Research Of Automatic Classification Technlolgy Based On Mdteorological Satellite Remote Sensing Imagery

Meteorological satellite can provide all-weather conditions of the atmosphere, ocean, and cloud information, and is becoming an important data source for weather monitoring and forecasting. With the development of satellite remote sensing technology and image processing technology, image processing technology of satellite cloud picture has become the main means of meteorological satellite data analysis. Satellite cloud images conveniently provide the cloud information on a wide variety of spatial and temporal scales, and show the types and forms of cloud which contains rich weather evolution information and comprehensively reflect the ongoing dynamic and thermodynamic processes in the atmosphere. Therefore, it is of great practical significance to improve the identification accuracy of satellite imageries cloud types and distribution, which is helpful to severe weather monitoring and weather forecasting.Currently, visual interpretation is still one of the main methods of satellite cloud image identification and analysis, which is based on experts’experience and a variety of materials. However, artificial interpretation contains a certain degree of subjective factors, and is not conducive to the satellite cloud image information extraction. Meanwhile, visual interpretation method is incompatible with the automation and quantified of weather forecast production. Therefore, exploring the satellite cloud images’automatic classification technology is of great significance for promoting and deepening its quantitative recognition work. By the means of the interdisciplinary cross-studies, such as digital image proeessing, data fusion, pattern recognition and satellite meteorology, et al, a research course of satellite cloud images’feature extraction and classification is presented in this thesis. The main research contents and results are as follows.(1) Research of cloud images fusion technology. By researching the image fusion of infrared cloud image and water vapor image, this thesis solves the problem that water vapor image is long-term by less attention in the process of cloud classificaton, and highlight the characteristics of water vapor image. Wavelet transform method is adopted in this thesis, and three fusion schemes are designed. Finally, some criterions are used to evaluate effect of fusion cloud image. It is con2eluded that the fusion cloud image is not only retains the original features of infrared cloud image information effectively, and water features in the fusion cloud image could also be better reflected.(2)Research of cloud feature extraction technology. By researching the spectral characteristics of the cloud type and texture features of fusion cloud image, this thesis proposes identification method based on the spectral and texture features, which solves the problem that previous studies only use a single spectral feature or texture feature for cloud recognition, and feature extraction is not comprehensive. In this thesis, Gray level cooccurrence matrix method is used to extract fusion cloud images’texture features, and digital image processing technology is used to extract the spectral characteristics of the multi-band cloud images. Results of the trial indicate that the selected cloud features are effective and can describe types of clouds and surface.(3)Research of cloud classificaton technology. The weighted minimum distance classification and multi-class SVM methods are designed to establish the automatic cloud type identification model. This thesis improves the problem of traditional minimum distance classifier, and proposes the weighted minimum distance classifier based on the normalized properties. A multi-class SVM method base on one-versus-one combination algorithm is also put forward. Results of the trial indicates that the weighted minimum distance classifier and multi-class SVM classifier both can effectively distinguish the types of clouds, and classification result is optimal.