Hyperspectral Image Classification Method Based on Targets Constraint and Spectral-Spatial Iteration

Chunyan Yu; Meng Zhao; Meiping Song; Sen Li; Yulei Wang

doi:10.3788/AOS201838.0628003

Acta Optica Sinica, Volume. 38, Issue 6, 0628003(2018)

Hyperspectral Image Classification Method Based on Targets Constraint and Spectral-Spatial Iteration

Chunyan Yu¹, Meng Zhao¹, Meiping Song^1,2、*, Sen Li¹, and Yulei Wang^1,2,3

Author Affiliations

¹ Information Science and Technology College, Dalian Maritime University, Dalian, Liaoning 116026, China

² State Key Laboratory of Integrated Services Networks, Xi'an, Shannxi 710071, China

³ Key Laboratory of Spectral Imaging Technology, Chinese Academy of Sciences, Xi'an, Shannxi 710071, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(18)

Fig. 1. Flow chart of classification method

Download full size

Fig. 2. Comparison of feedback stack methods

Download full size

Fig. 3. Image of Purdue. (a) False color image; (b) image of ground truth

Download full size

Fig. 4. Classification results of Purdue data by MTCC-1 method. (a) Iteration for 1 time; (b) iteration for 2 times; (c) iteration for 5 times; (d) final result

Download full size

Fig. 5. Classification results of Purdue data by MTCC-2 method. (a) Iteration for 1 time; (b) iteration for 2 times; (c) iteration for 5 times; (d) final result

Download full size

Fig. 6. Image of Salinas Valley. (a) False color image; (b) image of ground truth

Download full size

Fig. 7. Classification results of Salinas data by MTCC-1 method. (a) Iteration for 1 time; (b) iteration for 2 times; (c) iteration for 5 times; (d) final result

Download full size

Fig. 8. Classification results of Salinas data by MTCC-2 method. (a) Iteration for 1 time; (b) iteration for 2 times; (c) iteration for 5 times; (d) final result

Download full size

Fig. 9. Image of Pavia University. (a) False color image; (b) image of ground truth

Download full size

Fig. 10. Classification results of Pavia data by MTCC-1 method. (a) Iteration for 1 time; (b) iteration for 2 times; (c) iteration for 5 times; (d) final result

Download full size

Fig. 11. Classification results of Pavia data by MTCC-2 method. (a) Iteration for 1 time; (b) iteration for 2 times; (c) iteration for 5 times; (d) final result

Download full size

Table 1. Confusion matrix of C-class classification results
View table
Table 1. Confusion matrix of C-class classification results
Classification result Real result
C₁ C₂ … C_n BKG
C₁ n₁₁ n₁₂ … n₁_n n_1B
C₂ n₂₁ n₂₂ … n₂_n n_2B
︙ ︙ ︙ ︙ ︙ ︙
C_n n_n₁ n_n₂ … n_nn n_n_B
BKG n_B1 n_B2 … n_B_n n_BB

Table 2. Category name and pixel number of Purdue image

View table

Table 2. Category name and pixel number of Purdue image

Label	Class name	Sample
Class 1	Alfalfa	46
Class 2	Corn-notill	1428
Class 3	Corn-min	830
Class 4	Corn	237
Class 5	Grass/pasture	483
Class 6	Grass/trees	730
Class 7	Grass/pasture-mowed	28
Class 8	Hay-windrowed	478
Class 9	Oats	20
Class 10	Soybeans-notill	972
Class 11	Soybeans-min	2455
Class 12	Soybeans-clean	593
Class 13	Wheat	205
Class 14	Woods	1265
Class 15	Bldg-grass green-drives	386
Class 16	Stone-steel towers	93
BKG	Background	10776

Table 3. Comparison of classification evaluation results of Purdue data with different methods%

View table

Table 3. Comparison of classification evaluation results of Purdue data with different methods%

Class	MTCC-1		MTCC-2		EPF-B-g		EPF-B-c		EPF-G-g		EPF-G-c
Class	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre
1	91.30	97.67	95.65	95.65	100.00	57.50	97.83	57.69	97.83	58.44	100.00	60.53
2	97.76	94.58	97.69	91.96	85.01	67.03	84.94	66.91	85.22	67.09	84.45	65.76
3	99.40	93.96	98.80	92.34	93.13	74.40	94.10	73.54	92.41	73.12	92.41	74.11
4	100.00	99.16	100.00	99.58	99.16	64.74	99.16	64.56	99.16	60.26	99.16	59.95
5	97.10	94.94	93.79	95.77	93.58	39.65	93.37	41.45	94.00	42.23	93.58	43.55
6	97.40	95.05	98.63	91.25	100.00	49.06	99.73	48.92	99.73	46.82	100.00	48.70
7	100.00	100.00	100.00	100.00	96.43	62.79	96.43	65.85	96.43	56.25	96.43	56.25
8	99.58	99.37	99.58	100.00	100.00	69.88	100.00	70.40	100.00	71.77	100.00	71.34
9	100.00	64.52	100.00	62.50	95.00	38.78	100.00	42.55	100.00	55.56	95.00	63.33
10	97.53	93.03	97.53	90.46	82.30	62.70	82.82	61.97	81.79	62.21	82.51	62.51
11	98.78	95.62	98.37	94.19	95.23	76.71	95.64	76.63	94.46	77.02	94.70	77.45
12	96.96	94.73	96.80	97.95	98.82	62.47	98.65	63.38	98.48	61.67	98.65	61.13
13	98.54	98.54	97.07	98.51	99.02	79.61	99.02	77.19	99.51	76.69	99.51	73.12
14	97.55	97.70	95.57	96.95	98.26	31.91	98.50	31.58	98.10	31.90	98.50	31.97
15	97.15	98.17	99.74	98.72	96.63	7.85	96.89	7.94	94.30	7.67	99.48	7.99
16	96.77	95.74	96.77	96.77	96.77	54.88	98.92	57.86	100.00	52.25	100.00	52.54
A_O	98.09		97.70		94.83		95.33		94.99		94.60
P	96.84		96.00		46.23		46.47		46.31		46.12

Table 4. Category name and pixel number of Salinas image

View table

Table 4. Category name and pixel number of Salinas image

Label	Class name	Sample
Class 1	Weed 1	2009
Class 2	Weed 2	3726
Class 3	Fallow	1976
Class 4	Fallow rough plow	1394
Class 5	Fallow smooth	2678
Class 6	Stubble	3959
Class 7	Celery	3579
Class 8	Grapes untrained	11271
Class 9	Soil vineyard develop	6203
Class 10	Corn	3278
Class 11	Lettuce 4 weeks	1068
Class 12	Lettuce 5 weeks	1927
Class 13	Lettuce 6 weeks	916
Class 14	Lettuce 7 weeks	1070
Class 15	Vineyard untrained	7268
Class 16	Vineyard trellis	1807
BKG	Background	56975

Table 5. Comparison of classification evaluation results of Salinas data with different methods%

View table

Table 5. Comparison of classification evaluation results of Salinas data with different methods%

Class	MTCC-1		MTCC-2		EPF-B-g		EPF-B-c		EPF-G-g		EPF-G-c
Class	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre
1	98.61	93.22	99.15	90.46	100.00	73.19	100.00	74.05	100.00	74.00	100.00	71.22
2	98.28	95.22	99.06	94.01	100.00	56.22	99.97	56.21	100.00	56.72	100.00	54.67
3	92.91	92.73	95.24	88.69	100.00	12.20	100.00	12.22	100.00	12.13	100.00	12.10
4	94.12	85.47	94.62	83.75	100.00	23.13	100.00	23.25	100.00	23.04	100.00	22.64
5	95.22	86.03	96.34	85.63	98.51	55.95	98.36	55.66	98.47	56.29	98.84	57.61
6	98.81	89.97	99.27	89.32	100.00	81.75	100.00	81.18	100.00	81.58	100.00	80.65
7	98.44	89.87	99.39	88.35	100.00	79.34	100.00	80.14	100.00	80.21	99.97	79.48
8	98.97	95.61	98.05	94.88	81.52	87.11	81.47	86.84	82.37	88.33	83.52	88.82
9	94.97	96.70	94.23	96.50	99.85	35.25	99.84	35.33	99.87	35.19	99.87	34.90
10	95.79	95.88	96.03	94.93	96.19	28.06	96.06	27.77	96.49	27.96	97.86	30.16
11	94.10	93.75	95.32	89.85	100.00	23.57	99.91	23.58	100.00	23.71	100.00	24.09
12	96.94	90.59	97.82	90.36	100.00	27.42	100.00	27.36	100.00	27.20	100.00	27.23
13	97.38	76.96	96.83	76.27	99.13	67.16	99.45	67.38	99.56	66.91	99.78	66.52
14	97.57	83.92	98.60	82.17	100.00	65.97	100.00	65.64	100.00	65.40	100.00	67.85
15	98.35	98.18	96.86	96.89	93.82	76.17	93.26	75.86	94.32	76.86	96.19	77.94
16	97.68	92.80	98.56	92.23	99.61	69.47	99.56	69.19	99.39	70.35	99.67	70.38
A_O	97.33		97.29		95.87		95.70		96.01		96.55
P	95.32		94.81		46.71		46.63		46.77		47.04

Table 6. Category name and pixel number of Pavia image

View table

Table 6. Category name and pixel number of Pavia image

Label	Class name	Sample
Class 1	Asphalt	6631
Class 2	Meadows	18649
Class 3	Gravel	2099
Class 4	Trees	3064
Class 5	Painted metal sheets	1345
Class 6	Bare soil	5029
Class 7	Bitumen	1330
Class 8	Self-blocking bricks	3682
Class 9	Shadows	947
BKG	Background	164624

Table 7. Comparison of classification evaluation results of Pavia data with different methods%

View table

Table 7. Comparison of classification evaluation results of Pavia data with different methods%

Class	MTCC-1		MTCC-2		EPF-B-g		EPF-B-c		EPF-G-g		EPF-G-c
Class	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre	C_i_,OA	C_i_,Pre
1	83.56	32.66	85.06	23.00	97.07	18.28	97.10	18.20	96.53	18.56	96.95	18.72
2	85.05	85.17	84.63	76.47	98.10	36.51	98.09	36.48	98.13	37.10	98.16	37.42
3	80.04	42.60	76.46	33.66	91.47	34.19	91.71	34.55	91.76	35.11	91.81	35.02
4	83.86	22.37	84.93	19.18	95.04	10.03	93.99	9.93	94.35	10.21	98.14	10.69
5	99.12	61.43	98.03	60.02	100.00	42.92	100.00	44.42	100.00	41.81	100.00	44.63
6	85.41	89.91	88.24	85.92	100.00	9.69	100.00	9.74	100.00	9.73	100.00	9.72
7	87.37	70.25	85.29	66.51	100.00	39.82	100.00	39.19	100.00	40.82	100.00	41.34
8	82.03	27.09	90.52	22.79	99.02	19.19	98.72	19.23	98.91	19.28	99.51	19.31
9	80.15	24.55	73.22	19.06	100.00	9.41	100.00	9.15	100.00	7.44	100.00	7.00
A_O	84.68		85.43		98.97		98.95		98.84		99.17
P	79.13		71.44		20.41		20.41		20.39		20.45

Tools

Get Citation

Copy Citation Text

Chunyan Yu, Meng Zhao, Meiping Song, Sen Li, Yulei Wang. Hyperspectral Image Classification Method Based on Targets Constraint and Spectral-Spatial Iteration[J]. Acta Optica Sinica, 2018, 38(6): 0628003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Remote Sensing and Sensors

Received: Nov. 20, 2017

Accepted: --

Published Online: Jul. 9, 2018

The Author Email: Song Meiping (smping@163.com)

DOI:10.3788/AOS201838.0628003

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Confusion matrix of C-class classification results

Table 1. Confusion matrix of C-class classification results

Table 2. Category name and pixel number of Purdue image

Table 2. Category name and pixel number of Purdue image

Table 3. Comparison of classification evaluation results of Purdue data with different methods%

Table 3. Comparison of classification evaluation results of Purdue data with different methods%

Table 4. Category name and pixel number of Salinas image

Table 4. Category name and pixel number of Salinas image

Table 5. Comparison of classification evaluation results of Salinas data with different methods%

Table 5. Comparison of classification evaluation results of Salinas data with different methods%

Table 6. Category name and pixel number of Pavia image

Table 6. Category name and pixel number of Pavia image

Table 7. Comparison of classification evaluation results of Pavia data with different methods%

Table 7. Comparison of classification evaluation results of Pavia data with different methods%