Submarine Pipeline Leakage Detection Method Based On Deep Learning

Submarine pipeline plays an important role in offshore oil and gas resources exploitation,which not only ensures the safe transportation of oil and gas resources,but also greatly saves transportation costs.However,due to the adverse conditions of the ocean or human activities,pipelines may have various defects,such as deformation,cracks and leakage.Pipeline leakage will not only cause huge economic losses to enterprises,but also seriously pollute the marine environment.At present,there are two kinds of traditional detection methods: one is to use the detection of electrical,magnetic,acoustic or pressure wave signals to judge whether pipeline leakage occurs,which is vulnerable to working conditions.In addition,the suspected leakage alarm signals detected according to the signals also need to be reconfirmed by professionals through viewing the surveillance video obtained by ROV.The second method is to detect the optical image of submarine pipeline,and to detect and recognize the pipeline features by filtering and Hough transform.In terms of visual effect,optical image processing method is better,but the current optical image detection method can detect pipelines,but the detection effect for leakage is not ideal.Due to the deep learning method can automatically learn the characteristics of samples,and the detection effect is accurate and efficient,it has great advantages in the field of target segmentation and detection.Therefore,we propose a target segmentation method based on deep learning to determine pipeline leak mode and a leak point target detection method based on deep learning for submarine pipeline optical images,and study the leak point target detection method based on binocular vision in this paper.According to the foregoing,Specific research contents are as follows:Firstly,the method of target segmentation based on deep learning to determine pipeline leakage mode is studied.The pipeline repair methods used in different pipeline leakage modes will also be different.The pipeline leakage modes can be further judged by dividing the leaking areas from the target.Because the pipeline and the leaking areas can be easily distinguished from the seabed background,this paper chooses the semantic segmentation network FCN and U-net with simple network structure and good effect to compare the segmentation experiments.The result table shows that the network structure is simple and the effect is good.U-net segmentation accuracy is better than FCN,so this paper chooses U-net to segment pipeline and leak area,which can further determine the cause of pipeline leak according to the shape of leak area,and provide detailed information for underwater vehicle maintenance.Secondly,leak detection method based on deep learning is studied.Just knowing the type of pipeline and the cause of leakage,but we can not determine the location of the leakage point,and can not accurately repair the leakage pipeline.Therefore,in order to determine the location of the leakage point,in order to meet the real-time detection model and detection effect,the YOLOV3 network is studied and improved in this paper.Firstly,considering the large coverage area of pipeline in submarine pipeline images and the uneven thickness caused by viewing angle,the detection output network of YOLOV3 is improved.The union bounding boxes of several small detection bounding boxes are used to represent the detected pipeline area.In the output part of the network,the area control term is used to restrict the small detection bounding boxes,so that the union bounding can be achieved.The shape of box and pipe is more consistent.Secondly,the location of the leak point is determined according to the logical relationship of the leak point in the pipeline area.Experiments show that the improved YOLOV3 network has greatly improved the detection effect of submarine pipeline and leakage point.Thirdly,the leak location method based on binocular vision is studied.In addition,the maintenance robot also needs to know the spatial coordinates of the leaking point to repair the leaking pipeline accurately.Therefore,based on the above leaking point detection,this chapter uses binocular stereo vision to get the spatial location of the leaking point.Aiming at the problems of blurred submarine pipeline image,fewer feature points and disadvantageous to binocular stereo matching,this paper improves the feature point extraction algorithm SIFT.The experimental results show that the improved algorithm can extract more effective feature points from underwater images than SIFT algorithm.Using the improved SIFT and SGM fusion,stereo matching is carried out,and the distance measured by disparity map is more accurate.Finally,this paper verifies the feasibility of the deep learning-based detection and location method for submarine pipelines and leaks based on the submarine pipeline leakage detection simulation experiment system.