Review of Machine Learning Algorithms for Automatic Detection of Underground Objects in GPR Images

Abstract

Ground-penetrating radar (GPR) is a nondestructive tool that has gained popularity after giving promising results in different areas—such as utility engineering, transportation engineering, civil engineering, and geology—with relatively low cost. Even as the number of applications for GPR increases, the interpretation of GPR data is still challenging, in part due to varying ground conditions. Researchers are continuously working on the development of new analysis methods to address these challenges. Computer vision algorithms, including neural networks and convolution neural networks, have advanced significantly over the past decade, and researchers have utilized these algorithms to extract information from GPR images and thus improve the interpretation of GPR data. This paper presents a review of literature that employs computer vision and machine learning algorithms, such as YOLO V3, Viola–Jones, and AlexNet, for automatic extraction of information from GPR images. The uptake in the use of automatic detection algorithms for GPR is increased by the ability to rapidly quantify and locate buried targets that previously could only be identified by professionals with a high level of expertise and training.