Research And Application Of Intelligent Detection Technology Of Highway Pavement Disease

As an important transportation infrastructure,highways play a vital role in promoting the development of the national economy.In order to speed up the construction of a powerful country in transportation,the state has vigorously carried out the construction of transportation infrastructure,and the scale of highways has been continuously expanded.With a large number of highways are put into use,affected by traffic load,natural environment and other factors,the pavement diseases dominated by potholes and cracks are increasing,resulting in the reduction of highway durability and bearing capacity,and increasingly prominent safety hazards.Pavement disease detection is the key to pavement maintenance and repair,and has important guiding significance for pavement maintenance decision-making.Traditional pavement disease detection mainly relies on manual investigation,which has low efficiency and low safety,and cannot meet the actual needs of large-scale pavement maintenance.It is urgent to realize intelligent detection of pavement disease.In recent years,with the rapid development of image processing and deep learning technology,some intelligent detection technologies for pavement disease have been developed at home and abroad,but the research of these technologies is mainly based on cracks,the type of disease is relatively single.Affected by the complex background information interference,the detection accuracy is low.The model is large and difficult to meet the needs of actual deployment.In order to solve the above key problems,this thesis took the two main types of pavement diseases,potholes and cracks,as the research objects,and carried out the research on intelligent detection technology of highway pavement disease.The main research contents and results are as follows:1.Aiming at the problem that the existing pavement disease datasets have relatively single disease type and simple scene types,which are difficult to meet the actual and complex detection needs.Guided by the description of the "Highway Technical Condition Evaluation Standard"(JTG 5210-2018),this thesis constructed a pavement disease detection dataset HPDDD and a pavement disease segmentation dataset HPDSD with multiple categories,multiple scene types and containing category annotations and semantic annotations.Four types of pavement diseases such as pothole,lateral crack,longitudinal crack and mesh crack,and multiple scene types such as sunny days,rainy days and shadow interference,and multiple pavement types such as asphalt pavement,cement pavement and gravel pavement,as well as pavement disease images with different scale sample sizes were obtained by means of manual photography and web crawler,which enriched the categories,scene types and sample diversity of dataset.A normalized dataset was formed by data preprocessing and manual labeling,and a small number of category sample images were expanded by data augmentation method to complete the construction of HPDDD and HPDSD.2.Aiming at the problems that the existing pavement disease detection algorithms are easily interfered by complex background information such as illumination,shadow,lane line,side car and well cover,the detection accuracy is low,and the model size is large and difficult to deploy,a pavement disease detection model YOLOv5＿S+G+RS based on the improved YOLOv5 algorithm was proposed.Based on the constructed HPDDD dataset,this thesis firstly verified the superiority of the YOLOv5 algorithm in the problem of pavement disease detection by comparing the detection performance of the classical object detection algorithms.Then,in view of the problems of YOLOv5 algorithm’s poor ability to extract key features of pavement disease,high computational complexity and low model generalization,the YOLOv5 algorithm was improved by integrating attention module,model lightweight design and data resampling mechanism,and the YOLOv5＿S+G+RS model was designed and implemented.Experiments showed that the method in this thesis can realize the detection and identification of multiple types of pavement diseases under complex background conditions,and the m AP index is as high as 93.4%,with high detection accuracy,and the number of the model parameters is only about 5M,which makes it possible for actual deployment.3.Based on the detection of highway pavement disease,a pavement disease segmentation model UNet＿ECA+AG based on the improved UNet algorithm was proposed,and a quantitative analysis method of segmentation results based on morphological operations was explored to realize the segmentation,extraction and quantitative analysis of the detected pavement disease areas.Based on the constructed HPDSD dataset,this thesis used the model architecture of the classic image segmentation algorithm UNet to build a network framework for pavement disease segmentation.Aiming at the problems of poor segmentation effect caused by irregular shape of pavement disease,unclear outline,and external factors such as illumination and shadow,the UNet algorithm was improved by integrating spatial attention and channel attention methods,and the UNet＿ECA+AG model was designed and implemented.Experiments showed that the method in this thesis can realize more comprehensive feature attention and learning of the pavement disease area,and accurately segment the disease area.The length,width and area of the disease were obtained by morphological operation on the segmented pavement disease area,which can assist the damage assessment of pavement disease and provide objective data support for highway maintenance decision-making.4.An intelligent detection and management platform of highway pavement disease was designed and built.The platform was developed based on the mainstream system framework Vue+Element-plus+Spring Boot,and includes functions such as pavement disease data collection track,disease detection,disease segmentation,and disease information query,which realized effective unification of the data collection,visual analysis,data storage and data management of pavement disease,and the intelligent detection technology of highway pavement disease was pushed to practical application.