测绘通报 ›› 2023, Vol. 0 ›› Issue (6): 180-183.doi: 10.13474/j.cnki.11-2246.2023.0191

• 测绘地理信息技术应用案例 • 上一篇    

基于高分辨率遥感影像的建筑物提取

王丽梅1,2, 王延正3   

  1. 1. 河北省水文工程地质勘查院(河北省遥感中心), 河北 石家庄 050021;
    2. 河北省遥感中心, 河北 石家庄 050021;
    3. 北京图源科技有限公司, 北京 100192
  • 收稿日期:2023-03-16 修回日期:2023-04-19 发布日期:2023-07-05
  • 作者简介:王丽梅(1980-),女,高级工程师,主要从事空间地理信息与地图制图工作。E-mail:420545734@qq.com

Buildings extraction based on high-resolution remote sensing imagery

WANG Limei1,2, WANG Yanzheng3   

  1. 1. Hydrology Engineering Geological Exploration Institute of Hebei Province, Shijiazhuang 050021, China;
    2. Hebei Provincial Remote Sensing Center, Shijiazhuang 050021, China;
    3. Beijing MapCore Technology Co., Ltd., Beijing 100192, China
  • Received:2023-03-16 Revised:2023-04-19 Published:2023-07-05

摘要: 高分辨率遥感影像不仅具有丰富的光谱、空间分布、形状和纹理特征,也包含清晰的场景语义信息。本文以安徽省枞阳县枞阳镇为研究区域,以高分辨率影像为基础数据源,利用eCognition软件中深度学习与面向对象相结合的方法进行建筑物自动提取。结果表明,该方法具有更好的建筑物提取效果,总体分类精度达96.8%,可用于通过高分辨率影像进行建筑物提取的生产。

关键词: 深度学习, eCognition, 多尺度分割, 面向对象影像分析, 卷积神经网络

Abstract: High-resolution remote sensing images not only have rich spectrum, spatial distribution, shape and texture features, but also contain clear scene semantic information. Taking Zongyang town, Zongyang county, Anhui province as the research area, and using high-resolution images as the basic data source, the deep learning and object-oriented method in eCognition software is used to automatically extract buildings in this paper. The results show that the method of combining deep learning with object-oriented has a better effect of building extraction, and the overall classification accuracy reaches 96.8%, which can be used for building extraction production based on high-resolution images.

Key words: deep learning, eCognition, multiresolution segmentation, object-based image analysis, convolutional neural networks

中图分类号: