Research On Rapid Image Acquisition System And Disease Intelligent Recognition Algorithm Of Subway Tunnel

Subway is an important part of modern urban traffic.Because of various reasons,the shield subway tunnel will have surface defects such as cracks and leakage.It is very important to detect the surface diseases of the tunnel effectively.With the rapid growth of the operation mileage of subway in China,traditional manual visual inspection and other methods have been difficult to meet the detection needs.It is urgent to have an efficient and intelligent inspection method,which can realize the function of regular inspection and disease marking of subway tunnels.Based on this,this paper proposes a tunnel surface image acquisition system and disease intelligent recognition algorithm based on machine vision and image processing technology.Firstly,the paper designs the overall scheme of the tunnel surface image acquisition system,and expounds the working principle of the system.The image acquisition system can automatically adjust the trigger frequency according to the running speed of 20km/h～ 50km/h to meet the acquisition accuracy of 0.2mm/pixel.The whole system is divided into image acquisition module,synchronous trigger control module,data interaction and storage module,cabinet and wiring module.The parameters calculation and selection and design of hardware equipment are made.After the feasibility test in the laboratory environment,the acquisition system is built on the tunnel comprehensive inspection vehicle,and the overall debugging and acquisition parameter setting of the system are completed,which can capture high-quality images that meet the accuracy requirements.For the intelligent identification of tunnel diseases,this paper studies the intelligent identification algorithm of the disease based on semantic segmentation.Firstly,the tunnel multi-target semantic segmentation data set is established according to the collected images,and the existing semantic segmentation algorithms are used for experiments.Under the FCN network framework,two kinds of backbone networks,Alex Net and VGG16,are used;Under the Deep Lab V3+ network framework,three kinds of backbone networks,Mobile Net,Xception and Res Net-101,are used to compare and analyze the experimental results of different complexity backbone networks under the same network framework.Then,according to the particularity of tunnel specific scene and disease characteristics,a disease segmentation algorithm based on improved Deep Lab V3+network is proposed.The algorithm improves the network structure from three aspects:DCNN backbone network,hole convolution space pyramid module and multi-scale feature fusion.The improved network is compared with the original Deep Lab V3+network.,according to the improvement of accuracy of the experimental results and the intersection ratio data,as well as the image display of the segmentation results of the test set,the effectiveness and better segmentation performance of the disease segmentation algorithm proposed in this paper are illustrated.In order to realize the feature recognition and parameter calculation of disease,the image post-processing algorithm of semantic segmentation is designed.Firstly,the crack image is binarized based on the label value of the semantic segmentation image,and the criterion of eight connected regions of pixels is set.The skeleton extraction of cracks is completed by using the weighted sum mapping look-up table method.Then,according to the actual situation of the disease pixel label in the segmented image,the parameter calculation method of crack length and leakage area is set,and combined with the disease semantic segmentation algorithm,the whole intelligent recognition algorithm of the tunnel surface disease is designed.Finally,the tunnel surface image acquisition system designed in this paper,in the main line of the subway tunnel,according to the different running speed for many experiments,collected a large number of high-quality images in line with the design accuracy,confirmed the feasibility and reliability of the system.The disease intelligent recognition algorithm proposed in this paper,can process the collected image achieve 85.26% average accuracy and 73.62% average intersection and union ratio in offline,and can output semantic segmentation image and disease parameters,which has a certain engineering practical application value and provides new research ideas and theoretical support for the design of tunnel surface disease recognition algorithm.