中文摘要：

卫星云图的相似性检索可以帮助预报人员较快地定位历史相似天气状况，提高天气预报的可靠性。文中尝试采用Zemike矩和网格法来提取云图中云的形状区域特征，并对卫星云图进行相似性检索。其过程为：首先对云图进行云地分离的预处理，然后提取云的形状区域特征，并在此基础上生成云图库所对应的特征向量库，接着进行相似度的距离计算和相似性匹配，最后根据距离的排序顺序输出最终的检索结果。通过比较两种方法的检索效果，提出了一种改进的形状特征提取方法一网格内切圆法。实验表明该方法具有以上两种方法的优点，能够得到较好的检索效果，可以有效检索出视觉上相似的云图。

英文摘要：

Satellite cloud image retrieval can help forecasters position similar weather in the past quickly and improve the reliability of the weather forecast. It attempts to use the Zemike moment and grid based method to extract the cloud shape regional characteristics, making similar retrieval for satellite cloud image. The retrieval process is ： first of all, carry out the pretreatment of separation on the cloud image, extract the effective characteristics and generate characteristics of the cloud atlas library. Then carry out similarity measurement and simi- larity matching between a given cloud image and the cloud atlas. Finally, output the final result according to the order. By comparing re- trieval results of the two methods ,put forward an improved algorithm called grid based inscribed circle method. Experiments show that this method owns the advantages of the above two methods and can retrieve cloud images with similar visual characteristics from the his- tory cloud image atlas.