An Entropy-Based Index Evaluation Scheme for Multiple Sensor Fusion in Classification Process

Abstract

Sensor fusion aims to identify useful information to facilitate decision-making using data from multiple sensors. Signals from each sensor are usually processed, through feature extraction, into different indices by which knowledge can be better represented. However, cautions should be placed in decision-making when multiple indices are used, since each index may carry different information or different aspects of the knowledge for the process/system under study. To this end, a practical scheme for index evaluation based on entropy and information gain is presented. This procedure is useful when index ranking is needed in designing a classifier for a complex system or process. Both regional entropy and class entropy are introduced based on a set of training data. Application of this scheme is illustrated by using a data set for a tapping process.