EEG Signals De-Noising with Wavelet by Optimizing Threshold Based on Fruit Fly Optimization

EEG signal de-noising is the preprocessing part of brain-computer interface (BCI), which provides a relatively pure source for controlling external devices with EEG signals. In this paper, a new combination of threshold and threshold function based on wavelet threshold (WT) de-nosing method with undetermined coefficients is proposed. Using fruit fly optimization algorithm (FOA), these coefficients are determined by the combined fitness function of signal-to-noise ratio (SNR), mean square error (MSE) and smooth factor (S), and the noise in the signal is adaptively removed. Experimental results show that under different noise addition conditions, the wavelet threshold and threshold function determined by FOA are better than the combination of fixed threshold and traditional hard and soft thresholds, and other improved methods. The experiment is carried out using MATLAB simulation software. According to the wavelet basis function and the number of decomposition levels, two experimental conditions are designed to generate simulated EEG signals and add noise respectively, and then obtain reconstructed signals. The highest SNR of our method can reaches 18.0297 . In Condition 1, the overall average SNR of our method is increased by 27.98%, 38.29%, 31.96% 18.36% and 6.29%, respectively, compared with the above comparison methods. In Condition 2, the overall average SNR of our method is 23.67%, 31.13%, 35.33%, 12.53% and 7.45% respectively higher than the above same methods. In addition, FOA can help reconstruct a smoother signal. In Condition 1, the lowest S of our method drops to 0.1735, and the overall average S is 7.86% and 5.80% lower than particle swarm optimization algorithm (PSO) and artificial fish swarm algorithm (AFSA) respectively. The method proposed in this paper can better preprocess the EEG signal, so as to achieve a more accurate BCI.