云分类中逐个修改聚类和模糊聚类分类性能的对

利用卫星图像对各种云型进行识别在大气科学领域具有重要意义，为了深入了解云分类过程中逐个修改聚类和模糊聚类对各种云型的识别能力，采用极轨卫星EOS／MODIS图像资料和静止卫星GMS-5图像资料，在样本采集和特征提取的基础上。选择不同的光谱或纹理特征对两种分类器的分类性能进行测试和对比分析。结果发现，不管采用哪种图像资料，提取哪些特征量，逐个修改聚类的平均分类准确率总体上略高于模糊聚类。但就两种分类器对各种云型的识别能力而言，模糊聚类对低云和高云（如层云、薄卷云、密卷云、卷层云、积云）的分类准确率明显好于逐个修改聚类，而逐个修改聚类对积雨云的分类准确率稍高于模糊聚类。从各类别间混判的情形来看，积雨云和高中低混合云、低云之间及卷云子类之间混判的情形较多，模糊聚类与逐个修改聚类相比，混判的类别增多，相对比例减少。

A Comparative Study on Stepwise Cluster and Fuzzy Cluster in Cloud Classification Techniques

In order to profoundly understand abilities of two classifiers--stepwise cluster and fuzzy cluster in the cloud classification techniques, both EOS/MODIS and GMS-5 data set are used, spectral or textural features are drawn from samples randomly to identify various cloud/sur- face. The results show that the stepwise cluster gives higher accurades than fuzzy classifier on the whole. With regards to discriminating diverse cloud/surfaces, fuzzy duster dermnstrates its higher accuracies than stepwise cluster on the classes having similar characteristics such as strums, curnulostratus andcumulus; while stepwise cluster has better capabilities of distinguishing cumulonimbus and surfaces. As far as misclassification of cloud/surfaces, fuzzy cluster tends to show lower accuracies in more misclassified classes.