Remote Sensing Image Classification With Multiple Classifiers Integration Based On Voting

Remote sensing images provide a rapid, real-time geographic information with a widerange of areas in Geographical Conditions Monitoring. It is an important problem forGeographical Conditions Monitoring to extract features from a large number of images. Withthe increasing of precision requirements, the technology of remote sensing imageclassification has attracted extensive attention of researchers. As the difference in emphasisand theory, the classifiers have some differentiations between each other. Integration ofmultiple classifiers is a method to solve a given problem with several classifiers and can makegood use of the characteristics of the different classifiers, It also improve the learningperformance of the system through the combination with each other to make up the lack of asingle classifier. The application of remote sensing image classification has become a hottopic to improve the classification precision of remote sensing image.The key to realize multiple classifiers integration is the differences among the baseclassifiers and can complement each other. How to choose the base classifiers and how tointegrate them is a key problem to multiple classifiers integration. Therefore, in this paper, theintegration of multiple classifiers based on voting method is given, and do a lot of researchwith it.The main contents and achievements of the research in this paper includes the followingaspects:1. The paper introduces the background and the research progress of integration ofmultiple classifiers, and summarized the main problems at present in the research.2. The basic theory and working mechanism of multiple classifiers integration areexpounded comprehensively and based on this the differences between classifiers are alsostudied in the research and finally come to the conclusion that the performance of multipleclassifier integration is related to the differences between classifiers and complementationamong them. The measure index of differences between classifiers is also be studied in thispaper, including the double measure index and no double measure index, such as entropy,KW-difference and Kappa index. All what had done is to give a general method to measurethe performance of multiple classifier integration.3. Some algorithms which can be used to classify the remote sensing images arerecommended in this paper, the advantages and disadvantages of each algorithm are analyzedand the methods to realize them are given. At the same time, and the paper also give aresearch to the Object-oriented remote sensing image classification, and come out the generalmethods to Object-oriented remote sensing image classification. After that, do the experiments of Object-oriented remote sensing image classification by using differentalgorithms to validate the complementarity between each classifier.4. This paper gives a clear describe of the basic principle and method of the technologyof multiple classifier integration based on voting and realize it by changing structure of thesample and using different classifiers. Firstly, the author improved the AdaBoosting algorithmby using the sample data to train some basic classifiers which adapt different classes andcompleted the classifier integration with expert voting. Secondly, with the prior knowledgethe author put forward three multiple classifiers integration algorithm based on differentclassification algorithms: vote based on recognition performance matrix（VRPM）, vote basedon classes weight（VCW）and vote based on all of information relevance (VAIR).5. At the last, the author has done some experiment to prove the relationship between thedifference and the performance of multiple classifier integration and by using the index whichcan measure the difference between classifiers, some optimal classifier combinations waregotten. After this, the author used VRPM,VCW,VAIR and AdaBoosting to classify remotesensing images and proved the effectiveness and feasibility of multiple classifiers integrationbased on voting.