Hyperbolic Pattern Detection in Ground Penetrating Radar Images Using Faster R-CNN

Abstract

This study presents the application of Faster RCNN, a popular Region Based Convolutional Neural Network, for detecting hyperbolic patterns in Ground Penetrating Radar (GPR) images. GPR is an important tool for subsurface imaging in various fields such as geology, archaeology, and engineering. However, the analysis of GPR images can be challenging due to noise, small objects, and variations in object sizes. To evaluate the performance of the proposed method, 369 simulated GPR B-scan images were generated using GprMax simulation software. These images included single, double, and triple hyperbolic patterns. The results showed that preprocessing improved the detection accuracy and led to higher Intersection over Union (IoU) scores. The experimental results demonstrate that Faster R-CNN is an effective tool for hyperbolic pattern detection in GPR images and provides a promising direction for future research in the field.