A novel time-frequency feature extraction method of EEG signals utilizing fractional synchrosqueezing wavelet transform.

In order to improve the accuracy of Electroencephalogram (EEG) classification, Fractional Synchrosqueezing Wavelet Transform (FSSWT) is proposed to effectively overcome the contradiction between energy concentration and frequency separation in traditional time-frequency analysis methods. Firstly, the principle of FSSWT is introduced, and the time-frequency transformation equation for FSSWT applied to multi-frequency signals is established. The examples of synthetic signal and EEG signal show that the proposed method can suppress the mode aliasing of MI-EEG significantly while maintaining high resolution characteristics, and the energy concentration and related intermediate indexes perform well. The experimental results show that the proposed FSSWT-EEGDNN-ResNet model achieves an average classification accuracy of 95.17% under the condition of the MI-EEG signals processed by FSSWT of eight subjects, demonstrating the effectiveness of FSSWT in EEG signal feature extraction and classification.