Acta Scientiarum Naturalium Universitatis Pekinensis ›› 2019, Vol. 55 ›› Issue (4): 692-698.DOI: 10.13209/j.0479-8023.2019.045

Previous Articles     Next Articles

River Extraction from High-Resolution Satellite Images Combining Deep Learning and Multiple Chessboard Segmentation

FANG Haiquan1,†, JIANG Yunzhong2, YE Yuntao2, CAO Yin2   

  1. 1. School of Mathematical Sciences, Peking University, Beijing 100871 2. Institute of Water Resources, China Institute of Water Resources and Hydropower Research, Beijing 100038
  • Received:2018-05-29 Revised:2018-10-23 Online:2019-07-20 Published:2019-07-20
  • Contact: FANG Haiquan, E-mail: fanghaiquan22(at)


方海泉1,†, 蒋云钟2, 冶运涛2, 曹引2   

  1. 1. 北京大学数学科学学院, 北京 100871 2. 中国水利水电科学研究院水资源研究所, 北京 100038
  • 通讯作者: 方海泉, E-mail: fanghaiquan22(at)
  • 基金资助:


Using existing methods to extract rivers, especially the small river from remote sensing images, is liable to be interrupted. The combination of deep learning and multiple chessboard segmentation is applied to river extraction from high resolution remote sensing images. Three GF-2 satellite remote sensing images in mountain area, plain and city are used for experiment. The results show that compared with the existing methods, extracted river by proposed method is more continuous. The small rivers accounts for two pixel widths can also be extracted in GF-2 satellite remote sensing images.

Key words: deep learning, multiple chessboard segmentation, high resolution satellite images, river extraction; convolution neural network (CNN) 


针对目前从遥感影像中提取的河流, 尤其是细小河流容易出现中断的情况, 将深度学习与多次棋盘分割法相结合, 应用于高分辨率遥感影像的河流提取。基于对山区、平原和城市3景高分二号卫星遥感影像的实验表明, 与现有的方法相比, 该方法提取的河流更加连续, 并且能够提取高分二号卫星遥感影像中两个像元的细小河流。

关键词: 深度学习, 多次棋盘分割法, 高分辨率遥感影像, 河流提取, 卷积神经网络(CNN)