High-Resolution Hyperspectral Image Classification Based on Hybrid Convolutional Network

Bingzhi Shen; Ruomei Nie; Haipeng Jiang; Zhishuai Yang; Mingrui Song; Siqi Chen; Xinwei Li

doi:10.3788/LOP202259.2428006

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 24 >Page 2428006 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 24, 2428006 (2022)

High-Resolution Hyperspectral Image Classification Based on Hybrid Convolutional Network

Bingzhi Shen, Ruomei Nie, Haipeng Jiang, Zhishuai Yang, Mingrui Song, Siqi Chen, and Xinwei Li^*

Author Affiliations

College of Science, Beijing Forestry University, Beijing 100083, China

show less

DOI: 10.3788/LOP202259.2428006 Cite this Article Set citation alerts

Bingzhi Shen, Ruomei Nie, Haipeng Jiang, Zhishuai Yang, Mingrui Song, Siqi Chen, Xinwei Li. High-Resolution Hyperspectral Image Classification Based on Hybrid Convolutional Network[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2428006 Copy Citation Text

show less

Fig. 1. Schematic of lightweight multi-scale pyramid hybrid pooling hybrid convolution network

Download full size

Fig. 2. Classification results of different classification methods on WHU-Hi-LongKou dataset

Download full size

Fig. 3. Classification results of different classification methods on WHU-Hi-HanChuan dataset

Download full size

Fig. 4. Classification results of different classification methods on WHU-Hi-HongHu dataset

Download full size

No.	Class name	Number of training samples	Number of test samples	No.	Class name	Number of training samples	Number of test samples
C1	Corn	345	34166	C6	Rice	119	11735
C2	Cotton	84	8290	C7	Water	670	66386
C3	Sesame	30	3001	C8	Roads and houses	71	7053
C4	Broad-leaf soybean	632	62580	C9	Mixed weed	52	5177
C5	Narrow-leaf soybean	42	4109

Table 1. Class information of the WHU-Hi-LongKou dataset

No.	Class name	Number of training samples	Number of test samples	No.	Class name	Number of training samples	Number of test samples
C1	Strawberry	895	43840	C9	Grass	189	9280
C2	Cowpea	455	22298	C10	Red roof	210	10306
C3	Soybean	206	10081	C11	Gray roof	338	16593
C4	Sorghum	107	5246	C12	Plastic	74	3605
C5	Water spinach	24	1176	C13	Bare soil	182	8934
C6	Watermelon	91	4442	C14	Road	371	18189
C7	Greens	118	5785	C15	Bright object	23	1113
C8	Trees	359	17619	C16	Water	1508	73893

Table 2. Class information of the WHU-Hi-HanChuan dataset

No.	Class name	Number of training samples	Number of test samples	No.	Class name	Number of training samples	Number of test samples
C1	Red roof	281	13760	C12	Brassica chinensis	179	8775
C2	Road	70	3442	C13	Small Brassica chinensis	450	22057
C3	Bare soil	437	21384	C14	Lactuca sativa	147	7209
C4	Cotton	3265	160020	C15	Celtuce	20	982
C5	Cotton firewood	124	6094	C16	Film covered lettuce	145	7117
C6	Rape	891	43666	C17	Romaine lettuce	60	2950
C7	Chinese cabbage	482	23621	C18	Carrot	64	3153
C8	Pakchoi	81	3973	C19	White radish	174	8538
C9	Cabbage	217	10602	C20	Garlic sprout	70	3416
C10	Tuber mustard	248	12146	C21	Broad bean	27	1301
C11	Brassica parachinensis	220	10795	C22	Tree	81	3959

Table 3. Class information of the WHU-Hi-HongHu dataset

Parameter	Res-2D-CNN	3D-CNN	HybridSN	LHHN
Number of parameters	479114	1299306	1715078	114438

Table 4. Number of parameters of different models

No.	Class	Accuracy /%
No.	Class	Res-2D-CNN	3D-CNN	HybridSN	LHHN
C1	Corn	64.88	71.92	99.79	99.72
C2	Cotton	47.50	50.14	95.55	98.41
C3	Sesame	54.98	0	96.07	99.49
C4	Broad-leaf soybean	86.03	88.94	99.42	99.59
C5	Narrow-leaf soybeans	25.21	33.33	91.58	98.38
C6	Rice	72.07	43.91	99.57	99.84
C7	Water	99.93	99.59	99.91	99.90
C8	Roads and houses	86.35	72.42	95.04	92.53
C9	Mixed weed	50.74	85.53	96.01	96.55
Training time /s		171.49	837.99	55.17	29.37
Predicting time /s		247.11	736.40	22.27	25.10
OA /%		84.96	79.13	99.03	99.12
Kappa /%		80.00	73.00	98.72	99.33
AA /%		55.76	45.26	96.03	97.76

Table 5. Classification accuracy of different methods on WHU-Hi-LongKou dataset(1% samples)

No.	Class	Accuracy /%
No.	Class	Res-2D-CNN	3D-CNN	HybridSN	LHHN
C1	Strawberry	89.21	94.59	96.84	98.02
C2	Cowpea	66.01	90.05	98.41	98.72
C3	Soybean	58.19	60.95	97.69	97.27
C4	Sorghum	56.86	72.56	99.64	98.94
C5	Water spinach	24.49	83.75	88.98	97.78
C6	Watermelon	22.29	61.15	94.19	94.01
C7	Greens	71.51	95.07	92.66	94.03
C8	Trees	74.79	75.56	96.24	98.44
C9	Grass	58.76	71.59	95.75	98.04
C10	Red roof	94.16	97.58	98.51	98.89
C11	Gray roof	89.48	92.62	96.49	98.26
C12	Plastic	30.88	64.54	96.22	95.48
C13	Bare soil	63.09	69.92	94.70	95.51
C14	Road	82.41	82.69	97.00	98.27
C15	Bright object	69.33	61.53	97.33	95.83
C16	Water	99.31	97.61	99.95	99.92
Training time /s		441.68	2124.83	58.83	64.77
Predicting time /s		405.13	1298.87	33.86	35.45
OA /%		82.16	80.06	97.69	98.43
Kappa /%		79.07	76.87	97.30	98.16
AA /%		60.20	65.95	94.35	96.40

Table 6. Classification accuracy of different methods on WHU-Hi-HanChuan dataset(2% samples)

No.	Class	Accuracy /%
No.	Class	Res-2D-CNN	3D-CNN	HybridSN	LHHN
C1	Red roof	95.55	95.38	97.99	98.64
C2	Road	73.52	79.98	90.39	92.61
C3	Bare soil	86.68	88.43	96.73	99.1
C4	Cotton	97.86	99.00	99.57	99.74
C5	Cotton firewood	79.00	82.61	95.38	97.94
C6	Rape	92.32	88.71	99.02	99.57
C7	Chinese cabbage	75.41	84.47	95.38	97.11
C8	Pakchoi	41.85	65.76	94.47	95.94
C9	Cabbage	98.07	97.51	96.58	99.61
C10	Tuber mustard	66.36	80.92	96.97	97.89
C11	Brassica parachinensis	74.98	76.52	96.49	97.19
C12	Brassica chinensis	62.03	71.27	93.41	97.17
C13	Small Brassica chinensis	65.39	75.41	96.77	97.30
C14	Lactuca sativa	81.88	88.40	97.56	99.44
C15	Celtuce	61.57	98.54	97.57	65.91
C16	Film covered lettuce	88.48	87.45	88.86	98.59
C17	Romaine lettuce	63.77	88.36	80.95	98.53
C18	Carrot	60.85	64.01	93.27	96.09
C19	White radish	82.31	73.04	96.49	98.61
C20	Garlic sprout	80.37	61.30	96.81	96.62
C21	Broad bean	2.78	19.54	93.12	97.06
C22	Tree	69.74	66.90	96.12	96.12
Training time /s		674.38	3243.99	122.16	101.07
Predicting time /s		473.05	1519.56	55.54	61.31
OA /%		87.74	84.80	97.70	98.84
Kappa /%		84.47	80.93	97.09	98.53
AA /%		66.53	66.19	93.25	97.35

Table 7. Classification accuracy of different methods on WHU-Hi-HongHu dataset(2% samples)

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information