Research On Target Detection And Location Based On Hyperbolic-Shaped Signatures In GPR B-scan Image

Ground Penetrating Radar(GPR)is a popular non-destructive earth exploration technology that uses the reflection and scattering of radar waves emitted by antennas in different underground media to obtain target information.It has a wide range of applications,such as highway and bridge civil engineering quality inspection,underground hydrological quality inspection,urban underground pipeline inspection,glacier geological inspection,archaeology,etc.The working environment that GPR faces is an underground soil medium with an inconstant dielectric constant.The radar waves emitted by GPR cannot keep a straight line during the propagation process,and irregular scattering and refraction will occur,which will bring interference to the echo and put forward higher requirements for subsequent signal processing.However,most studies still use homogeneous media in forward modeling without taking this problem into consideration.In this project,the Peplinski realistic soil model is used to fully approximate the real scene in the modeling.In this paper,the response law of targets with different geometric features in GPR B-scan images is discussed,and the theoretical basis of hyperbolic features of point targets in echo is given.Based on this,the detection of underground targets is transformed into hyperbolic detection in echo images.However,in the real scene,the received two-dimensional echo will be disturbed by many factors,such as the inhomogeneity of underground soil medium,the noise of GPR system itself and the environmental noise.Therefore,this project takes the B-scan echo of the inhomogeneous soil medium as the processing object to realize the algorithm of detection and location of underground targets.Because the distance between the receiving and transmitting antennas of GPR is very close,it cannot be effectively isolated,and the mutual coupling wave between the antennas will be received.In addition,the surface reflection waves will be generated when the electromagnetic waves emitted by GPR pass through the interface of air and soil.In this paper,the average offset method is used to remove the direct wave at first.Then use image processing methods to perform a whole set of processes from threshold segmentation to clustering hyperbola to the image after removing the direct waves.This process makes full use of the characteristics of the GPR B-scan image itself.However,the GPR hyperbola detection method based on OSCA is difficult to deal with the echo data with more interference from background clutter interference.This paper introduces the target detection network Faster RCNN,which has achieved good results in deep learning,into the detection of GPR B-scan images.The improved GPR target location method based on Faster RCNN-OSCA is proposed in this paper,and the rectangular region detected by Faster RCNN is extracted,the hyperbola is fitted,the target position is deduced,and the location is finally completed,which is verified on both measured data and gpr Max simulation data.In addition,due to the influence of antenna ringing effect,the ringing signal will form multiple reflection waves of the target in the B-scan image,which has a high similarity with the target echo features formed by the actual emitted radar waves in the B-scan image,so the correlation information is used to remove multiple reflection waves.