Progressive Line Processing of Kernel RX Anomaly Detection Algorithm for Hyperspectral Imagery.

Area of Science:

• Computer Science
• Signal Processing
• Remote Sensing

Background:

• Kernel-RX detector (KRXD) is valuable for hyperspectral image processing due to its use of nonlinear information.
• The kernelization process in KRXD often results in inefficient execution, limiting its practical application.
• Hyperspectral data acquisition typically follows a line-by-line pattern, suggesting a processing approach aligned with this structure.

Purpose of the Study:

• To develop an efficient progressive line processing method for the Kernel-RX detector (PLP-KRXD).
• To address the computational inefficiency associated with kernelizing hyperspectral data in KRXD.
• To enable real-time or near-real-time hyperspectral image analysis using an optimized KRXD algorithm.

Main Methods:

• The proposed Progressive Line Processing for KRXD (PLP-KRXD) processes hyperspectral data line by line.
• Parallel causal sliding windows are implemented to maintain data causality during processing.
• The Woodbury matrix identity and matrix inversion lemma are employed for recursive updates of kernel matrices, minimizing redundant calculations.

Main Results:

• PLP-KRXD significantly reduces the computational complexity of KRXD.
• The recursive update mechanism avoids extensive repetitive matrix calculations, enhancing efficiency.
• Experimental results on three hyperspectral datasets demonstrate the effectiveness and usefulness of PLP-KRXD.

Conclusions:

• PLP-KRXD offers a computationally efficient solution for hyperspectral image analysis using the Kernel-RX detector.
• The line-by-line processing approach aligns with typical hyperspectral data acquisition patterns.
• This method enhances the practicality and applicability of KRXD in real-world scenarios.