Discovering Diverse Subset for Unsupervised Hyperspectral Band Selection.

Area of Science:

• Remote Sensing
• Computer Vision
• Data Science

Background:

• Hyperspectral band selection is crucial for reducing computational complexity and improving performance in hyperspectral data analysis.
• Existing methods struggle to effectively model complex relationships between spectral bands and efficiently search for optimal subsets.
• Key challenges include capturing high-dimensional spectral band interdependencies and developing robust, adaptable selection strategies.

Purpose of the Study:

• To propose a novel multigraph determinantal point process (MDPP) model for hyperspectral band selection.
• To address the limitations of existing methods in modeling spectral band relationships and optimizing subset selection.
• To provide an efficient and robust solution for selecting discriminative bands in hyperspectral applications.

Main Methods:

• Developed a multigraph determinantal point process (MDPP) model to represent relationships between hyperspectral bands.
• Designed multiple graphs to capture intrinsic spectral band dependencies.
• Utilized a mixture determinantal point process (DPP) for modeling multiple dependencies and enabling efficient band selection.

Main Results:

• The proposed MDPP model effectively captures the underlying structure of hyperspectral data.
• The mixture DPP provides an efficient search strategy for optimal band subsets.
• Experimental validation on hyperspectral classification, anomaly detection, and target detection demonstrates superior performance.

Conclusions:

• The MDPP model offers a powerful and flexible framework for hyperspectral band selection.
• The method demonstrates reliability and effectiveness across various hyperspectral tasks and datasets.
• This approach enhances the efficiency and accuracy of hyperspectral data analysis.