADP
ajst-23543	30	28	neurons	neuron	NOUN
ajst-23543	30	29	at	at	ADP
ajst-23543	30	30	the	the	DET
ajst-23543	30	31	layer	layer	NOUN
ajst-23543	30	32	is	be	AUX
ajst-23543	30	33	and	and	CCONJ
ajst-23543	30	34	the	the	DET
ajst-23543	30	35	number	number	NOUN
ajst-23543	30	36	of	of	ADP
ajst-23543	30	37	neurons	neuron	NOUN
ajst-23543	30	38	at	at	ADP
ajst-23543	30	39	the	the	DET
ajst-23543	30	40	−1	−1	NOUN
ajst-23543	30	41	layer	layer	NOUN
ajst-23543	30	42	is	be	AUX
ajst-23543	30	43	−1	−1	NOUN
ajst-23543	30	44	.	.	PUNCT
ajst-23543	31	1	then	then	ADV
ajst-23543	31	2	−1×	−1×	PUNCT
ajst-23543	31	3	connection	connection	NOUN
ajst-23543	31	4	edges	edge	NOUN
ajst-23543	31	5	are	be	AUX
ajst-23543	31	6	required	require	VERB
ajst-23543	31	7	to	to	PART
ajst-23543	31	8	connect	connect	VERB
ajst-23543	31	9	the	the	DET
ajst-23543	31	10	−1	−1	NOUN
ajst-23543	31	11	layer	layer	NOUN
ajst-23543	31	12	with	with	ADP
ajst-23543	31	13	the	the	DET
ajst-23543	31	14	layer	layer	NOUN
ajst-23543	31	15	,	,	PUNCT
ajst-23543	31	16	that	that	ADV
ajst-23543	31	17	is	is	ADV
ajst-23543	31	18	,	,	PUNCT
ajst-23543	31	19	the	the	DET
ajst-23543	31	20	number	number	NOUN
ajst-23543	31	21	of	of	ADP
ajst-23543	31	22	parameters	parameter	NOUN
ajst-23543	31	23	contained	contain	VERB
ajst-23543	31	24	in	in	ADP
ajst-23543	31	25	the	the	DET
ajst-23543	31	26	weight	weight	NOUN
ajst-23543	31	27	matrix	matrix	NOUN
ajst-23543	31	28	is	be	AUX
ajst-23543	31	29	86	86	NUM
ajst-23543	31	30	−1×	−1×	NOUN
ajst-23543	31	31	.	.	PUNCT
ajst-23543	32	1	when	when	SCONJ
ajst-23543	32	2	the	the	DET
ajst-23543	32	3	number	number	NOUN
ajst-23543	32	4	of	of	ADP
ajst-23543	32	5	neurons	neuron	NOUN
ajst-23543	32	6	is	be	AUX
ajst-23543	32	7	large	large	ADJ
ajst-23543	32	8	,	,	PUNCT
ajst-23543	32	9	the	the	DET
ajst-23543	32	10	number	number	NOUN
ajst-23543	32	11	of	of	ADP
ajst-23543	32	12	parameters	parameter	NOUN
ajst-23543	32	13	in	in	ADP
ajst-23543	32	14	the	the	DET
ajst-23543	32	15	weight	weight	NOUN
ajst-23543	32	16	matrix	matrix	NOUN
ajst-23543	32	17	will	will	AUX
ajst-23543	32	18	be	be	AUX
ajst-23543	32	19	large	large	ADJ
ajst-23543	32	20	,	,	PUNCT
ajst-23543	32	21	resulting	result	VERB
ajst-23543	32	22	in	in	ADP
ajst-23543	32	23	a	a	DET
ajst-23543	32	24	decrease	decrease	NOUN
ajst-23543	32	25	in	in	ADP
ajst-23543	32	26	the	the	DET
ajst-23543	32	27	training	training	NOUN
ajst-23543	32	28	efficiency	efficiency	NOUN
ajst-23543	32	29	of	of	ADP
ajst-23543	32	30	the	the	DET
ajst-23543	32	31	neural	neural	ADJ
ajst-23543	32	32	network	network	NOUN
ajst-23543	32	33	[	[	X
ajst-23543	32	34	5,6	5,6	NUM
ajst-23543	32	35	]	]	PUNCT
ajst-23543	32	36	.	.	PUNCT
ajst-23543	33	1	in	in	ADP
ajst-23543	33	2	the	the	DET
ajst-23543	33	3	convolutional	convolutional	ADJ
ajst-23543	33	4	layer	layer	NOUN
ajst-23543	33	5	,	,	PUNCT
ajst-23543	33	6	the	the	DET
ajst-23543	33	7	input	input	NOUN
ajst-23543	33	8	of	of	ADP
ajst-23543	33	9	layer	layer	NOUN
ajst-23543	33	10	is	be	AUX
ajst-23543	33	11	the	the	DET
ajst-23543	33	12	convolution	convolution	NOUN
ajst-23543	33	13	of	of	ADP
ajst-23543	33	14	the	the	DET
ajst-23543	33	15	activity	activity	NOUN
ajst-23543	33	16	value	value	NOUN
ajst-23543	33	17	−1	−1	NOUN
ajst-23543	33	18	of	of	ADP
ajst-23543	33	19	layer	layer	NOUN
ajst-23543	33	20	−1	−1	NOUN
ajst-23543	33	21	and	and	CCONJ
ajst-23543	33	22	the	the	DET
ajst-23543	33	23	filter	filter	NOUN
ajst-23543	33	24	,	,	PUNCT
ajst-23543	33	25	as	as	SCONJ
ajst-23543	33	26	follows	follow	VERB
ajst-23543	33	27	:	:	PUNCT
ajst-23543	33	28	1n	1n	NUM
ajst-23543	33	29	n	n	CCONJ
ajst-23543	33	30	n	n	PROPN
ajst-23543	33	31	nz	nz	PROPN
ajst-23543	33	32	w	w	PROPN
ajst-23543	33	33	a	a	PROPN
ajst-23543	33	34	b	b	PROPN
ajst-23543	33	35			ADJ
ajst-23543	33	36			X
ajst-23543	33	37	(	(	PUNCT
ajst-23543	33	38	1	1	X
ajst-23543	33	39	)	)	PUNCT
ajst-23543	33	40	where	where	SCONJ
ajst-23543	33	41	the	the	DET
ajst-23543	33	42	filter	filter	NOUN
ajst-23543	33	43	is	be	AUX
ajst-23543	33	44	the	the	DET
ajst-23543	33	45	weight	weight	NOUN
ajst-23543	33	46	vector	vector	NOUN
ajst-23543	33	47	and	and	CCONJ
ajst-23543	33	48	is	be	AUX
ajst-23543	33	49	the	the	DET
ajst-23543	33	50	bias	bias	NOUN
ajst-23543	33	51	quantity	quantity	NOUN
ajst-23543	33	52	.	.	PUNCT
ajst-23543	34	1	because	because	SCONJ
ajst-23543	34	2	of	of	ADP
ajst-23543	34	3	local	local	ADJ
ajst-23543	34	4	connections	connection	NOUN
ajst-23543	34	5	,	,	PUNCT
ajst-23543	34	6	each	each	DET
ajst-23543	34	7	neuron	neuron	NOUN
ajst-23543	34	8	in	in	ADP
ajst-23543	34	9	the	the	DET
ajst-23543	34	10	convolutional	convolutional	ADJ
ajst-23543	34	11	layer	layer	NOUN
ajst-23543	34	12	(	(	PUNCT
ajst-23543	34	13	we	we	PRON
ajst-23543	34	14	assume	assume	VERB
ajst-23543	34	15	layer	layer	NOUN
ajst-23543	34	16	−1	−1	NOUN
ajst-23543	34	17	)	)	PUNCT
ajst-23543	34	18	is	be	AUX
ajst-23543	34	19	connected	connect	VERB
ajst-23543	34	20	only	only	ADV
ajst-23543	34	21	to	to	ADP
ajst-23543	34	22	certain	certain	ADJ
ajst-23543	34	23	neurons	neuron	NOUN
ajst-23543	34	24	in	in	ADP
ajst-23543	34	25	the	the	DET
ajst-23543	34	26	next	next	ADJ
ajst-23543	34	27	layer	layer	NOUN
ajst-23543	34	28	(	(	PUNCT
ajst-23543	34	29	layer	layer	NOUN
ajst-23543	34	30	)	)	PUNCT
ajst-23543	34	31	,	,	PUNCT
ajst-23543	34	32	which	which	PRON
ajst-23543	34	33	are	be	AUX
ajst-23543	34	34	identified	identify	VERB
ajst-23543	34	35	by	by	ADP
ajst-23543	34	36	the	the	DET
ajst-23543	34	37	local	local	ADJ
ajst-23543	34	38	window	window	NOUN
ajst-23543	34	39	.	.	PUNCT
ajst-23543	35	1	compared	compare	VERB
ajst-23543	35	2	with	with	ADP
ajst-23543	35	3	the	the	DET
ajst-23543	35	4	fully	fully	ADV
ajst-23543	35	5	connected	connect	VERB
ajst-23543	35	6	layer	layer	NOUN
ajst-23543	35	7	,	,	PUNCT
ajst-23543	35	8	the	the	DET
ajst-23543	35	9	number	number	NOUN
ajst-23543	35	10	of	of	ADP
ajst-23543	35	11	connections	connection	NOUN
ajst-23543	35	12	between	between	ADP
ajst-23543	35	13	the	the	DET
ajst-23543	35	14	convolutional	convolutional	ADJ
ajst-23543	35	15	layer	layer	NOUN
ajst-23543	35	16	and	and	CCONJ
ajst-23543	35	17	the	the	DET
ajst-23543	35	18	next	next	ADJ
ajst-23543	35	19	layer	layer	NOUN
ajst-23543	35	20	is	be	AUX
ajst-23543	35	21	greatly	greatly	ADV
ajst-23543	35	22	reduced	reduce	VERB
ajst-23543	35	23	.	.	PUNCT
ajst-23543	36	1	the	the	DET
ajst-23543	36	2	number	number	NOUN
ajst-23543	36	3	of	of	ADP
ajst-23543	36	4	connections	connection	NOUN
ajst-23543	36	5	between	between	ADP
ajst-23543	36	6	the	the	DET
ajst-23543	36	7	original	original	ADJ
ajst-23543	36	8	−1×	−1×	NOUN
ajst-23543	36	9	is	be	AUX
ajst-23543	36	10	reduced	reduce	VERB
ajst-23543	36	11	to	to	ADP
ajst-23543	36	12	−1×	−1×	PROPN
ajst-23543	36	13	.	.	PUNCT
ajst-23543	37	1	the	the	DET
ajst-23543	37	2	size	size	NOUN
ajst-23543	37	3	of	of	ADP
ajst-23543	37	4	filters	filter	NOUN
ajst-23543	37	5	in	in	ADP
ajst-23543	37	6	the	the	DET
ajst-23543	37	7	convolutional	convolutional	ADJ
ajst-23543	37	8	layer	layer	NOUN
ajst-23543	37	9	is	be	AUX
ajst-23543	37	10	expressed	express	VERB
ajst-23543	37	11	as	as	ADP
ajst-23543	37	12	.	.	PUNCT
ajst-23543	38	1	because	because	SCONJ
ajst-23543	38	2	of	of	ADP
ajst-23543	38	3	the	the	DET
ajst-23543	38	4	weight	weight	NOUN
ajst-23543	38	5	sharing	sharing	NOUN
ajst-23543	38	6	,	,	PUNCT
ajst-23543	38	7	filters	filter	NOUN
ajst-23543	38	8	can	can	AUX
ajst-23543	38	9	act	act	VERB
ajst-23543	38	10	on	on	ADP
ajst-23543	38	11	all	all	DET
ajst-23543	38	12	neurons	neuron	NOUN
ajst-23543	38	13	together	together	ADV
ajst-23543	38	14	.	.	PUNCT
ajst-23543	39	1	as	as	SCONJ
ajst-23543	39	2	shown	show	VERB
ajst-23543	39	3	in	in	ADP
ajst-23543	39	4	figure	figure	NOUN
ajst-23543	39	5	1	1	NUM
ajst-23543	39	6	,	,	PUNCT
ajst-23543	39	7	connections	connection	NOUN
ajst-23543	39	8	of	of	ADP
ajst-23543	39	9	the	the	DET
ajst-23543	39	10	same	same	ADJ
ajst-23543	39	11	color	color	NOUN
ajst-23543	39	12	all	all	PRON
ajst-23543	39	13	have	have	VERB
ajst-23543	39	14	the	the	DET
ajst-23543	39	15	same	same	ADJ
ajst-23543	39	16	weight	weight	NOUN
ajst-23543	39	17	.	.	PUNCT
ajst-23543	40	1	however	however	ADV
ajst-23543	40	2	,	,	PUNCT
ajst-23543	40	3	in	in	ADP
ajst-23543	40	4	weight	weight	NOUN
ajst-23543	40	5	sharing	sharing	NOUN
ajst-23543	40	6	,	,	PUNCT
ajst-23543	40	7	one	one	NUM
ajst-23543	40	8	filter	filter	NOUN
ajst-23543	40	9	can	can	AUX
ajst-23543	40	10	only	only	ADV
ajst-23543	40	11	extract	extract	VERB
ajst-23543	40	12	one	one	NUM
ajst-23543	40	13	local	local	ADJ
ajst-23543	40	14	feature	feature	NOUN
ajst-23543	40	15	,	,	PUNCT
ajst-23543	40	16	so	so	SCONJ
ajst-23543	40	17	in	in	ADP
ajst-23543	40	18	order	order	NOUN
ajst-23543	40	19	to	to	PART
ajst-23543	40	20	extract	extract	VERB
ajst-23543	40	21	multiple	multiple	ADJ
ajst-23543	40	22	features	feature	NOUN
ajst-23543	40	23	,	,	PUNCT
ajst-23543	40	24	it	it	PRON
ajst-23543	40	25	is	be	AUX
ajst-23543	40	26	often	often	ADV
ajst-23543	40	27	necessary	necessary	ADJ
ajst-23543	40	28	to	to	PART
ajst-23543	40	29	use	use	VERB
ajst-23543	40	30	multiple	multiple	ADJ
ajst-23543	40	31	filters	filter	NOUN
ajst-23543	40	32	in	in	ADP
ajst-23543	40	33	the	the	DET
ajst-23543	40	34	convolution	convolution	NOUN
ajst-23543	40	35	layer	layer	NOUN
ajst-23543	40	36	.	.	PUNCT
ajst-23543	41	1	figure	figure	NOUN
ajst-23543	41	2	1	1	NUM
ajst-23543	41	3	.	.	PUNCT
ajst-23543	42	1	convolution	convolution	NOUN
ajst-23543	42	2	layer	layer	NOUN
ajst-23543	42	3	scheme	scheme	NOUN
ajst-23543	42	4	figure	figure	NOUN
ajst-23543	42	5	2	2	NUM
ajst-23543	42	6	.	.	PUNCT
ajst-23543	43	1	the	the	DET
ajst-23543	43	2	three	three	NUM
ajst-23543	43	3	-	-	PUNCT
ajst-23543	43	4	dimensional	dimensional	ADJ
ajst-23543	43	5	structure	structure	NOUN
ajst-23543	43	6	of	of	ADP
ajst-23543	43	7	the	the	DET
ajst-23543	43	8	convolution	convolution	NOUN
ajst-23543	43	9	layer	layer	NOUN
ajst-23543	43	10	convolutional	convolutional	ADJ
ajst-23543	43	11	layer	layer	NOUN
ajst-23543	43	12	(	(	PUNCT
ajst-23543	43	13	convolutional	convolutional	ADJ
ajst-23543	43	14	layer	layer	NOUN
ajst-23543	43	15	)	)	PUNCT
ajst-23543	43	16	is	be	AUX
ajst-23543	43	17	the	the	DET
ajst-23543	43	18	core	core	NOUN
ajst-23543	43	19	infrastructure	infrastructure	NOUN
ajst-23543	43	20	of	of	ADP
ajst-23543	43	21	convolutional	convolutional	ADJ
ajst-23543	43	22	neural	neural	ADJ
ajst-23543	43	23	networks	network	NOUN
ajst-23543	43	24	[	[	X
ajst-23543	43	25	7	7	NUM
ajst-23543	43	26	]	]	PUNCT
ajst-23543	43	27	.	.	PUNCT
ajst-23543	44	1	the	the	DET
ajst-23543	44	2	function	function	NOUN
ajst-23543	44	3	of	of	ADP
ajst-23543	44	4	convolution	convolution	NOUN
ajst-23543	44	5	is	be	AUX
ajst-23543	44	6	to	to	PART
ajst-23543	44	7	extract	extract	VERB
ajst-23543	44	8	a	a	DET
ajst-23543	44	9	local	local	ADJ
ajst-23543	44	10	feature	feature	NOUN
ajst-23543	44	11	of	of	ADP
ajst-23543	44	12	the	the	DET
ajst-23543	44	13	input	input	NOUN
ajst-23543	44	14	data	datum	NOUN
ajst-23543	44	15	.	.	PUNCT
ajst-23543	45	1	every	every	DET
ajst-23543	45	2	convolution	convolution	NOUN
ajst-23543	45	3	kernel	kernel	NOUN
ajst-23543	45	4	in	in	ADP
ajst-23543	45	5	the	the	DET
ajst-23543	45	6	convolution	convolution	NOUN
ajst-23543	45	7	layer	layer	NOUN
ajst-23543	45	8	is	be	AUX
ajst-23543	45	9	a	a	DET
ajst-23543	45	10	feature	feature	NOUN
ajst-23543	45	11	extractor	extractor	NOUN
ajst-23543	45	12	,	,	PUNCT
ajst-23543	45	13	so	so	SCONJ
ajst-23543	45	14	features	feature	NOUN
ajst-23543	45	15	can	can	AUX
ajst-23543	45	16	be	be	AUX
ajst-23543	45	17	extracted	extract	VERB
ajst-23543	45	18	fully	fully	ADV
ajst-23543	45	19	.	.	PUNCT
ajst-23543	46	1	at	at	ADP
ajst-23543	46	2	present	present	ADJ
ajst-23543	46	3	,	,	PUNCT
ajst-23543	46	4	convolutional	convolutional	ADJ
ajst-23543	46	5	neural	neural	ADJ
ajst-23543	46	6	networks	network	NOUN
ajst-23543	46	7	are	be	AUX
ajst-23543	46	8	mainly	mainly	ADV
ajst-23543	46	9	applied	apply	VERB
ajst-23543	46	10	in	in	ADP
ajst-23543	46	11	image	image	NOUN
ajst-23543	46	12	processing	processing	NOUN
ajst-23543	46	13	,	,	PUNCT
ajst-23543	46	14	while	while	SCONJ
ajst-23543	46	15	images	image	NOUN
ajst-23543	46	16	belong	belong	VERB
ajst-23543	46	17	to	to	ADP
ajst-23543	46	18	two	two	NUM
ajst-23543	46	19	-	-	PUNCT
ajst-23543	46	20	dimensional	dimensional	ADJ
ajst-23543	46	21	data	datum	NOUN
ajst-23543	46	22	in	in	ADP
ajst-23543	46	23	data	datum	NOUN
ajst-23543	46	24	processing	processing	NOUN
ajst-23543	46	25	.	.	PUNCT
ajst-23543	47	1	therefore	therefore	ADV
ajst-23543	47	2	,	,	PUNCT
ajst-23543	47	3	in	in	ADP
ajst-23543	47	4	order	order	NOUN
ajst-23543	47	5	to	to	PART
ajst-23543	47	6	fully	fully	ADV
ajst-23543	47	7	extract	extract	VERB
ajst-23543	47	8	image	image	NOUN
ajst-23543	47	9	data	datum	NOUN
ajst-23543	47	10	features	feature	NOUN
ajst-23543	47	11	,	,	PUNCT
ajst-23543	47	12	the	the	DET
ajst-23543	47	13	convolutional	convolutional	ADJ
ajst-23543	47	14	layer	layer	NOUN
ajst-23543	47	15	needs	need	VERB
ajst-23543	47	16	to	to	PART
ajst-23543	47	17	be	be	AUX
ajst-23543	47	18	reconstructed	reconstruct	VERB
ajst-23543	47	19	into	into	ADP
ajst-23543	47	20	a	a	DET
ajst-23543	47	21	three	three	NUM
ajst-23543	47	22	-	-	PUNCT
ajst-23543	47	23	dimensional	dimensional	ADJ
ajst-23543	47	24	structure	structure	NOUN
ajst-23543	47	25	with	with	ADP
ajst-23543	47	26	the	the	DET
ajst-23543	47	27	dimensions	dimension	NOUN
ajst-23543	47	28	×	×	NOUN
ajst-23543	47	29	×	×	NOUN
ajst-23543	47	30	,	,	PUNCT
ajst-23543	47	31	where	where	SCONJ
ajst-23543	47	32	is	be	AUX
ajst-23543	47	33	the	the	DET
ajst-23543	47	34	width	width	NOUN
ajst-23543	47	35	,	,	PUNCT
ajst-23543	47	36	is	be	AUX
ajst-23543	47	37	the	the	DET
ajst-23543	47	38	width	width	NOUN
ajst-23543	47	39	and	and	CCONJ
ajst-23543	47	40	is	be	AUX
ajst-23543	47	41	the	the	DET
ajst-23543	47	42	depth	depth	NOUN
ajst-23543	47	43	.	.	PUNCT
ajst-23543	48	1	figure	figure	NOUN
ajst-23543	48	2	2	2	NUM
ajst-23543	48	3	shows	show	VERB
ajst-23543	48	4	the	the	DET
ajst-23543	48	5	three	three	NUM
ajst-23543	48	6	-	-	PUNCT
ajst-23543	48	7	dimensional	dimensional	ADJ
ajst-23543	48	8	structure	structure	NOUN
ajst-23543	48	9	of	of	ADP
ajst-23543	48	10	the	the	DET
ajst-23543	48	11	convolutional	convolutional	ADJ
ajst-23543	48	12	layer	layer	NOUN
ajst-23543	48	13	.	.	PUNCT
ajst-23543	49	1	in	in	ADP
ajst-23543	49	2	fig	fig	NOUN
ajst-23543	49	3	.	.	PUNCT
ajst-23543	50	1	2	2	NUM
ajst-23543	50	2	,	,	PUNCT
ajst-23543	50	3	m×	m×	AUX
ajst-23543	50	4	feature	feature	VERB
ajst-23543	50	5	maps	map	NOUN
ajst-23543	50	6	together	together	ADV
ajst-23543	50	7	constitute	constitute	VERB
ajst-23543	50	8	the	the	DET
ajst-23543	50	9	basic	basic	ADJ
ajst-23543	50	10	structure	structure	NOUN
ajst-23543	50	11	of	of	ADP
ajst-23543	50	12	the	the	DET
ajst-23543	50	13	convolutional	convolutional	ADJ
ajst-23543	50	14	layer	layer	NOUN
ajst-23543	50	15	.	.	PUNCT
ajst-23543	51	1	feature	feature	NOUN
ajst-23543	51	2	mapping	mapping	NOUN
ajst-23543	51	3	(	(	PUNCT
ajst-23543	51	4	feature	feature	NOUN
ajst-23543	51	5	map	map	NOUN
ajst-23543	51	6	)	)	PUNCT
ajst-23543	51	7	is	be	AUX
ajst-23543	51	8	the	the	DET
ajst-23543	51	9	feature	feature	NOUN
ajst-23543	51	10	extracted	extract	VERB
ajst-23543	51	11	from	from	ADP
ajst-23543	51	12	the	the	DET
ajst-23543	51	13	image	image	NOUN
ajst-23543	51	14	after	after	ADP
ajst-23543	51	15	product	product	NOUN
ajst-23543	51	16	operation	operation	NOUN
ajst-23543	51	17	,	,	PUNCT
ajst-23543	51	18	and	and	CCONJ
ajst-23543	51	19	each	each	DET
ajst-23543	51	20	individual	individual	ADJ
ajst-23543	51	21	feature	feature	NOUN
ajst-23543	51	22	mapping	mapping	NOUN
ajst-23543	51	23	represents	represent	VERB
ajst-23543	51	24	a	a	DET
ajst-23543	51	25	feature	feature	NOUN
ajst-23543	51	26	of	of	ADP
ajst-23543	51	27	the	the	DET
ajst-23543	51	28	image	image	NOUN
ajst-23543	51	29	.	.	PUNCT
ajst-23543	52	1	therefore	therefore	ADV
ajst-23543	52	2	,	,	PUNCT
ajst-23543	52	3	it	it	PRON
ajst-23543	52	4	is	be	AUX
ajst-23543	52	5	necessary	necessary	ADJ
ajst-23543	52	6	to	to	PART
ajst-23543	52	7	use	use	VERB
ajst-23543	52	8	multiple	multiple	ADJ
ajst-23543	52	9	feature	feature	NOUN
ajst-23543	52	10	maps	map	NOUN
ajst-23543	52	11	in	in	ADP
ajst-23543	52	12	each	each	DET
ajst-23543	52	13	convolutional	convolutional	ADJ
ajst-23543	52	14	layer	layer	NOUN
ajst-23543	52	15	to	to	PART
ajst-23543	52	16	improve	improve	VERB
ajst-23543	52	17	the	the	DET
ajst-23543	52	18	feature	feature	NOUN
ajst-23543	52	19	representation	representation	NOUN
ajst-23543	52	20	capability	capability	NOUN
ajst-23543	52	21	of	of	ADP
ajst-23543	52	22	the	the	DET
ajst-23543	52	23	network	network	NOUN
ajst-23543	52	24	.	.	PUNCT
ajst-23543	53	1	one	one	NUM
ajst-23543	53	2	-	-	PUNCT
ajst-23543	53	3	dimensional	dimensional	ADJ
ajst-23543	53	4	convolution	convolution	NOUN
ajst-23543	53	5	was	be	AUX
ajst-23543	53	6	originally	originally	ADV
ajst-23543	53	7	used	use	VERB
ajst-23543	53	8	in	in	ADP
ajst-23543	53	9	signal	signal	ADJ
ajst-23543	53	10	processing	processing	NOUN
ajst-23543	53	11	,	,	PUNCT
ajst-23543	53	12	often	often	ADV
ajst-23543	53	13	to	to	PART
ajst-23543	53	14	calculate	calculate	VERB
ajst-23543	53	15	the	the	DET
ajst-23543	53	16	delay	delay	NOUN
ajst-23543	53	17	accumulation	accumulation	NOUN
ajst-23543	53	18	of	of	ADP
ajst-23543	53	19	various	various	ADJ
ajst-23543	53	20	signals	signal	NOUN
ajst-23543	53	21	.	.	PUNCT
ajst-23543	54	1	we	we	PRON
ajst-23543	54	2	assume	assume	VERB
ajst-23543	54	3	that	that	SCONJ
ajst-23543	54	4	a	a	DET
ajst-23543	54	5	one	one	NUM
ajst-23543	54	6	-	-	PUNCT
ajst-23543	54	7	dimensional	dimensional	ADJ
ajst-23543	54	8	convolution	convolution	NOUN
ajst-23543	54	9	of	of	ADP
ajst-23543	54	10	the	the	DET
ajst-23543	54	11	input	input	NOUN
ajst-23543	54	12	signal	signal	NOUN
ajst-23543	54	13	sequence	sequence	NOUN
ajst-23543	54	14	1	1	NUM
ajst-23543	54	15	,	,	PUNCT
ajst-23543	54	16	2	2	NUM
ajst-23543	54	17	,	,	PUNCT
ajst-23543	54	18	∙∙∙	∙∙∙	NOUN
ajst-23543	54	19	,	,	PUNCT
ajst-23543	54	20	,	,	PUNCT
ajst-23543	54	21	convolution	convolution	NOUN
ajst-23543	54	22	layer	layer	NOUN
ajst-23543	54	23	of	of	ADP
ajst-23543	54	24	convolution	convolution	NOUN
ajst-23543	54	25	kernels	kernel	NOUN
ajst-23543	54	26	or	or	CCONJ
ajst-23543	54	27	filter	filter	NOUN
ajst-23543	54	28	1	1	NUM
ajst-23543	54	29	,	,	PUNCT
ajst-23543	54	30	2,∙∙∙	2,∙∙∙	NUM
ajst-23543	54	31	,	,	PUNCT
ajst-23543	54	32	,	,	PUNCT
ajst-23543	54	33	then	then	ADV
ajst-23543	54	34	filter	filter	NOUN
ajst-23543	54	35	and	and	CCONJ
ajst-23543	54	36	convolution	convolution	NOUN
ajst-23543	54	37	sequence	sequence	NOUN
ajst-23543	54	38	of	of	ADP
ajst-23543	54	39	the	the	DET
ajst-23543	54	40	input	input	NOUN
ajst-23543	54	41	signal	signal	NOUN
ajst-23543	54	42	as	as	SCONJ
ajst-23543	54	43	shown	show	VERB
ajst-23543	54	44	in	in	ADP
ajst-23543	54	45	type	type	NOUN
ajst-23543	54	46	2	2	NUM
ajst-23543	54	47	:	:	SYM
ajst-23543	54	48	1	1	NUM
ajst-23543	54	49	1	1	NUM
ajst-23543	54	50	m	m	NOUN
ajst-23543	54	51	t	t	NOUN
ajst-23543	54	52	k	k	PROPN
ajst-23543	54	53	t	t	PROPN
ajst-23543	55	1	k	k	PROPN
ajst-23543	56	1	k	k	PROPN
ajst-23543	57	1	y	y	PROPN
ajst-23543	57	2	x	x	PROPN
ajst-23543	57	3			PROPN
ajst-23543	57	4			VERB
ajst-23543	57	5			PROPN
ajst-23543	57	6			PROPN
ajst-23543	57	7			X
ajst-23543	57	8	(	(	PUNCT
ajst-23543	57	9	2	2	NUM
ajst-23543	57	10	)	)	PUNCT
ajst-23543	57	11	the	the	DET
ajst-23543	57	12	one	one	NUM
ajst-23543	57	13	-	-	PUNCT
ajst-23543	57	14	dimensional	dimensional	ADJ
ajst-23543	57	15	convolution	convolution	NOUN
ajst-23543	57	16	of	of	ADP
ajst-23543	57	17	the	the	DET
ajst-23543	57	18	signal	signal	ADJ
ajst-23543	57	19	sequence	sequence	NOUN
ajst-23543	57	20	and	and	CCONJ
ajst-23543	57	21	the	the	PRON
ajst-23543	57	22	of	of	ADP
ajst-23543	57	23	the	the	DET
ajst-23543	57	24	filter	filter	NOUN
ajst-23543	57	25	is	be	AUX
ajst-23543	57	26	defined	define	VERB
ajst-23543	57	27	as	as	SCONJ
ajst-23543	57	28	follows	follow	VERB
ajst-23543	57	29	:	:	PUNCT
ajst-23543	57	30	*	*	PUNCT
ajst-23543	57	31	y	y	NOUN
ajst-23543	57	32	x	x	PUNCT
ajst-23543	57	33	(	(	PUNCT
ajst-23543	57	34	3	3	X
ajst-23543	57	35	)	)	PUNCT
ajst-23543	57	36	where	where	SCONJ
ajst-23543	57	37	*	*	PUNCT
ajst-23543	57	38	is	be	AUX
ajst-23543	57	39	the	the	DET
ajst-23543	57	40	convolution	convolution	NOUN
ajst-23543	57	41	operator	operator	NOUN
ajst-23543	57	42	.	.	PUNCT
ajst-23543	58	1	in	in	ADP
ajst-23543	58	2	image	image	NOUN
ajst-23543	58	3	processing	processing	NOUN
ajst-23543	58	4	,	,	PUNCT
ajst-23543	58	5	two	two	NUM
ajst-23543	58	6	-	-	PUNCT
ajst-23543	58	7	dimensional	dimensional	ADJ
ajst-23543	58	8	convolution	convolution	NOUN
ajst-23543	58	9	is	be	AUX
ajst-23543	58	10	more	more	ADV
ajst-23543	58	11	commonly	commonly	ADV
ajst-23543	58	12	used	use	VERB
ajst-23543	58	13	than	than	ADP
ajst-23543	58	14	one	one	NUM
ajst-23543	58	15	-	-	PUNCT
ajst-23543	58	16	dimensional	dimensional	ADJ
ajst-23543	58	17	convolution	convolution	NOUN
ajst-23543	58	18	because	because	SCONJ
ajst-23543	58	19	the	the	DET
ajst-23543	58	20	image	image	NOUN
ajst-23543	58	21	data	data	VERB
ajst-23543	58	22	itself	itself	PRON
ajst-23543	58	23	is	be	AUX
ajst-23543	58	24	a	a	DET
ajst-23543	58	25	two	two	NUM
ajst-23543	58	26	-	-	PUNCT
ajst-23543	58	27	dimensional	dimensional	ADJ
ajst-23543	58	28	structure	structure	NOUN
ajst-23543	58	29	.	.	PUNCT
ajst-23543	59	1	twodimensional	twodimensional	ADJ
ajst-23543	59	2	convolution	convolution	NOUN
ajst-23543	59	3	is	be	AUX
ajst-23543	59	4	the	the	DET
ajst-23543	59	5	extension	extension	NOUN
ajst-23543	59	6	for	for	ADP
ajst-23543	59	7	one	one	NUM
ajst-23543	59	8	-	-	PUNCT
ajst-23543	59	9	dimensional	dimensional	ADJ
ajst-23543	59	10	convolution	convolution	NOUN
ajst-23543	59	11	,	,	PUNCT
ajst-23543	59	12	assume	assume	VERB
ajst-23543	59	13	that	that	SCONJ
ajst-23543	59	14	the	the	DET
ajst-23543	59	15	input	input	NOUN
ajst-23543	59	16	image	image	NOUN
ajst-23543	59	17	data	datum	NOUN
ajst-23543	59	18	for	for	ADP
ajst-23543	59	19	∈	∈	NOUN
ajst-23543	59	20	ℝ	ℝ	PROPN
ajst-23543	59	21	x	x	NOUN
ajst-23543	59	22	,	,	PUNCT
ajst-23543	59	23	convolution	convolution	NOUN
ajst-23543	59	24	layer	layer	NOUN
ajst-23543	59	25	filter	filter	NOUN
ajst-23543	59	26	for	for	ADP
ajst-23543	59	27	∈	∈	NOUN
ajst-23543	59	28	ℝ	ℝ	PROPN
ajst-23543	59	29	x	x	X
ajst-23543	59	30	,	,	PUNCT
ajst-23543	59	31	and	and	CCONJ
ajst-23543	59	32	n	n	PRON
ajst-23543	59	33	is	be	AUX
ajst-23543	59	34	greater	great	ADJ
ajst-23543	59	35	than	than	ADP
ajst-23543	59	36	and	and	CCONJ
ajst-23543	59	37	its	its	PRON
ajst-23543	59	38	convolution	convolution	NOUN
ajst-23543	59	39	is	be	AUX
ajst-23543	59	40	:	:	PUNCT
ajst-23543	59	41	,	,	PUNCT
ajst-23543	59	42	1	1	X
ajst-23543	59	43	,	,	PUNCT
ajst-23543	59	44	1	1	NUM
ajst-23543	59	45	1	1	NUM
ajst-23543	59	46	1	1	NUM
ajst-23543	59	47	m	m	NOUN
ajst-23543	59	48	n	n	ADV
ajst-23543	60	1	i	i	PRON
ajst-23543	60	2	j	j	INTJ
ajst-23543	61	1	uv	uv	INTJ
ajst-23543	62	1	i	i	PRON
ajst-23543	62	2	u	u	X
ajst-23543	62	3	j	j	PROPN
ajst-23543	62	4	v	v	ADP
ajst-23543	62	5	u	u	NOUN
ajst-23543	62	6	v	v	ADP
ajst-23543	62	7	y	y	PROPN
ajst-23543	62	8	x	x	PROPN
ajst-23543	62	9			PROPN
ajst-23543	62	10			VERB
ajst-23543	62	11			PROPN
ajst-23543	62	12			VERB
ajst-23543	62	13			PROPN
ajst-23543	63	1			NUM
ajst-23543	63	2			NOUN
ajst-23543	63	3			SYM
ajst-23543	63	4	(	(	PUNCT
ajst-23543	63	5	4	4	X
ajst-23543	63	6	)	)	PUNCT
ajst-23543	63	7	an	an	DET
ajst-23543	63	8	example	example	NOUN
ajst-23543	63	9	of	of	ADP
ajst-23543	63	10	two	two	NUM
ajst-23543	63	11	-	-	PUNCT
ajst-23543	63	12	dimensional	dimensional	ADJ
ajst-23543	63	13	convolution	convolution	NOUN
ajst-23543	63	14	is	be	AUX
ajst-23543	63	15	shown	show	VERB
ajst-23543	63	16	in	in	ADP
ajst-23543	63	17	figure	figure	NOUN
ajst-23543	63	18	3	3	NUM
ajst-23543	63	19	.	.	PUNCT
ajst-23543	64	1	the	the	DET
ajst-23543	64	2	input	input	NOUN
ajst-23543	64	3	is	be	AUX
ajst-23543	64	4	a	a	DET
ajst-23543	64	5	4×4	4×4	NUM
ajst-23543	64	6	two	two	NUM
ajst-23543	64	7	-	-	PUNCT
ajst-23543	64	8	dimensional	dimensional	ADJ
ajst-23543	64	9	image	image	NOUN
ajst-23543	64	10	,	,	PUNCT
ajst-23543	64	11	and	and	CCONJ
ajst-23543	64	12	a	a	DET
ajst-23543	64	13	2×2	2×2	NUM
ajst-23543	64	14	feature	feature	NOUN
ajst-23543	64	15	map	map	NOUN
ajst-23543	64	16	is	be	AUX
ajst-23543	64	17	obtained	obtain	VERB
ajst-23543	64	18	by	by	ADP
ajst-23543	64	19	convolution	convolution	NOUN
ajst-23543	64	20	operation	operation	NOUN
ajst-23543	64	21	with	with	ADP
ajst-23543	64	22	a	a	DET
ajst-23543	64	23	convolution	convolution	NOUN
ajst-23543	64	24	kernel	kernel	NOUN
ajst-23543	64	25	of	of	ADP
ajst-23543	64	26	3×3	3×3	NUM
ajst-23543	64	27	.	.	PUNCT
ajst-23543	65	1	the	the	DET
ajst-23543	65	2	specific	specific	ADJ
ajst-23543	65	3	operation	operation	NOUN
ajst-23543	65	4	is	be	AUX
ajst-23543	65	5	as	as	SCONJ
ajst-23543	65	6	follows	follow	VERB
ajst-23543	65	7	:	:	PUNCT
ajst-23543	65	8	on	on	ADP
ajst-23543	65	9	the	the	DET
ajst-23543	65	10	feature	feature	NOUN
ajst-23543	65	11	plane	plane	NOUN
ajst-23543	65	12	of	of	ADP
ajst-23543	65	13	the	the	DET
ajst-23543	65	14	input	input	NOUN
ajst-23543	65	15	image	image	NOUN
ajst-23543	65	16	data	datum	NOUN
ajst-23543	65	17	,	,	PUNCT
ajst-23543	65	18	the	the	DET
ajst-23543	65	19	convolutional	convolutional	ADJ
ajst-23543	65	20	kernel	kernel	NOUN
ajst-23543	65	21	uses	use	VERB
ajst-23543	65	22	step	step	NOUN
ajst-23543	65	23	size	size	NOUN
ajst-23543	65	24	1	1	NUM
ajst-23543	65	25	to	to	PART
ajst-23543	65	26	slide	slide	VERB
ajst-23543	65	27	on	on	ADP
ajst-23543	65	28	the	the	DET
ajst-23543	65	29	feature	feature	NOUN
ajst-23543	65	30	plane	plane	NOUN
ajst-23543	65	31	in	in	ADP
ajst-23543	65	32	sequence	sequence	NOUN
ajst-23543	65	33	from	from	ADP
ajst-23543	65	34	left	left	ADJ
ajst-23543	65	35	to	to	ADP
ajst-23543	65	36	right	right	NOUN
ajst-23543	65	37	and	and	CCONJ
ajst-23543	65	38	from	from	ADP
ajst-23543	65	39	top	top	NOUN
ajst-23543	65	40	to	to	ADP
ajst-23543	65	41	bottom	bottom	NOUN
ajst-23543	65	42	,	,	PUNCT
ajst-23543	65	43	and	and	CCONJ
ajst-23543	65	44	then	then	ADV
ajst-23543	65	45	performs	performs	AUX
ajst-23543	65	46	weighted	weight	VERB
ajst-23543	65	47	sum	sum	NOUN
ajst-23543	65	48	to	to	PART
ajst-23543	65	49	obtain	obtain	VERB
ajst-23543	65	50	the	the	DET
ajst-23543	65	51	final	final	ADJ
ajst-23543	65	52	feature	feature	NOUN
ajst-23543	65	53	mapping	mapping	NOUN
ajst-23543	65	54	.	.	PUNCT
ajst-23543	66	1	figure	figure	VERB
ajst-23543	66	2	3	3	NUM
ajst-23543	66	3	.	.	X
ajst-23543	66	4	two	two	NUM
ajst-23543	66	5	-	-	PUNCT
ajst-23543	66	6	dimensional	dimensional	ADJ
ajst-23543	66	7	convolution	convolution	NOUN
ajst-23543	66	8	diagram	diagram	NOUN
ajst-23543	66	9	3	3	NUM
ajst-23543	66	10	.	.	PUNCT
ajst-23543	66	11	attention	attention	NOUN
ajst-23543	66	12	mechanisms	mechanism	NOUN
ajst-23543	66	13	the	the	DET
ajst-23543	66	14	essence	essence	NOUN
ajst-23543	66	15	of	of	ADP
ajst-23543	66	16	attention	attention	NOUN
ajst-23543	66	17	mechanism	mechanism	NOUN
ajst-23543	66	18	comes	come	VERB
ajst-23543	66	19	from	from	ADP
ajst-23543	66	20	human	human	ADJ
ajst-23543	66	21	visual	visual	ADJ
ajst-23543	66	22	attention	attention	NOUN
ajst-23543	66	23	,	,	PUNCT
ajst-23543	66	24	which	which	PRON
ajst-23543	66	25	was	be	AUX
ajst-23543	66	26	first	first	ADV
ajst-23543	66	27	proposed	propose	VERB
ajst-23543	66	28	in	in	ADP
ajst-23543	66	29	the	the	DET
ajst-23543	66	30	field	field	NOUN
ajst-23543	66	31	of	of	ADP
ajst-23543	66	32	visual	visual	ADJ
ajst-23543	66	33	image	image	NOUN
ajst-23543	66	34	.	.	PUNCT
ajst-23543	67	1	the	the	DET
ajst-23543	67	2	basic	basic	ADJ
ajst-23543	67	3	idea	idea	NOUN
ajst-23543	67	4	of	of	ADP
ajst-23543	67	5	the	the	DET
ajst-23543	67	6	attention	attention	NOUN
ajst-23543	67	7	mechanism	mechanism	NOUN
ajst-23543	67	8	is	be	AUX
ajst-23543	67	9	to	to	PART
ajst-23543	67	10	choose	choose	VERB
ajst-23543	67	11	from	from	ADP
ajst-23543	67	12	input	input	NOUN
ajst-23543	67	13	information	information	NOUN
ajst-23543	67	14	that	that	PRON
ajst-23543	67	15	is	be	AUX
ajst-23543	67	16	more	more	ADV
ajst-23543	67	17	critical	critical	ADJ
ajst-23543	67	18	to	to	ADP
ajst-23543	67	19	the	the	DET
ajst-23543	67	20	objective	objective	NOUN
ajst-23543	67	21	of	of	ADP
ajst-23543	67	22	the	the	DET
ajst-23543	67	23	task	task	NOUN
ajst-23543	67	24	at	at	ADP
ajst-23543	67	25	hand	hand	NOUN
ajst-23543	67	26	.	.	PUNCT
ajst-23543	68	1	according	accord	VERB
ajst-23543	68	2	to	to	ADP
ajst-23543	68	3	the	the	DET
ajst-23543	68	4	importance	importance	NOUN
ajst-23543	68	5	of	of	ADP
ajst-23543	68	6	each	each	DET
ajst-23543	68	7	input	input	NOUN
ajst-23543	68	8	information	information	NOUN
ajst-23543	68	9	,	,	PUNCT
ajst-23543	68	10	it	it	PRON
ajst-23543	68	11	assigns	assign	VERB
ajst-23543	68	12	a	a	DET
ajst-23543	68	13	corresponding	corresponding	ADJ
ajst-23543	68	14	weight	weight	NOUN
ajst-23543	68	15	,	,	PUNCT
ajst-23543	68	16	so	so	SCONJ
ajst-23543	68	17	as	as	SCONJ
ajst-23543	68	18	to	to	PART
ajst-23543	68	19	capture	capture	VERB
ajst-23543	68	20	more	more	ADV
ajst-23543	68	21	valuable	valuable	ADJ
ajst-23543	68	22	information	information	NOUN
ajst-23543	68	23	.	.	PUNCT
ajst-23543	69	1	with	with	ADP
ajst-23543	69	2	the	the	DET
ajst-23543	69	3	continuous	continuous	ADJ
ajst-23543	69	4	development	development	NOUN
ajst-23543	69	5	and	and	CCONJ
ajst-23543	69	6	improvement	improvement	NOUN
ajst-23543	69	7	of	of	ADP
ajst-23543	69	8	attention	attention	NOUN
ajst-23543	69	9	mechanism	mechanism	NOUN
ajst-23543	69	10	,	,	PUNCT
ajst-23543	69	11	attention	attention	NOUN
ajst-23543	69	12	mechanism	mechanism	NOUN
ajst-23543	69	13	has	have	AUX
ajst-23543	69	14	gradually	gradually	ADV
ajst-23543	69	15	become	become	VERB
ajst-23543	69	16	a	a	DET
ajst-23543	69	17	research	research	NOUN
ajst-23543	69	18	hotspot	hotspot	NOUN
ajst-23543	69	19	of	of	ADP
ajst-23543	69	20	scholars	scholar	NOUN
ajst-23543	69	21	around	around	ADP
ajst-23543	69	22	the	the	DET
ajst-23543	69	23	world	world	NOUN
ajst-23543	69	24	,	,	PUNCT
ajst-23543	69	25	and	and	CCONJ
ajst-23543	69	26	has	have	AUX
ajst-23543	69	27	been	be	AUX
ajst-23543	69	28	widely	widely	ADV
ajst-23543	69	29	used	use	VERB
ajst-23543	69	30	in	in	ADP
ajst-23543	69	31	image	image	NOUN
ajst-23543	69	32	recognition	recognition	NOUN
ajst-23543	69	33	,	,	PUNCT
ajst-23543	69	34	speech	speech	NOUN
ajst-23543	69	35	recognition	recognition	NOUN
ajst-23543	69	36	,	,	PUNCT
ajst-23543	69	37	machine	machine	NOUN
ajst-23543	69	38	translation	translation	NOUN
ajst-23543	69	39	and	and	CCONJ
ajst-23543	69	40	other	other	ADJ
ajst-23543	69	41	tasks	task	NOUN
ajst-23543	69	42	,	,	PUNCT
ajst-23543	69	43	and	and	CCONJ
ajst-23543	69	44	has	have	AUX
ajst-23543	69	45	achieved	achieve	VERB
ajst-23543	69	46	good	good	ADJ
ajst-23543	69	47	results	result	NOUN
ajst-23543	69	48	[	[	X
ajst-23543	69	49	8	8	NUM
ajst-23543	69	50	]	]	PUNCT
ajst-23543	69	51	.	.	PUNCT
ajst-23543	70	1	in	in	ADP
ajst-23543	70	2	2017	2017	NUM
ajst-23543	70	3	,	,	PUNCT
ajst-23543	70	4	87	87	NUM
ajst-23543	70	5	the	the	DET
ajst-23543	70	6	google	google	PROPN
ajst-23543	70	7	team	team	NOUN
ajst-23543	70	8	used	use	VERB
ajst-23543	70	9	self	self	NOUN
ajst-23543	70	10	-	-	PUNCT
ajst-23543	70	11	attention	attention	NOUN
ajst-23543	70	12	mechanisms	mechanism	NOUN
ajst-23543	70	13	to	to	PART
ajst-23543	70	14	learn	learn	VERB
ajst-23543	70	15	text	text	NOUN
ajst-23543	70	16	representations	representation	NOUN
ajst-23543	70	17	and	and	CCONJ
ajst-23543	70	18	achieved	achieve	VERB
ajst-23543	70	19	good	good	ADJ
ajst-23543	70	20	learning	learning	NOUN
ajst-23543	70	21	results	result	NOUN
ajst-23543	70	22	[	[	X
ajst-23543	70	23	9	9	NUM
ajst-23543	70	24	]	]	PUNCT
ajst-23543	70	25	.	.	PUNCT
ajst-23543	71	1	selfattention	selfattention	NOUN
ajst-23543	71	2	mechanism	mechanism	NOUN
ajst-23543	71	3	is	be	AUX
ajst-23543	71	4	an	an	DET
ajst-23543	71	5	improvement	improvement	NOUN
ajst-23543	71	6	of	of	ADP
ajst-23543	71	7	attention	attention	NOUN
ajst-23543	71	8	mechanism	mechanism	NOUN
ajst-23543	71	9	,	,	PUNCT
ajst-23543	71	10	which	which	PRON
ajst-23543	71	11	is	be	AUX
ajst-23543	71	12	less	less	ADV
ajst-23543	71	13	dependent	dependent	ADJ
ajst-23543	71	14	on	on	ADP
ajst-23543	71	15	external	external	ADJ
ajst-23543	71	16	information	information	NOUN
ajst-23543	71	17	and	and	CCONJ
ajst-23543	71	18	better	well	ADJ
ajst-23543	71	19	at	at	ADP
ajst-23543	71	20	capturing	capture	VERB
ajst-23543	71	21	internal	internal	ADJ
ajst-23543	71	22	correlation	correlation	NOUN
ajst-23543	71	23	of	of	ADP
ajst-23543	71	24	data	datum	NOUN
ajst-23543	71	25	.	.	PUNCT
ajst-23543	72	1	its	its	PRON
ajst-23543	72	2	structure	structure	NOUN
ajst-23543	72	3	is	be	AUX
ajst-23543	72	4	shown	show	VERB
ajst-23543	72	5	in	in	ADP
ajst-23543	72	6	figure	figure	NOUN
ajst-23543	72	7	4	4	NUM
ajst-23543	72	8	.	.	PUNCT
ajst-23543	72	9	attention	attention	NOUN
ajst-23543	72	10	mechanism	mechanism	NOUN
ajst-23543	72	11	.	.	PUNCT
ajst-23543	73	1	as	as	SCONJ
ajst-23543	73	2	shown	show	VERB
ajst-23543	73	3	in	in	ADP
ajst-23543	73	4	fig	fig	NOUN
ajst-23543	73	5	.	.	PUNCT
ajst-23543	74	1	4	4	NUM
ajst-23543	74	2	,	,	PUNCT
ajst-23543	74	3	ℎ	ℎ	NOUN
ajst-23543	74	4	represents	represent	VERB
ajst-23543	74	5	the	the	DET
ajst-23543	74	6	implicit	implicit	ADJ
ajst-23543	74	7	state	state	NOUN
ajst-23543	74	8	vector	vector	NOUN
ajst-23543	74	9	at	at	ADP
ajst-23543	74	10	moment	moment	NOUN
ajst-23543	74	11	t	t	PROPN
ajst-23543	74	12	,	,	PUNCT
ajst-23543	74	13	represents	represent	VERB
ajst-23543	74	14	the	the	DET
ajst-23543	74	15	similarity	similarity	NOUN
ajst-23543	74	16	weight	weight	NOUN
ajst-23543	74	17	of	of	ADP
ajst-23543	74	18	ℎ	ℎ	PROPN
ajst-23543	74	19	,	,	PUNCT
ajst-23543	74	20	represents	represent	VERB
ajst-23543	74	21	the	the	DET
ajst-23543	74	22	normalized	normalized	ADJ
ajst-23543	74	23	similarity	similarity	NOUN
ajst-23543	74	24	weight	weight	NOUN
ajst-23543	74	25	of	of	ADP
ajst-23543	74	26	ℎ	ℎ	PROPN
ajst-23543	74	27	,	,	PUNCT
ajst-23543	74	28	and	and	CCONJ
ajst-23543	74	29	r	r	NOUN
ajst-23543	74	30	represents	represent	VERB
ajst-23543	74	31	the	the	DET
ajst-23543	74	32	self	self	NOUN
ajst-23543	74	33	-	-	PUNCT
ajst-23543	74	34	attention	attention	NOUN
ajst-23543	74	35	output	output	NOUN
ajst-23543	74	36	matrix	matrix	NOUN
ajst-23543	74	37	obtained	obtain	VERB
ajst-23543	74	38	by	by	ADP
ajst-23543	74	39	weighted	weighted	ADJ
ajst-23543	74	40	summation	summation	NOUN
ajst-23543	74	41	.	.	PUNCT
ajst-23543	75	1	the	the	DET
ajst-23543	75	2	calculation	calculation	NOUN
ajst-23543	75	3	process	process	NOUN
ajst-23543	75	4	can	can	AUX
ajst-23543	75	5	be	be	AUX
ajst-23543	75	6	divided	divide	VERB
ajst-23543	75	7	into	into	ADP
ajst-23543	75	8	three	three	NUM
ajst-23543	75	9	steps	step	NOUN
ajst-23543	75	10	as	as	SCONJ
ajst-23543	75	11	follows	follow	VERB
ajst-23543	75	12	:	:	PUNCT
ajst-23543	75	13	(	(	PUNCT
ajst-23543	75	14	1	1	X
ajst-23543	75	15	)	)	PUNCT
ajst-23543	75	16	the	the	DET
ajst-23543	75	17	multi	multi	ADJ
ajst-23543	75	18	-	-	ADJ
ajst-23543	75	19	layer	layer	ADJ
ajst-23543	75	20	perceptron	perceptron	NOUN
ajst-23543	75	21	is	be	AUX
ajst-23543	75	22	used	use	VERB
ajst-23543	75	23	to	to	PART
ajst-23543	75	24	calculate	calculate	VERB
ajst-23543	75	25	the	the	DET
ajst-23543	75	26	similarity	similarity	NOUN
ajst-23543	75	27	weight	weight	NOUN
ajst-23543	75	28	of	of	ADP
ajst-23543	75	29	the	the	DET
ajst-23543	75	30	data	datum	NOUN
ajst-23543	75	31	at	at	ADP
ajst-23543	75	32	each	each	DET
ajst-23543	75	33	moment	moment	NOUN
ajst-23543	75	34	,	,	PUNCT
ajst-23543	75	35	and	and	CCONJ
ajst-23543	75	36	the	the	DET
ajst-23543	75	37	calculation	calculation	NOUN
ajst-23543	75	38	formula	formula	NOUN
ajst-23543	75	39	is	be	AUX
ajst-23543	75	40	as	as	SCONJ
ajst-23543	75	41	follows	follow	VERB
ajst-23543	75	42	:	:	PUNCT
ajst-23543	76	1	tanh	tanh	NOUN
ajst-23543	76	2	(	(	PUNCT
ajst-23543	76	3	[	[	PUNCT
ajst-23543	76	4	;	;	PUNCT
ajst-23543	76	5	]	]	X
ajst-23543	76	6	)	)	PUNCT
ajst-23543	76	7	t	t	PROPN
ajst-23543	76	8	b	b	PROPN
ajst-23543	76	9	a	a	PRON
ajst-23543	76	10	te	te	PROPN
ajst-23543	76	11	w	w	PROPN
ajst-23543	76	12	w	w	PROPN
ajst-23543	76	13	h	h	PROPN
ajst-23543	76	14	h	h	NOUN
ajst-23543	76	15	(	(	PUNCT
ajst-23543	76	16	5	5	NUM
ajst-23543	76	17	)	)	PUNCT
ajst-23543	76	18	among	among	ADP
ajst-23543	76	19	them	they	PRON
ajst-23543	76	20	,	,	PUNCT
ajst-23543	76	21	the	the	PRON
ajst-23543	76	22	said	say	VERB
ajst-23543	76	23	the	the	DET
ajst-23543	76	24	mlp	mlp	NOUN
ajst-23543	76	25	connection	connection	NOUN
ajst-23543	76	26	weights	weight	VERB
ajst-23543	76	27	between	between	ADP
ajst-23543	76	28	input	input	NOUN
ajst-23543	76	29	and	and	CCONJ
ajst-23543	76	30	hidden	hidden	ADJ
ajst-23543	76	31	layer	layer	NOUN
ajst-23543	76	32	,	,	PUNCT
ajst-23543	76	33	said	say	VERB
ajst-23543	76	34	mlp	mlp	NOUN
ajst-23543	76	35	hidden	hidden	ADJ
ajst-23543	76	36	layer	layer	NOUN
ajst-23543	76	37	and	and	CCONJ
ajst-23543	76	38	output	output	NOUN
ajst-23543	76	39	layer	layer	NOUN
ajst-23543	76	40	connection	connection	NOUN
ajst-23543	76	41	weights	weight	VERB
ajst-23543	76	42	between	between	ADP
ajst-23543	76	43	=	=	SYM
ajst-23543	76	44	(	(	PUNCT
ajst-23543	76	45	ℎ	ℎ	X
ajst-23543	76	46	−1	−1	VERB
ajst-23543	76	47	,	,	PUNCT
ajst-23543	76	48	ℎ	ℎ	X
ajst-23543	76	49	,	,	PUNCT
ajst-23543	76	50	...	...	PUNCT
ajst-23543	76	51	,	,	PUNCT
ajst-23543	76	52	ℎ	ℎ	X
ajst-23543	76	53	+	+	PUNCT
ajst-23543	76	54	)	)	PUNCT
ajst-23543	76	55	represents	represent	VERB
ajst-23543	76	56	the	the	DET
ajst-23543	76	57	hidden	hidden	ADJ
ajst-23543	76	58	state	state	NOUN
ajst-23543	76	59	matrix	matrix	NOUN
ajst-23543	76	60	.	.	PUNCT
ajst-23543	77	1	(	(	PUNCT
ajst-23543	77	2	2	2	X
ajst-23543	77	3	)	)	PUNCT
ajst-23543	77	4	the	the	DET
ajst-23543	77	5	softmax	softmax	NOUN
ajst-23543	77	6	function	function	NOUN
ajst-23543	77	7	is	be	AUX
ajst-23543	77	8	used	use	VERB
ajst-23543	77	9	to	to	PART
ajst-23543	77	10	normalize	normalize	VERB
ajst-23543	77	11	these	these	DET
ajst-23543	77	12	similarity	similarity	NOUN
ajst-23543	77	13	weights	weight	NOUN
ajst-23543	77	14	,	,	PUNCT
ajst-23543	77	15	and	and	CCONJ
ajst-23543	77	16	the	the	DET
ajst-23543	77	17	calculation	calculation	NOUN
ajst-23543	77	18	formula	formula	NOUN
ajst-23543	77	19	is	be	AUX
ajst-23543	77	20	as	as	SCONJ
ajst-23543	77	21	follows	follow	VERB
ajst-23543	77	22	:	:	PUNCT
ajst-23543	77	23	1	1	NUM
ajst-23543	77	24	(	(	PUNCT
ajst-23543	77	25	)	)	PUNCT
ajst-23543	77	26	max	max	PROPN
ajst-23543	77	27	(	(	PUNCT
ajst-23543	77	28	)	)	PUNCT
ajst-23543	77	29	(	(	PUNCT
ajst-23543	77	30	)	)	PUNCT
ajst-23543	77	31	t	t	NOUN
ajst-23543	77	32	t	t	PROPN
ajst-23543	77	33	t	t	PROPN
ajst-23543	77	34	t	t	PROPN
ajst-23543	78	1	i	i	PRON
ajst-23543	78	2	j	j	PROPN
ajst-23543	79	1	j	j	PROPN
ajst-23543	79	2	t	t	PROPN
ajst-23543	79	3	exp	exp	NOUN
ajst-23543	79	4	e	e	NOUN
ajst-23543	79	5	a	a	DET
ajst-23543	79	6	soft	soft	ADJ
ajst-23543	79	7	e	e	NOUN
ajst-23543	79	8	e	e	NOUN
ajst-23543	79	9			X
ajst-23543	79	10			PROPN
ajst-23543	79	11			PROPN
ajst-23543	79	12			NOUN
ajst-23543	79	13			NOUN
ajst-23543	79	14			X
ajst-23543	79	15	(	(	PUNCT
ajst-23543	79	16	6	6	NUM
ajst-23543	79	17	)	)	PUNCT
ajst-23543	79	18	where	where	SCONJ
ajst-23543	79	19	exp	exp	NOUN
ajst-23543	79	20	(	(	PUNCT
ajst-23543	79	21	)	)	PUNCT
ajst-23543	79	22	represents	represent	VERB
ajst-23543	79	23	an	an	DET
ajst-23543	79	24	exponential	exponential	ADJ
ajst-23543	79	25	function	function	NOUN
ajst-23543	79	26	based	base	VERB
ajst-23543	79	27	on	on	ADP
ajst-23543	79	28	the	the	DET
ajst-23543	79	29	natural	natural	ADJ
ajst-23543	79	30	constant	constant	ADJ
ajst-23543	79	31	e.	e.	PROPN
ajst-23543	79	32	(	(	PUNCT
ajst-23543	79	33	3	3	X
ajst-23543	79	34	)	)	PUNCT
ajst-23543	79	35	the	the	DET
ajst-23543	79	36	weighted	weight	VERB
ajst-23543	79	37	sum	sum	NOUN
ajst-23543	79	38	of	of	ADP
ajst-23543	79	39	the	the	DET
ajst-23543	79	40	normalized	normalize	VERB
ajst-23543	79	41	similarity	similarity	NOUN
ajst-23543	79	42	weight	weight	NOUN
ajst-23543	79	43	and	and	CCONJ
ajst-23543	79	44	the	the	DET
ajst-23543	79	45	corresponding	corresponding	ADJ
ajst-23543	79	46	data	datum	NOUN
ajst-23543	79	47	is	be	AUX
ajst-23543	79	48	carried	carry	VERB
ajst-23543	79	49	out	out	ADP
ajst-23543	79	50	to	to	PART
ajst-23543	79	51	obtain	obtain	VERB
ajst-23543	79	52	the	the	DET
ajst-23543	79	53	selfattention	selfattention	NOUN
ajst-23543	79	54	output	output	NOUN
ajst-23543	79	55	matrix	matrix	NOUN
ajst-23543	79	56	r.	r.	NOUN
ajst-23543	80	1	the	the	DET
ajst-23543	80	2	calculation	calculation	NOUN
ajst-23543	80	3	formula	formula	NOUN
ajst-23543	80	4	is	be	AUX
ajst-23543	80	5	as	as	SCONJ
ajst-23543	80	6	follows	follow	VERB
ajst-23543	80	7	:	:	PUNCT
ajst-23543	80	8	1	1	NUM
ajst-23543	80	9	d	d	X
ajst-23543	80	10	j	j	PROPN
ajst-23543	80	11	j	j	PROPN
ajst-23543	80	12	j	j	PROPN
ajst-23543	80	13	t	t	PROPN
ajst-23543	80	14	r	r	NOUN
ajst-23543	80	15	h	h	VERB
ajst-23543	80	16			ADJ
ajst-23543	80	17			PROPN
ajst-23543	80	18			NOUN
ajst-23543	80	19			X
ajst-23543	80	20	(	(	PUNCT
ajst-23543	80	21	7	7	NUM
ajst-23543	80	22	)	)	PUNCT
ajst-23543	80	23	where	where	SCONJ
ajst-23543	80	24	,	,	PUNCT
ajst-23543	80	25	ℎ	ℎ	PROPN
ajst-23543	80	26	represents	represent	VERB
ajst-23543	80	27	the	the	DET
ajst-23543	80	28	implicit	implicit	ADJ
ajst-23543	80	29	state	state	NOUN
ajst-23543	80	30	vector	vector	NOUN
ajst-23543	80	31	at	at	ADP
ajst-23543	80	32	time	time	NOUN
ajst-23543	80	33	j	j	PROPN
ajst-23543	80	34	,	,	PUNCT
ajst-23543	80	35	and	and	CCONJ
ajst-23543	80	36	represents	represent	VERB
ajst-23543	80	37	the	the	DET
ajst-23543	80	38	similarity	similarity	NOUN
ajst-23543	80	39	weight	weight	NOUN
ajst-23543	80	40	after	after	ADP
ajst-23543	80	41	normalization	normalization	NOUN
ajst-23543	80	42	of	of	ADP
ajst-23543	80	43	ℎ	ℎ	PROPN
ajst-23543	80	44	.	.	PUNCT
ajst-23543	81	1	figure	figure	VERB
ajst-23543	81	2	4	4	NUM
ajst-23543	81	3	.	.	PUNCT
ajst-23543	81	4	self	self	NOUN
ajst-23543	81	5	-	-	PUNCT
ajst-23543	81	6	attention	attention	NOUN
ajst-23543	81	7	mechanism	mechanism	NOUN
ajst-23543	81	8	4	4	NUM
ajst-23543	81	9	.	.	PUNCT
ajst-23543	81	10	experiment	experiment	NOUN
ajst-23543	81	11	in	in	ADP
ajst-23543	81	12	this	this	DET
ajst-23543	81	13	paper	paper	NOUN
ajst-23543	81	14	,	,	PUNCT
ajst-23543	81	15	the	the	DET
ajst-23543	81	16	performance	performance	NOUN
ajst-23543	81	17	of	of	ADP
ajst-23543	81	18	the	the	DET
ajst-23543	81	19	method	method	NOUN
ajst-23543	81	20	is	be	AUX
ajst-23543	81	21	verified	verify	VERB
ajst-23543	81	22	on	on	ADP
ajst-23543	81	23	the	the	DET
ajst-23543	81	24	imagenet	imagenet	NOUN
ajst-23543	81	25	dataset	dataset	NOUN
ajst-23543	81	26	,	,	PUNCT
ajst-23543	81	27	and	and	CCONJ
ajst-23543	81	28	this	this	DET
ajst-23543	81	29	experiment	experiment	NOUN
ajst-23543	81	30	is	be	AUX
ajst-23543	81	31	implemented	implement	VERB
ajst-23543	81	32	based	base	VERB
ajst-23543	81	33	on	on	ADP
ajst-23543	81	34	pytorch	pytorch	NOUN
ajst-23543	81	35	.	.	PUNCT
ajst-23543	82	1	all	all	DET
ajst-23543	82	2	deep	deep	ADJ
ajst-23543	82	3	convolutional	convolutional	ADJ
ajst-23543	82	4	neural	neural	ADJ
ajst-23543	82	5	network	network	NOUN
ajst-23543	82	6	models	model	NOUN
ajst-23543	82	7	in	in	ADP
ajst-23543	82	8	the	the	DET
ajst-23543	82	9	experiment	experiment	NOUN
ajst-23543	82	10	follow	follow	VERB
ajst-23543	82	11	the	the	DET
ajst-23543	82	12	same	same	ADJ
ajst-23543	82	13	architecture	architecture	NOUN
ajst-23543	82	14	,	,	PUNCT
ajst-23543	82	15	and	and	CCONJ
ajst-23543	82	16	the	the	DET
ajst-23543	82	17	initial	initial	ADJ
ajst-23543	82	18	learning	learning	NOUN
ajst-23543	82	19	rate	rate	NOUN
ajst-23543	82	20	is	be	AUX
ajst-23543	82	21	set	set	VERB
ajst-23543	82	22	as	as	ADP
ajst-23543	82	23	=	=	PROPN
ajst-23543	82	24	6	6	NUM
ajst-23543	82	25	×	×	NOUN
ajst-23543	82	26	10	10	NUM
ajst-23543	82	27	-	-	SYM
ajst-23543	82	28	1	1	NUM
ajst-23543	82	29	.	.	PUNCT
ajst-23543	83	1	the	the	DET
ajst-23543	83	2	learning	learning	NOUN
ajst-23543	83	3	rate	rate	NOUN
ajst-23543	83	4	is	be	AUX
ajst-23543	83	5	gradually	gradually	ADV
ajst-23543	83	6	adjusted	adjust	VERB
ajst-23543	83	7	according	accord	VERB
ajst-23543	83	8	to	to	ADP
ajst-23543	83	9	the	the	DET
ajst-23543	83	10	number	number	NOUN
ajst-23543	83	11	of	of	ADP
ajst-23543	83	12	iterations	iteration	NOUN
ajst-23543	83	13	.	.	PUNCT
ajst-23543	84	1	table	table	NOUN
ajst-23543	84	2	1	1	NUM
ajst-23543	84	3	.	.	PUNCT
ajst-23543	85	1	experimental	experimental	ADJ
ajst-23543	85	2	results	result	NOUN
ajst-23543	85	3	of	of	ADP
ajst-23543	85	4	different	different	ADJ
ajst-23543	85	5	models	model	NOUN
ajst-23543	85	6	model	model	PROPN
ajst-23543	85	7	macro	macro	PROPN
ajst-23543	85	8	micro	micro	PROPN
ajst-23543	85	9	cnn	cnn	PROPN
ajst-23543	85	10	78.5	78.5	NUM
ajst-23543	85	11	%	%	NOUN
ajst-23543	85	12	77.9	77.9	NUM
ajst-23543	85	13	%	%	NOUN
ajst-23543	85	14	resnet	resnet	VERB
ajst-23543	85	15	81.3	81.3	NUM
ajst-23543	85	16	%	%	NOUN
ajst-23543	85	17	81.7	81.7	NUM
ajst-23543	85	18	%	%	NOUN
ajst-23543	85	19	our	our	PRON
ajst-23543	85	20	83.8	83.8	NUM
ajst-23543	85	21	%	%	NOUN
ajst-23543	85	22	84.2	84.2	NUM
ajst-23543	85	23	%	%	NOUN
ajst-23543	85	24	according	accord	VERB
ajst-23543	85	25	to	to	ADP
ajst-23543	85	26	the	the	DET
ajst-23543	85	27	experimental	experimental	ADJ
ajst-23543	85	28	results	result	NOUN
ajst-23543	85	29	in	in	ADP
ajst-23543	85	30	table	table	NOUN
ajst-23543	85	31	i	i	PRON
ajst-23543	85	32	,	,	PUNCT
ajst-23543	85	33	the	the	DET
ajst-23543	85	34	three	three	NUM
ajst-23543	85	35	models	model	NOUN
ajst-23543	85	36	-cnn	-cnn	PUNCT
ajst-23543	85	37	,	,	PUNCT
ajst-23543	85	38	resnet	resnet	NOUN
ajst-23543	85	39	and	and	CCONJ
ajst-23543	85	40	our	our	PRON
ajst-23543	85	41	model	model	NOUN
ajst-23543	85	42	-show	-show	PROPN
ajst-23543	85	43	obvious	obvious	ADJ
ajst-23543	85	44	differences	difference	NOUN
ajst-23543	85	45	in	in	ADP
ajst-23543	85	46	performance	performance	NOUN
ajst-23543	85	47	under	under	ADP
ajst-23543	85	48	the	the	DET
ajst-23543	85	49	two	two	NUM
ajst-23543	85	50	evaluation	evaluation	NOUN
ajst-23543	85	51	indexes	index	NOUN
ajst-23543	85	52	"	"	PUNCT
ajst-23543	85	53	macro	macro	NOUN
ajst-23543	85	54	"	"	PUNCT
ajst-23543	85	55	and	and	CCONJ
ajst-23543	85	56	"	"	PUNCT
ajst-23543	85	57	micro	micro	NOUN
ajst-23543	85	58	"	"	PUNCT
ajst-23543	85	59	.	.	PUNCT
ajst-23543	86	1	the	the	DET
ajst-23543	86	2	"	"	PUNCT
ajst-23543	86	3	macro	macro	NOUN
ajst-23543	86	4	"	"	PUNCT
ajst-23543	86	5	indicator	indicator	NOUN
ajst-23543	86	6	reflects	reflect	VERB
ajst-23543	86	7	the	the	DET
ajst-23543	86	8	average	average	ADJ
ajst-23543	86	9	performance	performance	NOUN
ajst-23543	86	10	of	of	ADP
ajst-23543	86	11	the	the	DET
ajst-23543	86	12	model	model	NOUN
ajst-23543	86	13	across	across	ADP
ajst-23543	86	14	various	various	ADJ
ajst-23543	86	15	categories	category	NOUN
ajst-23543	86	16	,	,	PUNCT
ajst-23543	86	17	while	while	SCONJ
ajst-23543	86	18	the	the	DET
ajst-23543	86	19	"	"	PUNCT
ajst-23543	86	20	micro	micro	NOUN
ajst-23543	86	21	"	"	PUNCT
ajst-23543	86	22	indicator	indicator	NOUN
ajst-23543	86	23	measures	measure	VERB
ajst-23543	86	24	the	the	DET
ajst-23543	86	25	overall	overall	ADJ
ajst-23543	86	26	performance	performance	NOUN
ajst-23543	86	27	of	of	ADP
ajst-23543	86	28	the	the	DET
ajst-23543	86	29	model	model	NOUN
ajst-23543	86	30	across	across	ADP
ajst-23543	86	31	the	the	DET
ajst-23543	86	32	overall	overall	ADJ
ajst-23543	86	33	data	datum	NOUN
ajst-23543	86	34	set	set	VERB
ajst-23543	86	35	.	.	PUNCT
ajst-23543	87	1	specifically	specifically	ADV
ajst-23543	87	2	,	,	PUNCT
ajst-23543	87	3	the	the	DET
ajst-23543	87	4	accuracy	accuracy	NOUN
ajst-23543	87	5	of	of	ADP
ajst-23543	87	6	the	the	DET
ajst-23543	87	7	cnn	cnn	PROPN
ajst-23543	87	8	model	model	NOUN
ajst-23543	87	9	on	on	ADP
ajst-23543	87	10	the	the	DET
ajst-23543	87	11	"	"	PUNCT
ajst-23543	87	12	macro	macro	NOUN
ajst-23543	87	13	"	"	PUNCT
ajst-23543	87	14	and	and	CCONJ
ajst-23543	87	15	"	"	PUNCT
ajst-23543	87	16	micro	micro	NOUN
ajst-23543	87	17	"	"	PUNCT
ajst-23543	87	18	indexes	index	NOUN
ajst-23543	87	19	is	be	AUX
ajst-23543	87	20	78.5	78.5	NUM
ajst-23543	87	21	%	%	NOUN
ajst-23543	87	22	and	and	CCONJ
ajst-23543	87	23	77.9	77.9	NUM
ajst-23543	87	24	%	%	NOUN
ajst-23543	87	25	,	,	PUNCT
ajst-23543	87	26	respectively	respectively	ADV
ajst-23543	87	27	,	,	PUNCT
ajst-23543	87	28	showing	show	VERB
ajst-23543	87	29	its	its	PRON
ajst-23543	87	30	basic	basic	ADJ
ajst-23543	87	31	level	level	NOUN
ajst-23543	87	32	in	in	ADP
ajst-23543	87	33	the	the	DET
ajst-23543	87	34	multi	multi	ADJ
ajst-23543	87	35	-	-	ADJ
ajst-23543	87	36	classification	classification	ADJ
ajst-23543	87	37	task	task	NOUN
ajst-23543	87	38	.	.	PUNCT
ajst-23543	88	1	the	the	DET
ajst-23543	88	2	accuracy	accuracy	NOUN
ajst-23543	88	3	of	of	ADP
ajst-23543	88	4	resnet	resnet	NOUN
ajst-23543	88	5	model	model	NOUN
ajst-23543	88	6	in	in	ADP
ajst-23543	88	7	the	the	DET
ajst-23543	88	8	two	two	NUM
ajst-23543	88	9	indexes	index	NOUN
ajst-23543	88	10	increased	increase	VERB
ajst-23543	88	11	to	to	ADP
ajst-23543	88	12	81.3	81.3	NUM
ajst-23543	88	13	%	%	NOUN
ajst-23543	88	14	and	and	CCONJ
ajst-23543	88	15	81.7	81.7	NUM
ajst-23543	88	16	%	%	NOUN
ajst-23543	88	17	,	,	PUNCT
ajst-23543	88	18	indicating	indicate	VERB
ajst-23543	88	19	that	that	SCONJ
ajst-23543	88	20	the	the	DET
ajst-23543	88	21	generalization	generalization	NOUN
ajst-23543	88	22	ability	ability	NOUN
ajst-23543	88	23	and	and	CCONJ
ajst-23543	88	24	prediction	prediction	NOUN
ajst-23543	88	25	accuracy	accuracy	NOUN
ajst-23543	88	26	of	of	ADP
ajst-23543	88	27	the	the	DET
ajst-23543	88	28	model	model	NOUN
ajst-23543	88	29	were	be	AUX
ajst-23543	88	30	effectively	effectively	ADV
ajst-23543	88	31	improved	improve	VERB
ajst-23543	88	32	through	through	ADP
ajst-23543	88	33	innovative	innovative	ADJ
ajst-23543	88	34	design	design	NOUN
ajst-23543	88	35	such	such	ADJ
ajst-23543	88	36	as	as	ADP
ajst-23543	88	37	residual	residual	ADJ
ajst-23543	88	38	connection	connection	NOUN
ajst-23543	88	39	.	.	PUNCT
ajst-23543	89	1	on	on	ADP
ajst-23543	89	2	the	the	DET
ajst-23543	89	3	other	other	ADJ
ajst-23543	89	4	hand	hand	NOUN
ajst-23543	89	5	,	,	PUNCT
ajst-23543	89	6	our	our	PRON
ajst-23543	89	7	model	model	NOUN
ajst-23543	89	8	has	have	AUX
ajst-23543	89	9	reached	reach	VERB
ajst-23543	89	10	a	a	DET
ajst-23543	89	11	high	high	ADJ
ajst-23543	89	12	level	level	NOUN
ajst-23543	89	13	of	of	ADP
ajst-23543	89	14	83.8	83.8	NUM
ajst-23543	89	15	%	%	NOUN
ajst-23543	89	16	and	and	CCONJ
ajst-23543	89	17	84.2	84.2	NUM
ajst-23543	89	18	%	%	NOUN
ajst-23543	89	19	respectively	respectively	ADV
ajst-23543	89	20	in	in	ADP
ajst-23543	89	21	the	the	DET
ajst-23543	89	22	"	"	PUNCT
ajst-23543	89	23	macro	macro	NOUN
ajst-23543	89	24	"	"	PUNCT
ajst-23543	89	25	and	and	CCONJ
ajst-23543	89	26	"	"	PUNCT
ajst-23543	89	27	micro	micro	ADJ
ajst-23543	89	28	"	"	PUNCT
ajst-23543	89	29	indexes	index	NOUN
ajst-23543	89	30	,	,	PUNCT
ajst-23543	89	31	which	which	PRON
ajst-23543	89	32	not	not	PART
ajst-23543	89	33	only	only	ADV
ajst-23543	89	34	means	mean	VERB
ajst-23543	89	35	that	that	SCONJ
ajst-23543	89	36	our	our	PRON
ajst-23543	89	37	model	model	NOUN
ajst-23543	89	38	has	have	VERB
ajst-23543	89	39	excellent	excellent	ADJ
ajst-23543	89	40	average	average	ADJ
ajst-23543	89	41	performance	performance	NOUN
ajst-23543	89	42	in	in	ADP
ajst-23543	89	43	various	various	ADJ
ajst-23543	89	44	categories	category	NOUN
ajst-23543	89	45	,	,	PUNCT
ajst-23543	89	46	but	but	CCONJ
ajst-23543	89	47	also	also	ADV
ajst-23543	89	48	the	the	DET
ajst-23543	89	49	accuracy	accuracy	NOUN
ajst-23543	89	50	of	of	ADP
ajst-23543	89	51	the	the	DET
ajst-23543	89	52	overall	overall	ADJ
ajst-23543	89	53	sample	sample	NOUN
ajst-23543	89	54	prediction	prediction	NOUN
ajst-23543	89	55	is	be	AUX
ajst-23543	89	56	extremely	extremely	ADV
ajst-23543	89	57	outstanding	outstanding	ADJ
ajst-23543	89	58	,	,	PUNCT
ajst-23543	89	59	surpassing	surpass	VERB
ajst-23543	89	60	the	the	DET
ajst-23543	89	61	cnn	cnn	PROPN
ajst-23543	89	62	and	and	CCONJ
ajst-23543	89	63	resnet	resnet	NOUN
ajst-23543	89	64	models	model	NOUN
ajst-23543	89	65	,	,	PUNCT
ajst-23543	89	66	showing	show	VERB
ajst-23543	89	67	a	a	DET
ajst-23543	89	68	strong	strong	ADJ
ajst-23543	89	69	performance	performance	NOUN
ajst-23543	89	70	advantage	advantage	NOUN
ajst-23543	89	71	.	.	PUNCT
ajst-23543	90	1	figure	figure	NOUN
ajst-23543	90	2	5	5	NUM
ajst-23543	90	3	.	.	PUNCT
ajst-23543	91	1	the	the	DET
ajst-23543	91	2	influence	influence	NOUN
ajst-23543	91	3	of	of	ADP
ajst-23543	91	4	different	different	ADJ
ajst-23543	91	5	learning	learning	NOUN
ajst-23543	91	6	rate	rate	NOUN
ajst-23543	91	7	on	on	ADP
ajst-23543	91	8	model	model	NOUN
ajst-23543	91	9	loss	loss	NOUN
ajst-23543	91	10	figure	figure	NOUN
ajst-23543	91	11	5	5	NUM
ajst-23543	91	12	shows	show	VERB
ajst-23543	91	13	the	the	DET
ajst-23543	91	14	different	different	ADJ
ajst-23543	91	15	learning	learning	NOUN
ajst-23543	91	16	rates	rate	NOUN
ajst-23543	91	17	.	.	PUNCT
ajst-23543	92	1	the	the	DET
ajst-23543	92	2	chart	chart	NOUN
ajst-23543	92	3	shows	show	VERB
ajst-23543	92	4	the	the	DET
ajst-23543	92	5	changes	change	NOUN
ajst-23543	92	6	of	of	ADP
ajst-23543	92	7	the	the	DET
ajst-23543	92	8	loss	loss	NOUN
ajst-23543	92	9	function	function	NOUN
ajst-23543	92	10	in	in	ADP
ajst-23543	92	11	the	the	DET
ajst-23543	92	12	training	training	NOUN
ajst-23543	92	13	process	process	NOUN
ajst-23543	92	14	for	for	ADP
ajst-23543	92	15	three	three	NUM
ajst-23543	92	16	different	different	ADJ
ajst-23543	92	17	learning	learning	NOUN
ajst-23543	92	18	rates	rate	NOUN
ajst-23543	92	19	(	(	PUNCT
ajst-23543	92	20	0.01	0.01	NUM
ajst-23543	92	21	,	,	PUNCT
ajst-23543	92	22	0.005	0.005	NUM
ajst-23543	92	23	and	and	CCONJ
ajst-23543	92	24	0.001	0.001	NUM
ajst-23543	92	25	respectively	respectively	ADV
ajst-23543	92	26	)	)	PUNCT
ajst-23543	92	27	.	.	PUNCT
ajst-23543	93	1	as	as	ADP
ajst-23543	93	2	the	the	DET
ajst-23543	93	3	number	number	NOUN
ajst-23543	93	4	of	of	ADP
ajst-23543	93	5	iterations	iteration	NOUN
ajst-23543	93	6	increases	increase	NOUN
ajst-23543	93	7	,	,	PUNCT
ajst-23543	93	8	each	each	DET
ajst-23543	93	9	curve	curve	NOUN
ajst-23543	93	10	has	have	VERB
ajst-23543	93	11	its	its	PRON
ajst-23543	93	12	own	own	ADJ
ajst-23543	93	13	manifestation	manifestation	NOUN
ajst-23543	93	14	:	:	PUNCT
ajst-23543	93	15	the	the	DET
ajst-23543	93	16	learning	learning	NOUN
ajst-23543	93	17	rate	rate	NOUN
ajst-23543	93	18	is	be	AUX
ajst-23543	93	19	0.01	0.01	NUM
ajst-23543	93	20	(	(	PUNCT
ajst-23543	93	21	red	red	ADJ
ajst-23543	93	22	curve	curve	NOUN
ajst-23543	93	23	):	):	PUNCT
ajst-23543	93	24	it	it	PRON
ajst-23543	93	25	declines	decline	VERB
ajst-23543	93	26	rapidly	rapidly	ADV
ajst-23543	93	27	at	at	ADP
ajst-23543	93	28	the	the	DET
ajst-23543	93	29	beginning	beginning	NOUN
ajst-23543	93	30	,	,	PUNCT
ajst-23543	93	31	but	but	CCONJ
ajst-23543	93	32	due	due	ADP
ajst-23543	93	33	to	to	ADP
ajst-23543	93	34	the	the	DET
ajst-23543	93	35	high	high	ADJ
ajst-23543	93	36	learning	learning	NOUN
ajst-23543	93	37	rate	rate	NOUN
ajst-23543	93	38	,	,	PUNCT
ajst-23543	93	39	there	there	PRON
ajst-23543	93	40	are	be	VERB
ajst-23543	93	41	large	large	ADJ
ajst-23543	93	42	fluctuations	fluctuation	NOUN
ajst-23543	93	43	later	later	ADV
ajst-23543	93	44	,	,	PUNCT
ajst-23543	93	45	which	which	PRON
ajst-23543	93	46	may	may	AUX
ajst-23543	93	47	mean	mean	VERB
ajst-23543	93	48	that	that	SCONJ
ajst-23543	93	49	the	the	DET
ajst-23543	93	50	model	model	NOUN
ajst-23543	93	51	is	be	AUX
ajst-23543	93	52	difficult	difficult	ADJ
ajst-23543	93	53	to	to	PART
ajst-23543	93	54	converge	converge	VERB
ajst-23543	93	55	.	.	PUNCT
ajst-23543	94	1	the	the	DET
ajst-23543	94	2	learning	learning	NOUN
ajst-23543	94	3	rate	rate	NOUN
ajst-23543	94	4	is	be	AUX
ajst-23543	94	5	0.005	0.005	NUM
ajst-23543	94	6	(	(	PUNCT
ajst-23543	94	7	green	green	ADJ
ajst-23543	94	8	curve	curve	NOUN
ajst-23543	94	9	):	):	PUNCT
ajst-23543	94	10	compared	compare	VERB
ajst-23543	94	11	to	to	ADP
ajst-23543	94	12	the	the	DET
ajst-23543	94	13	red	red	ADJ
ajst-23543	94	14	curve	curve	NOUN
ajst-23543	94	15	,	,	PUNCT
ajst-23543	94	16	it	it	PRON
ajst-23543	94	17	declines	decline	VERB
ajst-23543	94	18	slightly	slightly	ADV
ajst-23543	94	19	more	more	ADV
ajst-23543	94	20	slowly	slowly	ADV
ajst-23543	94	21	,	,	PUNCT
ajst-23543	94	22	but	but	CCONJ
ajst-23543	94	23	behaves	behave	VERB
ajst-23543	94	24	more	more	ADV
ajst-23543	94	25	smoothly	smoothly	ADV
ajst-23543	94	26	in	in	ADP
ajst-23543	94	27	the	the	DET
ajst-23543	94	28	later	later	ADJ
ajst-23543	94	29	period	period	NOUN
ajst-23543	94	30	,	,	PUNCT
ajst-23543	94	31	with	with	ADP
ajst-23543	94	32	no	no	DET
ajst-23543	94	33	significant	significant	ADJ
ajst-23543	94	34	oscillations	oscillation	NOUN
ajst-23543	94	35	.	.	PUNCT
ajst-23543	95	1	the	the	DET
ajst-23543	95	2	learning	learning	NOUN
ajst-23543	95	3	rate	rate	NOUN
ajst-23543	95	4	is	be	AUX
ajst-23543	95	5	0.001	0.001	NUM
ajst-23543	95	6	(	(	PUNCT
ajst-23543	95	7	blue	blue	ADJ
ajst-23543	95	8	curve	curve	NOUN
ajst-23543	95	9	):	):	PUNCT
ajst-23543	95	10	although	although	SCONJ
ajst-23543	95	11	the	the	DET
ajst-23543	95	12	initial	initial	ADJ
ajst-23543	95	13	decline	decline	NOUN
ajst-23543	95	14	was	be	AUX
ajst-23543	95	15	slow	slow	ADJ
ajst-23543	95	16	,	,	PUNCT
ajst-23543	95	17	it	it	PRON
ajst-23543	95	18	eventually	eventually	ADV
ajst-23543	95	19	approached	approach	VERB
ajst-23543	95	20	the	the	DET
ajst-23543	95	21	minimum	minimum	NOUN
ajst-23543	95	22	and	and	CCONJ
ajst-23543	95	23	maintained	maintain	VERB
ajst-23543	95	24	good	good	ADJ
ajst-23543	95	25	stability	stability	NOUN
ajst-23543	95	26	.	.	PUNCT
ajst-23543	96	1	to	to	PART
ajst-23543	96	2	sum	sum	VERB
ajst-23543	96	3	up	up	ADP
ajst-23543	96	4	,	,	PUNCT
ajst-23543	96	5	it	it	PRON
ajst-23543	96	6	is	be	AUX
ajst-23543	96	7	very	very	ADV
ajst-23543	96	8	important	important	ADJ
ajst-23543	96	9	to	to	PART
ajst-23543	96	10	select	select	VERB
ajst-23543	96	11	a	a	DET
ajst-23543	96	12	suitable	suitable	ADJ
ajst-23543	96	13	learning	learning	NOUN
ajst-23543	96	14	rate	rate	NOUN
ajst-23543	96	15	for	for	ADP
ajst-23543	96	16	model	model	NOUN
ajst-23543	96	17	training	training	NOUN
ajst-23543	96	18	.	.	PUNCT
ajst-23543	97	1	too	too	ADV
ajst-23543	97	2	high	high	ADJ
ajst-23543	97	3	or	or	CCONJ
ajst-23543	97	4	too	too	ADV
ajst-23543	97	5	low	low	ADJ
ajst-23543	97	6	learning	learning	NOUN
ajst-23543	97	7	rate	rate	NOUN
ajst-23543	97	8	may	may	AUX
ajst-23543	97	9	affect	affect	VERB
ajst-23543	97	10	the	the	DET
ajst-23543	97	11	model	model	NOUN
ajst-23543	97	12	performance	performance	NOUN
ajst-23543	97	13	.	.	PUNCT
ajst-23543	98	1	5	5	X
ajst-23543	98	2	.	.	X
ajst-23543	98	3	conclusion	conclusion	NOUN
ajst-23543	98	4	in	in	ADP
ajst-23543	98	5	this	this	DET
ajst-23543	98	6	paper	paper	NOUN
ajst-23543	98	7	,	,	PUNCT
ajst-23543	98	8	attention	attention	NOUN
ajst-23543	98	9	mechanism	mechanism	NOUN
ajst-23543	98	10	is	be	AUX
ajst-23543	98	11	innovatively	innovatively	ADV
ajst-23543	98	12	applied	apply	VERB
ajst-23543	98	13	to	to	ADP
ajst-23543	98	14	image	image	NOUN
ajst-23543	98	15	feature	feature	NOUN
ajst-23543	98	16	selection	selection	NOUN
ajst-23543	98	17	,	,	PUNCT
ajst-23543	98	18	which	which	PRON
ajst-23543	98	19	solves	solve	VERB
ajst-23543	98	20	the	the	DET
ajst-23543	98	21	limitation	limitation	NOUN
ajst-23543	98	22	of	of	ADP
ajst-23543	98	23	88	88	NUM
ajst-23543	98	24	manual	manual	ADJ
ajst-23543	98	25	feature	feature	NOUN
ajst-23543	98	26	design	design	NOUN
ajst-23543	98	27	in	in	ADP
ajst-23543	98	28	traditional	traditional	ADJ
ajst-23543	98	29	methods	method	NOUN
ajst-23543	98	30	.	.	PUNCT
ajst-23543	99	1	the	the	DET
ajst-23543	99	2	proposed	propose	VERB
ajst-23543	99	3	cnn	cnn	PROPN
ajst-23543	99	4	framework	framework	NOUN
ajst-23543	99	5	based	base	VERB
ajst-23543	99	6	on	on	ADP
ajst-23543	99	7	attention	attention	NOUN
ajst-23543	99	8	mechanism	mechanism	NOUN
ajst-23543	99	9	effectively	effectively	ADV
ajst-23543	99	10	highlights	highlight	VERB
ajst-23543	99	11	the	the	DET
ajst-23543	99	12	key	key	ADJ
ajst-23543	99	13	information	information	NOUN
ajst-23543	99	14	in	in	ADP
ajst-23543	99	15	the	the	DET
ajst-23543	99	16	image	image	NOUN
ajst-23543	99	17	by	by	ADP
ajst-23543	99	18	adaptive	adaptive	ADJ
ajst-23543	99	19	weight	weight	NOUN
ajst-23543	99	20	adjustment	adjustment	NOUN
ajst-23543	99	21	,	,	PUNCT
ajst-23543	99	22	and	and	CCONJ
ajst-23543	99	23	ignores	ignore	VERB
ajst-23543	99	24	the	the	DET
ajst-23543	99	25	redundant	redundant	ADJ
ajst-23543	99	26	background	background	NOUN
ajst-23543	99	27	,	,	PUNCT
ajst-23543	99	28	thus	thus	ADV
ajst-23543	99	29	improving	improve	VERB
ajst-23543	99	30	the	the	DET
ajst-23543	99	31	accuracy	accuracy	NOUN
ajst-23543	99	32	and	and	CCONJ
ajst-23543	99	33	efficiency	efficiency	NOUN
ajst-23543	99	34	of	of	ADP
ajst-23543	99	35	image	image	NOUN
ajst-23543	99	36	classification	classification	NOUN
ajst-23543	99	37	and	and	CCONJ
ajst-23543	99	38	object	object	NOUN
ajst-23543	99	39	detection	detection	NOUN
ajst-23543	99	40	tasks	task	NOUN
ajst-23543	99	41	.	.	PUNCT
ajst-23543	100	1	the	the	DET
ajst-23543	100	2	experimental	experimental	ADJ
ajst-23543	100	3	results	result	NOUN
ajst-23543	100	4	show	show	VERB
ajst-23543	100	5	that	that	SCONJ
ajst-23543	100	6	the	the	DET
ajst-23543	100	7	model	model	NOUN
ajst-23543	100	8	shows	show	VERB
ajst-23543	100	9	excellent	excellent	ADJ
ajst-23543	100	10	performance	performance	NOUN
ajst-23543	100	11	on	on	ADP
ajst-23543	100	12	multiple	multiple	ADJ
ajst-23543	100	13	datasets	dataset	NOUN
ajst-23543	100	14	,	,	PUNCT
ajst-23543	100	15	which	which	PRON
ajst-23543	100	16	proves	prove	VERB
ajst-23543	100	17	the	the	DET
ajst-23543	100	18	remarkable	remarkable	ADJ
ajst-23543	100	19	advantages	advantage	NOUN
ajst-23543	100	20	of	of	ADP
ajst-23543	100	21	attention	attention	NOUN
ajst-23543	100	22	mechanism	mechanism	NOUN
ajst-23543	100	23	in	in	ADP
ajst-23543	100	24	the	the	DET
ajst-23543	100	25	field	field	NOUN
ajst-23543	100	26	of	of	ADP
ajst-23543	100	27	deep	deep	ADJ
ajst-23543	100	28	learning	learn	VERB
ajst-23543	100	29	image	image	NOUN
ajst-23543	100	30	processing	processing	NOUN
ajst-23543	100	31	,	,	PUNCT
ajst-23543	100	32	and	and	CCONJ
ajst-23543	100	33	provides	provide	VERB
ajst-23543	100	34	new	new	ADJ
ajst-23543	100	35	directions	direction	NOUN
ajst-23543	100	36	and	and	CCONJ
ajst-23543	100	37	ideas	idea	NOUN
ajst-23543	100	38	for	for	ADP
ajst-23543	100	39	subsequent	subsequent	ADJ
ajst-23543	100	40	research	research	NOUN
ajst-23543	100	41	.	.	PUNCT
ajst-23543	101	1	references	reference	NOUN
ajst-23543	101	2	[	[	X
ajst-23543	101	3	1	1	NUM
ajst-23543	101	4	]	]	X
ajst-23543	101	5	muruganantham	muruganantham	NOUN
ajst-23543	101	6	,	,	PUNCT
ajst-23543	101	7	priyanga	priyanga	NOUN
ajst-23543	101	8	,	,	PUNCT
ajst-23543	101	9	et	et	PROPN
ajst-23543	101	10	al	al	PROPN
ajst-23543	101	11	.	.	PUNCT
ajst-23543	102	1	"	"	PUNCT
ajst-23543	102	2	a	a	DET
ajst-23543	102	3	systematic	systematic	ADJ
ajst-23543	102	4	literature	literature	NOUN
ajst-23543	102	5	review	review	NOUN
ajst-23543	102	6	on	on	ADP
ajst-23543	102	7	crop	crop	NOUN
ajst-23543	102	8	yield	yield	NOUN
ajst-23543	102	9	prediction	prediction	NOUN
ajst-23543	102	10	with	with	ADP
ajst-23543	102	11	deep	deep	ADJ
ajst-23543	102	12	learning	learning	NOUN
ajst-23543	102	13	and	and	CCONJ
ajst-23543	102	14	remote	remote	ADJ
ajst-23543	102	15	sensing	sensing	NOUN
ajst-23543	102	16	.	.	PUNCT
ajst-23543	102	17	"	"	PUNCT
ajst-23543	103	1	remote	remote	ADJ
ajst-23543	103	2	sensing	sense	VERB
ajst-23543	103	3	14.9	14.9	NUM
ajst-23543	103	4	(	(	PUNCT
ajst-23543	103	5	2022	2022	NUM
ajst-23543	103	6	):	):	PUNCT
ajst-23543	103	7	1990	1990	NUM
ajst-23543	103	8	.	.	PUNCT
ajst-23543	104	1	[	[	X
ajst-23543	104	2	2	2	NUM
ajst-23543	104	3	]	]	PUNCT
ajst-23543	104	4	xu	xu	PROPN
ajst-23543	104	5	,	,	PUNCT
ajst-23543	104	6	yonghao	yonghao	PROPN
ajst-23543	104	7	,	,	PUNCT
ajst-23543	104	8	and	and	CCONJ
ajst-23543	104	9	pedram	pedram	PROPN
ajst-23543	104	10	ghamisi	ghamisi	PROPN
ajst-23543	104	11	.	.	PUNCT
ajst-23543	105	1	"	"	PUNCT
ajst-23543	105	2	universal	universal	ADJ
ajst-23543	105	3	adversarial	adversarial	ADJ
ajst-23543	105	4	examples	example	NOUN
ajst-23543	105	5	in	in	ADP
ajst-23543	105	6	remote	remote	ADJ
ajst-23543	105	7	sensing	sensing	NOUN
ajst-23543	105	8	:	:	PUNCT
ajst-23543	105	9	methodology	methodology	NOUN
ajst-23543	105	10	and	and	CCONJ
ajst-23543	105	11	benchmark	benchmark	NOUN
ajst-23543	105	12	.	.	PUNCT
ajst-23543	105	13	"	"	PUNCT
ajst-23543	106	1	ieee	ieee	NOUN
ajst-23543	106	2	transactions	transaction	NOUN
ajst-23543	106	3	on	on	ADP
ajst-23543	106	4	geoscience	geoscience	NOUN
ajst-23543	106	5	and	and	CCONJ
ajst-23543	106	6	remote	remote	ADJ
ajst-23543	106	7	sensing	sense	VERB
ajst-23543	106	8	60	60	NUM
ajst-23543	106	9	(	(	PUNCT
ajst-23543	106	10	2022	2022	NUM
ajst-23543	106	11	):	):	PUNCT
ajst-23543	106	12	1	1	NUM
ajst-23543	106	13	-	-	SYM
ajst-23543	106	14	15	15	NUM
ajst-23543	106	15	.	.	PUNCT
ajst-23543	107	1	[	[	X
ajst-23543	107	2	3	3	NUM
ajst-23543	107	3	]	]	X
ajst-23543	107	4	kattenborn	kattenborn	ADJ
ajst-23543	107	5	,	,	PUNCT
ajst-23543	107	6	teja	teja	PROPN
ajst-23543	107	7	,	,	PUNCT
ajst-23543	107	8	et	et	PROPN
ajst-23543	107	9	al	al	PROPN
ajst-23543	107	10	.	.	PROPN
ajst-23543	107	11	spatially	spatially	ADV
ajst-23543	107	12	autocorrelated	autocorrelated	ADJ
ajst-23543	107	13	training	training	NOUN
ajst-23543	107	14	and	and	CCONJ
ajst-23543	107	15	validation	validation	NOUN
ajst-23543	107	16	samples	sample	NOUN
ajst-23543	107	17	inflate	inflate	VERB
ajst-23543	107	18	performance	performance	NOUN
ajst-23543	107	19	assessment	assessment	NOUN
ajst-23543	107	20	of	of	ADP
ajst-23543	107	21	convolutional	convolutional	ADJ
ajst-23543	107	22	neural	neural	ADJ
ajst-23543	107	23	networks	network	NOUN
ajst-23543	107	24	.	.	PUNCT
ajst-23543	108	1	isprs	isprs	PROPN
ajst-23543	108	2	open	open	ADJ
ajst-23543	108	3	journal	journal	NOUN
ajst-23543	108	4	of	of	ADP
ajst-23543	108	5	photogrammetry	photogrammetry	NOUN
ajst-23543	108	6	and	and	CCONJ
ajst-23543	108	7	remote	remote	ADJ
ajst-23543	108	8	sensing	sensing	NOUN
ajst-23543	108	9	,	,	PUNCT
ajst-23543	108	10	2022	2022	NUM
ajst-23543	108	11	,	,	PUNCT
ajst-23543	108	12	5	5	NUM
ajst-23543	108	13	:	:	SYM
ajst-23543	108	14	100018	100018	NUM
ajst-23543	108	15	.	.	PUNCT
ajst-23543	109	1	[	[	X
ajst-23543	109	2	4	4	X
ajst-23543	109	3	]	]	X
ajst-23543	109	4	graves	grave	NOUN
ajst-23543	109	5	,	,	PUNCT
ajst-23543	109	6	alex	alex	PROPN
ajst-23543	109	7	,	,	PUNCT
ajst-23543	109	8	and	and	CCONJ
ajst-23543	109	9	jürgen	jürgen	PROPN
ajst-23543	109	10	schmidhuber	schmidhuber	PROPN
ajst-23543	109	11	.	.	PUNCT
ajst-23543	110	1	"	"	PUNCT
ajst-23543	110	2	framewise	framewise	ADJ
ajst-23543	110	3	phoneme	phoneme	NOUN
ajst-23543	110	4	classification	classification	NOUN
ajst-23543	110	5	with	with	ADP
ajst-23543	110	6	bidirectional	bidirectional	ADJ
ajst-23543	110	7	lstm	lstm	NOUN
ajst-23543	110	8	and	and	CCONJ
ajst-23543	110	9	other	other	ADJ
ajst-23543	110	10	neural	neural	ADJ
ajst-23543	110	11	network	network	NOUN
ajst-23543	110	12	architectures	architecture	NOUN
ajst-23543	110	13	.	.	PUNCT
ajst-23543	110	14	"	"	PUNCT
ajst-23543	111	1	neural	neural	ADJ
ajst-23543	111	2	networks	network	NOUN
ajst-23543	111	3	18.5	18.5	NUM
ajst-23543	111	4	-	-	SYM
ajst-23543	111	5	6	6	NUM
ajst-23543	111	6	(	(	PUNCT
ajst-23543	111	7	2005	2005	NUM
ajst-23543	111	8	):	):	PUNCT
ajst-23543	111	9	602610	602610	NUM
ajst-23543	111	10	.	.	PUNCT
ajst-23543	112	1	[	[	X
ajst-23543	112	2	5	5	NUM
ajst-23543	112	3	]	]	X
ajst-23543	112	4	zhu	zhu	PROPN
ajst-23543	112	5	,	,	PUNCT
ajst-23543	112	6	jiawei	jiawei	PROPN
ajst-23543	112	7	,	,	PUNCT
ajst-23543	112	8	et	et	PROPN
ajst-23543	112	9	al	al	PROPN
ajst-23543	112	10	.	.	PUNCT
ajst-23543	112	11	"	"	PUNCT
ajst-23543	112	12	ast	ast	NOUN
ajst-23543	112	13	-	-	PUNCT
ajst-23543	112	14	gcn	gcn	NOUN
ajst-23543	112	15	:	:	PUNCT
ajst-23543	112	16	attribute	attribute	NOUN
ajst-23543	112	17	-	-	PUNCT
ajst-23543	112	18	augmented	augment	VERB
ajst-23543	112	19	spatiotemporal	spatiotemporal	ADJ
ajst-23543	112	20	graph	graph	NOUN
ajst-23543	112	21	convolutional	convolutional	ADJ
ajst-23543	112	22	network	network	NOUN
ajst-23543	112	23	for	for	ADP
ajst-23543	112	24	traffic	traffic	NOUN
ajst-23543	112	25	forecasting	forecasting	NOUN
ajst-23543	112	26	.	.	PUNCT
ajst-23543	112	27	"	"	PUNCT
ajst-23543	113	1	ieee	ieee	NOUN
ajst-23543	113	2	access	access	NOUN
ajst-23543	113	3	9	9	NUM
ajst-23543	113	4	(	(	PUNCT
ajst-23543	113	5	2021	2021	NUM
ajst-23543	113	6	):	):	PUNCT
ajst-23543	113	7	35973	35973	NUM
ajst-23543	113	8	-	-	SYM
ajst-23543	113	9	35983	35983	NUM
ajst-23543	113	10	.	.	PUNCT
ajst-23543	114	1	[	[	X
ajst-23543	114	2	6	6	NUM
ajst-23543	114	3	]	]	X
ajst-23543	114	4	wen	wen	PROPN
ajst-23543	114	5	,	,	PUNCT
ajst-23543	114	6	guangqi	guangqi	PROPN
ajst-23543	114	7	,	,	PUNCT
ajst-23543	114	8	et	et	PROPN
ajst-23543	114	9	al	al	PROPN
ajst-23543	114	10	.	.	PUNCT
ajst-23543	115	1	"	"	PUNCT
ajst-23543	115	2	mvs	mvs	NOUN
ajst-23543	115	3	-	-	PUNCT
ajst-23543	115	4	gcn	gcn	NOUN
ajst-23543	115	5	:	:	PUNCT
ajst-23543	115	6	a	a	DET
ajst-23543	115	7	prior	prior	ADJ
ajst-23543	115	8	brain	brain	NOUN
ajst-23543	115	9	structure	structure	NOUN
ajst-23543	115	10	learning	learning	NOUN
ajst-23543	115	11	-	-	PUNCT
ajst-23543	115	12	guided	guide	VERB
ajst-23543	115	13	multi	multi	ADJ
ajst-23543	115	14	-	-	ADJ
ajst-23543	115	15	view	view	ADJ
ajst-23543	115	16	graph	graph	NOUN
ajst-23543	115	17	convolution	convolution	NOUN
ajst-23543	115	18	network	network	NOUN
ajst-23543	115	19	for	for	ADP
ajst-23543	115	20	autism	autism	NOUN
ajst-23543	115	21	spectrum	spectrum	NOUN
ajst-23543	115	22	disorder	disorder	NOUN
ajst-23543	115	23	diagnosis	diagnosis	NOUN
ajst-23543	115	24	.	.	PUNCT
ajst-23543	115	25	"	"	PUNCT
ajst-23543	116	1	computers	computer	NOUN
ajst-23543	116	2	in	in	ADP
ajst-23543	116	3	biology	biology	NOUN
ajst-23543	116	4	and	and	CCONJ
ajst-23543	116	5	medicine	medicine	NOUN
ajst-23543	116	6	142	142	NUM
ajst-23543	116	7	(	(	PUNCT
ajst-23543	116	8	2022	2022	NUM
ajst-23543	116	9	):	):	PUNCT
ajst-23543	116	10	105239	105239	NUM
ajst-23543	116	11	.	.	PUNCT
ajst-23543	117	1	[	[	X
ajst-23543	117	2	7	7	NUM
ajst-23543	117	3	]	]	X
ajst-23543	117	4	hou	hou	PROPN
ajst-23543	117	5	,	,	PUNCT
ajst-23543	117	6	jialu	jialu	NOUN
ajst-23543	117	7	,	,	PUNCT
ajst-23543	117	8	hang	hang	PROPN
ajst-23543	117	9	wei	wei	PROPN
ajst-23543	117	10	,	,	PUNCT
ajst-23543	117	11	and	and	CCONJ
ajst-23543	117	12	bin	bin	PROPN
ajst-23543	117	13	liu	liu	PROPN
ajst-23543	117	14	.	.	PUNCT
ajst-23543	118	1	"	"	PUNCT
ajst-23543	118	2	ipida	ipida	ADJ
ajst-23543	118	3	-	-	PUNCT
ajst-23543	118	4	gcn	gcn	NOUN
ajst-23543	118	5	:	:	PUNCT
ajst-23543	118	6	identification	identification	NOUN
ajst-23543	118	7	of	of	ADP
ajst-23543	118	8	pirna	pirna	NOUN
ajst-23543	118	9	-	-	PUNCT
ajst-23543	118	10	disease	disease	NOUN
ajst-23543	118	11	associations	association	NOUN
ajst-23543	118	12	based	base	VERB
ajst-23543	118	13	on	on	ADP
ajst-23543	118	14	graph	graph	NOUN
ajst-23543	118	15	convolutional	convolutional	ADJ
ajst-23543	118	16	network	network	NOUN
ajst-23543	118	17	.	.	PUNCT
ajst-23543	118	18	"	"	PUNCT
ajst-23543	119	1	plos	plos	PROPN
ajst-23543	119	2	computational	computational	ADJ
ajst-23543	119	3	biology	biology	NOUN
ajst-23543	119	4	18.10	18.10	NUM
ajst-23543	119	5	(	(	PUNCT
ajst-23543	119	6	2022	2022	NUM
ajst-23543	119	7	):	):	PUNCT
ajst-23543	119	8	e1010671	e1010671	NOUN
ajst-23543	119	9	.	.	PUNCT
ajst-23543	120	1	[	[	X
ajst-23543	120	2	8	8	NUM
ajst-23543	120	3	]	]	PUNCT
ajst-23543	120	4	eliasof	eliasof	NOUN
ajst-23543	120	5	,	,	PUNCT
ajst-23543	120	6	moshe	moshe	PROPN
ajst-23543	120	7	,	,	PUNCT
ajst-23543	120	8	eldad	eldad	PROPN
ajst-23543	120	9	haber	haber	PROPN
ajst-23543	120	10	,	,	PUNCT
ajst-23543	120	11	and	and	CCONJ
ajst-23543	120	12	eran	eran	ADJ
ajst-23543	120	13	treister	treister	NOUN
ajst-23543	120	14	.	.	PUNCT
ajst-23543	121	1	"	"	PUNCT
ajst-23543	121	2	pde	pde	NOUN
ajst-23543	121	3	-	-	PUNCT
ajst-23543	121	4	gcn	gcn	NOUN
ajst-23543	121	5	:	:	PUNCT
ajst-23543	121	6	novel	novel	ADJ
ajst-23543	121	7	architectures	architecture	NOUN
ajst-23543	121	8	for	for	ADP
ajst-23543	121	9	graph	graph	NOUN
ajst-23543	121	10	neural	neural	ADJ
ajst-23543	121	11	networks	network	NOUN
ajst-23543	121	12	motivated	motivate	VERB
ajst-23543	121	13	by	by	ADP
ajst-23543	121	14	partial	partial	ADJ
ajst-23543	121	15	differential	differential	ADJ
ajst-23543	121	16	equations	equation	NOUN
ajst-23543	121	17	.	.	PUNCT
ajst-23543	121	18	"	"	PUNCT
ajst-23543	121	19	advances	advance	NOUN
ajst-23543	121	20	in	in	ADP
ajst-23543	121	21	neural	neural	ADJ
ajst-23543	121	22	information	information	NOUN
ajst-23543	121	23	processing	processing	NOUN
ajst-23543	121	24	systems	system	NOUN
ajst-23543	121	25	34	34	NUM
ajst-23543	121	26	(	(	PUNCT
ajst-23543	121	27	2021	2021	NUM
ajst-23543	121	28	):	):	PUNCT
ajst-23543	121	29	3836	3836	NUM
ajst-23543	121	30	-	-	SYM
ajst-23543	121	31	3849	3849	NUM
ajst-23543	121	32	.	.	PUNCT
ajst-23543	122	1	[	[	X
ajst-23543	122	2	9	9	NUM
ajst-23543	122	3	]	]	X
ajst-23543	122	4	peng	peng	PROPN
ajst-23543	122	5	,	,	PUNCT
ajst-23543	122	6	shaowen	shaowen	PROPN
ajst-23543	122	7	,	,	PUNCT
ajst-23543	122	8	kazunari	kazunari	NOUN
ajst-23543	122	9	sugiyama	sugiyama	NOUN
ajst-23543	122	10	,	,	PUNCT
ajst-23543	122	11	and	and	CCONJ
ajst-23543	122	12	tsunenori	tsunenori	X
ajst-23543	122	13	mine	mine	NOUN
ajst-23543	122	14	.	.	PUNCT
ajst-23543	123	1	"	"	PUNCT
ajst-23543	123	2	svd	svd	PROPN
ajst-23543	123	3	-	-	NOUN
ajst-23543	123	4	gcn	gcn	NOUN
ajst-23543	123	5	:	:	PUNCT
ajst-23543	123	6	a	a	DET
ajst-23543	123	7	simplified	simplified	ADJ
ajst-23543	123	8	graph	graph	NOUN
ajst-23543	123	9	convolution	convolution	NOUN
ajst-23543	123	10	paradigm	paradigm	NOUN
ajst-23543	123	11	for	for	ADP
ajst-23543	123	12	recommendation	recommendation	NOUN
ajst-23543	123	13	.	.	PUNCT
ajst-23543	123	14	"	"	PUNCT
ajst-23543	124	1	proceedings	proceeding	NOUN
ajst-23543	124	2	of	of	ADP
ajst-23543	124	3	the	the	DET
ajst-23543	124	4	31st	31st	PROPN
ajst-23543	124	5	acm	acm	PROPN
ajst-23543	124	6	international	international	ADJ
ajst-23543	124	7	conference	conference	NOUN
ajst-23543	124	8	on	on	ADP
ajst-23543	124	9	information	information	NOUN
ajst-23543	124	10	&	&	CCONJ
ajst-23543	124	11	knowledge	knowledge	PROPN
ajst-23543	124	12	management	management	PROPN
ajst-23543	124	13	.	.	PUNCT
ajst-23543	125	1	2022	2022	NUM
ajst-23543	125	2	.	.	PUNCT