Dimensionality Reduction Algorithm for Hyperspectral Image Based on Self-Supervised Learning

Zheng Zhou; Yu Yang; Gan Zhang; Libing Xu; Mingqing Wang; Qibing Zhu

doi:10.3788/LOP231646

Laser & Optoelectronics Progress, Volume. 61, Issue 12, 1237001(2024)

Dimensionality Reduction Algorithm for Hyperspectral Image Based on Self-Supervised Learning

Zheng Zhou^1,3, Yu Yang^2、*, Gan Zhang⁴, Libing Xu³, Mingqing Wang^1,3, and Qibing Zhu²

Author Affiliations

¹Wuxi Nine Cosmos Technology Co., Ltd., Wuxi 214072, Jiangsu, China

²Key Laboratory of Advanced Process Control for Light Industry, Ministry of Education, School of Internet of Things Engineering, Jiangnan University, Wuxi 214122, Jiangsu, China

³Department of Earth System Science, Tsinghua University, Beijing 100084, China

⁴School of Electronic and Electrical Engineering, Zhengzhou University of Science and Technology, Zhengzhou 450052, Henan, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Diagram of HSI data dimension reduction based on SSLFE network

Download full size

Fig. 2. Diagram of external attention module

Download full size

Fig. 3. Performance evaluation flowchart of dimensionality reduction method

Download full size

Fig. 4. Visualization of the three HSI datasets. (a)‒(c) Ground truth of Indian Pines, Salinas, and Pavia University datasets; (d)‒(f) mean reflectance of Indian Pines, Salinas, and Pavia University datasets

Download full size

Fig. 5. Classification accuracy of four dimensionality reduction methods with LSSVM for each class of object

Download full size

Fig. 6. Classification performance of SSLFE combined with ELM on different size datasets

Download full size

Table 1. Classification performance of RF, LSSVM, and ELM on India Pines dataset

View table

Table 1. Classification performance of RF, LSSVM, and ELM on India Pines dataset

Classifier	Index	ICS	FCM-FA	EBM3GP	SSLFE
RF	OA /%	78.20 ± 0.39	78.96 ± 0.82	80.98 ± 0.32	81.35 ± 0.31
	AA /%	66.12 ± 1.09	71.12 ± 0.59	72.99 ± 1.02	74.99 ± 0.92
	KC /%	78.02 ± 0.51	82.58 ± 0.56	84.62 ± 0.36	85.76 ± 0.38
LSSVM	OA /%	85.02 ± 0.69	85.32 ± 0.52	87.50 ± 0.43	88.15 ± 0.33
	AA /%	84.38 ± 1.02	84.62 ± 0.82	86.46 ± 0.93	87.06 ± 0.82
	KC /%	79.98 ± 0.52	85.68 ± 0.73	87.61 ± 0.30	88.21 ± 0.36
ELM	OA /%	80.12 ± 0.59	84.62 ± 0.46	86.58 ± 0.32	87.18 ± 0.62
	AA /%	71.35 ± 1.12	79.95 ± 0.82	82.55 ± 1.12	83.25 ± 0.82
	KC /%	78.08 ± 0.63	82.68 ± 0.69	84.63 ± 0.35	85.13 ± 0.50

Table 2. Classification performance of RF, LSSVM, and ELM on Salinas dataset

View table

Table 2. Classification performance of RF, LSSVM, and ELM on Salinas dataset

Classifier	Index	ICS	FCM-FA	EBM3GP	SSLFE
RF	OA /%	91.30 ± 0.40	92.06 ± 0.58	92.87 ± 0.36	93.25 ± 0.51
	AA /%	94.56 ± 0.46	94.98 ± 0.71	96.09 ± 0.26	96.89 ± 0.52
	KC /%	90.29 ± 0.53	90.92 ± 0.69	91.98 ± 0.28	92.56 ± 0.58
LSSVM	OA /%	93.08 ± 0.79	93.85 ± 0.64	94.93 ± 0.31	95.55 ± 0.33
	AA /%	96.22 ± 0.52	96.98 ± 0.59	97.68 ± 0.27	98.56 ± 0.62
	KC /%	92.46 ± 0.39	93.06 ± 0.53	94.50 ± 0.29	95.81 ± 0.46
ELM	OA /%	90.78 ± 0.62	91.22 ± 0.52	93.42 ± 0.25	94.18 ± 0.52
	AA /%	93.11 ± 0.68	93.83 ± 0.68	96.52 ± 0.28	96.65 ± 0.64
	KC /%	89.88 ± 0.50	90.65 ± 0.81	92.66 ± 0.18	93.13 ± 0.63

Table 3. Classification performance of RF, LSSVM, and ELM on Pavia University dataset

View table

Table 3. Classification performance of RF, LSSVM, and ELM on Pavia University dataset

Classifier	Index	ICS	FCM-FA	EBM3GP	SSLFE
RF	OA /%	90.82 ± 0.23	90.48 ± 0.22	93.28 ± 0.29	94.03 ± 0.51
	AA /%	87.66 ± 0.32	87.22 ± 0.52	91.19 ± 0.19	92.36 ± 0.22
	KC /%	87.68 ± 0.22	87.19 ± 0.29	91.06 ± 0.39	92.36 ± 0.28
LSSVM	OA /%	94.08 ± 0.39	94.65 ± 0.56	95.54 ± 0.25	96.62 ± 0.29
	AA /%	92.38 ± 0.32	93.29 ± 0.62	93.96 ± 0.33	94.69 ± 0.52
	KC /%	92.68 ± 0.28	92.86 ± 0.39	94.23 ± 0.38	95.32 ± 0.53
ELM	OA /%	90.06 ± 0.28	91.56 ± 0.52	92.96 ± 0.36	93.62 ± 0.52
	AA /%	82.12 ± 0.38	83.60 ± 0.64	87.87 ± 0.62	89.92 ± 0.72
	KC /%	86.55 ± 0.46	88.40 ± 0.59	90.55 ± 0.55	91.13 ± 0.63

Table 4. Classification performance of SSLFE and SAMFE with LSSVM on three datasets

View table

Table 4. Classification performance of SSLFE and SAMFE with LSSVM on three datasets

Dataset	Index	SAMFE	SSLFE
Indian Pines	OA /%	92.38 ± 0.59	94.03 ± 0.51
	AA /%	90.09 ± 0.39	92.36 ± 0.22
	KC /%	90.02 ± 0.23	92.36 ± 0.28
Salinas	OA /%	93.14 ± 0.35	96.62 ± 0.29
	AA /%	92.04 ± 0.53	94.69 ± 0.52
	KC /%	92.33 ± 0.68	95.32 ± 0.53
Pavia University	OA /%	91.06 ± 0.26	93.62 ± 0.52
	AA /%	85.57 ± 0.52	89.92 ± 0.72
	KC /%	88.36 ± 0.62	91.13 ± 0.63

Tools

Get Citation

Copy Citation Text

Zheng Zhou, Yu Yang, Gan Zhang, Libing Xu, Mingqing Wang, Qibing Zhu. Dimensionality Reduction Algorithm for Hyperspectral Image Based on Self-Supervised Learning[J]. Laser & Optoelectronics Progress, 2024, 61(12): 1237001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Digital Image Processing

Received: Jul. 3, 2023

Accepted: Aug. 22, 2023

Published Online: Jun. 5, 2024

The Author Email: Yu Yang (yangxiangyu1168@163.com)

DOI:10.3788/LOP231646

CSTR:32186.14.LOP231646

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology