FBDM based time-frequency representation for sleep stages classification using EEG signals

In this paper, we have proposed a new method of time-frequency representation (TFR) which is based on the Fourier-Bessel decomposition method (FBDM). This proposed method is an advanced version of the existing Fourier decomposition method (FDM). The proposed method decomposes the non-stationary signal into a finite number of Fourier-Bessel intrinsic band functions (FBIBFs). The FBIBFs are the real parts of analytic FBIBFs (AFBIBFs) which are obtained from an analytic signal during frequency scanning (FS) operations. The Hilbert transform (HT) is used to generate an analytic signal from the Fourier-Bessel series (FBS) expansion of an arbitrary signal. In addition to FBDM, we have also proposed zero-phase filter-bank based FBDM in order to get fix number of FBIBFs in this work. The performance of the proposed FBDM has been evaluated with the help of Poverall measure and TFR analysis of synthesized signals. The experimental results and performance measures show that the proposed FBDM is more capable for analysis of non-stationary multi-component signals such as linear frequency modulated and nonlinear frequency modulated signals as compared to the existing methods. The developed FBDM has also been used for the classification of six different sleep stages using electroencephalogram (EEG) signals. The convolutional neural network (CNN) classifier has been utilized for the classification of TFR images, which were obtained with the application of FBDM on a publicly available sleep EEG signals database. The developed classification system has achieved 91.90% classification accuracy for the classification of six different sleep stages using EEG signals.

Automated summary

This paper presents a new method of time-frequency representation based on Fourier-Bessel decomposition, which decomposes non-stationary signals into a finite number of Fourier-Bessel intrinsic band functions using zero-phase filter-bank. Experimental results show that this method is more effective for analyzing non-stationary multi-component signals.