Research On Detecting Method For Diseases Of Tunnel Lining Based On Ground Penetrating Radar

During the operation of the tunnel,due to the long-term use and the influence of bad weather and environment,diseases such as voids,voids,water leakage,concrete corrosion or spalling and cracking will appear inside the lining,and the existence of these diseases has a serious impact on the safety of the train..Therefore,it is of great significance to accurately identify tunnel diseases and maintain them.Ground Penetrating Radar is a nondestructive testing method that detects the internal shape distribution and characteristics of the medium by transmitting high-frequency electromagnetic waves into the medium to receive the reflected echoes.The detection is fast and efficient,and the results are intuitive.However,in the process of detecting tunnel lining diseases by ground penetrating radar,due to the complex engineering situation and the high threshold of radar technology,the actual detection still has problems such as low efficiency,low precision,and low automation.To solve this problem,this paper applies convolutional neural network to the feature extraction of ground penetrating radar images,which realizes the target detection of tunnel lining diseases to a certain extent.The research content of the full text is mainly divided into the following points:(1)Numerical forward modeling experiments of ground penetrating radar were carried out.The electromagnetic field distribution in two-dimensional space is deduced based on the time-domain finite difference algorithm,and the numerical forward modeling experiment of ground penetrating radar is carried out for some simple underground targets,and the experimental results are analyzed and evaluated.(2)The radar image dataset based on gprMax is given.Three sets of radar image datasets with single target under concrete,multiple targets,and multiple diseases in the tunnel lining are respectively constructed.The complexity of each dataset sample increases in turn.In the radar image dataset with multiple diseases in the tunnel lining,manually Six lining diseases were designed and the effect of reinforcing mesh of different densities was added.The construction of these three datasets fully prepared for the subsequent experiments.(3)A single target recognition algorithm based on the LeNet-5 network is given in the ground penetrating radar image.This paper analyzes and utilizes the characteristics of Convolutional Neural Network(CNN),applies CNN to target recognition in ground penetrating radar images,and designs a CNN structure for target recognition in radar images based on the LeNet-5 network structure.(4)The multi-target recognition and localization algorithm based on FasterRCNN ground penetrating radar image is given.For the radar image data set of multiple targets,the FasterRCNN algorithm using different feature extraction networks and other target detection algorithms are compared and analyzed,and the advantages of the FasterRCNN algorithm are analyzed from the aspects of precision,recall and mAP.(5)The identification and localization algorithm of multiple diseases in tunnel lining based on the improved FasterRCNN ground penetrating radar image is given.Aiming at the complex situation of multiple diseases in the tunnel lining,this paper adds a feature fusion mechanism and online difficult case mining on the basis of the FasterRCNN algorithm,and replaces the ROI Pooling operation of the original algorithm with ROI Align,to a certain extent,the detection effect of FasterRCNN.At the same time,according to the influence of different densities of steel mesh,the algorithm designs a CNN layer to identify the characteristics of steel bars,which is combined with the improved FasterRCNN to further improve the performance of the algorithm.In this paper,starting from the principle of ground penetrating radar,aiming at the numerical forward modeling of ground penetrating radar,various radar data sets are established for studying the diseases of tunnel lining,and the recognition and positioning of each target in the radar image are studied by the method of deep learning.An automatic detection algorithm for tunnel lining disease is proposed,which has certain application value.