Study On The Granular Computing Method Of Remote Sensing Image Classification Based On Cloud Model

Remote sensing image classification is the image classification processing based on the spectral, texture, shape and other characteristics of the remote sensing image. It is the premise and basis of remote sensing image information extraction. Classification results have a direct impact on the accuracy of subsequent processing. Remote sensing image classification is a classic problem of remote sensing information extraction, and it is recognized as a challenging area of research. It can be found that granularity and uncertainty are the two key issues which affect the remote sensing image classification results by summary of the characteristics of remote sensing and improvements of the classification method in recent years.Granularity used to be a physics concept which is used to measure the average size of particle. Granular computing theory borrows the concept which is used to measure the concept or knowledge at different levels and different perspectives. Theory of granular computing simulates problem solving strategies of human which deal with problems in different granularity space. It has a strong ability to quickly deal with the uncertainty of data. Rough sets, quotient space and fuzzy information granularity theory have been improved and developed under the granularity theory. At present, granular computing is a new concept and method of information processing including all relevant theories, methods and techniques of granularity theory. It is mainly used to describe and deal with fuzzy, random, incomplete, and large amounts of information. It provides a kind of problem-solving methods based on granular computing and researchs of relationship between granularities.Based on the research and analysis of basic principles, methods and commonly used model of granular computing, the paper introduces cloud model into the theory systems of granular computing. The framework is built from granulating, granular layer construction and granular computing. The key issues and techniques are discussed and improved. For remote sensing image classification, the method of classification based cloud model is proposed. The low resolution remote sensing images and high resolution remote sensing image classification experiments verify the effectiveness of the method.The main work and innovation are as follows:(1) The paper introduces cloud model into the theory systems of granular computing. The framework is built from granulating, granular layer construction and granular computing, and the application to remote sensing image classification is proposed.(2) Cloud transform algorithm is improved. Heuristic cloud transform algorithm can not calculate He and it uses the membership curve to fit the frequency distribution curve. In order to overcome drawbacks, a cloud transform algorithm based on histogram analysis is proposed. A cloud transform algorithm based on Gaussian mixture model is presented to realize granulating of the multi-dimension data. For dealing with the spatial and structure information of the high resolution remote sensing image, the region segmentation method based on cloud model is proposed to realize region granulating of the high resolution remote sensing image.(3) Three key technologies of granular layer construction cloud model distance calculation, cloud synthesis and termination condition of granular layer construction are improved. A new cloud model distance calculation method is proposed based on fuzzy degree of nearness, and the amplitude cloud synthesis is introduced, termination condition of granular layer construction is used to avoid cloud model atomizing utilizing cloud model distance calculation and amplitude cloud synthesis algorithm.(4) Classification method based on cloud model and fuzzy pattern recognition is proposed for the low resolution remote sensing images classification, and classification method based on cloud model and fuzzy rule reasoning is presented for the high resolution remote sensing images classification.