Research On Detection Of Submarine Pipeline And Leak Point

Our country is rich in marine oil and gas resources.Submarine pipelines play an important role in the transportation of these resources,which not only reduces the transportation cost,but also makes the transportation process safer.However,when the pipeline is on the sea floor for a long time,it will inevitably be affected by the erosion of sea water and marine organisms,resulting in leakage,which will not only damage the environment,but also cause huge economic losses.At present,there is no fixed and effective ocean pipeline detection and maintenance technology.The traditional technology based on electromagnetic wave and other signals detection has a low accuracy,and it is easy to miss detection.It needs to do secondary inspection,which is not conducive to the work.The new detection method based on seabed optical image,which relies on the traditional optical image detection algorithm to find the line features in the image,is easy to be interfered by external objects,and the accuracy is not satisfactory.However,the deep learning algorithm can automatically extract image features with high accuracy and low detection time,which adds development power to target recognition technology.Therefore,this paper proposes the detection algorithm and segmentation algorithm of underwater pipeline image based on deep learning,and also proposes the stereo matching positioning of underwater image based on binocular vision.The main research contents are as follows:1)Research on underwater environment target detection method based on deep learning.The research content of this part is to use the underwater robot to cruise detect the submarine pipeline in the underwater environment and to deal with the leakage point in time.Due to the low accuracy and slow detection speed of the traditional underwater environment target detection,this paper studies the detection of submarine pipeline based on deep learning,using Faster R-CNN and YOLOv3 detection methods to detect submarine pipeline and leakage point respectively,and the results show that YOLOv3 is significantly better than faster in speed when the accuracy is almost the same.Therefore,this paper chooses the YOLOv3 algorithm as the underwater environment target detection method.2)Research on underwater environment target segmentation method based on deep learning.After detecting the leakage of subsea pipeline in the underwater environment,it is necessary to find out the location of the leakage point accurately for maintenance.In this paper,the image segmentation algorithm based on deep learning is studied.Due to the need to locate and repair the leakage point,each pipeline in the figure needs to be labeled separately.Therefore,the instance segmentation algorithm is used,and MASK R-CNN and YOLACT algorithms are used to segment submarine pipelines and leaks,respectively,and improved on the basis of the two methods,combining the advantages of the two methods,adding a multi-layer feature map to the feature extraction layer to reduce the distance of information flow between high-level and low-level feature maps,using a new loss function and non-maximum suppression method in the segmentation layer,and verifying through experiments that the improved method in this paper has certain advantages over the previous two methods.3)Based on binocular vision,underwater environment submarine pipeline leakage target location research.After finding the leakage point of the pipeline,the actual location of the leakage point should be carried out.In this paper,the target location based on binocular vision is used.Because the underwater brightness is low,we choose the stereo matching algorithm which is not sensitive to the brightness: the census algorithm and the NCC algorithm.By fusing and improving the two algorithms,the matching effect is optimized.The experimental results show that the distance measured by the improved algorithm is closer to the actual distance.