基于深度学习和高分二号卫星影像的城市植被信息提取  

Urban vegetation information extraction based on deep learning and Gaofen 2 satellite images

作　　者：段永涛 DUAN Yongtao(Beijing Huaxing Exploration New Technology Co.,Ltd.,Tongzhou,Beijing 101102,China)

机构地区：[1]北京华星勘查新技术有限公司,北京通州101102

出　　处：《测绘技术装备》2024年第2期6-9,共4页Geomatics Technology and Equipment

摘　　要：城市植被是生态文明建设的重要组成部分,准确获取城市植被的分布信息是研究城市气候和地表能量平衡的重要内容。传统的植被信息提取方法精度较低,为解决此问题,本文选取郑州市中原区为研究区,利用高分二号卫星影像,基于Deeplab V3+深度学习模型提取研究区内的植被信息,并与传统的支持向量机、最大似然分类方法和常用的SegNet深度学习模型进行对比。结果表明,利用本文深度学习模型提取植被信息的准确率为87.96%,召回率为91.35%,F1分数为0.90,提取结果明显优于传统分类方法和SegNet深度学习模型,可为城市生态环境评价和规划管理提供一定的技术支持。Urban vegetation is an important part of ecological civilization construction.Accurately obtaining the distribution of urban vegetation is an important content in the study of urban climate and surface energy balance.For the problems that the accuracy from traditional vegetation information extraction methods is not high,Zhongyuan District of Zhengzhou City is selected as the study area in this paper,the vegetation information in the study area is extracted based on Deeplab V3+deep learning model using Gaofen-2 satellite images,and it is compared with traditional support vector machine,maximum likelihood classification method,and commonly used SegNet deep learning model.The results show that the vegetation information extracted by using the deep learning method proposed in this paper is significantly better than traditional classification methods and SegNet deep learning model,its accuracy is 87.96%,the recall rate is 91.35%,and the F1-score is 0.90,it can provide some technical support for urban ecological environment evaluation and planning management.

关 键 词：深度学习 高分二号 城市植被 语义分割模型 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]