Research On Pavement Crack Detection Method Based On Deep Learning And Image Processing Technology

Highway is an important part of the transportation system,and road cracks are a common problem on highways.Cracks can cause corrosion of internal steel bars,induce more cracking,reduce structural durability,and in severe cases,lead to safety accidents.Traditional manual detection methods have low efficiency,subjectivity,and safety issues,while digital image processing methods are complex and susceptible to environmental influences.The rise of deep learning technology has made crack detection more automated,but there are still many problems to be solved,such as difficult detection under complex backgrounds,accuracy needs to be improved,loss of detail information,thick predicted edges,and class imbalance.The lack of standard definitions and calculation methods for crack feature parameters is also a problem.Therefore,this paper studies the crack detection method based on deep learning and image processing technology,and the main research contents are as follows:(1)To solve the problem of difficult crack detection and low accuracy under complex backgrounds,a road crack segmentation method based on attention mechanism and deformable convolution is proposed.The network uses an encoder-decoder structure,adds deformable convolution in the encoding stage to enhance the network’s representation ability for crack features,and uses dense connection strategy to strengthen feature information.In the decoding stage,transpose convolution and bridging method are used to gradually fuse the features of the encoding stage,and the idea of multi-level feature fusion is combined to improve detection accuracy.An attention mechanism is added to enhance target features and suppress noise.The experimental results indicate that the performance of the algorithm outperforms the comparison algorithm in all evaluation metrics.On the Crack500 dataset,the MPA and MIoU are 89.0% and 81.4%,respectively,and on the CFD dataset,the MPA and MIoU are 89.9% and 80.0%,respectively.The performance of complex background crack detection is excellent.(2)To solve the problems of lost detail features,thick predicted edges,and class imbalance,a road crack segmentation method based on dense connection strategy and image gradient is proposed.The network uses dense connection strategy to enhance the transmission between feature layers,adds 3D attention mechanism to capture low-level feature target information,designs feature enhancer to better represent crack features and capture context information,and proposes edge refinement to adapt image gradient to more accurately locate crack boundaries for crack detection.At the same time,an improved weighted cross-entropy loss and Dice loss are combined to solve the class imbalance problem.Experimental results show that this algorithm has more accurate edge positioning and better segmentation effect.It performs well on multiple datasets,and the average precision on the public dataset can reach over 90%,which can meet certain engineering detection needs.(3)To solve the problem of the lack of standard definitions and calculation methods for crack-related parameters,a crack feature parameter calculation method based on the combination of deep learning and image processing technology is proposed.First,the deep learning algorithm is used to obtain the crack binary image.Then,an image processing technology is constructed,including connected domain denoising,fracture connection,edge detection,crack skeletonization,and crack feature parameter calculation,to automatically obtain crack feature parameter information.Through experiments,it is verified that the calculated crack feature parameters are close to the actual values,with a relative error of less than 10%,and the crack feature parameter calculation accuracy meets the engineering application requirements.