Research On Bridge Crack Identification Technology By UAV Image And Convolutional Neural Network

With the rapid development of economy and transportation in China,the traffic flow is increasing year by year,the aging and breakage of bridges built in the early stage become serious,some bridges have insufficient bearing capacity or have reached the design reference period,the bridge collapse accidents occur from time to time,and China has gradually shifted from focusing on bridge construction to paying equal attention to construction and maintenance,the bridge health detection is paid more attention by the transportation department.In order to solve the problem during the traditional bridge health detection which is labor-intensive,inefficient,high-detection-rate and high-cost,this paper proposes a framework for obtaining the bridge surface image by UAV and processing the bridge crack information by computer vision technology,which provides a reference for bridge safety evaluation.Firstly,this paper introduces the research status of bridge health detection at home and abroad.Then,the images of bridge are collected by using Chan Si H20 and Jing Wei M300 RTK,with the images preprocessed by weighted average method,piecewise linear transformation method and median filter to obtain gray image with stronger contrast and less noise.Aiming at the problem of crack image recognition,this paper proposes a bridge image automatic recognition and classification model(BI-ARC)based on convolution neural network.Its function is mainly to identify crack images and divide cracks into four types:transverse,longitudinal,oblique and mesh.The test results show that the classification accuracy of BI-ARC convolution model is 0.988.Then,based on the traditional crack segmentation,a crack segmentation model(BCI-AS)based on full convolution neural network is proposed.The test results show that the BCI-AS full convolution model is accurate enough for crack location,and the crack continuity after segmentation is well enough,which performs better than threshold segmentation and edge detection algorithm,establishing a solid foundation for later geometric parameter calculation.Finally,this paper uses morphology and connected domain marking technology to process binary image of crack,eliminate crack binary image noise,calculate crack length by image refinement algorithm,and present a least square method based on projection technology to calculate the width of crack.To sum up,the proposed bridge crack detection method based on UAV images and convolution neural network has a high recognition rate,which provides a reference for intelligent detection and analysis of bridge crack.