Hyperspectral-Image Classification Combining Spatial-Spectral Self-Attention and Multigranularity Feature Extraction

Fig. 10. Classification visualization comparison of Trento dataset. (a) Label diagram; (b) SVM; (c) HybridSN; (d) SSRN; (e) FDSSC; (f) DBDA; (g) VIT; (h) morphFormer; (i) MCSSA

Download full size

Fig. 11. Classification visualization comparison of UH dataset. (a) Label diagram; (b) SVM; (c) HybridSN; (d) SSRN; (e) FDSSC; (f) DBDA; (g) VIT; (h) morphFormer; (i) MCSSA

Download full size

Fig. 12. Classification visualization comparison of MUFFL dataset. (a) Label diagram; (b) SVM; (c) HybridSN; (d) SSRN; (e) FDSSC;(f) DBDA; (g) VIT; (h) morphFormer; (i) MCSSA

Download full size

Fig. 13. Classification visualization comparison of IP dataset. (a) Label diagram; (b) SVM; (c) HybridSN; (d) SSRN; (e) FDSSC; (f) DBDA; (g) VIT; (h) morphFormer; (i) MCSSA

Download full size

Table 1. Results of MCSSA ablation comparison experiments on four datasets

View table

Table 1. Results of MCSSA ablation comparison experiments on four datasets

Multi-granularity Conv	Dual-channel DSConv	SSSA	Trento	Indian pines	Houston	MUUFL
√			94.57	95.51	81.83	89.76
√			97.69	99.17	85.43	91.61
√	√	√	97.83	96.95	81.61	91.82
√	√	√	97.80	95.72	83.25	91.26
√	√		98.28	99.19	87.53	93.06
√	√	√	98.30	98.32	87.33	92.68
√	√	√	98.24	98.74	89.10	94.08
√	√	√	98.72	99.32	89.69	94.61

Table 2. Comparative experimental results on Trento dataset

View table

Table 2. Comparative experimental results on Trento dataset

Class No.	SVM	HybridSN	SSRN	FDSSC	DBDA	VIT	morphFormer	Proposed algorithm
1	71.56±4.08	96.33±0.03	99.31±0.49	99.73±0.07	99.81±0.03	91.91±0.93	99.18±0.24	99.25±0.01
2	76.04±2.58	81.67±0.05	75.06±11.83	88.75±6.98	85.49±9.09	74.74±3.22	86.74±1.87	94.52±0.29
3	90.37±6.02	92.50±0.08	91.98±2.35	99.91±0.01	95.58±1.62	94.30±0.73	92.32±4.93	89.03±1.27
4	94.91±0.89	98.33±0.01	98.26±0.89	99.91±0.06	99.95±0.01	99.58±0.05	99.98±0.01	99.99±0.00
5	87.60±2.03	98.67±0.01	99.58±0.17	99.66±0.13	99.52±0.29	98.59±0.31	99.74±0.10	99.85±0.14
6	75.92±2.84	89.83±0.02	91.06±5.53	88.38±9.25	88.18±6.30	85.90±2.36	93.44±2.19	95.93±1.42
OA /%	85.10±1.09	95.49±1.05	94.84±3.94	97.19±1.17	96.75±1.03	94.31±1.26	97.72±0.11	98.72±0.15
AA /%	82.74±1.41	91.21±1.79	92.50±4.41	96.00±0.08	94.75±1.54	90.84±1.46	95.24±0.79	96.43±0.01
Kappa /%	80.18±1.42	93.98±1.39	93.12±5.22	96.25±1.56	95.66±1.37	92.41±1.98	96.96±0.15	98.29±0.20

Table 3. Comparative experimental results on UH dataset

View table

Table 3. Comparative experimental results on UH dataset

Class No.	SVM	HybridSN	SSRN	FDSSC	DBDA	VIT	morphFormer	Proposed algorithm
1	90.32±3.69	91.80±1.47	90.86±5.56	94.77±0.04	88.23±7.76	92.54±1.08	94.89±1.98	95.02±1.52
2	92.16±5.15	82.40±4.84	93.78±0.63	84.51±3.97	81.60±1.39	95.21±0.17	94.63±0.73	95.31±1.43
3	94.15±5.92	92.28±5.67	96.59±0.91	89.87±0.29	100.00±0.00	95.98±0.93	97.57±0.86	98.81±0.14
4	96.11±3.12	92.81±1.33	99.49±0.12	98.96±0.01	98.28±0.65	82.58±4.19	91.36±0.99	92.18±0.61
5	90.86±3.51	98.67±0.01	93.30±4.30	95.76±3.73	97.13±0.24	95.07±0.27	96.34±0.50	99.12±0.30
6	99.72±0.03	85.49±6.45	91.20±1.79	90.00±3.00	86.61±2.68	56.05±13.25	66.77±3.47	75.40±5.12
7	70.08±6.58	87.40±4.45	81.44±8.65	63.63±9.96	79.66±6.67	80.71±3.98	87.95±1.33	94.96±2.30
8	65.70±10.81	81.02±4.43	87.68±1.74	81.42±13.04	90.29±8.88	73.41±6.24	65.54±2.58	78.75±4.11
9	63.95±7.84	66.20±4.92	72.76±7.58	71.45±9.50	75.10±2.17	75.82±5.13	89.31±2.31	82.03±0.82
10	63.24±9.57	81.43±3.44	73.91±7.59	77.76±7.66	72.64±11.37	85.64±2.33	91.80±1.16	91.93±1.54
11	62.29±8.66	72.82±4.52	75.04±7.58	61.14±10.51	72.93±5.37	68.95±3.48	75.58±0.97	78.63±7.75
12	61.41±5.84	68.55±10.92	76.99±9.20	77.14±13.76	79.93±5.96	87.98±1.03	79.95±4.98	89.30±2.29
13	40.76±10.86	83.60±2.32	87.47±1.26	39.06±10.79	93.72±5.11	19.64±15.96	82.28±4.24	78.62±4.88
14	81.42±0.75	97.87±0.04	93.35±4.76	99.54±0.03	98.71±0.18	82.56±4.71	64.91±2.73	81.08±4.11
15	99.27±0.52	95.88±0.49	98.08±1.36	97.64±1.11	98.71±0.19	80.69±7.26	95.01±3.64	99.60±0.25
OA /%	76.74±3.28	81.92±1.07	85.26±3.67	80.86±1.22	83.75±2.83	81.61±1.89	86.40±0.38	89.69±0.57
AA /%	78.10±3.06	84.16±0.91	87.46±1.42	81.58±0.83	87.33±1.23	78.21±1.91	84.90±0.17	88.72±0.94
Kappa /%	74.83±3.56	80.47±1.16	84.06±4.04	84.06±4.04	82.42±3.08	80.10±2.11	85.29±0.41	88.85±0.63

Table 4. Comparative experimental results on MUFFL dataset

View table

Table 4. Comparative experimental results on MUFFL dataset

Class No.	SVM	HybridSN	SSRN	FDSSC	DBDA	VIT	morphFormer	Proposed algorithm
1	93.94±0.58	96.50±0.05	97.26±0.56	96.88±1.05	96.64±0.86	96.34±0.58	97.52±0.09	98.04±0.23
2	70.88±2.10	82.75±0.06	84.30±4.52	85.03±5.71	81.84±2.68	77.12±3.67	87.23±1.05	90.62±1.20
3	73.47±1.74	86.25±1.03	88.07±3.24	85.08±7.19	87.88±3.10	80.81±1.09	92.12±0.28	93.70±0.80
4	73.24±4.96	87.24±0.25	91.56±2.56	87.53±3.82	87.42±3.79	82.59±2.26	93.33±0.45	94.48±0.68
5	84.36±1.06	92.01±0.14	94.22±0.25	92.39±1.32	90.05±1.79	91.32±2.17	94.57±0.62	94.88±0.61
6	93.26±1.68	92.50±0.29	93.46±3.71	78.45±1.13	90.62±5.35	75.39±8.19	85.38±2.95	97.07±1.26
7	67.20±20.71	80.32±0.87	81.88±6.63	81.91±5.13	81.12±5.51	84.29±3.15	87.74±1.36	88.79±1.13
8	92.88±0.46	92.75±0.39	97.24±0.52	94.75±2.18	96.47±1.66	90.38±2.98	97.51±0.38	97.94±0.46
9	60.67±3.58	73.50±0.87	75.23±7.79	85.54±8.97	63.83±6.71	63.52±1.48	51.88±1.89	59.37±3.11
10	81.78±11.10	38.52±13.46	36.43±24.25	10.00±30.00	25.36±21.92	11.49±3.14	21.09±18.62	9.58±3.82
11	95.10±2.70	83.75±13.22	86.77±10.76	97.23±1.19	80.87±6.19	79.60±6.29	82.35±3.37	85.10±6.35
OA /%	85.19±0.18	91.27±1.24	92.94±1.10	92.07±1.13	91.51±0.90	89.13±0.35	93.46±0.07	94.61±0.07
AA /%	80.62±0.15	78.99±1.75	84.22±1.42	81.35±1.28	80.19±1.62	75.71±1.92	80.04±0.55	82.81±0.29
Kappa /%	80.44±0.21	88.47±1.58	90.66±1.37	89.50±2.21	88.77±1.20	85.61±0.45	91.30±0.09	92.86±0.11

Table 5. Comparative experimental results on IP dataset

View table

Table 5. Comparative experimental results on IP dataset

Class No.	SVM	HybridSN	SSRN	FDSSC	DBDA	VIT	morphFormer	Proposed algorithm
1	61.33±23.62	100.00±0.00	100.00±0.00	99.00±0.02	98.13±0.37	89.63±5.55	96.34±2.11	100.00±0.00
2	71.15±1.65	90.12±1.71	98.42±0.56	99.37±0.05	95.10±3.94	93.87±0.93	97.56±0.27	98.92±0.34
3	75.18±2.01	96.82±1.82	98.49±0.48	98.62±0.08	93.63±0.10	96.95±2.49	98.56±0.52	99.43±0.40
4	59.43±8.24	96.35±0.51	96.98±1.23	99.16±0.07	98.31±0.26	81.92±4.15	93.19±2.66	98.75±2.15
5	90.43±1.57	99.12±0.07	98.46±0.24	98.16±0.14	98.27±0.14	89.42±6.12	95.74±0.11	97.74±0.73
6	88.12±1.78	98.31±0.65	99.82±0.02	99.06±0.51	97.76±0.25	98.09±0.43	99.54±0.18	99.87±0.07
7	85.32±9.59	100.00±0.00	100.00±0.00	97.39±0.52	80.84±10.13	80.01±0.32	100.00±0.00	100.00±0.00
8	89.61±1.80	99.86±0.02	97.96±0.39	99.89±0.21	98.78±0.24	99.53±0.16	100.00±0.00	100.00±0.00
9	73.58±6.74	100.00±0.00	100.00±0.00	96.47±0.70	77.63±18.90	73.61±13.81	87.50±10.15	95.58±2.54
10	74.85±2.17	97.97±0.04	93.42±1.00	97.92±1.40	90.72±0.14	86.31±3.13	94.80±1.44	98.78±0.60
11	77.56±0.96	97.68±1.89	98.96±0.16	99.23±0.07	98.74±0.89	98.46±0.72	98.97±0.25	99.72±0.14
12	71.27±3.84	93.15±2.13	98.49±0.45	99.24±0.07	90.65±8.98	85.06±11.84	94.71±1.14	99.35±0.45
13	91.52±3.41	100.00±0.00	100.00±0.00	99.87±0.02	98.59±0.22	97.02±2.71	99.05±0.23	100.00±0.00
14	91.68±0.87	99.07±0.17	98.92±0.38	99.28±0.07	98.57±0.10	98.52±1.49	99.89±0.03	99.97±0.04
15	75.87±2.72	98.69±0.45	94.33±3.88	98.59±0.19	98.56±0.16	82.63±12.13	98.91±0.71	98.93±1.17
16	97.24±2.83	89.42±2.33	94.80±4.21	95.17±0.55	93.79±0.79	94.64±1.97	99.70±0.05	93.98±2.88
OA /%	79.72±0.75	96.15±0.64	98.02±0.20	98.97±0.04	95.73±0.51	94.20±0.38	97.99±0.26	99.32±0.09
AA /%	79.63±2.97	94.28±0.85	98.06±0.18	98.53±0.08	94.26±0.62	90.35±1.01	97.15±1.14	98.57±0.07
Kappa /%	76.75±0.86	95.60±0.31	97.75±0.25	98.83±0.05	95.19±0.57	93.36±0.44	97.70±0.29	99.22±0.03

Tools

Get Citation

Copy Citation Text

Lin Wei, Zhe Chen, Yuping Yin. Hyperspectral-Image Classification Combining Spatial-Spectral Self-Attention and Multigranularity Feature Extraction[J]. Laser & Optoelectronics Progress, 2024, 61(24): 2428002

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: Mar. 6, 2024

Accepted: Apr. 18, 2024

Published Online: Dec. 17, 2024

The Author Email: Zhe Chen (415899149@qq.com)

DOI:10.3788/LOP240832

CSTR:32186.14.LOP240832

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology