Advanced Buried Threat Detection Using Convolutional Neural Networks and Recurrent Neural Networks

Abstract

Ground Penetrating Data is a remote geophysical sensory system that has been used intensively and
researched. GPR uses radar pulses to image the subsurface and is used to find underground threats and
objects. Existing model detectors use specific models or handcrafted features. To improve on them, algorithms
are being evolved to predict the presence of such threats. Through this paper, we propose using vast and
authentic GPR data collections and methods driven by discrimination algorithms for BTD's that capitalize on
deep convolutional neural networks by combining CNNs and RNNs to investigate two- dimensional groundpenetrating
data (GPR) scans in the x-axis and y-axis cartesian directions as well as 3-dimensional GPR
volumes. The data utilizes a vast collection of BEO's including various shapes, metallic substances, and
underground internment profundities. We also provide a quantitative analysis that compares the different
results found using the algorithms and models on CNN's and conventional learning methods.