Method of Shared Nearest Neighbors in Hyperspectral Image Band Selection Considering Local Density

Yuxuan Wang; Xiaobing Sun; Rufang Ti; Honglian Huang; Xiao Liu

doi:10.3788/AOS240843

Acta Optica Sinica, Volume. 44, Issue 24, 2428010(2024)

Method of Shared Nearest Neighbors in Hyperspectral Image Band Selection Considering Local Density

Yuxuan Wang^1,2, Xiaobing Sun^1、*, Rufang Ti¹, Honglian Huang¹, and Xiao Liu¹

¹Key Laboratory of General Optical Calibration and Characterization Technology, Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei 230031, Anhui , China

²University of Science and Technology of China, Hefei 230026, Anhui , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Flowchart of SNBS

Download full size

Fig. 2. Process of optimizing segmentation points based on the Euclidean distances from the subspace center to the remaining bands

Download full size

Fig. 3. Optimization of subspace interval points based on band sharing neighbors

Download full size

Fig. 4. Estimation of single-band noise in the Salinas dataset

Download full size

Fig. 5. Color image and ground truth map on the Indian Pines dataset, with the ground truth transparently overlaid on the color image

Download full size

Fig. 6. Color image and ground truth map on the Salinas dataset

Download full size

Fig. 7. Color image and ground truth map on the Pavia University Scene dataset

Download full size

Fig. 8. Replace different methods for the selection part and calculate the OA values of SVM and KNN on three datasets. (a)(d) Indian Pines dataset; (b)(e) Salinas dataset; (c)(f) Pavia University Scene dataset

Download full size

Fig. 9. OA values of SVM and KNN for both the proposed method and the comparative methods on three datasets. (a)(d) Indian Pines dataset; (b)(e) Salinas dataset; (c)(f) Pavia University Scene dataset.

Download full size

Table 1. OA values for SVM classification using different parameter settings on the Indian Pines dataset

View table

Table 1. OA values for SVM classification using different parameter settings on the Indian Pines dataset

Z	n	Band number
Z	n	5	10	15	20	25	30	35	40	45	50
5	2	62.06	73.71	76.07	80.35	80.38	79.31	79.39	80.46	80.61	80.61
	3	61.15	72.80	75.74	79.29	79.80	78.49	79.78	80.48	78.87	79.07
	4	61.29	72.57	77.59	78.70	79.72	79.07	80.27	79.59	81.03	81.52
10	2	61.61	72.00	76.39	77.88	80.22	78.63	79.85	80.11	80.04	81.42
	3	60.62	72.11	76.10	78.27	79.11	80.10	80.12	79.90	79.55	81.53
	4	58.81	72.30	76.12	77.53	79.70	79.23	80.69	79.26	80.55	81.78
15	2	62.09	75.22	77.38	79.68	79.47	80.19	79.89	80.01	79.58	81.04
	3	61.46	74.06	76.81	79.67	80.26	80.41	81.54	80.51	78.97	79.65
	4	62.59	75.04	76.83	80.14	80.22	80.10	80.75	80.00	80.01	80.23

Table 2. OA values for SVM classification using different parameter settings on the Salinas dataset

View table

Table 2. OA values for SVM classification using different parameter settings on the Salinas dataset

Z	n	Band number
Z	n	5	10	15	20	25	30	35	40	45	50
5	2	88.87	91.36	91.30	92.11	92.25	92.67	92.44	92.66	92.76	92.96
	3	88.80	91.31	92.02	92.27	92.18	92.38	92.23	92.51	92.69	92.89
	4	88.71	91.57	91.78	92.02	92.03	92.26	92.33	92.57	92.73	92.63
10	2	88.83	91.12	91.64	91.94	92.34	92.61	92.13	92.59	92.50	92.83
	3	88.70	90.89	91.68	92.20	92.09	92.39	92.13	92.30	92.48	92.61
	4	88.51	91.23	91.35	91.73	92.18	92.18	92.21	92.48	92.58	92.46
15	2	87.68	91.27	91.73	91.99	92.08	92.40	92.29	92.58	92.72	92.87
	3	87.68	91.12	91.55	91.66	92.39	92.19	92.28	92.44	92.71	92.87
	4	87.17	91.07	91.32	91.84	91.96	92.19	92.26	92.38	92.65	92.80

Table 3. OA values for SVM classification using different parameter settings on the Pavia University Scene dataset

View table

Table 3. OA values for SVM classification using different parameter settings on the Pavia University Scene dataset

Z	n	Band number
Z	n	5	10	15	20	25	30	35	40	45	50
5	2	84.33	86.75	92.70	93.03	92.86	92.39	93.30	93.36	93.16	93.70
	3	84.25	88.49	91.80	92.97	92.97	93.35	93.53	93.35	93.16	93.55
	4	84.81	88.38	92.01	90.17	92.96	93.48	93.17	93.20	93.44	93.64
10	2	84.13	88.05	88.87	92.76	92.73	92.81	93.33	93.48	93.57	93.66
	3	85.16	88.50	91.03	92.54	92.44	92.94	93.61	93.43	93.74	93.64
	4	84.75	88.56	90.81	90.97	93.25	93.20	93.40	93.61	93.58	93.20
15	2	83.51	88.38	89.27	92.70	93.49	92.67	93.05	93.40	93.51	93.72
	3	83.17	87.55	90.86	92.77	92.86	93.39	93.83	93.64	93.52	93.87
	4	82.74	88.80	90.48	91.84	93.16	93.39	93.50	93.37	93.84	93.51

Table 4. Best AOA values for different methods

View table

Table 4. Best AOA values for different methods

Dataset	Method	SVM	KNN
Indian Pines	SNBS	81.02±0.19	69.53±0.13
	DIG	81.51±0.23	69.41±0.15
	ASPS-MN	80.59±0.25	68.88±0.12
	ASPS-EI	80.32±0.34	68.22±0.29
	PIENL	79.59±0.26	68.21±0.17
	UBS	79.46±0.16	67.91±0.11
Salinas	SNBS	88.23±0.14	91.68±0.10
	DIG	87.56±0.11	91.54±0.07
	ASPS-MN	87.34±0.15	91.49±0.11
	ASPS-EI	87.56±0.19	91.24±0.13
	PIENL	87.32±0.17	91.13±0.14
	UBS	87.41±0.13	91.34±0.09
Pavia University Scene	SNBS	92.16±0.21	87.45±0.12
	DIG	91.81±0.13	87.43±0.06
	ASPS-MN	92.02±0.18	86.77±0.08
	ASPS-EI	91.76±0.14	86.75±0.08
	PIENL	91.75±0.15	86.37±0.10
	UBS	91.46±0.09	86.35±0.04

Table 5. Comparison of runtime for different methods when selecting 20 bands
View table
Table 5. Comparison of runtime for different methods when selecting 20 bands
Dataset SNBS DIG APSP-EI APSP-MN PIENL UBS
Indian Pines 0.214 0.205 0.327 0.310 0.412 0.001
Salinas 0.528 0.512 1.118 1.114 1.264 0.001
Pavia University
Scene
0.597 0.576 1.234 1.259 1.347 0.001

Tools

Get Citation

Copy Citation Text

Yuxuan Wang, Xiaobing Sun, Rufang Ti, Honglian Huang, Xiao Liu. Method of Shared Nearest Neighbors in Hyperspectral Image Band Selection Considering Local Density[J]. Acta Optica Sinica, 2024, 44(24): 2428010

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: Apr. 11, 2024

Accepted: Jun. 4, 2024

Published Online: Dec. 17, 2024

The Author Email: Sun Xiaobing (xbsun@aiofm.ac.cn)

DOI:10.3788/AOS240843

CSTR:32393.14.AOS240843

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. OA values for SVM classification using different parameter settings on the Indian Pines dataset

Table 1. OA values for SVM classification using different parameter settings on the Indian Pines dataset

Table 2. OA values for SVM classification using different parameter settings on the Salinas dataset

Table 2. OA values for SVM classification using different parameter settings on the Salinas dataset

Table 3. OA values for SVM classification using different parameter settings on the Pavia University Scene dataset

Table 3. OA values for SVM classification using different parameter settings on the Pavia University Scene dataset

Table 4. Best AOA values for different methods

Table 4. Best AOA values for different methods

Table 5. Comparison of runtime for different methods when selecting 20 bands

Table 5. Comparison of runtime for different methods when selecting 20 bands