Efficient Underground Object Detection for Ground Penetrating Radar Signals

摘要

Ground penetrating radar (GPR) is one of the common sensor system for underground inspection. GPR emits electromagnetic waves which can pass through objects. The reflecting waves are recorded and digitised, and then, the B-scan images are formed. According to the properties of scanning object, GPR creates higher or lower intensity values on the object regions. Thus, these changes in signal represent the properties of scanning object. This paper proposes a 3-step method to detect and discriminate landmines: n-row average-subtraction (NRAS); Min-max normalisation; and image scaling. Proposed method has been tested using 3 common algorithms from the literature. According to the results, it has increased object detection ratio and positive object discrimination (POD) significantly. For artificial neural networks (ANN), POD has increased from 77.4 per cent to 87.7 per cent. And, it has increased from 37.8 per cent to 80.2 per cent, for support vector machines (SVM).