Modern GPR Target Recognition Methods

TL;DR

This paper reviews modern GPR target recognition techniques, emphasizing adaptive processing, learning-based methods, sparse representations, advanced classification, and convolutional coding, demonstrated through landmine detection applications.

Contribution

It provides a comprehensive overview of recent advanced GPR processing methods, highlighting their advantages over traditional techniques in target recognition.

Findings

Enhanced detection accuracy in noisy environments

Improved target classification using learning-based methods

Effective landmine detection with sparse and convolutional features

Abstract

Traditional GPR target recognition methods include pre-processing the data by removal of noisy signatures, dewowing (high-pass filtering to remove low-frequency noise), filtering, deconvolution, migration (correction of the effect of survey geometry), and can rely on the simulation of GPR responses. The techniques usually suffer from the loss of information, inability to adapt from prior results, and inefficient performance in the presence of strong clutter and noise. To address these challenges, several advanced processing methods have been developed over the past decade to enhance GPR target recognition. In this chapter, we provide an overview of these modern GPR processing techniques. In particular, we focus on the following methods: adaptive receive processing of range profiles depending on the target environment; adoption of learning-based methods so that the radar utilizes the results from prior measurements; application of methods that exploit the fact that the target scene is sparse in some domain or dictionary; application of advanced classification techniques; and convolutional coding which provides succinct and representatives features of the targets. We describe each of these techniques or their combinations through a representative application of landmine detection.