Research On Crack Detection System And Method Of Vehicle-mounted Tunnel Lining Based On Image Processing And Deep Learning

With the vigorous development of the highway infrastructure construction,the domestic highway tunnel’s mileage increasing,which brought great pressure to the maintenance and management of the tunnel,the tunnel diseases brought serious security hidden danger to traffic safety,especially the lining crack,it directly reflects the force of the tunnel,so the research on tunnel lining crack detection system is of great significance.At present,the detection of lining cracks mainly relies on manual detection,and the detection efficiency is low and the detection accuracy cannot be guaranteed.Therefore,it is urgent to develop automatic detection technology of lining cracks to meet the growing detection needs.In view of the current detection status,this paper studies the crack detection system of vehicle-type tunnel lining based on image processing and deep learning.In this paper,first of all,according to the relevant national standards of tunnel construction and the function of the detection system overall structure are studied,in accordance with the requirements for system design and work principle of the construction of the test platform,to study the related hardware such as lighting of the modeling and installation,industrial side of platform array camera and lens selection and the determination of the number and image acquisition card,auxiliary lighting,dot laser,work-managing machine,rotary encoder on-board equipment selection,etc.In the process of detection,the vehicle tunnel lining crack detection system in this paper collects images of the tunnel lining surface by carrying several industrial array cameras,determines the specific location of the tunnel corresponding to each image with an encoder and saves them in real time.Finally,the identification and detection of lining cracks are realized through offline image processing.Before using semantic segmentation model to simulate the fracture identification,this paper,by using image processing methods on the lining of the image preprocessing,aiming at the condition of the tunnel more noise exist in the image,this paper compares the multiple filtering methods in the tunnel on the image,the effect of filter to solve the problem of uneven illumination of the image,the Mask well light algorithm is improved and in addition to the lining of image shadow,and by using image processing methods to achieve the removal of the seam of lining,image contrast enhanced lining,finally has carried on the construction of data set;In using the deep learning method,which can identify cracks in tunnel in this paper,based on improved VGG19 model puts forward a whole convolution network structure,the contrast analysis of the variety of the optimizer model used in this paper,the cracks in the segmentation effect,realized the recognition of cracks,and then use image thinning algorithm to get the width of the crack skeleton,single pixel according to the characters of related parameters achieved for the classification of the cracks and combined with laser virtual scale method for measuring the length and width of cracks,and finally the system presented in this paper to test the real tunnel experiment,the experimental results show that this system of recognition and detection precision meet the expected effect.To sum up,the research on the truck-type tunnel crack detection system based on image processing and deep learning in this paper provides a useful reference for the automatic detection of cracks in tunnel lining.