Research On Surface Damage Detection Technology Of Bridge Cable Based On Machine Vision

Cable-stayed bridge is an important part of the traffic network,and its health monitoring is the main work of the maintenance of cable-stayed bridge at present.The surface damage detection of bridge cables is the focus of the inspection.The structural and functional integrity of bridge cables directly affects the operation safety of cable-stayed bridge.The traditional cable surface damage detection method relies on the lifting machine to detect the cable surface defect manually,which is low in efficiency and has a big hidden danger.At present,there are few researches on automatic detection methods of cable surface damage based on machine vision,and the detection algorithms used are mostly traditional image processing methods or ineffective deep learning models.Meanwhile,the detected damage cannot be further measured in the current research.Aiming at the difficulties of bridge cable detection,this paper proposes a method of bridge cable surface damage detection based on machine vision.The main work of this paper is as follows:(1)Camera selection was completed and the visual inspection system of the cable crawling robot was constructed.The characteristics and precision requirements of cable detection are analyzed.And A suitable camera system is selected for the image acquisition,including camera selection,camera position setting and camera parameters selection.(2)Automatic detection of cable surface defects was realized based on deep learning technology.Firstly,the cable surface damage data set was enhanced to solve the problem of small sample size.Secondly,the YOLOv5 depth model was used to train the model for damage detection,and the YOLOv5 model was improved and optimized according to the characteristics of cable surface damage.The introduction of CBAM attention mechanism effectively improves the information extraction ability of the model,and the K-means clustering algorithm is used to make the anchor frame of the model more suitable for the detection of cable surface damage.The optimized YOLOv5 model effectively improves the accuracy of damage detection.(3)In order to realize perspective imaging of bridge cables,a perspective imaging distortion correction algorithm for cylindrical objects is proposed,and the cylindrical surface texture rendering is realized.The defect detection results were processed by this method to restore the real texture of cable surface damage.Based on this method,the damage detection results can be quantitatively analyzed,so as to judge the severity of surface damage.(4)Cable surface damage detection software was developed,which can adjust camera parameters in the data acquisition stage,and monitor the robot acquisition state by using LAN communication technology;In the damage detection stage,optimized YOLOv5 is used to detect cable image data,and perspective imaging distortion correction is carried out on damaged images to realize damage measurement,damage statistics and result preservation,etc.,integrating the whole detection process is convenient for deployment and use of the detection system.