Research On Underwater Structure Cracks Detection Method

As the on-sea infrastructure projects develop,more and more underwater structures have been put into use.However,underwater structures are inevitably damaged in different degree due to the long-term exposure to complicated environment,among which,crack is the main and the most serious defect.Considering the poor visibility and complicated underwater environment,Unmanned Underwater Vehicles(UUV)are used to detect cracks rather than relying on manual detection way.The deep learning method will be applied to the underwater images enhancement detection to improve the underwater image quality and accuracy of underwater structure crack detection.The main research works as follow:Firstly,research on enhancement algorithm of underwater images is made based on underwater image degradation features.Sophisticated image is generated on the increasing number network layers of Pix2Pix generative adversarial network,then upper and lower feature modules are added to the generator to improve generative image quality.Smooth L1 is also introduced to loss function to make the model easier to converge.On the basis of above mentioned things,the underwater image enhancement algorithm framework is developed.The underwater image enhancement data set built on UIEB and EUVP dataset is adopted to test and prove the effectiveness of the underwater image enhancement algorithm framework.Secondly,the target detection method based on deep learning is studied.Based on improved YOLOv4,the algorithm for underwater structure crack detection method is built.The algorithm replaces the backbone feature extraction network of YOLOv4 with the ResNet-50D network,which helps to reduce the amount of network parameters and to improve the detection rate of the model,and introduce DCNv2 to enhance the network’s feature extraction ability for long and narrow irregular cracks.To improve the model detection accuracy,the CBAM is applied to the algorithm.Collected underwater structure crack data set will be employed to verify and make an analysis of the effectiveness of the improved algorithm.Finally,the research on cascade method for crack enhancement detection of underwater structures is made.Cascade model for crack enhancement detection of underwater structures will be completed as the improved underwater enhancement algorithm is embedded to the algorithm of underwater structure crack detection method.The model will automatically enhance image quality and detect cracks on underwater structures.The experiments are respectively carried out on the collected underwater structure crack data set and the synthetic underwater structure crack data set with artificially generated images,and the results show that the cascade model for crack enhancement detection of underwater structures could effectively reduce the leak detection and improve the accuracy.