Research On Pavement Crack Detection Technology Based On Improved Canny Algorithm And Deep Learnin

Road traffic has a significant impact on the national economy and people’s lives,gradually becoming a crucial infrastructure supporting economic development and transportation.With the continuous increase in road mileage,maintenance and upkeep of roads face growing challenges,particularly in areas such as crack detection and repair,which require increasingly heavy workloads.However,traditional image processing methods still face difficulties in providing accurate results,making it challenging even for experienced professionals to make precise judgments.To address this issue,this paper focuses on two improvements: first,using the improved OTSU algorithm with the wolf pack algorithm to enhance the traditional Canny algorithm for image processing;second,combining Mask R-CNN to improve image recognition efficiency.The key improvements and innovations proposed in this paper include the following aspects:1.Adopting the optimized two-dimensional OTSU(two-dimensional maximum interclass segmentation)algorithm through the wolf pack algorithm,replacing the manual threshold setting method.By comparing the tracking speed between the wolf pack algorithm and particle swarm optimization algorithm,this paper selects the wolf pack algorithm to improve the two-dimensional OTSU algorithm,significantly enhancing the pixel traversal speed in the image and enabling more effective and faster identification of crack pixels.At the same time,the automatic setting of high and low thresholds through the two-dimensional OTSU algorithm effectively removes subjective interference caused by human factors,making threshold selection more objective.2.Introducing a combined approach of multiple morphological filtering and bilateral filtering to replace the traditional single Gaussian filtering for image crack processing.Through comparisons among morphological opening and closing operations,dilation algorithms,erosion algorithms,median filtering,mean filtering,and bilateral filtering,this novel processing method helps solve problems like pixel loss and crack fragmentation,thus improving the accuracy of detection results.3.This paper utilizes ResNet 101(ResidualNetwork)and FPN(Feature PyramidNetworks)as the main components of Mask R-CNN,optimizing the network model and backbone,and enhancing the optimization of anchors.Under the same conditions,different sizes of anchors are selected for training and testing.After experimental comparison,anchors with sizes {20,40,80,100,120} are used as the final configuration,resulting in an accuracy improvement of 10.898% in precision and 7.710% in recall compared to RCNN.In terms of speed,the processing time per image is reduced from 0.16834 s/patch to 0.12853 s/patch,reducing the workload of manual identification.Building on the improved Canny algorithm,this paper innovatively proposes the use of the improved Mask R-CNN for crack identification after image processing,which not only improves the overall efficiency of road crack detection but also reduces reliance on manual inspection.As a result,maintenance teams can allocate resources more reasonably and promptly address road damage issues,providing the public with safer and smoother transportation infrastructure.By improving image processing techniques,this paper presents a novel and efficient method for road crack detection.These improvements not only enhance detection accuracy but also significantly reduce detection time and dependency on manual inspection.