基于静止气象卫星（GMS-5）多光谱云图的天气采样数据，分别对各样本数据在红外、水汽及可见光通道的灰度、梯度和纹理高维特征空间的投影点进行聚类分析，以确定诸天气样本在特征空间中的类属区域，进而用其对云图进行天气区的判别分类。针对传统聚类方法存在的缺点，本文采用了模糊C均值聚类（FCM）、遗传算法（GA）和模糊减法聚类（FSC）相互交叉、优势互补的思想，既克服了GA／FCM算法局部／全局寻优的不足，又可客观确定出聚类中心数目。对高维特征空间中的重叠和交叉部分的样本点类属，通过计算其与空间中各聚类中心点的欧氏距离来予以甄别，最后得到高维特征空间中各天气的类属域，实况云图中诸像素点通过计算和判断其灰度-梯度特征量在高维空间中的投影点落区位置，即可确定其天气类属，进而实现对天气区的自动分类。试验结果表明，该方法具有良好的分类效果，判别结果与天气实况基本一致。

According to weather sampling data from static GMS-5 images, the projections of IR1, IR2, VS WV in high-dimension feature spaces such as gray degree, grade degree and veins can be clustered. In this way, we can get to know the subject area of each weather sample in the feature spaces, so that we can get the weather classification of each nephogram. In view of the disadvantages of the conventional clustering algorithm, we develop an idea to combine FCM, GA with FSC mutually. In this way, we can not only overcome the local/the global optimum of GA/FCM algorithm, but also confirm the number of clustering centers objectively. Especially to estimate the classifications of the overlapped samples in high-dimensional feature spaces, we can calculate the distance between these samples and the clustering centers to determine their classifications. The type of the pixels in original cloud images can be found out which group in high-dimensional feature spaces the pixels belong to. So that we can make sure its weather area to accomplish the automatic classifications of the weather area. A lot of experimental test to our method has shown good classification effect and the estimated outcome basically conforms to the weather fact.