Maximum Overlap Discrete Wavelet Transform and Improved Cyclegan Based EEG Signal Processing

To evaluate brain electrical activity EEG signals are utilized and diagnose a variety of neurological conditions. Although because of its modest amplitudes, it is frequently contaminated by noise, particularly when it comes to physiological distortions from recording systems, such as myogenic, cardiac, and ocular artifacts. This kind of physiological artifact has the capability to seriously impair information of cerebral, which could be misinterpreted in practical applications such as brain-computer interfaces as normal phenomena. Especially for real-world applications, obtaining high-quality training data for deep learning techniques based on EEG denoising is very expensive. A ship rescue optimization-based cycle Gan is developed in this proposed approach for EEG signal denoising. Using Common Averaged Reference (CAR) and Extended Kalman Filter (EKF) the input EEG signal is initially pre-processed for motion artifacts removal and baseline removal. Higher-Order-Linear Moment-based Singular Spectrum Analysis (HOL-SSA) technique is used for decomposing the EEG signal. To denoise the first level of the decomposed signals Maximum Overlap Discrete Wavelet Transform (MODWT) is used, such as high and low frequency signals. For secondlevel denoising deep learning approach termed Cycle GAN with Ship Rescue Optimization (SRO) is used. Lastly for reconstruction to generate the final denoised signal the Inverse MODWT is applied to the signal. The proposed method achieves performance such as PSNR 25.13 dB and SNR 17.86 dB. In reducing motion irregularities from EEG data to improve the analysis of responses to auditory stimuli this proposed method performs better.