Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (13): 1-16.DOI: 10.3778/j.issn.1002-8331.2210-0037

• Research Hotspots and Reviews • Previous Articles     Next Articles

Research Progress of Ground Cloud Image Segmentation Method

SHI Chaojun, LI Xingkuan, ZHANG Ke, HAN Leile, YANG Shifang   

  1. 1.Department of Electronic and Communication Engineering, North China Electric Power University, Baoding, Hebei 071003, China
    2.Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding, Hebei 071003, China
    3.Department of Electrical Engineering, North China Electric Power University, Baoding, Hebei 071003, China
  • Online:2023-07-01 Published:2023-07-01



  1. 1.华北电力大学 电子与通信工程系,河北 保定 071003
    2.华北电力大学 河北省电力物联网技术重点实验室,河北 保定 071003
    3.华北电力大学 电力工程系,河北 保定 071003

Abstract: The change and distribution of cloud cover have an important influence on photovoltaic power forecast, astronomical telescope station location and weather forecast. Cloud-based observation is an important way of cloud observation, which mainly reflects the information of cloud base and the distribution, change and movement of clouds in local areas of the sky. Segmentation of foundation cloud image is the foundation of building an automatic observation system of foundation cloud image, so the related research is of great significance. Thanks to the rapid development of deep learning, the general semantic segmentation model has been continuously expanded and applied to the field of ground-based cloud image segmentation, and achieved good segmentation performance. However, considering that different types of clouds have different thicknesses and edges are difficult to distinguish, the segmentation method of ground cloud images based on deep learning still faces severe challenges in accuracy and efficiency. Firstly, from three aspects:threshold, traditional machine learning and deep learning, the segmentation methods of foundation cloud image are comprehensively summarized. Secondly, the data sets commonly used in the segmentation of foundation cloud images are summarized. In addition, the performances of various ground cloud image segmentation methods on GDNCI and WSISEG datasets are compared, and the advantages and disadvantages of various methods in the two datasets are analyzed. Finally, the problems to be solved and the future research direction in the segmentation of foundation cloud images are prospected.

Key words: segmentation of foundation cloud image, threshold segmentation, machine learning, deep learning, ground cloud image segmentation dataset

摘要: 云量的变化和分布对光伏发电功率预测、天文望远镜观测站选址和气象预报等均具有重要影响。地基云观测是云观测的重要方式,是对卫星云观测数据的有效补充,其主要反映天空局部区域云底信息和云层分布、变化及运动情况。地基云图分割是构建地基云图自动观测系统的基础,因此相关研究具有重要意义。得益于深度学习的飞速发展,深度卷积神经网络的通用语义分割模型被不断拓展应用到地基云图分割领域,并取得了良好分割性能。然而由于地基云图内在的特殊性和复杂性,特别是考虑到不同类别云层厚度不同并且边缘难以区分等问题,基于深度学习的地基云图分割方法仍面临着精度及效率等方面的严峻挑战。从阈值、传统机器学习和深度学习三方面出发,对地基云图分割方法进行全面综述;总结了当前地基云图分割常用的数据集;对比了各类地基云图分割方法在GDNCI和WSISEG两种数据集上的性能,并分析了各类方法在两种数据集中的优劣;最后进行了全面总结,并对地基云图分割中待解决的问题与未来的研究方向进行了展望。

关键词: 地基云图分割, 阈值分割, 机器学习, 深度学习, 地基云图分割数据集