基于深度学习语义模型的大尺度泥炭沼泽遥感识别  

Large-scale Remote Sensing Identification of Peat Mires Based on Deep Learning Semantic Models

作　　者：傅显浩 吴艳飞 邱俊杰 王佳珂 杜兰珠 徐俊锋[1,2] FU Xianhao;WU Yanfei;QIU Junjie;WANG Jiake;DU Lanzhu;XU Junfeng(Institute of Remote Sensing and Earth Sciences,Hangzhou Normal University,Hangzhou 311121,China;Zhejiang Provincial Key Laboratory of Urban Wetlands and Regional Change,Hangzhou 311121,China)

机构地区：[1]杭州师范大学遥感与地球科学研究院,浙江杭州311121 [2]浙江省城市湿地与区域变化研究重点实验室,浙江杭州311121

出　　处：《杭州师范大学学报(自然科学版)》2025年第1期27-34,共8页Journal of Hangzhou Normal University(Natural Science Edition)

基　　金：浙江省公益技术研究项目(LGF22D010008).

摘　　要：随着气候变化和人类活动的影响,泥炭沼泽正面临退化的挑战,亟需准确的空间分布信息以便于进行保护、修复和管理.为了精确描绘泥炭沼泽的空间分布,选择芬兰中部2个行政分区作为研究区.采用U-Net、ResUnet和DeeplabV3+3种深度学习语义分割模型,对泥炭沼泽进行高精度的空间信息提取.结果显示,ResUnet模型在划分土路和泥炭沼泽边界方面表现优异,F1指数达到了96.3%.基于ResUnet模型计算得出研究区内泥炭沼泽的总面积约为3348.6 km2,占总面积的17.2%,制图精度达到了88.4%.研究结果为泥炭沼泽的有效保护和可持续管理提供了数据支持和新视角,展示了深度学习技术在环境监测和生态系统管理中的应用潜力.With the impact of climate change and human activities,peat mires are facing the challenge of degradation.There is an urgent need for accurate spatial distribution information for conservation,restoration and management.In order to accurately depict the spatial distribution of peat mires,the study selected two administrative subdivisions in central Finland as the study areas.Three deep learning semantic segmentation models,including U-Net,ResUnet and DeeplabV3+,were used to extract high-precision spatial information from peat mires.The results showed that the ResUnet model performed well in dividing the boundary between dirt roads and peat mires,achieving an F 1 index of 96.3%.Using the ResUnet model,the area of peat mire in the study area was about 3348.6 km 2,accounting for 17.2%of the total area,and the mapping accuracy reached 88.4%.This study not only provides important data support and a new perspective for the effective conservation and sustainable management of peat mires,but also highlights the application potential of deep learning technology in environmental monitoring and ecosystem management.

关 键 词：泥炭沼泽 深度学习 大尺度制图 语义分割 

分 类 号：X87[环境科学与工程—环境工程]