Online Dictionary Learning Aided Target Recognition In Cognitive GPR

TL;DR

This paper demonstrates that online dictionary learning significantly accelerates target recognition in ground penetration radar, reducing training time and improving detection accuracy for landmines using real data.

Contribution

It introduces an online dictionary learning approach for GPR that decreases learning time and enhances target detection compared to traditional methods.

Findings

Reduces learning time by 94%

Increases clutter detection by 10%

Effective in real GPR data for landmine identification

Abstract

Sparse decomposition of ground penetration radar (GPR) signals facilitates the use of compressed sensing techniques for faster data acquisition and enhanced feature extraction for target classification. In this paper, we investigate the application of an online dictionary learning (ODL) technique in the context of GPR to bring down the learning time as well as improve identification of abandoned anti-personnel landmines. Our experimental results using real data from an L-band GPR for PMN/PMA2, ERA and T72 mines show that ODL reduces learning time by 94\% and increases clutter detection by 10\% over the classical K-SVD algorithm. Moreover, the proposed methodology could be helpful in cognitive operation of the GPR where the system adapts the range sampling based on the learned dictionary.