An improved multi-scale feature extraction method for nonlinear signals.

Abstract

This paper proposes an innovative multi-scale feature extraction method for analyzing electroencephalogram (EEG) and electrocardiogram (ECG) signals. The method utilizes an energy functional derived from the Cahn-Hilliard (CH) phase field equation to extract features, aiming to improve classification accuracy. To validate its effectiveness, we integrate the extracted features with a Support Vector Machine (SVM) classifier, forming the CH-SVM model for both EEG and ECG classification. The proposed method achieves an accuracy of 97.14% for EEG and 92.65% for ECG. Compared to conventional convolutional neural network (CNN) models, it demonstrates a significant reduction in computational cost. Furthermore, in comparison to the traditional multi-scale feature extraction method-Multifractal Detrended Fluctuation Analysis (MF-DFA)-the proposed method improves EEG classification accuracy by 5.84% and ECG classification accuracy by 5.15%. These results highlight the superior performance of the CH-SVM method in biomedical signal classification, offering both enhanced accuracy and computational efficiency.