一种新颖的探地雷达快速正演模拟及埋地目标探测机器学习方法

A novel method for fast forward simulation of ground penetrating radar and buried target detection based on machine learning

探地雷达正演模拟在真实雷达数据解译及全波形反演中扮演着重要的角色,针对传统探地雷达(Ground Penetrating Radar,GPR)正演模拟计算量巨大、耗时、不利于实时探测等问题,提出一种基于机器学习框架的近实时GPR正演模拟方法.以混凝土中的钢筋探测作为GPR应用场景,混凝土的含水量、钢筋半径及埋地深度作为模型参数,利用时域有限差分数值模拟散射回波信号;运用主成分分析对回波数据进行降维处理得到相应的主成分权值系数,并以此作为机器学习网络的输出;设计了一种基于随机森林的多层循环网络架构和学习策略,不仅充分挖掘学习模型参数和主成分权值系数之间的内在因果关系,也共享主成分间的相互联系,并具有对每个预测主成分完善和修正的功能,以此实现基于机器学习的探地雷达快速正演模拟,与传统机器学习相比,有效提高了正演模拟的精度.在此基础上将两个深度神经网络与随机森林相结合,以回波数据主成分系数为输入,建立了基于机器学习的场景参数预测模型,实现了近实时的埋地目标探测,预测的混凝土含水量最大误差为2%,钢筋埋地深度最大误差为6.7%.

Ground penetrating radar forward simulation plays an important role in real radar data interpretation and full waveform inversion.A near real-time GPR forward simulation method based on machine learning framework is proposed for the problems of huge computation,time-consuming and unfavorable to timely detection in traditional ground-penetrating radar forward simulation.Using the detection of rebar in concrete as a GPR application scenario,the water content of concrete,the radius and the burial depth of the rebar as model parameters,and numerically simulating the scattered echo signal using FDTD.Principal component analysis is applied to reduce the dimensionality of the echo data to obtain the corresponding principal component weight coefficients,which are used as the output of the machine learning network.A multilayer recurrent network architecture and learning strategy based on random forest is designed,which not only fully explores the intrinsic causal relationship between learning model parameters and principal component weight coefficients,but also shares the interconnection between principal components,and has the function of refining and correcting for each predicted principal component,so as to realize the fast forward simulation of ground-penetrating radar based on machine learning,which effectively improves the accuracy of forward simulation compared with traditional machine learning.On this basis two deep neural networks are combined with random forest to establish a machine learning-based scene parameter prediction model with the principal component weight coefficients of echo data as input,which achieves near real-time buried target detection with a maximum error of 2%for the predicted concrete water content and 6.7%for the maximum error of the buried depth of the rebar.