Research On Deep Sea Pipeline Damage Identification Technology Based On Deep Learning

Offshore oil and natural gas are important resources for human competition,and the deep-sea pipeline is one of the essential basic equipment for offshore oil and natural gas projects.Welding technology is widely used in the production and assembly of deep-sea pipelines,but due to the instability of the welding process and the interference of environmental factors,welding damage defects will inevitably occur in the welds,thereby damaging the quality of deep-sea pipelines.This project has brought huge hidden dangers to future projects.Existing weld defect detection technology combines manual extraction and machine learning,and usually has the problems of low efficiency and low accuracy.Therefore,it is of great significance to research more effective and accurate methods for identifying defects in weld damage.In this paper,based on the ray image of deep sea pipeline weld defects,the method of identifying damage defects is studied.The purpose is to propose an algorithm for identifying damage defects with excellent performance.First,with the help of Internet search and related units,a certain number of deep sea pipeline damage and defect images were obtained,and then they were expanded using data augmentation.Then pre-process the image: wavelet filtering can better eliminate the interference in the image,the histogram normalization method can improve the contrast of the blurred image,the Otsu method can be used to separate the weld and the background area,and the Canny edge detection method The edge of the weld can be detected more clearly,and the Hough transform can realize the detection of the weld area.The area growth method can segment and extract a more complete image of damaged defects,and the size normalization method can unify the weld size.Damaged defective images,thereby creating a sample set of damaged defective images.Using the established image sample set,using traditional machine learning and deep learning methods for research.When using traditional machine learning,first select the eight geometric features of damage defects as the necessary feature parameters of traditional machine learning,and then construct a support vector machine based on RBF kernel function to conduct deep sea damage defect identification research,the experiment reached an average recognition rate of 92.6%.When using convolutional neural networks in deep learning,after exploring the effects of the number of layers and convolutional kernels on the convolutional neural network,an optimal network structure was constructed,and a deep sea pipeline damage defect image recognition test was conducted,the average recognition rate reached 96.8%.Finally,the ZCA whitening and genetic algorithm are used to optimize the convolutional neural network,and then the recognition experiment is carried out,the average recognition rate was 97.8%.The comparison of experimental results proves the superiority of deep learning compared with traditional machine learning,and verifies the effectiveness of the optimization method in this paper.