Research On The Processing Method Of Concrete Bridge Surface Crack Images

Many highway bridges in China have gradually entered the maintenance and repair stage.As an early manifestation of bridge surface damage,cracks are the critical monitoring object for bridge maintenance.Traditional crack detection or aided detection by equipment such as UAVs relies on manual identification of crack damage,which is a time-consuming and laborious process,and the results are affected by the subjective factors of the inspector.In order to improve the efficiency of crack detection and reduce the influence of subjective factors on the results,digital image processing techniques and deep learning are used to complete a series of processing of bridge surface crack images.The complex environment on the bridge surface,with various disturbing factors such as uneven lighting,paint lines,potholes and stains in addition to crack areas,affects the processing effect of crack images.Therefore,this paper takes concrete bridge crack images in complex contexts as the object of study and explores the processing method of crack images at various stages to improve the digitalization of bridge maintenance.The details of the research are as follows:(1)Pre-processing of bridge crack images.By comparing the processing effects of various image grayscale and filtering denoising methods,a suitable method is determined to improve the image quality.For the band stripe noise caused by uneven illumination in the line-array image,an automatic dodging method based on the idea of chunking is proposed.The cumulative slope is introduced to chunk the crack image according to the sharpness of the image grayscale change.The image signal-to-noise ratio is used as the filter window size for sub-block dodging.The statistical value of the sub-block image is used to adj ust the grayscale difference.The above processing is used to improve the dodging effect of the method and enhance the adaptiveness.The original contrast stretching method is improved by reducing the stretching origin value to make it more suitable for increasing the crack image contrast.(2)Classification of bridge crack images.Produce a crack image classification dataset for classification model training,validation,and testing.A simple convolutional neural network classification model is constructed for the crack image classification problem.The size of the input layer images,the number of network layers,and the number of convolutional kernels are changed to reduce the model parameters to improve the model training efficiency.(3)Extraction of bridge crack feature parameters.The minimum path selection method is improved to segment crack target regions.Its running time is shortened by reducing the number of path endpoints,modifying the path search strategy,and the calculation process of the minimum path.The processing efficiency is improved when the crack segmentation accuracy is guaranteed.For the protruding points and short burrs in the crack refinement results,the corresponding deletion conditions are set to ensure the single-pixel width of the crack skeleton,the vertex selection rules of the outer polygon of the mesh crack and the calculation method of the maximum crack width are determined to obtain the crack-related feature parameters.(4)The design of the image processing system for bridge surface cracks.The processing methods proposed in this paper are combined to design functional modules for image import,pre-processing,feature parameter extraction and information storage.And provide two processing modes,batch processing and single test.The graphical interface of the processing system is designed based on PyQt5,which makes the processing process at each stage intuitive and easy to use by operators.