Fluctuation-based reverse dispersion entropy and its applications to signal classification

The recently proposed fluctuation-based dispersion entropy (FDE) can distinguish various physiological states of biomedical time series, and is usually used in the field of biomedicine. Inspired by the theory of FDE, we redefine FDE and reverse dispersion entropy (RDE), and propose fluctuation-based reverse dispersion entropy (FRDE), which is an improved method of FDE and RDE. As a complexity feature, FRDE is first applied to signal classification combined with K-Nearest Neighbor (KNN), and then a novel signal classification method is proposed based on FRDE and KNN, called FRDE-KNN. We combine dispersion entropy (DE), permutation entropy (PE) and FDE with KNN to get three classification methods of DE-KNN, PE-KNN and FDE-KNN respectively, and then comparative experiments based on these four classification methods are carried out, the experimental results show that FRDE can represent the complexity of signals and have the better separability; and FRDE-KNN has higher classification recognition rate than DE-KNN, PE-KNN and FDE-KNN, which can better classify the ship signals and gear fault signals. (C) 2020 Elsevier Ltd. All rights reserved.

Automated summary

The study introduced fluctuation-based reverse dispersion entropy (FRDE) as a complexity feature for signal classification, combined with K-Nearest Neighbor (KNN), showing better separability and higher classification recognition rate. Compared to other classification methods, FRDE-KNN is more suitable for classifying ship signals and gear fault signals.