Research And Application Of Cable Tension Test Based On Computer Visual Using Convolutional Neural Network

In order to ensure that the cables in the bridge structure are maintained in an optimal condition,it is necessary to conduct tension tests on them during both the construction and operation phases.However,the existing methods for tension testing suffer from various issues such as expensive equipment,difficult installation,complex operation,and poor adaptability to different environments.To address these challenges,this paper proposes a computer vision-based tension testing method using the non-contact nature of computer visual techniques and the multi-scale,multi-level feature learning advantages of convolutional neural networks.The accuracy and stability of this method are demonstrated through tension testing experiments conducted on both a cable-stayed bridge and a suspension bridge,validating the efficacy of the proposed approach.In addition,a computer vision-based tension testing system has been designed and developed to facilitate the development of computer vision in the field of tension testing.The main research contents are as follows:(1)Summarizing the basic theories of tension testing and computer vision,this paper analyzes the shortcomings of traditional tension testing methods in practical applications.Meanwhile,the basic theories of frequency method for tension testing and convolutional neural networks are introduced to provide a theoretical foundation for this research.(2)This paper proposes a tension testing method based on the edge detection algorithm of convolutional neural networks,namely the improved HED-based tension testing method.This method addresses the poor adaptability and low robustness of tension testing methods based on traditional computer vision techniques.The accuracy and stability of this method are verified through tension testing experiments conducted on both a cable-stayed bridge and a suspension bridge.(3)This proposes a tension testing method based on the image differential algorithm of convolutional neural networks,namely the IDACNN-based tension testing method.This method addresses the dependence of the improved HED method on obtaining clear cable edges.Field experiments show that this method has higher accuracy and stability compared to the HED-based tension testing method.(4)A computer vision-based tension testing system has been designed and developed to integrate various computer vision-based tension testing methods,simplify the tension testing procedure,and promote the development of computer vision in the field of tension testing.