Prior-Based Tensor Approximation for Anomaly Detection in Hyperspectral Imagery.

Area of Science:

• Remote Sensing
• Computer Vision
• Data Science

Background:

• Distinguishing anomalies from complex backgrounds is crucial in hyperspectral anomaly detection.
• Hyperspectral imagery (HSI) presents unique challenges due to its high dimensionality and complex spectral-spatial information.

Purpose of the Study:

• To propose a novel prior-based tensor approximation (PTA) method for enhanced hyperspectral anomaly detection.
• To effectively differentiate weak anomalies from intricate background signatures in HSI data.

Main Methods:

• Representing HSI as a third-order tensor integrating spectral and spatial information.
• Decomposing the HSI tensor into background and anomaly tensors using convex optimization.
• Incorporating low-rank prior (spectral dimension) and piecewise-smooth prior (spatial dimension) for background tensor.
• Applying spatial group sparse prior (l2,1-norm regularization) for the anomaly tensor.

Main Results:

• The proposed PTA method integrates multiple priors into a unified convex framework.
• Experimental validation on real hyperspectral datasets confirms the algorithm's effectiveness.
• The PTA method demonstrates superior performance compared to existing state-of-the-art anomaly detection techniques.

Conclusions:

• The developed prior-based tensor approximation method offers a robust solution for hyperspectral anomaly detection.
• The PTA method successfully addresses the challenge of detecting weak anomalies in complex background scenarios.
• This approach advances the field of hyperspectral data analysis and anomaly identification.