2020年青藏高原湖泊语义分割数据集  

A dataset of semantic segmentation for the Qinghai-Tibet Plateau in 2020

作　　者：王兆滨[1,2] 高雄 康建芳 艾鸣浩 WANG Zhaobin;GAO Xiong;KANG Jianfang;AI Minghao(School of Information Science and Engineering,Lanzhou Univeristy,Lanzhou 730000,P.R.China;Northwest Institute of Eco-Environment and Resources,Chinese Academy of Sciences,Lanzhou 730000,P.R.China;National Cryosphere Desert Data Center,Lanzhou 730000,P.R.China)

机构地区：[1]兰州大学信息科学与工程学院,兰州730000 [2]中国科学院西北生态环境资源研究院,兰州730000 [3]国家冰川冻土沙漠科学数据中心,兰州730000

出　　处：《中国科学数据（中英文网络版）》2021年第4期69-79,共11页China Scientific Data

基　　金：寒旱区环境演变研究“科技领域云”的建设与应用(XXH13506-103);国家冰川冻土沙漠科学数据中心(No.E01Z7902);中国科学院冰川冻土沙漠科学数据中心能力提升(No.Y9298302)。

摘　　要：青藏高原是我国湖泊分布最密集的两个地区之一。随着全球气候变暖,冰川径流量增加,青藏高原湖泊存在着明显的扩张。湖泊水体的自动提取是进行灾害监测的基础性工作,也对水资源和生态环境保护具有重要的意义。本研究以青藏高原(空间范围大致介于26°00′N–39°47′N,73°19′E–104°47′E之间)为研究区域,以14级(空间分辨率为17 m)谷歌地球遥感影像为基础构建小型的青藏高原湖泊语义分割数据集。本数据集共包含随机选取的6774张大小为256×256的RGB图像,以及结合深度学习数据集标注软件labelme和目视解译方法获得的真值图像(红色为湖泊水体,黑色为背景)。本数据集一方面可以用作青藏高原湖泊水体的提取,为该区域工程建设提供基础的数据支撑;另一方面可以用来预训练语义分割模型的权重,提高其他区域水体提取的精度。The Qinghai-Tibet Platea is one of China’s top 2 regions with the most densely distributed lakes.Due to effect of global warming and increasing glacial runoff,there is a significant expansion the lakes on the plateau has significantly expanded.The automatic extraction of lake water bodies is fundamental for disaster monitoring,and for the protection of water resources and ecological environment.In this research,we studied the Qinghai-Tibet Plateau(spatial range roughly between 26°00′N–39°47′N latitude and 73°19′E–104°47′E longitude),and used the 14-level(17 m spatial resolution)Google Earth remote sensing images as the basis for constructing a small-scale dataset of lake semantic segmentation for the Qinghai-Tibet Plateau.This dataset consists of 6,774 randomly selected RGB images(256×256 in size),as well as ground truth images(lake water bodies in red,background in black)obtained by combining the deep learning dataset annotation software—labelme and visual interpretation methods.On the one hand,this dataset can be used for the extraction of lake water bodies on the Qinghai-Tibet Plateau,supporting the engineering construction in this region.On the other,it can be used to pre-train the weight of the semantic segmentation model and improve the accuracy of water extraction in other regions.

关 键 词：谷歌地球 青藏高原 深度学习 语义分割 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置] P343.3[自动化与计算机技术—控制科学与工程]