Research On Crack Detection Technology Of Subway Tunnel Based On YOLOv5

With the rapid development of urban rail transit in China,the demand for safety detection of tunnels is getting higher and higher.However,the traditional manual detection can no longer meet the current requirements of fast and accurate detection,so it has been gradually replaced by detection method based on deep learning and digital image processing technology.Therefore,a tunnel crack detection system based on YOLOv5 and digital image processing is proposed in this paper.The YOLOv5 target detection model is built,and the established dataset is used for training to realize fast positioning of the crack area.A crack extraction algorithm is proposed to extract the crack features and realize the automatic detection.Firstly,this paper designs the tunnel crack acquisition vehicle.Acquisition card,linear array camera,signal generator and other equipment are designed and selected for the task of tunnel crack acquisition.The acquisition vehicle is equipped with four linear array cameras for synchronous acquisition,and a series of pre-processing operations are carried out on the collected original images.Image distortion correction technology is used to stretch the original image to eliminate the imbalance of the ratio between horizontal and vertical in the process of acquisition.In order to solve the problem of too much noise and interference in subway tunnel,Wiener filter is used to filter most background noise.Secondly,due to the complex environment inside the tunnel and the elongated structure of cracks,cracks only account for a small proportion in the image,and the image classification model alone is not enough for crack detection.In order to achieve more accurate detection of crack disease,this paper proposes a crack disease detection system based on target detection,builds an improved YOLOv5 neural network,and constructs a multi-category target detection dataset for subway tunnels.After training the model,automatic detection of crack disease is completed.After training,the m AP value of the improved YOLOv5 target detection model algorithm reaches 0.861,among which the AP values of cracks and water seepage are0.777 and 0.887,respectively.The detection accuracy reaches 94.07% and 95.21%,which better meets the detection need of subway tunnel diseases.Thirdly,a crack feature extraction algorithm based on image processing is proposed on base of the target detection model to determine the crack area.The improved Otsu was proposed to segment the crack and the background,and the connected domain of the crack was extracted by filtering out the noise with the morphological operation.Finally,the skeleton extraction algorithm and the adaptive burr removal algorithm were used to calculate the quantitative characteristics of the crack,which provided technical support for the subsequent tunnel disease processing.