Automatic detection of reflexion hyperbolas in gpr data with neural networks

In order to locate cylindrical objects like pipes and cables buried underground using ground penetrating radar it is necessary to detect reflexion hyperbolas in the measured radargrams. In practice, this task is in many cases complicated due to different geological environments, incomplete or disturbed hyperbolas, and first of all the fact that nearby objects lead to hyperbolas interfering with each other. In this paper we present an automatic detection system based on a specially connected neural network using receptive fields. We show that with an adequate definition of training data the system is capable of reliably detecting reflexion hyperbolas even in those challenging situations.