Multi-Scale Superpixels Dimension Reduction Hyperspectral Image Classification Algorithm Based on Low Rank Sparse...

Area of Science:

• Remote Sensing
• Image Processing
• Machine Learning

Background:

• Hyperspectral images (HSI) suffer from Hughes phenomenon and noise, degrading classification accuracy.
• Effective utilization of spatial-spectral information is crucial for HSI analysis.
• Existing methods often struggle with noise reduction and comprehensive feature extraction.

Purpose of the Study:

• To propose a novel dual feature extraction framework for HSI classification.
• To address noise and Hughes phenomenon in HSI data.
• To fully leverage spatial-spectral joint information for improved classification accuracy.

Main Methods:

• A dual feature extraction framework (LRS-HRFMSuperPCA) combining transform and spatial domain filtering.
• Low-rank structure and sparse representation for HSI noise repair and denoising using Block-Matching 3D.
• Principal Component Analysis (PCA) for dimensionality reduction, multi-scale entropy rate superpixels for segmentation, and hierarchical domain transform recursive filtering.
• Support Vector Machine (SVM) for decision fusion and classification.

Main Results:

• The proposed LRS-HRFMSuperPCA method demonstrated superior performance across three datasets (Indian Pines, University of Pavia, Salinas).
• Quantitative evaluation using Overall Accuracy (OA), Average Accuracy (AA), and Kappa coefficient confirmed the method's effectiveness.
• The framework successfully denoises, reconstructs HSI, and extracts joint spatial-spectral information.

Conclusions:

• The LRS-HRFMSuperPCA framework effectively denoises and reconstructs Hyperspectral images.
• The method fully extracts spatial-spectral joint information, leading to enhanced classification accuracy.
• This approach offers a significant improvement over existing state-of-the-art methods for HSI classification.