Ocular Artifact Removal Model Based On Adaptive Genetic Algorithm Optimization And Stack Sparse Auto-encoder

In recent years,deep learning has been developed rapidly and widely applied in the fields of image processing,speech recognition and natural language processing.In this thesis,the deep learning model is applied to EEG signal processing,and EEG signal denoising is performed using adaptive genetic algorithm-optimized Stack Sparse Auto-encoder network.Since EEG signals can be affected by ocular artifacts,ocular artifacts removal is one of the key steps in the processing of EEG signals.This thesis investigates using the adaptive genetic algorithm-optimized Stack Sparse Auto-encoder to remove ocular artifacts in EEG signals.After the model is trained,the ocular artifacts can be quickly and automatically removed by the model.The work of this thesis is divided into three parts.Firstly,the fuzzy c-means clustering method is used to classify the collected EEG signals,distinguish the periods of strong ocular artifacts from the periods of weak ocular artifacts,and divide the training set and test set of the model.Then,the sparse and stacking operations are introduced into the Auto-encoder to form a Stack Sparse Auto-encoder network.The genetic algorithm is used to optimize the network structure parameters to obtain appropriate parameter combinations and to accelerate the convergence speed of network training.On the basis of the network structure,with the goal of reconstructing the EEG signal,the model is trained with the EEG signal without ocular artifacts.The trained model can achieve the effect of removing the ocular artifacts during the test.Finally,this proposed method used on the EEG data recorded from human subjects and its effect compared with that of Wavelet Transform,Independent Component Analysis(ICA),Principal Component Analysis(PCA)and shallow networks.The results show that the proposed method has the best performance in EEG signal reconstruction and ocular artifacts removal.For the EEG segments containing ocular artifacts,the method can effectively remove ocular artifacts,and the method can preserve the original details of the waveform well in the EEG segments without ocular artifacts.In addition,the method has good generalization ability across subjects.