Research And Application On Pavement Crack Detection Algorithm Based On Deep Learning

With the continuous increase of road mileage,the demand for road surface maintenance and repair is also increasing.Road surface cracks are the most common road surface disease,which not only affects the appearance of the road surface,but may also endanger road traffic safety.Therefore,timely detection and repair of road surface cracks can prevent further development of cracks,reduce traffic accidents,and save road maintenance costs.Compared with manual on-site inspection,image-based road surface crack detection technology is safer,more efficient and economical.Image-based road surface detection has experienced a transition from manual discrimination to automatic discrimination based on computer vision technology.The development of deep learning technology has made significant breakthroughs in road surface crack detection technology.However,due to the diversity of road surface cracks and the complexity of road environments,road surface crack detection remains a challenging problem.In order to achieve fast and accurate detection of cracks based on images,this thesis conducts research and application of deep learning-based road surface detection algorithms from three aspects: dataset construction,model design and system development.The main work of this thesis can be summarized as follows:(1)Firstly,this article constructs a dataset of road crack images.The road crack images were collected and manually annotated with cracks using on-board cameras and smartphones.The types of cracks were divided into four categories: longitudinal cracks(D00),transverse cracks(D10),mesh cracks(D20),and block cracks(D40).In addition,the images were preprocessed using a Gaussian bilateral filter to reduce noise and highlight key information,in order to improve the effectiveness of subsequent crack detection.(2)Secondly,this thesis designs crack detection algorithms from two perspectives.In order to improve the accuracy of the crack detection network,this thesis designs a crack detection network based on Transformer and multi-scale structures,which combines the modeling ability of global context of transformer and the feature fusion advantage of multi-scale structures;to improve the efficiency of the model,this thesis modifies the model to reduce the number of parameters while maintaining the accuracy as much as possible,and uses Transformer network to guide the lightweight model based on knowledge distillation technology,improving the performance of the lightweight model.The lightweight model can be deployed on mobile or embedded devices,enriching the usage scenarios of road surface crack automatic detection technology.Experimental results show that the two algorithms proposed in this thesis can achieve good results in crack detection.(3)Finally,based on the algorithm in this thesis,the author uses the PyQt5 graphical interface framework to develop a road surface crack automatic detection system based on images.Using the trained model and integrating relevant optimization strategies,this system can automatically detect cracks in input road surface images.In summary,this thesis has conducted research and application of deep learning-based road surface detection algorithms,which has certain practical engineering application value.