Mutual Information-Driven Feature Reduction for Hyperspectral Image Classification.

Area of Science:

• Remote Sensing
• Computer Vision
• Data Science

Background:

• Hyperspectral images (HSIs) offer rich spectral information for ground cover analysis.
• The curse of dimensionality, high correlation, and computational costs hinder direct HSI classification.
• Existing methods like Principal Component Analysis (PCA) struggle with noisy data and high computational demands.

Purpose of the Study:

• To develop an efficient feature extraction and selection approach for HSI classification.
• To overcome the limitations of traditional methods like PCA in HSI analysis.
• To improve classification accuracy and reduce computational cost in HSI analysis.

Main Methods:

• Proposed a normalized mutual information (NMI)-based band grouping strategy for feature extraction.
• Applied classical PCA within each band subgroup for intrinsic feature extraction.
• Utilized NMI-based minimum redundancy and maximum relevance (mRMR) criteria for feature selection.
• Classified the reduced feature subspace using a kernel support vector machine (KSVM).

Main Results:

• Achieved high classification accuracies: 94.93% for AVIRIS and 99.026% for HYDICE datasets.
• Demonstrated significant improvement in classification performance compared to existing methods.
• Effectively reduced computational cost associated with HSI classification.

Conclusions:

• The proposed NMI-based feature reduction approach is highly effective for HSI classification.
• The method successfully mitigates the curse of dimensionality and improves classification accuracy.
• This approach offers a cost-effective solution for hyperspectral data analysis.