Hyperspectral Image Classification Combined with Convolutional Neural Network and Sparse Coding

Jinguang Sun; Yanbei Li; Xian Wei; Wanli Wang

doi:10.3788/LOP57.182802

Journals >Laser & Optoelectronics Progress >Volume 57 >Issue 18 >Page 182802 > Article

Laser & Optoelectronics Progress
Vol. 57, Issue 18, 182802 (2020)

Hyperspectral Image Classification Combined with Convolutional Neural Network and Sparse Coding

Jinguang Sun¹, Yanbei Li^1、2、*, Xian Wei², and Wanli Wang^1、2

Author Affiliations

¹College of Electronic and Information Engineering, Liaoning University of Engineering and Technology, Huludao, Liaoning 125100 China

²Quanzhou Institute of Equipment Manufacturing Haixi Institutes, Chinese Academy of Sciences, Quanzhou, Fujian 362000, China

show less

DOI: 10.3788/LOP57.182802 Cite this Article Set citation alerts

Jinguang Sun, Yanbei Li, Xian Wei, Wanli Wang. Hyperspectral Image Classification Combined with Convolutional Neural Network and Sparse Coding[J]. Laser & Optoelectronics Progress, 2020, 57(18): 182802 Copy Citation Text

show less

Fig. 1. Structure diagram of residual module

Download full size

Fig. 2. Structure of hyperspectral image classification combined with convolutional neural network and sparse coding

Download full size

Fig. 3. Indian Pines dataset and label drawing

Download full size

Fig. 4. Salinas dataset and label diagram

Download full size

Fig. 5. Pavia University dataset and label diagram

Download full size

Fig. 6. Classification of each algorithm in Indian Pines dataset. (a) Lable; (b) RBF-SVM; (c) Se-2D-CNN; (d) 3D-CNN; (e) SSC; (f) SOMP; (g) our algorithm

Download full size

Fig. 7. Classification of each algorithm in Salinas dataset. (a) Lable; (b) RBF-SVM; (c) Se-2D-CNN; (d) 3D-CNN; (e) SSC; (f) SOMP; (g) our algorithm

Download full size

Fig. 8. Classification of each algorithm in Pavia University dataset. (a) Lable; (b) FBF-SVM; (c) Se-2D-CNN; (d) 3D-CNN; (e) SSC; (f) SOMP; (g) our algorithm

Download full size

Label	Class	Sample quantity
Label	Class	Training	Test
1	Alfalfa	9	37
2	Corn-no	285	1143
3	Corn-min	166	664
4	Corn	47	190
5	Grass/pasture	97	386
6	Grass/trees	146	584
7	G/pasture-mo	6	22
8	Hay-win	96	382
9	Oats	4	16
10	Soy-no	194	778
11	Soy-min	491	1964
12	Soy-cle	118	475
13	Wheat	41	164
14	Woods	253	1012
15	BGTD	77	309
16	SST	19	74
Total	-	2049	8200

Table 1. Quantity of training and test samples in Indian Pines dataset

Label	Class	Sample quantity
Label	Class	Training	Test
1	Brocoli-gw-1	402	1607
2	Brocoli-gw-2	745	2981
3	Fallow	395	1581
4	Fallow-rp	279	1115
5	Fallow-sm	536	2142
6	Stubble	792	3167
7	Celery	716	2863
8	Grapes-un	2254	9017
9	Soil-vd	1240	4963
10	CSGW	656	2622
11	Lettuce-ro-4wk	214	854
12	Lettuce-ro-5wk	385	1542
13	Lettuce-ro-6wk	183	733
14	Lettuce-ro-7wk	214	856
15	Vinyard-un	1453	5815
16	Vinyard-vt	361	1446
Total	-	10825	43304

Table 2. Quantity of training and test samples in Salinas dataset

Label	Class	Sample quantity
Label	Class	Training	Test
1	Asphalt	1326	5305
2	Meadows	3730	14919
3	Gravel	420	1679
4	Trees	613	2451
5	Painted-ms	269	1076
6	Bare Soil	1006	4023
7	Bitumen	269	1064
8	Self-b Bricks	736	2946
9	Shadows	189	758
Total	-	8555	34221

Table 3. Quantity of training and test sample in Pavia University dataset

Label	Classification accuracy /%
Label	RBF-SVM	Se-2D-CNN	3D-CNN	SSC	SOMP	Ours
1	49.58	57.72	70.31	55.42	86.88	99.94
2	78.36	88.74	82.02	74.90	93.78	95.38
3	64.60	81.89	89.88	60.91	91.84	97.65
4	60.05	77.42	97.23	48.33	93.48	99.82
5	94.03	95.68	94.56	91.59	92.48	98.33
6	96.09	98.43	99.30	95.54	99.23	99.59
7	61.30	77.56	93.65	24.35	51.74	98.67
8	98.73	95.01	99.99	99.27	99.95	98.56
9	42.22	57.54	54.55	7.78	3.33	99.99
10	60.14	85.02	96.23	64.02	87.22	99.83
11	87.39	88.41	98.97	82.27	97.35	99.78
12	73.06	71.33	93.36	74.26	87.63	98.56
13	99.32	97.91	98.32	99.26	98.26	98.93
14	96.91	97.37	99.67	95.64	98.54	99.33
15	54.30	91.45	97.82	58.98	97.46	99.79
16	90.82	90.30	97.35	91.65	96.00	97.72
OA /%	81.68	88.93	97.24	79.61	94.64	98.94
AA /%	75.43	84.49	91.32	70.26	85.95	98.89
Kappa	0.786	0.873	0.957	0.764	0.938	0.981

Table 4. Classification accuracy of each algorithm in Indian Pines dataset

Label	Classification accuracy /%
Label	RBF-SVM	Se-2D-CNN	3D-CNN	SSC	SOMP	Ours
1	98.54	99.99	98.84	99.68	99.98	99.99
2	99.06	99.99	97.07	99.49	98.78	99.99
3	93.76	99.30	99.97	93.12	93.87	99.92
4	99.22	99.82	99.04	99.41	98.55	99.74
5	97.93	99.84	99.72	98.56	91.74	99.95
6	99.76	99.99	99.74	99.75	99.97	99.99
7	99.53	99.91	99.96	99.73	99.98	99.99
8	88.50	91.00	94.26	77.86	97.40	99.97
9	99.51	99.99	99.96	99.81	99.94	99.99
10	93.92	98.32	98.81	94.71	96.01	99.82
11	95.35	97.65	99.58	96.23	99.47	98.73
12	89.96	99.06	99.56	99.60	91.81	99.74
13	98.28	99.31	99.13	99.35	87.20	99.99
14	94.49	98.66	98.66	94.97	94.22	99.78
15	62.58	76.13	99.36	63.57	67.25	99.96
16	98.24	98.73	99.15	98.54	99.62	99.92
OA /%	91.34	95.95	98.87	89.49	93.48	98.99
AA /%	94.59	97.17	98.93	94.65	94.74	99.78
Kappa	0.903	0.955	0.987	0.882	0.926	0.988

Table 5. Classification accuracy of each algorithm in Salinas dataset

Label	Classification accuracy /%
Label	RBF-SVM	Se-2D-CNN	3D-CNN	SSC	SOMP	Ours
1	81.76	98.13	98.52	84.62	80.39	98.15
2	84.91	83.49	98.91	84.00	92.75	99.78
3	76.61	98.36	98.84	75.55	94.91	99.54
4	96.08	77.85	97.98	94.85	95.26	99.57
5	99.57	99.93	99.98	99.62	99.94	99.80
6	83.16	91.93	98.46	84.86	85.29	99.97
7	89.65	98.30	99.45	78.24	99.27	97.26
8	81.79	92.14	95.76	68.39	87.34	98.21
9	97.99	89.21	96.48	97.98	85.14	98.93
OA /%	85.20	98.82	98.94	83.87	89.21	99.01
AA /%	87.95	92.27	96.61	85.45	91.11	99.05
Kappa	0.810	0.984	0.986	0.793	0.858	0.988

Table 6. Classification accuracy of each algorithm in Pavia University dataset

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information