Hyperspectral Anomaly Detection With Tensor Average Rank and Piecewise Smoothness Constraints

Area of Science:

• Remote Sensing
• Computer Vision
• Data Science

Background:

• Anomaly detection in hyperspectral images (HSIs) is crucial for remote sensing.
• Current methods often degrade performance by losing spatial-spectral information.
• Preserving spatial-spectral correlations is key for accurate anomaly identification.

Purpose of the Study:

• To develop a tensor-based anomaly detection algorithm for HSIs.
• To effectively preserve spatial-spectral information lost in traditional methods.
• To improve the accuracy and robustness of anomaly detection in remote-sensing data.

Main Methods:

• Separating HSI data into background and anomaly tensors.
• Utilizing tensor nuclear norm and tensor singular value decomposition (SVD) for background characterization.
• Incorporating total variation (TV) regularization and l_2.1 norm for anomaly component analysis.
• Designing a robust background dictionary using spectral similarity, spatial distance, and SVD-based pixel selection.

Main Results:

• The proposed tensor-based method effectively preserves spatial-spectral information.
• The algorithm demonstrates superior performance in anomaly detection compared to state-of-the-art methods.
• Experimental results on real hyperspectral datasets validate the method's effectiveness.

Conclusions:

• The developed tensor-based algorithm offers a significant advancement in HSI anomaly detection.
• Preserving spatial-spectral information is vital for enhancing detection performance.
• The method provides a robust and effective solution for identifying anomalies in remote-sensing applications.