Interband Consistency-Driven Structural Subspace Clustering for Unsupervised Hyperspectral Band Selection.

Area of Science:

• Remote Sensing
• Computer Vision
• Data Science

Background:

• Band selection is vital for hyperspectral image classification, addressing high dimensionality and preserving spectral information.
• Existing clustering methods often ignore physical properties, leading to suboptimal band combinations due to noise and redundancy.

Purpose of the Study:

• To propose a novel Interband Consistency-Constrained Structural Subspace Clustering (ICC-SSC) method for hyperspectral image band selection.
• To leverage the inherent low-dimensional subspace structure and physical consistency of spectral characteristics in land cover analysis.

Main Methods:

• Employed the l1,2 norm in a self-representation model to identify physically informative basis bands.
• Incorporated total variance (TV) regularization to ensure coherence and smoothing characteristics between adjacent bands.
• Developed an efficient algorithm using the Alternating Direction Method of Multipliers (ADMM) for model optimization.

Main Results:

• The proposed ICC-SSC method effectively discovers inherent grouping structures among hyperspectral bands.
• Physics-based constraints enhance the consistency of subspace representations across all bands.
• Experimental results on three real hyperspectral datasets show significant performance improvements over state-of-the-art methods.

Conclusions:

• ICC-SSC offers a physically-informed approach to band selection in hyperspectral imaging.
• The method successfully mitigates the curse of dimensionality while retaining crucial spectral features.
• ICC-SSC demonstrates superior performance in hyperspectral image classification tasks.