Spectral-Spatial Feature Extraction of Hyperspectral Images Based on Propagation Filter.

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Hyperspectral image (HSI) feature extraction using image filtering faces challenges like limited spatial resolution and complex feature distributions.
• Existing methods struggle with cross-region pixel mixing and spectral discriminative reduction after filtering.

Purpose of the Study:

• To propose a novel spectral-spatial propagation filter (PF) for effective HSI feature extraction.
• To address the limitations of existing filtering methods in handling cross-region mixing and spectral feature degradation.

Main Methods:

• Applied Principal Component Analysis (PCA) for initial HSI dimensionality reduction.
• Developed and applied a spectral-spatial propagation filter (PF) to principal components.
• Utilized a filter template within the PF to manage weights of cross-region mixed pixels.

Main Results:

• The proposed PF method effectively extracts spectral-spatial features from HSIs.
• Experimental validation using Support Vector Machine (SVM) classifiers showed significant improvements in HSI classification accuracy.
• The PF successfully mitigated issues related to cross-region pixel mixing.

Conclusions:

• The spectral-spatial propagation filter (PF) offers a robust solution for HSI feature extraction.
• The method demonstrates superior performance in enhancing HSI classification accuracy compared to existing approaches.