基于无人机正射影像的玉龙雪山白水河1号冰川末端冰裂隙提取

Crevasse extraction at the end of Baishui River Glacier No. 1 in Yulong SnowMountain based on orthophoto unmanned aerial vehicle( UAV) image

冰川是冰冻圈的重要部分,对局地乃至全球气候变化响应敏感。冰裂隙作为冰川表面上显著的特征,对于认识冰川的状态、稳定性以及内部应力有着重要的作用。针对冰川冰裂隙高精度快速识别提取问题,本文以玉龙雪山白水河1号冰川为研究对象,使用无人机航拍获取的0.12 m分辨率的正射影像,应用U-Net深度学习网络开展白水河1号冰川的冰裂隙的智能提取研究。使用U-Net网络提取冰裂隙的精度高于传统的Canny算子以及SVM算法,总体精度可高达93%,且U-Net网络泛化能力强。提取结果表明,白水河1号冰川冰裂隙主要为横向裂隙,伸展裂隙以及雁行裂隙,呈现随海拔降低,冰裂隙逐渐由横向裂隙变化为伸展裂隙的趋势,通过不同时期提取结果对比,发现冰裂隙数量和平均长度均有增加。基于无人机影像和深度学习方法的冰裂隙智能提取研究,可为监测冰川变化及其与气候变化的关系提供技术支撑。

Glaciers,as an integral part of the cryosphere,are highly susceptible to both local and global climate change.Ice crevasses,which are prominent features on the surface of glaciers and the important channels for gla-cier meltwater,play a crucial role in understanding the condition,stability,internal stress and mass balance of glaciers.Mountain glaciers are subject to cloud cover and area limitation,and the spatial resolution of traditional satellite remote sensing data is low,which is difficult to be used for extracting ice crevasses,so there are fewer studies related to ice crevasses on mountain glaciers.In this study,the objective was to address the challenge of identifying and extracting glacier crevasses quickly and accurately.This research takes the mountain glacier:Baishui River Glacier No.1 in Yulong Snow Mountain in Lijiang,Yunnan Province as the research object,and takes the cloud-free orthophotos of the glacier surface with a resolution of 0.12 m in 2021 and 2022 acquired by aerial photography of the DJI M300RTK drone as the data source,and applies the U-Net Deep Learning Net-work to carry out the extraction of ice crevasses of the Baishui River Glacier No.1.The results demonstrate that the U-Net network outperforms traditional methods such as the Canny operator and SVM algorithm in terms of crevasse extraction accuracy.The overall accuracy can be as high as 93%.Fur-thermore,the U-Net network exhibits strong generalization capabilities,which can be used to automatically ex-tract unmanned aerial imagery from different time periods.From the perspective of spatial distribution of ice cre-vasses,the crevasses observed on BRG1 predominantly consist of transverse crevasses,splaying crevasses,and Enéchelon crevasses,which show the typical characteristics of mountain glacier ice crevasses in the low advec-tion lifecycle.As the altitude decreases,there is a gradual transition from transverse crevasses to splaying cre-vasses.From the perspective of temporal change of ice crevasses,comparing the extraction results from differ-ent time periods reveals an increase in the number and average length of crevasses.This proves that the ablation of BRG1 is intensifying,and the glacier mass is gradually losing.The orientation of the ice crevasses was al-most unchanged,indicating that the stress inside the glacier didn’tchange dramatically.In summary,the study of intelligent extraction of ice crevasses based on UAV images and deep learning methods creates new possibili-ties for extracting ice crevasses from mountain glaciers,and can provide technical support for monitoring glacier changes and their relationship with climate change.