Decision level fusion for pulse signal classification using multiple features

With the progress in sensing and analysis techniques, computerized pulse diagnosis has been developed to improve the reliability and consistency in traditional Chinese pulse diagnosis. A number of feature extraction methods have been proposed to extract spatial, frequency features from pulse signal. In this paper, we first extract three kinds of features, spatial, frequency, and similarity features, and then use support vector machine to train three individual classifiers. Finally, we propose a decision level fusion approach to combine these three classifiers for pulse signal classification by using different fusion rules. The proposed method is evaluated on a data set which includes 135 healthy people and 98 patients. Experimental results show that the proposed approach achieves an average classification accuracy of 93.13%.