Underwater Disease Detection Of Pier Based On Deep Learning And 3D Point Cloud

The foundation of the underwater pier is the main bearing structure of the bridge,and its health condition affects the service life of the bridge,so it is necessary to test the underwater pier regularly.The traditional detection method is to detect the structural disease of the pier with the way of exploring artificially,and the detection result is susceptible to the subjective influence of individuals,resulting in the phenomenon of misjudgment and omission,and the detection efficiency is low,in the area with large water flow,bringing hidden trouble to the life safety of divers.With the rapid development of computer technology,more and more scholars pay attention to deep learning technology and apply it to object recognition,detection and other aspects.In order to overcome the defects of traditional detection methods,a lightweight convolutional neural network is built based on the existing technology to detect bridge underwater diseases.At the same time,a bridge underwater disease measurement algorithm based on 3D point cloud is proposed.The 3d point cloud of the disease is obtained by depth camera,and the point cloud is input into the algorithm,and the 3D information of the underwater disease can be calculated.The main contents include the following parts:(1)Preprocessing of pier disease dataFirstly,the size and format of the collected underwater pictures of bridge pier are unified.In order to enable convolutional neural network to better learn the detail features of bridge disease images and improve detection accuracy,this thesis proposes an underwater image enhancement algorithm.Based on MSRCR,CLAHE algorithm is introduced to effectively improve the quality of underwater images and establish high-quality underwater disease data sets.Due to the small number of disease images collected,deep convolution adversarial network DCGAN was built to expand the data set in order to improve the accuracy of network training prediction in the later stage.Labelimg software was used to make the preprocessed samples into data sets.(2)The construction of disease detection network of pierIn order to solve the problem that the original YOLOv4 network is difficult to deploy in mobile equipment for disease detection due to its large number of parameters and large computation,a lightweight network model lite-YOLOV4 is built to improve the YOLOv4 network,which uses deep detachable convolution instead of ordinary convolution to perform multi-feature fusion of prior frames.The underwater diseases under different working conditions are established,and the detection results of this network are compared with those of other common networks.The results show that this network has higher detection accuracy.(3)Construction of binocular vision systemThe types of depth cameras are introduced,and the binocular structured light camera is selected according to the research purpose of this thesis.The camera is designed with an underwater capsule for underwater use.When the camera is used underwater,light refracts at the junction of air and water,which will lead to errors in point cloud information obtained by the camera and affect the measurement results.Therefore,an underwater imaging model of the camera is established,and a high-order distortion correction polynomial is introduced to eliminate the distortion of the underwater image.The underwater calibration of the camera is completed by using Zhang Zhengyou calibration method.According to the solved parameters,the corresponding relationship between 3d points in underwater space and image pixels is determined.(4)Research on disease measurement algorithm of pierAn underwater bridge disease measurement algorithm based on 3D point cloud was proposed.A calibrated camera was used to obtain 3d point cloud of underwater disease,and the point cloud was filtered and denoised.According to the difference of 3D points in X,Y and Z directions,the length,width and depth of the disease were obtained.Poisson surface reconstruction algorithm was used to reconstruct the disease,and Helen formula was used to calculate the area of the disease.Three dimensional convex hull algorithm is used to calculate the volume of damaged pier.The experimental results of real underwater concrete damage show that this method has high calculation accuracy.