Hyperspectral Image Target Detection Improvement Based on Total Variation.

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Hyperspectral target detection relies on spatial homogeneity, where neighboring pixels share similar characteristics.
• Existing methods often struggle with preserving spatial smoothness while accurately identifying targets, especially small ones.

Purpose of the Study:

• To develop a novel supervised hyperspectral target detection algorithm.
• To effectively utilize spatial homogeneity in hyperspectral images for improved target detection.
• To ensure spectral signature preservation of the target within the detection output.

Main Methods:

• Proposed a supervised algorithm based on Total Variation (TV) to enforce spatial smoothness.
• Incorporated a constraint to guarantee the spectral signature of the target remains unsuppressed.
• Formulated the detection model as an ℓ1-norm convex optimization problem, solved efficiently using the split Bregman algorithm.

Main Results:

• The proposed algorithm demonstrated superior performance compared to other methods on both synthetic and real hyperspectral datasets.
• Achieved successful target detection even when the target occupied only a single pixel.
• The method effectively smoothed the background while allowing for sharp edges in the detection output.

Conclusions:

• The novel TV-based algorithm effectively leverages spatial homogeneity for robust hyperspectral target detection.
• The algorithm's ability to handle single-pixel targets highlights its sensitivity and accuracy.
• This approach offers an efficient and effective solution for hyperspectral target detection, outperforming existing methods.