浅剖管道图像零样本深度学习自动检测研究  

Zero-sample automatic detection of SBP pipeline images using deep learning

作　　者：高兴国[1] 秦毅超 常增亮 GAO Xingguo;QIN Yichao;CHANG Zengliang(Shandong Electric Power engineering Consulting Institute Corp.,LTD,Jinan 250013,China;School of Geodesy and Geomatics,Wuhan University,Wuhan 430079,China;Institute of Marine Science and Technology,Wuhan University,Wuhan 430079,China)

机构地区：[1]山东电力工程咨询院有限公司,山东济南250013 [2]武汉大学测绘学院,湖北武汉430079 [3]武汉大学海洋研究院,湖北武汉430079

出　　处：《海洋测绘》2024年第5期43-47,共5页Hydrographic Surveying and Charting

基　　金：国家自然科学基金(42176186);国家重点研发计划(2022YFC2808303)。

摘　　要：针对现有海底管道检测通过人工判读浅地层剖面仪(sub-bottom profiler,SBP)影像带来的检测低效和低精度问题,提出了一种顾及成像机理的基于零样本深度学习SBP图像中海底管道自动检测方法。首先研究了SBP工作原理及管道成像特点;其次顾及成像背景及各种实际影响因素,基于成像机理生成了管道样本图形;之后利用生成的管道样本,训练YOLOv5神经网络,构建了SBP图形中管道的检测模型,实现了SBP图形中管道的自动检测,取得了优于90%的正确检测率,提出的方法为基于SBP图形的海底管道自动检测提供了一种新途径。Aiming at the problems of low efficiency and low accuracy in the detection of existing submarine pipeline by manually interpreting the Sub-bottom profiler(SBP)image,this paper proposes an automatic detection method of submarine pipeline in SBP image based on zero sample deep learning,which takes into account the imaging mechanism.Firstly,the working principle of SBP and the characteristics of pipeline imaging are studied;Then,taking into account the imaging background and various practical factors,the pipeline sample graph is generated based on the imaging mechanism;After that,the generated pipeline samples are used to train YOLOv5 neural network,and the pipeline detection model in SBP graph is constructed.The automatic detection of pipeline in SBP graph is realized,and the correct detection rate is better than 90%.The proposed method provides a new way for the detection of submarine pipeline in SBP graph.

关 键 词：海底管道检测 深度学习 浅地层剖面仪 样本生成 YOLOv5神经网络 

分 类 号：P229.52[天文地球—大地测量学与测量工程]