Research And Implementation Of Pavement Crack Detection Algorithm Based On Adversarial And Depth-Guided Network

Pavement cracks are an extremely serious problem for roads.Cracks will not only aggravate the pavement disease,shorten the service life of the road,but also affect the comfort and safety of driving.With the increase of road network and traffic flow,it becomes more and more important to locate and repair pavement cracks in time.Therefore,automatic identification and detection of pavement cracks has become an important research topic in the sustainable development of public transportation system.In recent years,deep learning algorithms represented by convolutional neural networks have performed well in pavement crack detection,and their detection speed and accuracy are far higher than traditional pavement crack detection methods.However,in practical engineering application scenarios,due to the difficulty in making pixel-level pavement crack labeled images,the existing pavement crack detection methods are often based on block-level pavement crack labels,which is often accompanied by the difficulty in identifying pavement fine cracks.This paper focuses on the above problems.The achievements and innovations are as follows:(1)A pavement crack detection method based on adversarial and depth guided network is proposedThe method consists of two components:pixel-level pavement crack label extraction algorithm based on edge detection;pavement crack detection algorithm based on adversarial and depth guided network(UCRGNet).The former can extract pixel-level crack labels from block-level crack labels,so as to effectively improve the problem that it is difficult to make pixel-level crack labeled images and the problem that labeling granularity is too coarse.Based on the idea of generative adversarial network,the latter enhanced the feedback of the network to the small crack area by adding necessary supervision to the generated pavement crack segmentation image.At the same time,it adds guiding filter module and attention mechanism to alleviate the problem of information loss,so as to improve the recognition effect of fine cracks.The pavement crack detection method proposed by this paper was tested on the NCDataset.The results showed that the accuracy rate,precision rate and recall rate of the pavement crack detection method were higher than other similar algorithms,reaching 95.89%,67.96%and 65.93%.(2)A complete set of pixel-level pavement crack detection system is designed and implementedThe system consists of six modules:data acquisition,data processing,model training,reasoning recognition,visual display and index evaluation.The test results show that the function and performance of the system can meet the needs of daily work of highway maintenance department.