Cascade Superpixel Regularized Gabor Feature Fusion for Hyperspectral Image Classification.

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Hyperspectral imaging captures detailed spectral and spatial information.
• Gabor wavelets are effective for extracting spectral-spatial-fused features.
• Existing methods often process Gabor magnitude and phase features separately.

Purpose of the Study:

• To propose a novel Cascade Superpixel Regularized Gabor Feature Fusion (CSRGFF) approach.
• To enhance material classification and recognition in hyperspectral images.
• To effectively fuse Gabor magnitude and phase features for improved performance.

Main Methods:

• Utilizing Gabor filters to extract magnitude and phase features from hyperspectral images.
• Employing a Support Vector Machine (SVM)-based probability representation for feature fusion.
• Applying quadrant bit coding and Hamming distance for Gabor phase feature encoding and similarity measurement.
• Implementing superpixel graphs for progressive regularization of feature weighting cubes.

Main Results:

• The proposed CSRGFF approach demonstrates superior performance compared to state-of-the-art methods.
• Experimental results on four real hyperspectral datasets consistently show the effectiveness of CSRGFF.
• Classification performance improves with decreasing superpixel numbers during regularization.

Conclusions:

• The CSRGFF method offers an effective strategy for spectral-spatial feature fusion in hyperspectral imagery.
• The integrated fusion of Gabor magnitude and phase, regularized by superpixels, significantly boosts classification accuracy.
• This approach provides a robust framework for hyperspectral material classification and recognition.