基于DeepLab v3+模型的高分辨率影像养殖用海信息提取  

Extraction of Aquaculture Sea from High-Resolution Remote Sensing Imagery Based on DeepLabv3+Model

作　　者：何显锦 蒙方鑫 HE Xian-jin;MENG Fang-xin(Key Laboratory of Environment Change and Resource Use in Beibu Gulf,Ministry of Education(Nanning Normal University),Nanning,Guangxi 530001;Guangxi Key Laboratory of Earth Surface Processes and Intelligent Simulation,Nanning,Guangxi 530001;School of Geography and Planning,Nanning Normal University,Nanning,Guangxi 530001)

机构地区：[1]北部湾环境演变与资源利用教育部重点实验室(南宁师范大学),广西南宁530001 [2]广西地表过程与智能模拟重点实验室,广西南宁530001 [3]南宁师范大学地理科学与规划学院,广西南宁530001

出　　处：《安徽农业科学》2022年第12期208-213,共6页Journal of Anhui Agricultural Sciences

基　　金：广西高校中青年教师科研基础能力提升项目(2019KY0406);广西自然科学基金项目(2020GXNSFAA159060);广西创新驱动发展专项资金项目(桂科AA18118025)。

摘　　要：及时、精确的养殖用海信息探测对于海洋利用开发与管理至关重要,而传统机器学习因不能自动提取原始信息特征,常导致信息提取精度不高、边界不完整等问题。鉴于此,提出采用DeepLab v3+模型进行养殖用海信息提取,重点关注该模型在高分辨率复杂影像中养殖用海信息的提取效果。结果表明,无论是单一时相影像还是由多时相构成的拼接影像,DeepLab v3+模型对养殖用海信息提取的精度均达到了97%以上,明显优于基于传统机器学习方法分类精度;在养殖网箱、排筏等非均质地物提取细节上,DeepLab v3+模型能够完整提取地物,边界也较为规则,可为用海监测与管理提供科学技术手段。Timely and accurate detection of aquaculture sea information was essential for the utilization and management of marine.However,traditional machine learning cannot automatically extract the original information features of objects,leading to low precision and incomplete boundaries of extracted objects.A method to extract aquaculture sea information was proposed by DeepLab v3+model.The extraction effects of this model in high-resolution complex images were mainly researched.The results indicated that the accuracy of the DeepLab v3+model for extracting marine information for aquaculture reached more than 97%for both a single-phase image and a mosaic image composed of multiple time phases.The classification accuracy was obviously better than that of the Maximum Likelihood Classification in ENVI.For the extraction of aquaculture cages and rafts,DeepLab v3+model could completely extract these objects with regular boundaries.This research provided methodological guidance for sea monitoring and management.

关 键 词：DeepLabv3+模型 信息提取 养殖用海 钦州湾 

分 类 号：S951[农业科学—水产养殖]