Research On Content-based Images Classification Method For Satellite Nephograms

Nephograms, obtained by continuous observing the earth surface and cloud through satellites, can be used to analyze the distribution of the cloud system in a large area, and to study the evolvement rules of weather system. However, with the abundance of the nephograms, the researches and applications of the relevant tools for analyzing and processing the data are severely laggard and we still rely on the manually qualitative analysis. Therefore, it becomes a hotspot that how to extract the implied meteorologic pattern and discover the meteorologic knowledge automatically, speedily and effectively from the masses of nephograms. In allusion to the problem, this thesis studies the satellite nephogram classification, concludes the basic framework and develops a series of researches according to the relative techniques.First, as the basic of nephogram classification, the meteorology and image mining knowledge are summarized, in which some of the important contents are emphasized, such as the classes and characteristics of nephogram, the types of cloud, the criteria to recognize the cloud, the concepts and methods of image mining.Based on that, measures of image clustering quality (MICQ) are developed in view of the definition of clustering analysis and the bi-factors of intra and inter class in data mining. Then the particle swarm optimizer (PSO) is introduced into nephogram classification area. In conjunction with MICQ, a nephogram clustering algorithm based on PSO is proposed, of which the coding and fitness function are projected.In addition, nowadays, many researches of nephogram classification usually focus on some single feature, and the classifications are achieved based on these unique features, correspondingly. This strategy would probably ignore the potential which is provided by the combined classification based on multiple features. In fact, combined classification is not only a trend in pattern recognition field, but an effective method for sure that has been proved. Based on these problems, multiple features of the nephogram data have been extracted in this paper, and according to the theory of information fusion, a classification method of multiple features of nephogram has been constructed. Recurring to the idea of information fusion, the performance of classification has been remarkably enhanced.Finally, the methods are validated through experiments. The results show that they have important sense in the nephogram classification field.