ANN-based Sub-surface Objects Identification And Inversion Of Dielectric Properties By Means Of GPR

Ground penetrating radar (GPR) is an effective tool to detect buried targets which has been used to fields of military, police, urban construction, construction of highway, railway, airport, water conservation project, tunnel etc. Therefore, buried target parameters inversion technology in GPR has great significance to the development of the national economy. Since the amplitude and delay of receiving signal could correspondingly change due to varying of the position, dimension, material, shape of targets, permittivity of background. In this paper, a new framework integrated image processing with nonlinear parameter inversion is presented to reconstruct their non-linear relationship and implement targets location and parameter inversion. The algorithm system contains the following sections:Firstly, object localization:the extraction the hyperbolic vertex(x0,t0)can be implemented by symmetric algorithm. A single target position and statistics the number of targets can be estimated by symmetric algorithm.Secondly, permittivity of background estimation:electromagnetic wave velocity v is estimated by the combination of different velocity models match algorithm and least squares method evaluation criteria in two state of antenna measuring:antenna affixed to the ground and antenna off the ground to lay the foundation for subsequent target radius estimates.Thirdly, object size estimation:based on parameters (x0,t0) and v, suitable constraint formula will be chosen according to height of radar off the ground. The dimension of cylindrical target and hyperbolic-fitting can be achieved by making use of modified Hough Transform. According to traditional hyperbola mathematical model of object, the accuracy of object hyperbola-fitting is low and error of object inversion is large due to assumption that radar emitter is close to ground or ignorance of height of radar emitter influencing on propagation path. Aimed at these problems, a modified Hough model for object hyperbola is proposed in this paper when radar working off the ground. The simulation result shows presented algorithm has better performance of target inversion accuracy. Finally, object material recognition:different feature extraction techniques and different nonlinear parameter inversion algorithm are analyzed in this paper. The same target echo characteristics are input to the neural network and S VR training, and then a higher inversion accuracy of pattern recognition algorithms will be chosen. Different target echo characteristics are input to the same algorithm for inversion of the target dielectric constant, and then a higher inversion accuracy of characteristics will be selected.