Research And Application Of Automatic Deformation Detection Technology Of Subway Tunnel Based On Deep Learning

With the continuous advancement of urbanization,human activities become more frequent and diversified,and the requirements for traffic convenience and punctuality are increasing.In order to alleviate the travel pressure,many cities are more committed to improving urban rail transit in improving transportation.The subway construction site is underground and the environment is relatively poor.Uncontrollable situations such as power failure may occur in the construction process.Relying on manual monitoring has great limitations.In addition,the monitoring of the subway in the operation process needs to go through a long period of process instructions,and there are certain risks in manual monitoring.Therefore,the research on automatic monitoring of subway tunnel is more and more important.At present,the measurement robot is mainly used for automatic monitoring in the subway tunnel,but this technology has some shortcomings: the tunnel is at the bottom of the ground and the distance is long.The instrument orientation in the first measurement process of the measurement robot can not be fully automated,and it needs to find the target prism manually,resulting in low efficiency.Secondly,there are emergencies such as power failure during underground construction,which makes it impossible to carry out real-time monitoring.To solve the above problems,this paper proposes automatic monitoring based on deep learning.Taking TS60 measurement robot as an example,the prism recognition model is constructed with the help of yolov5 algorithm in deep learning to realize target object recognition.During this period,a large number of video frequencies or pictures with prisms are collected by the measurement robot camera,and the collected data are processed and converted into coco data set,With the help of prism recognition model,the confidence is set to be 0.35,0.5 and 0.65.The test is carried out under different conditions,and the average accuracy is more than 90%,which meets the requirements for prism recognition accuracy in subway tunnel monitoring scene,and solves the operation of manual prism recognition for the first time in the monitoring process,so as to improve the efficiency of automatic monitoring.