Research Of FY-2E Night Satellite Cloud Images Classification Method Based On GHSOM Network Model

Cloud images which provided by Meteorological satellite has rich climate information, itplays an important role in monitoring weather changes. Various types and distribution ofcloud which presents in cloud images is contributed to improve the accuracy of weatherforecast and the effectiveness of climate monitoring. It is also avoiding the phenomena ofdisaster and providing great convenience for human daily life and travel. Because of thecomplex diversity of cloud, there is a lager discrepancy of cloud types in different areas, sothe cloud classification methods which are applied in satellite cloud images has muchdiversity.Traditional classification methods mainly uses visual features to classify cloud images,however, due to the visual images has no chance to get at night, so the night classified resultsare not very well. This paper is based on the lack of visual features, from infrared andwater-vapor features, using the network-model of Growing Hierarchical Self-Organizing Mapto classify cloud images at night, and the classified results are compared with the traditionalclassification model Self-Organizing Map.This paper firstly introduces stationary meteorological satellite FY-2E, the developmenthistory of GHSOM network model and cloud classification, briefly introduces the kinds ofcloud and its performance features in cloud images, then gives an overview of the basicprinciple and construction process of SOM and GHSOM. The study design in this paper isfocused on the night cloud images and classifier, next using cloud images to introduce theclassification process of two network models, meanwhile, counting and contrasting networkclassification accuracy. At last, this paper lists seven cloud examples from different time,longitude and latitude, after a full of consideration of cloud location, color, outline, size,texture and so on, through contrasting with the cloud classification products of nationalmeteorological center, verifies the classification performance of GHSOM.The thesis is divided into six chapters, and the paper gives classification process ofGHSOM for cloud images at night a detailed description and explanation. In the end, furtherresearch work is prospected.