EEG denoising through a wide and deep echo state network optimized by UPSO algorithm

In the complex environment of telemedicine, Electroencephalogram (EEG) signals are easily overwhelmed by noise, which affects the intelligent diagnosis of diseases. Since the time-frequency domain characteristics of some noise in EEG signals are complex and the distribution is unknown, and the spectrum of some noise overlaps with the original EEG signal spectrum, it is difficult to filter those noise by traditional methods. To tackle this problem, and considering the large data characteristics of EEG signals in the context of telemedicine, a wide-deep echo state networks (WDESN) with multiple reservoirs in parallel and stacked configuration is proposed for multivariate time series denoising. Firstly, stacking and paralleling multiple reservoirs, the deep features of signals can be extracted to complete the task of reconstructing signal from the noisy signal. Then, the Uniform Search Particle Swarm Optimization (UPSO) is used to optimize reservoir parameters of WDESN. Finally, the effectiveness of UPSO-WDESN is verified through experiments. Experiment results show that compared with existing models, the proposed UPSO-WDESN model can achieve better noise removal performance, while keeping more nonlinear feature of signals.