Decision boundaries in two dimensions for target detection in hyperspectral imagery

Area of Science:

• Remote Sensing
• Signal Processing
• Machine Learning

Background:

• Hyperspectral imagery presents challenges for detecting subtle targets like weak plumes and sub-pixel objects.
• Traditional detection methods often struggle with the high dimensionality and noise inherent in hyperspectral data.

Purpose of the Study:

• To develop a novel, robust approach for weak plume and sub-pixel target detection in hyperspectral imagery.
• To leverage a two-dimensional feature space for improved signal detection and machine learning application.

Main Methods:

• A two-dimensional space was constructed using matched-filter projection and residual magnitude.
• Well-established signal detection algorithms were mapped into this 2D space.
• Support Vector Machines (SVM) were employed for non-linear boundary learning within the 2D space.

Main Results:

• The 2D matched-filter-residual space effectively incorporates existing detection algorithms.
• The SVM-based detector demonstrated competitive performance against established methods.
• The approach mitigates the curse of dimensionality for adaptive machine learning in hyperspectral analysis.

Conclusions:

• The proposed 2D approach offers a powerful framework for hyperspectral target detection.
• Machine learning, particularly SVM, can be effectively applied in this reduced dimensional space.
• This method enhances the detection of weak plumes and sub-pixel targets, improving hyperspectral data analysis.