Preliminary Study On Real-time Detection System Of Tunnel Lining Diseases Based On Computer Vision

With the increase of mileage and number of tunnels in railway,highway,municipal and other fields,the importance of healthy service and long-life operation of tunnel lining structure is increasingly prominent.During the operation period of tunnel,there are many kinds of diseases such as cracks,holes,water leakage and falling blocks in the lining,which seriously affect the service life and safety of tunnel engineering.The research on the real-time detection method of tunnel lining diseases based on computer vision is of great scientific significance and engineering value to ensure the safe operation of tunnel and realize the scientific maintenance of tunnel lining.As an important branch of artificial intelligence,computer vision has a very unique advantage and efficiency.Through the general image processing technology and intelligent algorithm,the computer has a human like visual perception system,which can understand,analyze and judge the image and video information input from the outside world,so as to realize the recognition and detection of specific objects.In order to identify and detect the tunnel lining diseases quickly,efficiently and accurately,the application of computer vision technology in the field of tunnel engineering has become a trend.In this paper,computer vision technology based on artificial intelligence is used to research and develop the experimental device,software system and experimental application of tunnel lining disease identification and detection.The main achievements are as follows:(1)Aiming at the simulation environment of tunnel lining detection in the research of tunnel lining disease identification and detection,an intelligent detection simulation experiment device of tunnel lining defect based on computer vision is designed and developed.The device can meet the continuous relative motion relationship between the detection equipment and the tunnel lining in the process of tunnel lining disease detection,at the same time,it can reduce the space needed for the experiment as much as possible,and provide convenience for the initial exploration experiment in the laboratory.(2)Based on python programming language and tensorflow deep learning framework,a software system for real-time detection of tunnel lining diseases based on computer vision is developed.The functions of the software include the marking and construction of deep learning picture set,the generation of files needed for deep learning,the selection and training of deep learning model,and the recognition and detection of static pictures,video files and real-time feature objects based on camera by using the results of deep learning training.(3)Using the self-designed simulation experimental device and self-developed software system,based on the prepared disease simulation samples and lining pictures with disease,the research work of identification and detection effect is carried out,and the debugging of software and hardware system is completed.At the same time,the experimental research on the factors of illumination and observation angle that interfere with the image processing effect is carried out.The concepts of "illuminance offset" and "observation angle offset" are put forward.It is found that the detection rate,recognition accuracy rate,illuminance offset and observation angle offset of the tunnel lining disease recognition and detection by using the depth learning method are negatively correlated,while the recognition efficiency is always constant.The thesis includes 32 figures,8 tables and 99 references.