Research On Underground Target Location And Dielectric Constant Estimation Based On Convolutional Neural Network

The main uses of ground penetrating radar include underground target location,geometry identification,electrical performance parameter estimation and classification.Traditional machine learning method for classification,location and parameter estimation of underground targets,the extraction of the feature of ground penetrating radar signals requires artificial design,which involves the subjective experience of the experimental personnel and brings a lot of uncertainty to the experimental results.Therefore,how to construct the classification,location and parameter estimation framework of underground targets with automatic feature extraction is of great significance.In view of the above problems,this paper proposes method of the shape recognition,location and dielectric constant estimation of underground targets based on Convolutional Neural Networks(CNN).The main contents of this article are:Firstly,a classification method based on CNN for underground target shape is stdudied.For the shortcomings of traditional artificial design feature extraction methods,in view of CNN's advantages of automatic features learning,the CNN framework is used to classify shallow underground target shapes.The Support Vector Machine(SVM)is used to replace the softmax layer,and the CNN-SVM is used to identify the shape of the underground target.Compared with the traditional feature extraction method combined with SVM classification results,CNN-SVM method has higher classification accuracy and precision.Secondly,an underground target location method based on fully convolutional network(FCN)is stdudied.For the defect of the feature extraction method in underground target location,in view of the powerful feature extraction capability of AlexNet network model,fully connected layer of AlexNet which is substituted by convolutional layer is changed to a fully-convolution network,joint improved nonmaximum suppression method for underground target location.Compared with the traditional target extraction method,FCN has better location performance.Finally,a method for estimating the dielectric constant of underground targets based on cascaded CNN is researched.The traditional method of location and parameter estimation of underground targets is carried out independently.In this paper,based on the underground target location of the fully convolutional network,the dielectric constant estimation of the underground target is achieved by cascading a CNN.Compared with the dielectric constant estimation results of a single CNN,the cascaded CNN model has lower estimation error and implements the integration of the location of the underground target and the estimation of the dielectric constant.Therefore,the CNN method of this paper has reference value in the field of subsurface target classification,location and parameter estimation.