Weighted Sparseness-Based Anomaly Detection for Hyperspectral Imagery.

Area of Science:

• Remote Sensing
• Image Processing
• Data Science

Background:

• Anomaly detection in hyperspectral remote sensing data is crucial for image analysis.
• Existing low-rank and sparse matrix decomposition methods (LRaSMD) struggle with complex scenes, exhibiting poor detection due to background edges and noise.

Purpose of the Study:

• To propose a novel weighted sparse hyperspectral anomaly detection method.
• To enhance the detection performance in complex hyperspectral scenes with improved noise and background suppression.

Main Methods:

• Reconstructed hyperspectral data into low-rank, sparse, and noise sub-matrices using matrix decomposition.
• Utilized a low-rank background image and a sliding window strategy to build local spectral-spatial dictionaries.
• Introduced a sparse coefficient divergence evaluation index (SCDI) for weighting anomaly maps, suppressing residues and enhancing targets.

Main Results:

• The proposed method effectively suppresses background edges and noise in complex scenes.
• Experimental results on a real-scene hyperspectral dataset show superior detection performance compared to existing algorithms.
• Enhanced anomaly maps were obtained by weighting sparse anomaly maps with SCDI.

Conclusions:

• The weighted sparse hyperspectral anomaly detection method significantly improves detection accuracy in challenging environments.
• The SCDI weighting factor is effective in refining anomaly detection by reducing false positives from background clutter.
• This method offers a robust solution for anomaly detection in hyperspectral image processing.