Variational Methods Applied Research In The Gms-5 Meteorological Satellite Cloud Processing

In this thesis, we study methods of cloud differentiation and cloud classification based on variational PDEs.First, we discuss characteristics of satellite cloud image and methods of cloud differentiation and classification. We review the geometric curve evolution method, the level set image segmentation method proposed by Osher and Sethian, the Mumford-Shah image segmentation method proposed by Mumford and Shah, the level set solution of Mumford-Shah method proposed by Chan and Vese, and the segmentation method of Vector-Valued image based on Mumford-Shah model.Then, for the one-channel image, we put forward an impoved Mumford-Shah model. This model get the more precise and more significative result of image segmention by replaceing the average gray with the kernel gray of object, and adjusting coefficients of the kernel gray. By using this model, we distinguish and classify high cloud and middle cloud or low cloud from cloud image.For the multichannel image, we use the vector-valued image segmentation model based on Mumford-Shah model to deal with the artifical two-channel image, and get the reunion and set difference image of the two channels. Further, we improve this method and get the more precise position of middle cloud and low cloud in the infrared channel and visible-light channel, and distinguish between cumulonimbus and cirrus.