The Research On Image Enhancement And Anomaly Detection Method For Bridge Cable Anchorage Zone

Suspension bridges and cable-stayed bridges are types of bridges whose main load-bearing structures are suspended from and anchored to cables attached to both ends or both sides of the bridge.The cable anchor head is a critical component that affects the safety of the entire bridge,and therefore regular inspection of the cable anchor head is necessary.Traditional inspection methods require personnel to remove the protective cover of the anchor head on-site and visually inspect the anchorage zone for any abnormalities.This method is not only timeconsuming and labor-intensive,but also affects the sealing of the anchor head cover.Installing a small machine vision system inside the cable anchor head cover to detect and analyze anomalies based on images of the cable anchorage zone can help maintain the sealing effect of the anchor head cover,and also improve the efficiency and accuracy of the inspection results.This paper proposes an optimization scheme for imaging conditions in dark environments for cable anchorage zones based on the structure of bridge cable anchorage systems.It designs an image acquisition device and image enhancement method suitable for cable anchorage zone inspection and uses deep learning semantic segmentation to detect abnormal areas in the cable anchorage zone.It also develops visualization and anomaly detection software for cable anchorage zones.The main work of the paper is as follows:(1)A visual monitoring system for cable anchorage zone is constructed.First,the structure of the bridge cable anchorage system is analyzed,and then a visual monitoring system for the cable anchorage zone is designed.In the system design process,suitable cameras,image processing equipment,power supplies,and networking modules are selected considering environmental factors,and the operation strategy of the equipment is formulated.(2)A cable anchorage zone image enhancement method is proposed to solve the problem of uneven lighting and high reflectivity of metal components that cause local darkness and highlights in image acquisition.This method can solve the problem of local overexposure or underexposure that occurs when shooting images by fusing sequence images of cable anchorage zone with different exposure times.Experimental results demonstrate that the proposed method outperforms the method proposed by Mertens et al.,with the enhanced image entropy mean values of the cable anchorage zone ranging from 0.0196 to 0.2425 and the mean gradient mean values ranging from 0.4075 to 3.5593.The proposed method is more conducive to visual inspection and anomaly detection.(3)Using Deep Labv3+ as the basic network,a bridge cable anchorage zone anomaly detection network model is constructed.First,cable anchorage zone images are collected and abnormal areas are labeled.Then,cable anchorage zone image datasets are constructed through geometric transformations,color adjustments,and other methods.Deep Labv3+ is selected as the model for bridge cable anchorage zone anomaly detection and trained through transfer learning.The feature extraction network suitable for cable anchorage zone anomaly detection is determined through experiments,and the segmentation performance indicators of the model,MPA and MIo U,are 83.11% and 77.38%,respectively,achieving automatic recognition of cable anchorage zone anomalies.(4)Visualization and anomaly detection software for cable anchorage zones are developed.The software can remotely control the cable anchorage zone image acquisition device and integrates image enhancement algorithms and anomaly detection methods,laying the foundation for the engineering application of cable anchorage zone anomaly detection.