云的变化复杂多样，在天气预测、灾难预警中发挥着重大作用，影响着人们的日常生活。对云的观测主要通过雷达、遥感卫星和全天空成像仪，记录的云图分为雷达云图、卫星云图和地基云图，三者都是云观测中不可或缺的部分。随着机器学习在多领域的发展，逐渐被运用到云图分割中去并取得了很大的进步。通过广泛调研相关领域的文献和成果，将机器学习的云图分割分为基于神经网络的云图分割方法、基于迁移学习的云图分割方法和基于轻量级模型的云图分割方法这3种类型，对每种类型中近几年提出的方法进行了对比，并进一步总结了云图分割中面对不同问题的改进方法，给出了几个改进方案供参考。

The changes in clouds are complex and diverse, playing a significant role in weather forecast and disaster warning, and affecting our daily lives. The observation of clouds is mainly carried out through radar, remote sensing satellites, and all-sky imagers. The recorded cloud images are divided into radar cloud images, satellite cloud images, and ground-based cloud images, all of which are indispensable parts of cloud observation. With the development of machine learning in multiple fields, it has gradually been applied to cloud segmentation and has made great progress. Through extensive research on literature and achievements in related fields, machine learning cloud segmentation was divided into three types: cloud segmentation methods based on neural networks, cloud segmentation methods based on transfer learning, and cloud segmentation methods based on lightweight models. The methods proposed in recent years for each type were compared, and improvement methods for different problems in cloud segmentation were further summarized. Several improvement schemes were provided for reference.