The Key Technology Of Road Surface Disease Detection Research And System Implementation

This paper starts from the key technologies of road surface disease detection,and focuses on the low detection efficiency and poor accuracy of the current road surface disease detection system.The research focuses on real-time road pothole detection,small size crack detection and other contents,and builds an accurate and efficient road surface disease detection system on this basis.The main work of this paper is as follows:(1)A lightweight road pothole detection algorithm is proposed.Aiming at the problems of the existing road pothole detection model such as large number of parameters,low detection accuracy and poor real-time detection,a lightweight YOLOv5 road pothole detection model is proposed in this paper.The Ghost C3 and GSConv modules were used to make lightweight improvements to the YOLOv5 backbone network and Neck network respectively.The CA attention mechanism was combined to improve the feature extraction ability of the model for image spatial information and location information.The CIo U loss function was replaced by SIo U loss function to accelerate network convergence and improve the accuracy of network detection.Experimental results on Savit public data set show that compared with YOLOv5 network,the improved network accuracy improves by 2.8%,recall improves by 2.3%,m AP improves by 1.3%,and floating-point arithmetic reduces by 25%.Compared with the mainstream target detection networks,the accuracy and speed of the proposed algorithm are greatly improved,which provides technical support for the real-time detection of road potholes.(2)A precise and efficient road crack detection algorithm is proposed.Aiming at the problems of poor detection accuracy and incomplete feature extraction of existing road crack detection models,this paper proposes a road crack detection algorithm based on improved Trans Unet.Based on Trans Unet network,Swin Transformer structure with double-branch input is introduced to complete feature extraction of different semantic scales.Combined with MFFM multi-scale feature fusion module,global dependence relationship between different scale features can be effectively established,so as to improve the model’s ability to combine context information and facilitate the segmentation of small cracks.Compared with the Trans Unet network,the accuracy rate,recall rate and F1-score tested on the Crack500 dataset improved by 2.95%,2.65% and 2.79%,respectively.Compared with the mainstream semantic segmentation algorithms,the proposed method performs better in the task of road crack detection.(3)A stable and reliable road disease detection system is designed.At present,the road disease detection is still based on manual inspection,which is time-consuming and laborious,subjective and not standardized.In this paper,the functional and non-functional requirements of the system are analyzed in detail and comprehensively,and the overall design of the system and the integrated development of hardware and software are completed.The system function test shows that the system can realize the road section information management,image acquisition,road pothole and crack detection and other functions.Through the visual interface to provide users with convenient services to ensure the system interaction and practicability.In summary,this paper adopts the method based on deep learning to complete the task of road pothole and road crack detection,and completes the intuitive and accurate display through the software interface,which provides strong technical support for road maintenance work.