The Study Of City TM Remote Image Classification Methods

Remote sensing(RS)image classification is always a pivotal part of remote sensing study. How to improve the accuracy of RS interpretation is an urgent problem in RS application.Classification of remote sensing image using computer is data disposal of remote sensing image pixels by computer. There are mainly two methods: non-supervised classification, and supervised classification. Non-supervised classification is a clustering process, while supervised classification is a study and training process, and it needs preliminary knowledge. Because non-supervised classification can not ascertain attribute of sorts, its value is very little by direct used and applied. While supervised classification methods emerged in endlessly with development of new methods and technique. From traditional maximum likelihood classification(MLC) based on Byes to decision tree classification and artificial neural network(ANN) classification being studied and used now, they all enhance accuracy, improve effect and application to a great extent. Whereas different ways are different advantage and disadvantage, classification effect is affected by some factors.Based on world RS image classification methods analysis, this paper applied MLC, decision tree and BPNN classification to study classification of the Beijing city's TM RS image. Within classification carried out, first classification samples and features have been researched in detail, which are two absolute necessary steps that affect classification in the process .Afterward this paper concretely introduced experiment of three classification process. Finally this paper evaluated accuracy of different ways according to analysis of classification image, the confusion matrix and Kappa coefficient. Three ways'overall accuracy are all more 80% and so satisfied. By comparison, we find BPNN classification is not prior to else. So it is necessary for us to research of BPNN classification to improve its accuracy.The thesis divides into six parts:The first part is exordium that introduces purpose and meaning of the study, development and status quo of RS image classification, and presents contents and methods of the study. The second is the principal theories of classification technology for RS image that introduces basic principle of process about RS image classification, basic arithmetic of traditional non-supervised classification and supervised classification, methods about appraisal of classification precision that provide theories for research. The third introduces principles and methods ascertaining swatch of typical object, and brings forward methods selecting experimental swatch for classification. The forth introduces the process analyzing and selecting features for classification. The fifth is concrete methods and results exhibiting the whole process of RS image classification and analyzing the efficiency of different classification methods. The sixth is conclusion and expectation summarizing the work and shortcoming that is useful to the future research.