Underground Pipeline Identification Reserach Based On Deep Learning And GPR 3D Image

The construction of city and the maintenance of old pipelines require to realize the rapid and accurate detection of underground pipelines.As a nondestructive and fast detection technology,ground penetrating radar(GPR)has been widely used in pipeline detection in recent years.Especially after 2018,vehicle-mounted array ground penetrating radar can easily collect three-dimensional images,which can better reflect the morphological characteristics of underground targets compared with two-dimensional images.This paper mainly uses the means of depth learning to study and realize the fast identification of underground pipelines and detect the position direction of the pipelines according to the 3D image of GPR.This paper studies the process of making the GPR data set.First,the hardware implementation of array GPR is completed and actual 3D data is collected.Considering the problem of insufficient data,the forward simulation is completed,and the mass generation of simulation 3D data is realized.Data augmentation based on the 3D matrix rotation method is proposed.In view of direct wave interference in real data and simulation data,various methods are compared and time threshold interception method is adopted.In view of the noise problem of real data,wavelet threshold denoising method is introduced.This paper focuses on the deep learning neural network classification algorithm for 3D images of GPR,and compares and analyzes three different neural network structures: 2.5D-CNN that considers three-dimensional image as multi-channel color image,three-dimensional convolution neural network(3D-CNN)which directly extracts the spatial features,and the combination of convolution neural network and circular neural network(CNN+RNN).It is proved that the classification effect of 3D-CNN is the best,but the parameter is too much.Based on the depth separable convolution block,the depth separable convolution block with dimension fusion module is proposed.It can ensure good classification effect with reduced parameter and calculation amount.Finally,for the pipeline after classification,the curve fitting method based on Canny operator is adopted in this paper.With the help of different characteristics of horizontal and vertical pipelines,the location and direction of underground pipelines are determined.