Region-Aware Hierarchical Latent Feature Representation Learning-Guided Clustering for Hyperspectral Band Selection

Area of Science:

• Remote Sensing
• Computer Vision
• Data Science

Background:

• Hyperspectral band selection is crucial for reducing data redundancy in hyperspectral images (HSIs).
• Existing clustering-based methods often overlook spatial information and regional importance, leading to suboptimal band selection.
• Pixelwise feature extraction fails to capture the complex spatial-spectral characteristics of HSIs.

Purpose of the Study:

• To propose a novel region-aware hierarchical latent feature representation learning-guided clustering (HLFC) method for improved hyperspectral band selection.
• To address the limitations of existing methods by incorporating spatial information and regional importance.
• To achieve superior performance in identifying optimal band subsets for HSIs.

Main Methods:

• Utilized superpixel segmentation to preserve spatial information and segment HSIs into meaningful regions.
• Constructed similarity graphs and generated Laplacian matrices for hierarchical learning of low-dimensional latent features within each region.
• Fused latent features to create a unified representation and employed k-means clustering to select bands with maximum information entropy.

Main Results:

• The proposed HLFC method demonstrated superior performance compared to state-of-the-art band selection techniques.
• Experimental results validated the effectiveness of the region-aware and hierarchical feature learning approach.
• The method successfully identified optimal band subsets, enhancing the analysis of HSIs.

Conclusions:

• The HLFC method offers a significant advancement in hyperspectral band selection by effectively integrating spatial information and hierarchical feature learning.
• The approach overcomes the limitations of traditional pixelwise methods, leading to more accurate and representative band subsets.
• The proposed method provides a robust framework for analyzing complex hyperspectral data.