Research On Detection Method Of Pavement Disease Based On Deep Learning

Pavement disease is a key factor affecting road safety.Regular inspection and maintenance play an important role in maintaining pavement performance.At present,pavement disease detection mainly uses manual inspection detection,manual detection labor intensity,slow speed,there are security risks in the detection process,and the detection accuracy is easily affected by the experience and knowledge of inspectors.Research on fast,safe and automated pavement disease detection methods is of great significance to repair.In this study,a fast and accurate detection method of YOLOv5-pavement pavement disease based on deep learning is proposed,a data acquisition method and a complex scene data enhancement algorithm are proposed,and the structure of YOLOv5s algorithm model is improved and optimized.The road disease detection performance of YOLOv5-pavement is tested.The main research contents are as follows:(1)Data acquisition and complex scene data enhancement methods.According to the environment of the road,the characteristics of the road itself,and the requirements of the road disease detection work,this paper proposed an automatic road image acquisition scheme based on vehicle photography and supplemented by UAV equipment.The scheme can collect the pavement image data completely and continuously.Aiming at the problem of unbalanced number of different disease samples and low complexity,this paper proposed a complex scene data enhancement technology.Modules such as Gaussian noise,fog,rain,snow,and mud are used to imitate the road environment.Then a rich and complete training dataset is constructed by calling modules such as motion blur,brightness adjustment and saturation adjustment to imitate the field brightness and acquisition state.The experimental results show that the model accuracy of transverse crack,longitudinal crack,mesh crack,pothole and repaired pavement increased by 0.4%,0.6%,0.6%,5.0%and 2.2%respectively,and mAP increased by 1.76%after the enhancement of complex scene data enhancement technology.(2)Structural improvement and optimization of algorithm model based on YOLOv5s.Aiming at the problems of low accuracy and insufficient feature extraction ability of YOLOv5s algorithm model,the structure of YOLOv5s algorithm model is improved and optimized.Firstly,the standard convolution is optimized into the depth wise separable convolution,and the improved attention module SCBAM is added to further enhance the feature extraction ability of the algorithm model while maintaining the lightweight of the algorithm model.Then,SPP was improved to SPPF structure,which enriched feature information and improved inference speed.Finally,the classification Loss function and regression Loss function were improved into Focal Loss and EIOU Loss to solve the adverse impact of positive and negative sample imbalance on the optimization of the model training process,and to avoid the problem of penalty failure when the ratio of height and width between the prediction box and the true box is linear.Ablation experimental results show that compared with the original model,the improved YOLOv5s algorithm model has the accuracy,recall rate and average precision increased by 5.5%,4.5%and 4.2%respectively.(3)Algorithm model testing and practicability research.In order to explore the pavement disease detection performance of the improved algorithm model,the model was tested and evaluated after completing the training.The test results show that the pavement disease detection method of the improved algorithm model YOLOv5-pavement can accurately and quickly detect the diseases in the pavement image,the precision,recall rate and mAP reach 93.6%,93.0%,93.1%,and the FPS reaches 89.Compared with several mainstream algorithm models,the improved algorithm model mAP has advantages,and the scale is only 12.6%,7.2%,5.8%of Faster R-CNN,SSD,YOLOv3.The road practicality test results show that the improved algorithm model in this study has obvious advantages in accuracy and speed compared with the manual method and other algorithm models.The pavement disease detection method based on YOLOv5-pavement proposed in this study can effectively detect a variety of pavement diseases.The detection accuracy of the algorithm model meets the detection requirements of pavement diseases,and has high engineering application value.