Research On Pavement Multi-feature Detection System Based On Deep Learning

With the rapid development of China’s highway,some problems inevitably appear in the process of construction and operation,such as the damage and cracks in the construction process,the pavement structure damage caused by long-term use,etc.these problems will not only cause the waste of resources,but also endanger life safety in serious cases.Therefore,how to quickly and accurately evaluate the pavement quality and prepare for pavement maintenance has become an important research topic.Under the above background,it is particularly important for road detection equipment to detect and grade the road quickly and accurately,but the existing detection system and equipment are expensive,difficult to maintain and difficult to popularize.In view of these problems,this thesis studies a pavement detection system based on deep learning,which integrates the pavement evenness index and pavement disease data to grade the pavement quality.In this thesis,the depth neural network model is used to predict the evenness index,identify and classify the pavement diseases,and realize the on-board and pavement monitoring information platform of the pavement detection system.The system can quickly and accurately detect the road condition and display it on the information platform,providing intelligent decision support for the road management personnel.The main contents of this thesis are as follows:(1)The software and hardware design of road detection system.In the hardware part,the Jetson nano system board is used as the core control module,and the deep learning model is loaded into the system board to realize the collection of pavement roughness and pavement diseases,and the feature collection equipment is selected;The software part divides the detection system into several subsystems,including data acquisition and analysis,data storage and user interaction,which improves the convenience of the detection system.(2)Road roughness index prediction.The LSTM network is used to predict the international road roughness index,and the prediction results are analyzed.The convolution neural network and LSTM network are combined to obtain the spatial characteristics of radar range data and improve the accuracy of prediction.The experimental results show that this method can improve the accuracy of roughness index prediction,and can complete the grading of pavement roughness.(3)Test and improvement of pavement disease identification model.The SSD model is used to identify and classify the pavement diseases,analyze the characteristics of the pavement diseases,and propose an improved method for the SSD model.The basic network of the SSD model is replaced by the dense net network,and the attention mechanism is added to improve the detection efficiency and accuracy.The experimental results show that the accuracy of the improved model on the two data sets can reach 93.5% and 90.28%,which is 4.8% and 6.36% higher than that of the original model respectively.(4)The realization of road monitoring information platform.The road monitoring information platform establishes the connection between the management personnel and the road detection system.The actual test of the platform shows that the platform can receive the relevant data of the road detection system in real time and display it graphically,and provides the staff with the functions of account management and remote access.