A feature extraction method suitable for multi-channel sensor data

In order to obtain more significant information, the increasing hyper-dimensional data is acquired from multi-channel sensors, but the amount of data becomes very large. The quality of the data must be reduced for the data processing and transmission. The central problem is how to extract the significant features from these data for these purposes. Optimal discrimination plane (ODP) technique based on Fisher´s criterion method was developed to reduce the data in the paper. The patterns were projected onto the two orthogonal vectors that built up the ODP, and two-dimensional feature vectors were attained and utilized as features to represent the patterns. Electrocardiogram signals were applied to the analysis as an example in this study. A quadratic discriminant function (QDF) based classifier and a threshold vector based classifier were employed to measure the performance of the extracted features, respectively. The results show the proposed ODP is an effective and feasible technique to extract the features from the hyper-dimensional time series data.