格陵兰冰盖典型冰川冰裂隙数据集（2018-2020）

我们提出利用U-net网络进行冰裂隙识别探测的算法，可以实现格陵兰冰盖典型冰川冰裂隙的自动化探测。基于Sentinel-1 IW每年7、8月的数据，为了抑制SAR图像的相干斑噪声，选择Probabilistic Patch-Based Weights (PPB)算法进行滤波，然后选择具有代表性的样本输入U-net网络进行模型训练，根据训练的模型进行冰裂隙的预测。以格陵兰2个典型冰川（Jakobshavn、Kangerdlussuaq）为例分类结果的平均准确率可达94.5%，其中裂隙区域的局部准确率可达78.6%，召回率为89.4%。We propose an algorithm for glacial ice crack detection using the U-net network, which enables automated detection of typical glacial ice cracks on the Greenland Ice Sheet. Based on Sentinel-1 IW data collected in July and August each year, to suppress the speckle noise in SAR images, the Probabilistic Patch-Based Weights (PPB) algorithm is selected for filtering. Then, representative samples are selected and fed into the U-net network for model training, and ice crack prediction is conducted using the trained model. Taking two typical glaciers in Greenland (Jakobshavn, Kangerdlussuaq) as case studies, the average accuracy of the classification results reaches up to 94.5%, with the local accuracy for crack regions up to 78.6% and the recall rate at 89.4%.

搜集汇总

背景与挑战

背景概述

该数据集基于Sentinel-1 SAR数据，利用U-net网络和PPB滤波算法实现了格陵兰冰盖典型冰川（如Jakobshavn和Kangerdlussuaq）冰裂隙的自动化探测，覆盖2018年至2020年，平均准确率达94.5%。数据以shpfile格式提供，空间分辨率为1m-10m，适用于冰川变化研究，共享方式为申请获取。