Hyperspectral Remote Sensing Classification Using Multi-Scale Adaptive Capsule Network

Gen Zhang; Xiaohui Ding; Ji Yang; Hua Wang

doi:10.3788/LOP202259.2428004

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 24 >Page 2428004 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 24, 2428004 (2022)

Hyperspectral Remote Sensing Classification Using Multi-Scale Adaptive Capsule Network

Gen Zhang^1,2,3, Xiaohui Ding^1,3,*, Ji Yang^1,3, and Hua Wang²

Author Affiliations

¹Guangzhou Institute of Geography, Guangdong Academy of Sciences, Guangzhou 510070, Guangdong, China

²School of Civil and Transportation Engineering, Guangdong University of Technology, Guangzhou 510006, Guangdong, China

³Southern Marine Science and Engineering Guangdong Laboratory (Guangzhou), Guangzhou 511458, Guangdong, China

show less

DOI: 10.3788/LOP202259.2428004 Cite this Article Set citation alerts

Gen Zhang, Xiaohui Ding, Ji Yang, Hua Wang. Hyperspectral Remote Sensing Classification Using Multi-Scale Adaptive Capsule Network[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2428004 Copy Citation Text

show less

Fig. 1. Flowchart of dynamic routing algorithm

Download full size

Fig. 2. MSCaps architecture

Download full size

Fig. 3. Flowchart of adaptive routing algorithm without iteration

Download full size

Fig. 4. Experimental dataset 1. (a) PU dataset; (b) ground truth of PU dataset

Download full size

Fig. 5. Experimental dataset 2. (a) SA dataset; (b) ground truth of SA dataset

Download full size

Fig. 6. Network architectures of MSCNN and MSCaps

Download full size

Fig. 7. Classification accuracy under different input sizes

Download full size

Fig. 8. Classification results of different algorithms on PU dataset

Download full size

Fig. 9. Classification results of different algorithms on SA dataset

Download full size

Fig. 10. Training time of each model on PU and SA datasets

Download full size

Class No.	Land cover	Training	Validation	Test
1	Asphalt	1000	1000	4631
2	Meadows	1000	1001	16650
3	Gravel	460	461	1180
4	Trees	890	891	1285
5	Painted metal sheets	400	401	546
6	Bare Soil	1000	1001	3030
7	Bitumen	400	401	531
8	Self-Blocking Bricks	1000	1001	1683
9	Shadows	260	261	428

Table 1. Number of training, verification, and test samples of various types of objects on PU dataset

Class No.	Land cover	Training	Validation	Test
1	Brocoli_green_weeds_1	100	100	191
2	Corn_senesced_green_weeds	390	390	563
3	Lettuce_romaine_4wk	150	150	316
4	Lettuce_romaine_5wk	470	470	585
5	Lettuce_romaine_6wk	210	210	254
6	Lettuce_romaine_7wk	250	250	299

Table 2. Number of training, verification, and test samples of various types of objects on SA dataset

No.	Accuracy /%
No.	SVM	RF	PCA-SVM	PCA-RF	CNN	CapsNet	MCaps	ARWI-Caps	MSCNN	MSCaps
1	99.21	98.89	99.38	97.56	99.37	98.05	99.09	99.42	99.44	99.49
2	98.99	98.32	98.87	98.24	99.78	99.96	99.93	99.97	99.51	99.97
3	76.32	79.94	76.83	89.77	76.39	95.35	96.19	98.49	98.45	98.31
4	95.91	89.58	95.86	87.10	98.96	92.76	95.46	95.77	99.74	97.15
5	99.85	99.48	99.85	99.85	99.93	96.13	98.82	99.70	99.56	99.85
6	90.12	81.42	89.86	82.98	98.13	98.45	99.41	99.83	99.23	99.84
7	92.38	88.35	92.52	96.38	98.30	99.33	99.25	99.89	98.81	99.92
8	93.39	91.31	93.45	89.68	97.64	97.88	99.48	99.51	98.17	99.62
9	100	99.79	100	99.89	99.58	97.15	97.95	99.68	98.03	99.89
OA /%	91.60	87.38	91.86	88.44	95.84	96.73	98.22	98.96	98.71	99.14
K	0.82	0.74	0.84	0.76	0.91	0.93	0.96	0.97	0.97	0.99
p	$≪ 0.05$	0.0	$≪ 0.05$	$≪ 0.05$	$≪ 0.05$	$≪ 0.05$	$≪ 0.05$	0.011	1.501×10^-5

Table 3. Accuracy, overall accuracy, Kappa coefficient (K), and p-value of different algorithms on PU dataset

No.	Accuracy /%
No.	SVM	RF	PCA-SVM	PCA-RF	CNN	CapsNet	MCaps	ARWI-Caps	MSCNN	MSCaps
1	100	100	100	100	99.74	100	98.73	96.54	100	100
2	100	97.08	99.26	96.99	93.26	100	100	100	100	100
3	99.84	99.83	99.88	100	93.92	100	100	100	100	100
4	99.35	99.09	99.35	98.58	99.08	90.55	99.74	99.87	99.54	99.87
5	87.53	65.31	86.86	71.17	93.48	98.39	89.99	96.70	96.70	100
6	72.70	80.75	73.64	92.26	84.36	87.71	89.92	92.15	88.96	88.88
OA/%	81.57	73.41	82.13	82.21	85.01	87.27	92.16	95.10	94.07	95.38
K	0.66	0.52	0.69	0.67	0.72	0.76	0.85	0.90	0.89	0.91
p	$≪ 0.05$	$≪ 0.05$	$≪ 0.05$	$≪ 0.05$	$≪ 0.05$	$≪ 0.05$	$≪ 0.05$	0.033	0.025

Table 4. Accuracy, overall accuracy, Kappa coefficient (K), and p-value of different algorithms on SA dataset

Gen Zhang, Xiaohui Ding, Ji Yang, Hua Wang. Hyperspectral Remote Sensing Classification Using Multi-Scale Adaptive Capsule Network[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2428004

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享