Research On Highway Pavement Disease Detection Method Based On Deep Learning

As a public infrastructure,highways are closely related to the development of the national economy and people’s livelihood.In recent years,with the rapid economic growth,China’s highway construction has made remarkable achievements.However,the road is affected by environmental temperature changes,traffic loads,rainwater erosion and other factors in the operation process,the road pavement will produce different types of diseases.This not only affects the aesthetics of the road surface and driving comfort,but also poses significant safety hazards to traffic in severe cases.Therefore,timely detection and maintenance of diseases is particularly important.Early manual inspection has low efficiency,high false inspection rate,and the detection effect is easily affected by subjective factors of inspectors.The detection methods based on traditional image processing have poor robustness and generalization performance for disease detection tasks.Therefore,the research on highway pavement disease detection method based on deep learning has been conducted to achieve accurate and efficient identification of highway pavement diseases.The main research contents are as follows:(1)Due to the characteristics of multiple categories and significant scale differences of highway pavement diseases,and susceptible to environmental factors such as uneven illumination and shadow obscuration,which leads to less-than-ideal detection results.In addition,many existing highway pavement disease detection methods cannot achieve a good balance between detection accuracy,detection speed,and model size.Therefore,an enhanced lightweight network E-Efficient Det is proposed in this thesis.Firstly,a feature extraction enhancement module is designed to increase the receptive field of the network and improve the feature expression capability of the network,which can extract more abundant multi-scale feature information.Secondly,a feature pyramid module is proposed based on the idea of semi-dense connection,which is more suitable for pavement disease detection and more effectively fuses multi-scale contextual semantic information.Finally,in order to meet the pavement disease detection tasks under different hardware resource constraints,the E-Efficient Det-D0～D2 network is proposed based on the compound scaling strategy.The experimental results show that the detection accuracy of E-Efficient Det-D0 proposed in this thesis improves 2.41% over the original Efficient Det-D0 on the publicly available road pavement disease dataset and outperforms networks such as YOLOv5 s,YOLOv4-tiny,Faster R-CNN,SSD,etc.The detection speed can reach 27.08 frames per second,meeting the real-time detection requirements,and the model size is 32.31 MB.It is suitable for deployment on mobile devices such as unmanned inspection carts,UAVs,and smartphones.In addition,the detection accuracy of E-Efficient Det-D2 is 4.39% higher than that of E-Efficient Det-D0,and the model size is 61.78 MB,which is suitable for practical application scenarios that require higher detection accuracy and better hardware performance.(2)In order to improve the detection speed while ensuring that the highway pavement disease detection network has high detection accuracy,the YOLOv7-tiny network is taken as the baseline network and improved in this thesis.Firstly,an IECA module is designed,which can effectively suppress background noise,improve the contrast between the disease area and the background,and make the network pay more attention to the learning of disease target features.Secondly,to better utilize the rich detailed texture features of shallow layers,the original PANet structure is improved based on the dense connection idea,which promotes the more effective fusion of feature information at different scales.Finally,the CIo U loss in the original YOLOv7-tiny network is replaced by EIo U loss to improve the detection accuracy of the disease.The experimental results show that compared to the original YOLOv7-tiny,the detection accuracy of the improved YOLOv7-tiny is improved by 1.65%,and is superior to the Efficient Det-D0,YOLOv5 s,YOLOv4-tiny and other networks.The detection speed can reach 93.27 frames per second,which is superior to networks such as Efficient Det-D0,YOLOv5 s,Mobile Netv3-YOLOv4,and E-Efficient Det-D0.The model size is 25.07 MB,which is suitable for deployment on mobile devices.