基于深度学习的山地冰川表面裂隙提取方法——以青藏高原东南部雅弄冰川为例  

Deep learning-based method for the extraction of surface crevasse in mountain glaciers:a case study of Yanong Glacier in the southeastern Qinghai-Xizang Plateau

作　　者：段云鹏 吴坤鹏 刘时银 扈芷婷 杨欣 肖乐天 高永鹏 DUAN Yunpeng;WU Kunpeng;LIU Shiyin;HU Zhiting;YANG Xin;XIAO Letian;GAO Yongpeng(Institute of International Rivers and Eco-security,Yunnan University,Kunming 650091,China;Yunnan Key Laboratory of International Rivers and Transboundary Eco-security,Kunming 650091,China;Field Science Observation and Research Station of Yulong Snow Mountain Cryosphere and Sustainable Development,Northwest Institute of Eco-Environment and Resources,Chinese Academy of Sciences,Lanzhou 730000,China;Yunnan International Joint Laboratory of China-Laos-Bangladesh-Myanmar Natural Resources Remote Sensing Monitoring,Kunming 650500,China;Faculty of Geography,Yunnan Normal University,Kunming 650500,China)

机构地区：[1]云南大学国际河流与生态安全研究院,云南昆明650091 [2]云南省国际河流与跨境生态安全重点实验室,云南昆明650091 [3]中国科学院西北生态环境资源研究院玉龙雪山冰冻圈与可持续发展野外科学观测研究站,甘肃兰州730000 [4]云南省中老孟缅自然资源遥感监测国际联合实验室,云南昆明650091 [5]云南师范大学地理学部,云南昆明650500

出　　处：《冰川冻土》2024年第5期1506-1521,共16页Journal of Glaciology and Geocryology

基　　金：国家自然科学基金项目(42361021);云南省自然科学基金项目(202301AT070417,202205AM070011);国家重点研发计划项目(2022YFF0711702-03);云南大学人才引进项目资助。

摘　　要：冰川表面裂隙对于分析冰川的运动状态、稳定性、物质平衡以及内部和表面应力有着非常重要的作用。利用无人机搭载的高分辨率相机进行航测,可以精确获取复杂冰川表面地形的细微特征。本文以青藏高原东南部岗日嘎布山的雅弄冰川为研究对象,使用大疆M300 RTK无人机航拍获取到0.03 m空间分辨率的正射影像,采用深度学习网络模型开展冰川表面裂隙的自动提取研究。结果表明,使用本文提出的CBAM-UNet模型提取冰裂隙的性能优于经典U-Net、DeepLabV3+、PSPNet和HRNet等模型,提取的冰裂隙精确率可达到90.74%。受雅弄冰川运动的影响和地形因素制约,冰川末端主要涵盖了横向裂隙、伸展裂隙、雁列裂隙、边缘裂隙等四种冰裂隙类型以及少量分布的冰崖裂隙,基本成片分布在变化剧烈的冰川区。基于高分辨率无人机影像和深度学习的冰川表面裂隙智能提取方法,可为监测冰川变化及其与气候变化的响应提供新的技术手段。Understanding the dynamics of glaciers is of paramount importance in the context of global climate change,as glaciers serve as sensitive indicators of environmental shifts.They are key components of the Earth’s freshwater reserves,influencing sea level,water resources,and the global climate system.The study of gla⁃ciers,therefore,extends beyond pure scientific inquiry;it has critical implications for predicting future water availability,assessing potential sea-level rise,and developing strategies for mitigating the impacts of climate change.Within this broader framework,glacier surface crevasses offer a window into the deeper processes at work within glaciers.These features are not merely physical anomalies but are indicative of the glacier’s internal and external dynamics.They provide insights into the stress and strain glaciers undergo as they move and de⁃form,offering clues about the glacier’s velocity,the distribution of stresses within the ice,and the interaction between the glacier and its underlying bedrock.Moreover,crevasses influence the energy balance of a glacier’s surface by altering its albedo and affecting its meltwater dynamics,which in turn affects the glacier’s overall mass balance and movement.Thus,the detailed study of crevasses contributes significantly to our understanding of glacier mechanics and their response to climatic variations.This study focuses on the Yanong Glacier,located in the Kangri Karpo Mountain of the southeastern Qinghai-Xizang Plateau,an area noted for its complex terrain and significant glaciological interest.The deployment of Unmanned Aerial Vehicles(UAVs),particularly the DJI M300 RTK UAV,has facilitated the acquisition of high-resolution orthophotos with a ground resolution of 0.03 m.This technological advancement enables the detailed mapping of glacier surface features,which is cru⁃cial for identifying and analyzing crevasse patterns and their distribution across the glacier’s surface.To address the challenges associated with crevasse detection,the st

关 键 词：冰裂隙 正射影像 雅弄冰川 无人机遥感 深度学习 

分 类 号：P343.6[天文地球—水文科学]