Research And Realization Of Tunnel Lining Defect Identification

Due to the complicated construction geology,the lining structure of tunnel is easily affected by many factors.The apparent detect of lining will reduce the supporting strength of concrete,resulting in internal defects of lining and in serious cases will cause safety accidents.Hence,it is vital to carry out efficient defect detection on lining surface and interior.The detection of tunnel lining surface and internal defects is realized by 3D Laser Measurement Technology and Geological Radar Detection Technology respectively.The reading and judging of these two kinds of image data are not only time-consuming and labor-consuming,but also depend on the professional ability of engineers.As a result,in this thesis,a Multi-Scale Feature Fusion and ROI Align-Region with CNN Features detection model(MR-RCNN)and Improved YOLO with Multi-Scale Feature Fusion and Attention model(YOLO-MA)for tunnel lining internal defect detection(YOLO-MA)are proposed.In the MR-RCNN model,a set of feature fusion region candidate network is designed in order to transfer the high level has good semantic information and location information at the same time.Meanwhile,in order to better implement and improve the model for small-scale crack detection,a set of model generation frame is designed in the model for tunnel lining apparent detects.In addition,the recognition accuracy of the model is improved by aggregating the region features through floating-point bilinear interpolation method.In the YOLO-MA model,a bottom-up enhancement is designed to enhance the detection ability of different sensory fields,so that the top feature map can also more comprehensively extract the rich location information brought by the bottom layer.Meanwhile,to enhance the learning ability of the model and let the network focus on important features,different weights are assigned to each channel of the model.To improve the regression speed and accuracy rate of the bounding box,the bounding box loss function is improved in this model.In this thesis,the feasibility of MR-RCNN and YOLO-MA models is demonstrated through extensive experiments that the models have excellent identification performance.Based on this,a tunnel lining defect identification system is designed to identify both apparent and internal tunnel lining defects through intelligent detection,thus generating reports that can efficiently assist surveyors in planning tunnel defect repair plans.