张军　贺婷婷　侯谨毅　王萍

摘   要：风暴单体是形成各类强对流灾害天气的基本单元，它们的雷达回波形状复杂、内部分布不一、外层相互交织，从而造成单体分割困难. 提出了一种基于形态结构特征的对流单体自动迭代分割方法. 以雷达图片中区域树结构单体分割结果为初始输入，在每次迭代分割过程中，首先计算各个分割结果的3个形态结构特征，然后通过一个预先训练的支持向量机（Support Vector Machine，SVM）分类器判断分割结果是否为对流单体，对非单体的分割结果进行再次分割. 通过3种不同类型的风暴案例进行测试，结果表明，本文方法能够有效地识别出聚集的单体和处于分裂/合并状态的单体，并且能够获得单体的完整结构. 在定量评估测试中，本文算法获得了0.84的临界成功指数评分，高于传统的风暴单体识别与跟踪算法（Storm Cell Identification and Tracking，SCIT）方法（0.55）和单阈值方法（0.49）.

关键词：天气雷达;风暴单体;形态学特征;迭代分割;计算机视觉

中图分类号：TP391.4                        文献标志码：A

Research on Automatic Segmentation Method of Convective

Cell Based on Morphological Structure Characteristics

ZHANG Jun，HE Tingting，HOU Jinyi WANG Ping

（School of Electrical and Information Engineering，Tianjin University，Tianjin 300072，China）

Abstract：Storm cells are the basic units that form various types of severe convective weather. Their radar echoes have complex shapes，uneven internal distribution，and intertwined outer layers，which makes cell segmentation difficult. This paper proposes an automatic iterative segmentation method for convective cells based on their morphological structure characteristics. Taking the cell segmentation results based on the region-tree structure on the radar image as the initial input，in each iterative segmentation process，the three morphological structure features of each segmentation result are first calculated，and then a pre-trained SVM classifier is used to determine whether the segmentation result is a convective cell. The segmentation results that are not cells are segmented again. The method in this paper was tested through three storm cases with different types. The results show that the method can effectively identify aggregated cells and cells in a split/merged state，and can obtain the complete structure of cells. In the quantitative evaluation test，the algorithm presented in this paper obtained a Critical Success Index score of 0.84，which is higher than that of the traditional SCIT method （0.55） and the single threshold method （0.49）.

Key words：weather radar;storm cell;morphological characteristics;iterative segmentation;computer vision

由于氣候变化，近年来对流风暴的频率和强度都显著增加[1-3].  极端的对流风暴事件会造成严重的社会经济损失. 由于对流风暴高度动态的空间和时间过程，对流事件的研究仍然是一个具有挑战性的问题. 天气雷达是监测强对流天气（冰雹、大风、龙卷和暴洪）的主要工具之一，利用天气雷达，可以更详细地分析对流风暴的形成和运动过程.

在天气雷达的反射率强度图像上，对流风暴经常表现为多单体共存的……