Research On Bridge Crack Detection And Identification Model And Quantification Method Investigate

In the operation process of reinforced concrete bridge,due to the coupling effect of load,rain,temperature and other factors,cracks,spalling and other appearance defects will be produced,which will lead to the overall performance of the bridge structure decline.However,the traditional manual detection requires closed traffic and is not safe.At present,the automatic fracture detection technology is limited by complex background,irregular fracture boundary and other factors,resulting in low accuracy of target detection model.In addition,crack quantification is of great significance to determine the severity of bridge cracks and the selection of crack repair strategies,but there are few such studies.Based on this,this paper proposes an automatic bridge crack detection and recognition model and quantitative method.The main research contents and conclusions are as follows:(1)In order to verify the method proposed in this paper,the detection and recognition ability of the simulation model in multi-scale bridge crack and complex background was obtained.By collecting bridge crack images under complex background,data enhancement technology was used to enrich the data set,and a bridge crack image data set containing 3500 images was established.(2)Through the analysis and comparison of various image processing and image segmentation algorithms,the experimental results show that: weighted average method obtained gray image error is smaller;Gamma transform and bilateral filtering can reduce image noise and retain crack edge information better.OTSU algorithm has more advantages in the realization of fracture binarization segmentation and extraction of fracture targets.(3)In order to realize efficient,accurate and contactless identification and detection of bridge cracks,a bridge crack detection model YOLOV5-SA based on improved YOLOv5 is proposed.Four different attention mechanisms were introduced in the Head section of YOLOv5 s.Adaptively Spatial Feature Fusion(ASFF)was used on the basis of pyramid feature notation.It is found that Selective Kernel Networks(SKNet)can improve the representation ability of crack characteristics.ASFF can enhance the network feature fusion and improve the small target detection ability of bridge cracks.By comparing the YOLOv5-SA model with the original model,Faster-RCNN and YOLOv4 tiny model,the results show that the accuracy of the YOLOv5-SA model is 88.3%,which is 2.5% higher than that of the original model.m AP0.5 and m AP0.5-0.95 increased by 2.0%and2.2%,respectively,to 90.2%and 62.3%.Compared with the other two models,it also has comparable or better detection performance.(4)On the basis of determining the crack area,dividing the crack and the background information,combining the axial transformation and orthogonal measurement to describe the crack shape,extracting the ideal crack skeleton information,and quantitatively analyzing the characteristic information such as the length,width and area of the bridge crack,which can provide the basis for the later crack repair strategy.