Maximum Likelihood Estimation-Based Joint Sparse Representation for the Classification of Hyperspectral Remote...

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Joint Sparse Representation (JSR) is effective for Hyperspectral Image Classification (HSI).
• Traditional JSR is sensitive to outliers in HSI spatial neighborhoods, compromising robustness.
• There is a need for more robust JSR methods to handle noisy HSI data.

Purpose of the Study:

• To propose a Maximum Likelihood Estimation-based Joint Sparse Representation (MLEJSR) model.
• To improve the robustness of JSR against outliers and noise in HSI classification.
• To provide theoretical and empirical validation of the proposed MLEJSR method.

Main Methods:

• Developed an MLEJSR model replacing the quadratic loss function with an MLE-like estimator for joint approximation error.
• Converted the MLEJSR model to an iteratively reweighted JSR problem using priors on coding residuals.
• Employed a weight function to mitigate the impact of inhomogeneous pixels and outliers.

Main Results:

• Theoretical analysis demonstrated MLEJSR's effectiveness in terms of recovery error.
• Empirical evaluation on three public HSI datasets confirmed MLEJSR's superior performance.
• The MLEJSR method showed significant robustness, particularly in high-noise environments.

Conclusions:

• The proposed MLEJSR method significantly enhances the robustness of hyperspectral image classification.
• MLEJSR effectively reduces the influence of outliers and noise compared to traditional JSR.
• This approach offers a promising solution for accurate HSI classification under challenging conditions.