Greedy Unsupervised Hyperspectral Image Band Selection Method Based on Variable Precision Rough Set

Jing CHEN; Zhenxing ZHANG

doi:10.3788/gzxb20215002.0210004

Journals >Acta Photonica Sinica >Volume 50 >Issue 2 >Page 103 > Article

Acta Photonica Sinica
Vol. 50, Issue 2, 103 (2021)

Greedy Unsupervised Hyperspectral Image Band Selection Method Based on Variable Precision Rough Set

Jing CHEN and Zhenxing ZHANG

Author Affiliations

College of Information and Electrical Engineering， Ludong University， Yantai， Shandong264000， China

show less

DOI: 10.3788/gzxb20215002.0210004 Cite this Article

Jing CHEN, Zhenxing ZHANG. Greedy Unsupervised Hyperspectral Image Band Selection Method Based on Variable Precision Rough Set[J]. Acta Photonica Sinica, 2021, 50(2): 103 Copy Citation Text

show less

Fig. 1. Hyperspectral Botswana image and its reference map

Download full size | View in the Article

Fig. 2. Hyperspectral KSC image and its reference map

Download full size | View in the Article

Fig. 3. Hyperspectral Indian Pine image and its reference map

Download full size | View in the Article

Fig. 4. Comparison of average total classification accuracy

Download full size | View in the Article

Fig. 5. Classification and comparison results of 20 band IP data sets (β = 0.1)

Download full size | View in the Article

Fig. 6. Parameter sensitivity comparison of traditional VPRS

Download full size | View in the Article

Fig. 7. Parameter sensitivity comparison of proposed method

Download full size | View in the Article

Fig. 8. Accuracy of classification results generated by different sample numbers

Download full size | View in the Article

Number of bands	Proposed method（β=0.1）			SVD			ID			WaLuDi			WaLuMI			RRS
Number of bands	OA	KA	STD	OA	KA	STD	OA	KA	STD	OA	KA	STD	OA	KA	STD	OA	KA	STD
5	86.64	0.855	0.577	86.21	0.851	0.488	73.64	0.714	0.329	79.76	0.781	0.861	86.12	0.85	0.425	88.77	0.878	0.481
10	90.73	0.9	0.268	94	0.935	0.293	91.14	0.904	0.192	90.85	0.901	0.407	90.88	0.901	0.32	92.89	0.923	0.227
15	93.31	0.928	0.275	94.64	0.942	0.329	93.2	0.926	0.341	93.15	0.926	0.276	92.19	0.915	0.336	94.67	0.929	0.284
20	94.08	0.936	0.262	95.3	0.949	0.37	94.23	0.938	0.251	94.64	0.942	0.316	93.46	0.929	0.287	95.12	0.936	0.249
25	95.51	0.951	0.322	95.83	0.955	0.247	94.63	0.942	0.249	95.39	0.95	0.281	94.32	0.938	0.355	95.78	0.943	0.226
30	96.02	0.957	0.343	96.72	0.964	0.373	94.97	0.946	0.191	96.18	0.959	0.187	95.02	0.946	0.37	96.95	0.954	0.252

Table 1. Comparison of classification performance of Botswana dataset

View in the Article

Number of bands	Proposed method（β=0.1）			SVD			ID			WaLuDi			WaLuMI			RRS
Number of bands	OA	KA	STD	OA	KA	STD	OA	KA	STD	OA	KA	STD	OA	KA	STD	OA	KA	STD
5	84.91	0.831	0.241	58.25	0.53	0.958	52.98	0.465	0.383	81.96	0.798	0.505	81.38	0.791	0.216	80.77	0.785	0.322
10	88.97	0.877	0.205	66.08	0.618	0.541	57.25	0.515	0.459	85.84	0.842	0.295	90.04	0.889	0.442	90.31	0.892	0.172
15	91.74	0.908	0.198	69.45	0.657	0.498	61.26	0.561	0.42	89.54	0.883	0.359	92.49	0.916	0.307	92.76	0.919	0.269
20	92.99	0.922	0.275	70.94	0.674	0.389	63.67	0.589	0.466	90.2	0.891	0.287	93.32	0.926	0.183	93.48	0.927	0.273
25	94.12	0.934	0.14	72.66	0.693	0.287	68.57	0.646	0.341	90.97	0.899	0.439	93.9	0.932	0.253	94.11	0.934	0.306
30	94.44	0.938	0.099	76.54	0.737	0.505	71.17	0.676	0.414	92.2	0.911	0.316	94.51	0.939	0.245	94.44	0.938	0.308

Table 2. Comparison of classification performance of KSC dataset

View in the Article

Number of bands	Proposed method（β=0.1）			SVD			ID			WaLuDi			WaLuMI			RRS
Number of bands	OA	KA	STD	OA	KA	STD	OA	KA	STD	OA	KA	STD	OA	KA	STD	OA	KA	STD
5	69.61	0.649	0.272	69.88	0.65	0.517	52.46	0.443	0.416	60.59	0.535	0.356	60.9	0.541	0.341	67.03	0.616	0.496
10	80.78	0.78	0.183	80.16	0.773	0.2	55.87	0.483	0.651	75.15	0.714	0.216	77.85	0.745	0.221	75.19	0.714	0.236
15	83.92	0.816	0.237	84.62	0.824	0.222	58.71	0.518	0.584	84.06	0.818	0.232	80.24	0.773	0.17	80	0.771	0.256
20	86.38	0.844	0.242	86.24	0.843	0.219	62.31	0.557	0.465	85.78	0.837	0.223	85.39	0.833	0.206	83.17	0.807	0.408
25	87.42	0.856	0.192	86.92	0.851	0.193	64.93	0.593	0.381	87.75	0.86	0.206	86.44	0.845	0.173	84.83	0.826	0.229
30	88.7	0.871	0.252	87.93	0.862	0.305	69.24	0.647	0.205	88.73	0.871	0.168	87.69	0.859	0.259	86.65	0.847	0.275

Table 3. Comparison of classification performance of Indian pine dataset

View in the Article

Datasets	SVD	ID	WaLuDi	WaLuMI	RRS
Botswana	3 290.1	1 228.4	175.44	4 675.3	437.5
KSC	6 425.8	12 547.8	2 445.65	131.5	9.67
IP	2 265.6	135 636.3	136.05	19 661.1	19 500

Table 4. Comparison of |z| scores obtained

View in the Article

Datasets	SVD	ID	WaLuDi	WaLuMI	RRS	VPRS(β=0.1)
Botswana	357.9	656.2	647.8	700.4	514.9	515.8
KSC	316.6	542.3	540.7	585.9	433.6	430.7
IP	21.8	36.9	36.8	39.8	29.5	29.3

Table 5. Comparison of calculation time

Jing CHEN, Zhenxing ZHANG. Greedy Unsupervised Hyperspectral Image Band Selection Method Based on Variable Precision Rough Set[J]. Acta Photonica Sinica, 2021, 50(2): 103

Download Citation

Tools

Save the article for my favorites

Paper Information