卫星云图样本集的FCM优化调整与云类判别

基于云类样本的红外-可见光二维灰度空间投影,采用模糊聚类方法调整优化云类样本特征区域,消除采样误差。针对常规模糊C均值聚类(FCM)方法在处理上述问题时表现出的局限性,提出用样本特征均值替代FCM中随机初始中心的改进办法,既避免了常规FCM方法对初始中心敏感的缺陷,又可纠正其聚类结果对云类样本特征结构的歪曲。改进后的聚类结果既消除了采样误差,又保持了云类样本的基本特征属性,基于该判据的分类结果可较为准确地分辨出陆地、水体、低云、中云、卷云和积雨云,分割判别效果符合客观实际。

The FCM Optimized Adjustment and Cloud Classification Based on Satellite Images Samples Sets

Based on the projection in 2-dimension characteristic space of cloud samples, a fuzzy clustering method was presented to optimize and adjust the characteristic region of cloud samples and remove their sampling error. Aiming at the shortcoming of general FCM, a improved method of taking the characteristic value of samples to replace the random initial center of FCM was introduced. Using this method can both avoid the shortcoming of general FCM′sensitivity on initial center and can correct the characteristic structure distortion of cloud samples produced by general FCM clustering. The sampling error was removed and cloud′s basic property was kept in the improved clustering results. Accordingly, the land, water,low-layer cloud,middle-layer cloud, cirrus and cumulonimbus can be clearly classified, and the cloud classification results accorded with the actual weather.