Resrarch On Intelligent Identification Method Of Bridge Cracks Based On YOLOv5 And U-Net3+ Algorithm

Bridge structural health monitoring and intelligent detection technology based on UAV and artificial intelligence have become the current focus of in the field of bridge engineering.In order to overcome the problems of low efficiency and poor effect when using traditional digital image processing methods to detect bridge cracks,this paper introduced deep learning algorithm to detect bridge cracks.Firstly,the object detection algorithm is used to realize the intelligent identification and location of cracks;then,based on the crack identification results,the image segmentation algorithm is used to extract crack pixels,and calculate the crack width and length information to achieve the integration of high-precision cracks identification,location and information extraction.The method proposed in this paper can realize high-efficiency,high-precision and intelligent detection of bridge cracks,effectively reduce the risk and cost of bridge detection work,and has strong engineering application prospects and social value.The main research contents are as follows:(1)The training data set of the deep learning model is constructed,and the test images library of the crack detection model is established by studying the test image acquisition and preprocessing methods.Using the public data set of cracks and manually annotating the crack information,a crack identification data set containing4414 crack images and a crack segmentation data set containing 708 crack images were constructed,and the data enhancement method was used to improve the richness of crack training samples.Plan the UAV flight route,use the M210 RTK UAV platform to collect the tower test images.Different grayscale and filter denoising methods are used to process the UAV images,and compare and analyze the results.Finally,the weighted average method,median filter and 25% overlap cropping were used to preprocess the test image.(2)The deep learning object detection algorithm was studied,and a bridge crack identification and localization method based on the YOLOv5 algorithm is constructed.By adjusting the width and depth parameters and optimizing the bounding box loss function,a crack identification and location model based on YOLOv5 object detection algorithm is trained and constructed using the crack identification data set.the cable tower images of Hongshan bridge were used as the test object of the crack identification and location model,the overall precision,recall and F1 score are91.55%,95.15%,and 93.32%,realizing high-precision identification and localization of cracks.(3)The deep learning image segmentation algorithm was studied,and a crack segmentation model based on the U-Net3+ algorithm was constructed.And an eight-direction crack width measurement method combined with connected domain denoising,edge detection and morphology processing was developed.Using the crack segmentation data set,a U-Net3+ image segmentation algorithm combined with deep supervision strategy and classification-guided module was introduced to train and build a crack segmentation model.The cable tower crack images successfully identified by the crack identification and localization model were used to evaluate the segmentation accuracy of crack segmentation model,the overall precision,recall and F1 score are 93.02%,92.22% and 92.22%,achieving high-precision extraction of crack pixels.Based on the U-Net3+ crack segmentation results,pixel-level measurement of cracks length and width is finally achieved by the eight-direction search method,after image denoising and the extraction of crack edges and skeletons.