Neural detection for buried pipe using fully polarimetric GPR

This paper presents an automatic buried pipe detection algorithm applying the two-step Artificial Neural Networks (ANN) into the fully polarimetric ground penetrating radar (GPR). This algorithm searches a target response using neural networks and discriminates the target response of a buried pipe among clutters by analyzing the polarization characteristics. Linearity and orientation of the buried object are estimated by analyzing the fully-polarimetric GPR data. The estimated orientation is utilized to normalize the spatial variation pattern before applying the Step-2 ANN when the scan direction is not perpendicular to the pipe. The detection algorithm then provides location, depth, and orientation of a buried pipe as a result. The detection performance of the detection algorithm was tested by Monte Carlo simulations in the presence of different signal-to-noise and signal-to-clutter ratios. Finally, the developed ANN algorithm was applied to detect drainage pipes from actual GPR data collected at farmlands. The detection algorithm was able to effectively detect pipes buried up to one meter with only few radar scans.