Matched Shrunken Cone Detector (MSCD): Bayesian Derivations and Case Studies for Hyperspectral Target Detection.

Area of Science:

• Remote Sensing
• Signal Processing
• Computer Vision

Background:

• Hyperspectral images (HSIs) have non-negative properties suitable for cone-based modeling.
• Cone-based methods are underutilized in hyperspectral target detection.

Purpose of the Study:

• Propose a novel regularized cone-based representation for hyperspectral target detection.
• Introduce the Matched Shrunken Cone Detector (MSCD) and its variants.
• Provide Bayesian derivations for the MSCD approach.

Main Methods:

• Developed a regularized cone-based representation incorporating l2-norm and l1-norm.
• Derived the MSCD from Bayesian principles using half-Gaussian and half-Laplace priors.
• Evaluated MSCD against subspace and sparse representation methods.

Main Results:

• The proposed MSCD outperformed existing subspace and sparse representation methods.
• MSCD demonstrated effectiveness in detecting both sub-pixel and full-pixel targets.
• Experimental results validated the competitiveness of regularized cone-based representation.

Conclusions:

• The regularized cone-based representation offers a competitive approach to hyperspectral target detection.
• MSCD provides a principled and effective solution for HSI target detection.
• The Bayesian derivation offers theoretical grounding for the proposed method.