Group Sparse Representation Based on Nonlocal Spatial and Local Spectral Similarity for Hyperspectral Imagery...

Area of Science:

• Remote Sensing
• Computer Vision
• Data Science

Background:

• Hyperspectral imagery analysis is crucial for remote sensing applications.
• Traditional classification methods often overlook nonlocal spatial similarities.
• Existing nonlocal methods may not fully exploit spectral information.

Purpose of the Study:

• To develop a novel spectral-spatial classification framework for hyperspectral imagery.
• To effectively integrate nonlocal spatial and local spectral similarities.
• To enhance classification accuracy by fusing diverse information sources.

Main Methods:

• Exploiting nonlocal spatial similarities through non-overlapped patch searching.
• Analyzing spectral similarity locally within discovered patches.
• Applying group sparse representation (GSR) with a group structured prior for classification.

Main Results:

• The proposed method demonstrates significant improvements in classification accuracy on three real hyperspectral datasets.
• The fusion of nonlocal and local information outperforms methods relying on only one type of similarity.
• Experimental results validate the efficiency of the coupled spectral-spatial approach.

Conclusions:

• The proposed framework effectively couples nonlocal spatial and local spectral similarities for hyperspectral image classification.
• Integrating nonlocal and local information provides a more robust and accurate classification.
• This approach offers a promising direction for advancing hyperspectral data analysis.