基于卫星云图的DBSCAN聚类云团分类方法

DBSCAN Clustering Cloud Classification Method Based on Satellite Images

针对云分类问题提出一种新的云团分类方法.该方法先利用风云二号静止气象卫星实时云图图像资料建立多种云和地表类型的样本库,提取分析已知样本的光谱特征和纹理特征;再使用中值滤波器对云图进行预处理,并采用具有噪声的基于密度的聚类算法对云区聚类;最后对聚类得到的云团光谱特征和纹理特征进行匹配,确定云团所属的云类别.实验结果表明,该方法以云团为单位进行划分,易实现云团分类自动化.

According to the cloud classfication problem,we put forward a new cloud classification method.Firstly,we established a sample database of multiple clouds and surface types by using realtime cloud image data of FY2 geostationary meteorological satellite,and extracted the spectral features and texture features of known samples.After pretreating the cloud image by median filter,we clustered on the cloud area by using an algorithm based on density clustering algorithm with noise.Finally,we matched spectral features and texture features of the cloud,and determined the type of cloud.The experiment shows that the method,with clouds as the unit,is easy to realize automation of cloud classification.