A Robust Sparse Representation Model for Hyperspectral Image Classification.

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Sparse representation classification (SRC) has improved hyperspectral image (HSI) classification.
• Existing SRC methods often assume Gaussian noise, which is unrealistic for HSIs.
• HSIs are frequently affected by mixed noise, including Gaussian and sparse noise.

Purpose of the Study:

• To develop a robust HSI classification model that accommodates mixed noise.
• To enhance classification performance by addressing noise limitations in current SRC techniques.

Main Methods:

• A unified framework combining a mixed-noise model with priors on representation coefficients.
• Development of three robust classification methods: SRC, joint SRC, and super-pixel level joint SRC.
• Validation using both simulated and real hyperspectral image data.

Main Results:

• The proposed mixed-noise model significantly improves HSI classification robustness.
• Experimental results demonstrate clear benefits over traditional Gaussian noise-assuming methods.
• The developed methods show effectiveness on diverse datasets.

Conclusions:

• The novel classification framework effectively handles mixed noise in HSIs.
• The proposed methods offer a more realistic and accurate approach to HSI classification.
• This work advances robust feature extraction and classification for HSI data.