Investigation Of Crack Detection Of Bridge Structures Using Image Processing And Deep Learning

Bridge health condition structure is closely related to economy and safety.In-service bridges are encountered with the coupling effect of harsh natural environment,stochastic external load,material deterioration,fatigue effect,etc.As time goes on,bridge cracks will extend and spread under the load effect or the external environment,which will reduce the service life.Traditional ways of crack inspection are time-consuming and inefficient.And they are easily affected by the noise.A method based on the image processing and deep learning is proposed to detection the bridge crack in this thesis.And the detection results could be the foundation for the bridge safety and durability condition assessment.The main research contents are as follows:(1)The purpose and importance of bridge crack detection is stated.The research progress of crack detection is reviewed and the research work of this thesis is given.(2)The image reconstruction process based on the Bayesian compressed sensing approach is presented.The crack image reconstruction is conducted based on the discrete cosine transform and discrete wavelet transform.The reconstruction performance is estimated with the different sampling ratio using the peak signal to nosie and the reconstruction error.(3)The bridge crack detection is conducted based on the digital image processing method.Preprocessing the image with the gray level transformation and filtering,image segmentation is realized by the Ostu and the edge detection.The crack is distinguished through the feature extraction.Comparing the algorithm performance of the Sobel detector,LOG(Laplacian of Gaussian)detector and the Canny detector based on the crack images and the lighting's effect on the detection result is analyzed.(4)Automatic crack detection is conducted by the deep learning method.A fully convolutional network(FCN)called Ci-Net for structural crack identification is proposed.Pixel-level labeled image training data are obtained from the online dataset.Precision rate,recall rate,intersection over union(IoU)and F-measure are adopted to evaluate the performance of the trained Ci-Net.The proposed network is verified using the in-service bridge crack images.(5)The recognition results of the Ostu algorithm,edge detection approach and Ci-Net are compared,employing the crack images from an indoor concrete beam test and in-service.The sensitivity analysis to the disturb factor also is carried out.It indicates that Ci-Net exhibits a better performance over the edge detection methods and the Ostu algorithm in bridge crack detection.