融合残差模块与SegNet的多波束影像人工鱼礁检测  

Artificial reef detection method for multi-beam sonar imagery by combining residual module with SegNet

作　　者：冯义楷 董志鹏 刘焱雄 杨龙 王艳丽[3] FENG Yikai;DONG Zhipeng;LIU Yanxiong;YANG Long;WANG Yanli(The First Institute of Oceanography,Ministry of Natural Resources,Qingdao 266061,China;National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology,Xi’an 710072,China;The College of Geodesy and Geomatics,Shandong University of Science and Technology,Qingdao 266590,China)

机构地区：[1]自然资源部第一海洋研究所,山东青岛266061 [2]空天地海一体化大数据应用技术国家工程实验室,陕西西安710072 [3]山东科技大学测绘与空间信息学院,山东青岛266590

出　　处：《海洋测绘》2024年第5期14-17,23,共5页Hydrographic Surveying and Charting

基　　金：山东省自然科学基金(ZR2022QD104;ZR2023QD113);青岛市自然科学基金(23-2-1-73-zyyd-jch);山东省博士后基金(SDCX-ZG-202202041)。

摘　　要：针对如何从多波束声纳影像中自动、准确地提取人工鱼礁区域的问题,给出一种融合残差模块与SegNet的多波束声纳影像人工鱼礁检测方法。在该方法中,残差模块用于优化训练网络参数,防止网络训练出现“梯度弥散”和“梯度爆炸”问题;SegNet架构具有网络编码阶段与解码阶段结构对称优势,可实现网络输入影像与输出分类结果的等尺度对称。该方法通过融合残差模块优化训练网络参数与SegNet架构结构对称优势,实现多波束声纳影像中人工鱼礁检测。通过对不同类型的多波束人工鱼礁影像测试验证,该方法可获得86.85%F1-socre和76.75%IoU的检测结果,准确地检测出多波束声纳影像中人工鱼礁区域。With respect to how to automatically and accurately extract artificial reef areas from multi-beam sonar images,an artificial reef detection method for multi-beam sonar images by combining residual module with SegNet is proposed in this paper.The residual module can optimize the training network parameters to prevent the gradient dispersion and explosion in network training.The SegNet framework has the advantage of symmetry in the structure of the network encoding stage and decoding stage,and can achieve proportional symmetry between the network input image and output classification result.In this paper,the artificial reef detection in multi-beam sonar images is realized by integrating the residual module optimization network parameters and the symmetrical advantage of the SegNet structure.By testing different types of multi-beam artificial reef images,the proposed method can obtain 86.85%F1-socre and 76.75%IOU detection results,and accurately detect the artificial reef areas in the images.

关 键 词：多波束声纳影像 人工鱼礁检测 卷积神经网络 残差模块 海洋牧场 

分 类 号：P229.3[天文地球—大地测量学与测量工程]