Image processing and neural network techniques for automatic detection and interpretation of ground penetrating radar data

Abstract

Ground penetrating radar (GPR) has gained a distinguished place during recent years as a tool for investigating subsurface objects, yet its output is of low resolution, and in need of further processing in order to make its output readily interpretable. Furthermore, GPR data collected in a typical survey is usually in large quantities, and dealing with such quantities manually to produce final interpretations would result in human inaccuracy, and time-consumption problems. In this paper we present an automatic target-detection and localisation system based on unsupervised neural network classifier along with some image processing techniques to extract useful data representing targets (such as pipes, tanks, reinforcement bars, and voids) and discard undesirable data (such as noise and clutter). The neural classifier is capable of returning 3-dimensional images outlining regions of extended targets and pinpointing the location of localised targets such as mines and pipes. This classifier was applied to a variety of GPR data sets gathered from a number of sites, and it achieved rapid and accurate results. Key-Words: Ground penetrating radar, feature extraction, image processing, edge detection, neural networks.