Multi-Classification and Recognition of Hyperspectral Remote Sensing Objects Based on Convolutional Neural Network

Miao Yan; Hongdong Zhao; Yuhai Li; Jie Zhang; Zetong Zhao

doi:10.3788/LOP56.021702

Laser & Optoelectronics Progress, Volume. 56, Issue 2, 021702(2019)

Multi-Classification and Recognition of Hyperspectral Remote Sensing Objects Based on Convolutional Neural Network

Miao Yan^1,2, Hongdong Zhao^1、*, Yuhai Li², Jie Zhang^1,2, and Zetong Zhao^1,2

Author Affiliations

¹ School of Electronics and Information Engineering, Hebei University of Technology, Tianjin 300401, China

² Electronics Technology Group Corporation No.53 Research Institute, Key Laboratory of Electro-Optical Information Control and Security Technology, Tianjin 300308, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(9)

Fig. 1. Flow chart of proposed remote sensing image classification

Download full size

Fig. 2. Image examples of remote sensing ground objects

Download full size

Fig. 3. Learning rate comparison of neural network models with different pooling layers and classifiers. (a) Max pooling, Softmax classifier; (b) Max pooling, Sigmoid classifier; (c) Mean pooling, Softmax classifier; (d) Mean pooling, Sigmoid classifier

Download full size

Fig. 4. Recognition rate comparison of four models for different iteration times

Download full size

Fig. 5. Accuracy comparison of three datasets under optimal parameters

Download full size

Table 1. Comparison of neural network models

View table

Table 1. Comparison of neural network models

Model	ModelⅠ	ModelⅡ	ModelⅢ	ModelⅣ
Networkmodel	Convl (6@5×5)	Convl (6@5×5)	Convl (6@5×5)	Convl (6@5×5)
	Max pooling (2×2)	Max pooling (2×2)	Mean pooling (2×2)	Mean pooling (2×2)
	Convl (12@5×5)	Convl (12@5×5)	Convl (12@5×5)	Convl (12@5×5)
	Max pooling (2×2)	Max pooling (2×2)	Mean pooling (2×2)	Mean pooling (2×2)
	Softmax classifier	Sigmoid classifier	Softmax classifier	Sigmoid classifier
Optimal learn rate	0.03	0.3	0.6	0.9
Accuracy	0.9658	0.9583	0.9604	0.91

Table 2. Number of unclassified images in test set for dataset-I
View table
Table 2. Number of unclassified images in test set for dataset-I
Iteration times A B C D E F Total
20 0 180 0 54 4 0 238
40 0 98 0 50 0 0 148
60 0 1 0 55 0 0 56
80 0 0 0 68 0 0 68
100 0 0 0 63 0 4 67
120 0 0 0 36 0 0 36
140 0 0 0 33 0 0 33
160 0 1 0 28 0 0 29
180 0 0 0 36 0 0 36
200 0 0 0 35 0 0 35
220 0 0 0 36 0 0 36

Table 3. Number of unclassified images in test set for dataset-Ⅱ

View table

Table 3. Number of unclassified images in test set for dataset-Ⅱ

Iteration times	B	D	E	G	I	J	Total
20	191	41	2	54	111	168	567
40	17	63	1	22	85	88	276
60	15	36	1	10	11	62	135
80	14	56	2	8	5	31	116
100	5	39	2	4	1	33	84
120	1	63	1	4	8	34	111
140	0	61	1	6	10	20	98
160	0	50	2	2	6	26	86
180	4	58	1	1	1	29	94
200	0	49	1	1	11	25	87
220	2	45	1	1	2	21	72

Table 4. Number of unclassified images in test set for dataset-Ⅲ

View table

Table 4. Number of unclassified images in test set for dataset-Ⅲ

Iteration times	B	D	E	F	G	I	J	K	L	M	N	Total
20	160	29	4	40	78	168	183	116	67	54	24	923
40	25	25	5	0	27	16	90	63	18	14	10	293
60	0	91	4	9	34	75	38	45	1	13	8	318
80	25	33	4	0	8	7	63	30	10	8	6	194
100	0	44	0	0	16	19	40	37	3	7	6	172
120	72	8	3	38	7	0	47	34	19	7	3	238
140	0	56	2	0	16	11	32	31	4	7	5	164
160	0	58	1	0	5	12	36	30	5	5	5	157
180	0	106	0	0	12	68	28	31	3	5	5	258
200	3	30	0	0	9	0	34	32	4	6	5	123
220	0	57	0	0	15	16	24	30	4	5	5	156

Tools

Get Citation

Copy Citation Text

Miao Yan, Hongdong Zhao, Yuhai Li, Jie Zhang, Zetong Zhao. Multi-Classification and Recognition of Hyperspectral Remote Sensing Objects Based on Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2019, 56(2): 021702

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites