Data-oriented composite kernel-based support vector machine for image classification

One novel composite kernel based support vector machine (SVM), which is called DOCKSVM (Data Oriented Composite Kernel based Support Vector Machine) is proposed in the paper. SVM have been proved good potential in various studies, and tried to application for pattern classification problems such as text categorization, image classification, objects detection etc. Recently, more and more researches show that SVM is promising in remote sensing image classification. Unlike traditional SVM method, DOCKSVM could integrate the bio-geophysical character into final classification through the composite kernels, which lead to the accuracy improvement of classification results. Firstly method of DOCKSVM is described in detail, then the novel method according to information entropy of training data to evaluate the weighted value of kernels is proposed, finally, preliminary results of application to remote sensing image classification is given which show that it’s good potential tool for remote sensing image classification.