Research On Road Quality Detection Based On Deep Learning

In recent years,with the rapid development of road traffic in China,the construction of a large number of highways,national highways,and provincial highway networks has increased the pressure on road maintenance.In the daily operation and use of roads,it is inevitable to receive natural and man-made damage,resulting in cracks,collapses,potholes,and other structural damage.This requires regular detection and maintenance of the road surface.However,due to its high price and difficulty in maintenance,the existing road surface detection system equipment has not been widely used,which leads to the need to waste a lot of manpower and material resources for daily maintenance of the road surface.In this context,this thesis introduces a deep learning model to design an automated pavement quality detection system.First,the construction of the pavement dataset is completed through the generation of a deep learning network model,and then the detection of pavement diseases and roughness is completed through the deep learning detection network model.Finally,the operation of the pavement automatic detection system is realized on the embedded platform and the pavement monitoring platform.The main research content of this article is summarized as follows:(1)Aiming at the characteristics of rich spatial information and high correlation of feature information in pavement disease data,a pavement disease image generation algorithm based on improved LSGAN was proposed.Firstly,to ensure the diversity of generated images,a new loss term is introduced and the objective function of the generator is reconstructed.Secondly,the coding structure of CAE is fused to enhance the generator’s learning of image spatial information and improve the convergence speed and generation quality of the model.Finally,a lightweight residual projection expansion projection expansion module(RPEPX)was constructed and spectral normalization was introduced to further improve the quality of generated images and ensure the stability of model training.The experimental results show that each evaluation index FID,SSIM,and PSNR of the generated image has significantly improved,and the YOLOv5 s detection network verification shows that the method in this chapter achieves the optimal detection results compared to traditional data enhancement methods.(2)Aiming at the difficulty of detecting complex pavement features and small target features,a pavement disease detection algorithm based on improved YOLOv5 s is proposed.Firstly,in view of the poor detection effect of YOLOv5 s on targets with unclear disease features,the idea of Bi FPN is introduced to reconstruct the network structure and enhance the ability of image multiscale feature fusion.Secondly,in order to alleviate the pressure of network learning,the backbone network adds a Sin AM attention model to improve the model detection accuracy.Finally,the EIOU loss function is introduced to improve the regression accuracy and convergence speed of the model.Experiments show that the improved model improves the average detection accuracy by 15.2% compared to the improved model,while hardly increasing the amount of model parameters.(3)A pavement quality automatic detection system and monitoring platform are designed.First,use the LSTM model to complete the prediction of road roughness,then confirm the hardware selection with the Jetson TX2 development board as the core,select a CSI camera and laser radar to collect road data,and finally load the trained deep learning detection model into the control module of the development board.The software part designs multiple functional subsystems,including data acquisition,data processing,and data storage modules.The monitoring interaction platform has designed a graphical interaction interface,enabling users to query and feedback pavement data through the monitoring platform,and managers to manage pavement data and account information.