Research On Bridge Disease Detection Method Based On UAV Imaging

With the continuous development of the economy and society,China’s infrastructure construction capacity is also improving rapidly.The bridge structure not only has the basic functions of communication and transportation,but also is an important tool to promote economic and social development.However,with the increase of service life,the bridge will be damaged to varying degrees under the influence of natural environment,load,construction defects,material aging and other factors.At present,the disease problem of bridges in service in China is becoming increasingly serious,and the requirements for the efficiency and safety of bridge detection are getting higher and higher.In this thesis,the bridge image is collected by UAV,and the crack disease and area disease of bridges are studied by using digital image processing technology and neural network recognition technology.The purpose is to improve the automation of bridge detection and provide a basis for bridge disease evaluation.The specific research contents are as follows:(1)The UAV equipment is refitted to realize the image acquisition of the bridge bottom,and a complete UAV bridge image acquisition scheme is designed,mainly including the determination of the shooting distance,image acquisition process,precautions,and pre and post flight inspection.(2)Based on digital image processing technology,the bridge disease identification method is designed,including graying,grayscale correction,image denoising and image segmentation process.An improved median filtering denoising algorithm based on the statistical characteristics of bridge diseases is proposed,which can preserve the crack characteristics when denoising the crack image,and achieve a better denoising effect.At the same time,the weighted edge preserving smoothing algorithm is adopted for the area disease,which significantly reduces the influence of the smoothing process on the disease edge.(3)Based on convolutional neural network technology in depth learning,disease identification model is established by using GoogLeNet-V3 network,and the accuracy of disease image recognition is more than 95%.The SCD network sliding detection method is proposed for the crack image,and five size detection templates are designed,among which the 32×32 window model has the best sliding recognition effect.Corresponding of the identified image,an adaptive threshold calculation method is designed for global threshold segmentation to improve the accuracy of crack image segmentation.(4)According to the disease characteristics,the disease target is extracted to achieve the quantitative analysis of the bridge apparent disease.Calculate the crack trend according to the projection method,calculate the crack length based on the freeman code marking method,calculate the average width of the crack from the crack length and crack area,design the maximum width algorithm to calculate the maximum width of the crack,and use the image connectivity to calculate the size of the area disease.(5)According to the research results of bridge disease identification and feature calculation,the UAV bridge disease identification calculation software is developed and verified by an example.The UAV bridge disease identification system designed in this thesis can realize image acquisition of bridge apparent diseases,can accurately identify and calculate bridge diseases,improve the automation of bridge detection,have a good engineering application prospect,and provide a certain technical reference for disease detection in other fields.