基于语义分割的侧扫声纳管线目标检测方法  

A target detection method for underwater pipeline in side scan sonar images based on semantic segmentation

作　　者：郑根 徐会希[3] 赵建虎[4] 杨文林 ZHENG Gen;XU Huixi;ZHAO Jianhu;Yang Wenlin(Guangzhou Industrial Intelligence Research Institute,Guangzhou 511458,China;Guangdong Institute of Intelligent Unmanned System(Nansha),Guangzhou 511458,China;Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang 110169,China;School of Geodesy and Geomatics,Wuhan University,Wuhan 430079,China)

机构地区：[1]广州工业智能研究院,广东广州511458 [2]广东智能无人系统研究院(南沙),广东广州511458 [3]中国科学院沈阳自动化研究所,辽宁沈阳110169 [4]武汉大学测绘学院,湖北武汉430079

出　　处：《海洋测绘》2024年第2期9-13,共5页Hydrographic Surveying and Charting

基　　金：广东省自然资源厅海洋六大产业专项项目(GDNRC[2023]32)。

摘　　要：为提高侧扫声纳图像中管线目标检测的自动化程度及效率,提出了一种基于语义分割的水下管线目标检测方法。首先通过构建高效语义分割网络主干,提高网络计算速度并降低网络对计算机硬件性能的需求;其次给出了一种针对管线目标特点的加权交叉熵损失函数,解决了因类间数量不均衡导致的网络训练困难问题。以多种复杂条件下侧扫声纳实测数据进行了水下管线检测试验,结果表明,该方法在取得和经典网络相近精度的情况下,速度提升了2.7倍,可达52.6FPS,实现了水下管线的快速、准确检测。To improve the automation and efficiency of pipeline target detection in side scan sonar images,a underwater pipeline target detection method based on semantic segmentation is proposed.Firstly,the network computing speed is improved and the demand for computer hardware performance is reduced by building more efficient network backbone.Secondly,a weighted cross entropy loss function is proposed to solve the problem of network training difficulties caused by the imbalance of the number of classes according to the characteristics of pipeline targets.Underwater pipeline detection experiments were conducted using measured data under various complex conditions.The results showed that the proposed method achieved a speed increase of 2.7 times,reaching 52.6 FPS,and achieved fast and accurate detection of underwater pipelines with similar accuracy as classical networks.

关 键 词：水下目标检测 侧扫声纳图像 深度学习 语义分割 网络优化 类间不平衡 

分 类 号：P229[天文地球—大地测量学与测量工程]