Semi-supervised Classification of Seismic Attributes

Multi-attributes classification has become a standard technique in seismic interpretation. In this paper, we apply transductive support vector machine (TSVM), which is one of semi-supervised classification methods, on seismic attributes for reservoir characterization and hydrocarbon detection. In order to apply the TSVM on classification of real seismic attributes, an unlabelled samples selection strategy is proposed to reduce unlabelled samples by prior knowledge to improve efficiency. Our method exploits both precious well information and soft unlabelled samples to obtain a more reliable classifier. Experimental results on a real data show that the TSVM outperforms the traditional SVM.