A novel parallelized remote sensing image SVM classifier algorithm

As an effective classifier, Support Vector Machine (SVM) has received more and more attention in remote sensing communities. Spatial data and remote sensing image may have a lot of features, but too many spatial-features usually lead to the reduction of training speed and the decrease in classification accuracy. To solve this problem, our study presents a novel parallelized remote sensing classifier (NRSC) by utilizing rough set theory and SVM algorithm. In our algorithm, feature set is firstly split sub-feature sets, each sub-SVM classifier is then trained paralleled by these sub-feature sets, and sub-SVMs vote their decisions to avoid the instable performance a single classifier. The experimental results show that both the classification accuracy and training speed, compared with the traditional SVM method, are all improved in remote sensing image classification.