Water Seepage Detection In Subway Tunnel Point Cloud Images Based On Improved Mask R-CN

The problem of water seepage in the inner wall of subway tunnel is very common,which will do serious harm to the safe and normal operation of subway.Hence,identification and detection of water seepage disease is of vital importance and plays a very cruical role.The speed of how artificial intelligencedevelops and evolves is really stunning,and deep learning has been applied extensively and maturely.It can achieve the effect of recognition and detection more efficiently and accurately.Compared with the traditional artificial seepage detection method,the deep learning technology has high accuracy and strong generalization ability.At the same time,it can reduce the labor cost and reduce the error caused by the subjectivity of manual identification.For the problem of water seepage in the inner wall of subway tunnel,this paper aims to identify water seepage with depth learning correlation algorithm.The paper puts forward a new CNN model based on the full-blown Mask R-CNN,as well as an algorithm for identification and detection of seepage water on the inner wall of subway tunnel.Moreoever,the paper elaborates on deepened exploration on them.Details are given as below:(1)A 3D laser scanner is applied to help gather point cloud data on the internal walls of subway tunnels.The paper compares this methodology with old ones,exploiting and analyzing the advantages and superiority of data collection of this device.(2)Preprocess and apply point cloud data.After a simple analysis of the point cloud data,denoise,increase sampling,projection and other processing operations are carried out.(3)2D image of 3D plane point cloud;Open CV is used to denoise and sharpen the image to highlight the features and improve the recognition accuracy.Manually mark the water seepage in the picture with labelme software tool.(4)The research compared structures of multiple models,including CNN,R-CNN,Fast R-CNN,Faster R-CNN and Mask R-CNN,and the way of how each of them works,highlighting and underlining the advantages and superiority of the Mask R-CNN network structure in target detecting with comparisons between and study on the course of their network development.Therefore,this paper uses Mask R-CNN to train and identify the water seepage in the processed water seepage image of tunnel inner wall,and realizes the automatic identification and annotation of diseases.Using the test set test,the average accuracy of the original mask r-cnn recognition is 91%.(5)The mask r-cnn network model is improved to obtain S-mask r-cnn,and the average accuracy of the new model is 97%.Compared with the original mask r-cnn recognition method,S-mask r-cnn has clear structure,stronger feature learning ability and higher accuracy.