Tensor decomposition-based sparsity divergence index for hyperspectral anomaly detection.

Area of Science:

• Remote Sensing
• Data Science
• Signal Processing

Background:

• Hyperspectral anomaly detection (AD) increasingly utilizes spatial and spectral features.
• Tensor decomposition-based (TenB) algorithms offer improved AD performance by treating data as a three-order tensor.
• Existing methods show promise but advancements are sought for enhanced accuracy.

Purpose of the Study:

• To propose a novel hyperspectral anomaly detection method named sparsity divergence index based on tensor decomposition (SDI-TD).
• To enhance AD performance by effectively integrating spatial and spectral information through tensor decomposition.
• To evaluate the efficacy of the proposed SDI-TD method against established algorithms.

Main Methods:

• Hyperspectral data is decomposed into three modes using tensor decomposition.
• Low-rank and sparse matrix decomposition is applied along each mode to obtain sparse matrices.
• Sparsity divergence indices (SDIs) are calculated from these sparse matrices.
• A joint SDI is computed to generate the final anomaly detection result.

Main Results:

• The proposed SDI-TD method successfully extracts features from the three modes of hyperspectral data.
• Experimental results on real and synthetic datasets show superior anomaly detection performance of SDI-TD.
• The method effectively identifies anomalies by analyzing sparsity divergence across spatial and spectral dimensions.

Conclusions:

• The SDI-TD method represents a significant advancement in hyperspectral anomaly detection.
• The integration of tensor decomposition and sparsity divergence provides a robust framework for AD.
• The proposed method offers a promising alternative for accurate anomaly identification in hyperspectral imagery.