Research On Remote Sensing Image Fusion And Classification Methods Using Support Vector Machine

With the abundance of remote sensing data covering the same earth surface including spatial multi-resolution, multi-spectral, multi-temporal and multi-radiometric resolution since more and more sensor borned on the launched satellites, how to using these multi-source remote sensing data to get information of earth is becoming an important research task internationally.Remote sensing image fusion and classification is thoroughly researched in the thesis, and for the three level of image fusion, i.e., pixel, feature and decision-level, support vector machine (SVM) is chosen. The new methods for image fusion based on DOCKSVM, which is one composite kernel SVM were presented, also, the experiments of remote sensing image fusion were carried on, the results of experiments showed that the new methods have obviously advantages over the classical methods such as Maximum likelihood and BP artificial neural network.Generally, the novelties of the research work in the thesis are as follows:A. One new method for remote sensing image fusion based on support vector machine of DOCKSVM, which is one composite kernel SVM is presented for three level of remote sensing image fusion, i.e., pixel, feature and decision-level.B. Preliminary experiments were carried out to evaluate the new image fusion methods and for application to regional change detection, which results showed that the proposed method of image fusion and classification based DOCKSVM is very promising.The research results of the thesis showed that the new methods based on support vector machine of DOCKSVM have good potential in remote sensing image fusion and classification applications, and is promising to be used to regional change detection. The work done and described in the thesis is preliminary, however, is meaningful for relative research on image fusion, classification and applications in the future.