Research On Intelligent Pavement Crack Detection Method Based On Yolov5 Tiny Algorithm

China’s rapidly developing transportation industry has led to a year-on-year increase in total highway mileage.Total road mileage has increased to 5.28 million km by 2022.As time gose by,factors such as material aging,vehicle overload,weather and so on will cause different degrees of pavement diseases.The common early pavement diseases mainly include cracks and pits.If not handled in time,it will not only affect people’s driving safety,but also affect people’s driving comfort.For this reason,rapid,accurate,and real-time detection of early pavement disease plays a vital role in the current development of road traffic.Aiming at the characteristic information of pavement cracks,this paper proposes an intelligent detection method based on computer deep learning.Based on the deep learning target detection algorithm,this paper analyzes and researches how to improve the speed and accuracy of multi-scale and multi-category pavement cracks detection.The research contents of this paper include:(1)Analyze the deep convolutional neural network’s fundamental architecture and underlying principles;contrast and compare the two-stage and one-stage training networks;optimize and design the deep learning network structure from the standpoint of the network training approach,loss function enhancement,sample data pre-processing,etc.(2)In view of the issue that the feature message in the pavement crack samples is not obvious and the background noise is large,the paper first compares and analyzes the filtering algorithms of different image processing,and preprocesses the samples by comparing and selecting the method that can increase the characteristic information of the pavement crack edge and reduce the background noise by combining the mean shift filtering algorithm and the CLAHE algorithm,so as to improve the detection accuracy of the final model.Aiming at the problem of uneven positive and negative samples of pavement cracks,the paper uses contrast change,color transformation,mirror image,flip and other methods to expand the pavement crack data set.(3)In order to improve the speed and accuracy of deep learning network in crack detection,a new crack detection way based on increase YOLOv5＿tiny algorithm is proposed.First,to solve the problem of slow model detection and large calculation parameters,in the YOLOv5 network,Replace the backbone network in YOLOv5 network with Shuffle Netv2 lightweight network,which decreases the number of Calculating parameters and increase model discerning speed.Then,for network recognition accuracy,the paper then introduces the idea of a dense link,replaces the feature fusion network with PA-Dense Net network,and reuses the feature information,thus increasing the accuracy of the network in multi-scale and multi-class pavement crack detection.The original Re LU activation function is substituted by the Mish activation function which is relatively stable in the forward propagation of gradient flow to improve the stability of the model and avoid the network over-fitting caused by gradient explosion and gradient disappearance.Finally,the binary cross-entropy loss function is used to replace the original classification loss function,thus ensuring the stable regression of the boundary box during network training.(4)In order to meet the demand of human-computer interaction,the paper finally designed a user system interface based on Py Side2.Through calling the Open CV module,it realized the detection of images,videos,real-time pavement cracks and the output of detection results.After testing,it can basically meet the requirements of early pavement crack detection.By the experiment and analysis,Based on the improved YOLOv5 tiny algorithm,the proposed pavement crack detection method can significantly improve the detection accuracy and speed of the original YOLOv5 network,and can provide scientific basis for the formulation of early road maintenance plan by the road maintenance department.