A Generalized Non-Convex Surrogated Framework for Anomaly Detection on Blurred Hyperspectral Images

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Hyperspectral imaging offers high spectral resolution for land cover discrimination and anomaly detection.
• Image blurring significantly degrades hyperspectral image quality, complicating anomaly detection due to entangled neighboring pixels.
• Existing hyperspectral anomaly detection models often overlook the impact of blurring effects.

Purpose of the Study:

• To develop a robust hyperspectral anomaly detection method that effectively addresses image blurring.
• To propose a generalized non-convex framework capable of handling blurred hyperspectral data for anomaly detection.
• To enhance the accuracy and reliability of anomaly detection in the presence of blurring artifacts.

Main Methods:

• A generalized non-convex surrogated tensor framework is proposed.
• The framework employs Block Term Decomposition for adaptive spatial and spectral low-rankness.
• It considers uneven multi-linear low-rankness and utilizes non-convex surrogates for tighter prior modeling.

Main Results:

• The proposed framework demonstrates robust performance in anomaly detection on blurred hyperspectral images.
• Experimental results show superiority over state-of-the-art methods in both deblurring and anomaly detection tasks.
• The method effectively models the low-dimensional prior of hyperspectral images even with blurring.

Conclusions:

• The developed framework provides a significant advancement for anomaly detection in blurred hyperspectral imagery.
• It offers a unified approach guaranteeing convergence for various non-convex surrogates.
• The method enhances the practical applicability of hyperspectral anomaly detection in real-world scenarios.