GPR Target Identification Using CNN Based Classifier

In this study, we present the results of multi- class classification of buried objects using a Convolutional Neural Networks (CNN)-based Deep Learning (DL) approach, leading not only to successful target detection but also accurate and effective target identification. Our CNN is trained using a real dataset with 1040 images, which includes Ground Penetrating Radar (GPR) B-Scan images of clutter objects and surrogate mines. While clutter classes consist of stones, cans, bottles, nails and similar clutter objects, the surrogate mine classes consist of metallic and non-metallic anti-personnel and anti-tank surrogate mines. We illustrate the effect of simple but powerful pre-processing techniques on the performance of the DL-based final classifier. The results illustrate that using different pre-processing steps on the raw GPR data, it is possible to improve the multi-class classification performance of the DL model to nearly 100%.