Feature extraction for distance-based classification of signal sources

In this paper we present a solution for the classification of different patterns from seismic signals generated by different human activities for which an automatic recognition is required. Some a priori known signals were available, which gave us the possibility to represent them in feature space in order to capture their global characteristics. The classification of signals is based on computing the Mahalanobis distance in feature space and finding the smallest distance. At the end of the paper we present the results obtained for 108 signals acquired from 8 different vibration sources, the source of each signal being already known. The approach to feature engineering and feature selection presented in this paper is generic and can be applied to a large class of classification problems.