| Brain Computer Interface(BCI)is one of the most representative and cutting-edge technologies in the field of science and technology,which is widely used in many fields and has strong practical significance.At present,EEG signals are random and complex,and a single model cannot take into account the feature extraction of EEG signals in multiple dimensions,resulting in low recognition and classification accuracy.At the same time,the deep learning network model of EEG signal is more complex,which is not conducive to the future application of BCI system in embedded direction.To address the current problems,this paper investigates the EEG signal classification and recognition algorithm.The main work of this paper is as follows:(1)In this paper,a hybrid CBCNN-Bi LSTM model based on the attention mechanism is proposed to address the problem that the current single model cannot take into account the feature extraction of EEG signals in multiple dimensions,which leads to the low accuracy of EEG signal classification.By constructing the hybrid model and introducing the idea of attention mechanism at the same time,the model can be realized to capture both channel attention and spatial attention to improve the classification accuracy of this model.The BCI Competition IV public dataset is used for the experiments,and band-pass filtering and independent component analysis are chosen to eliminate the noise interference existing in the experimental data itself.The model is compared with the experimental results of other four types of classical deep learning models by ablation experiments,and the average accuracy of the experimental results on the two data is 74.78% and 76.45%,respectively,and the classification effect is better than the existing algorithms.(2)In this paper,a CBMobile-Bi LSTM network model is proposed to address the problem that the deep learning network model of EEG signals is more complex and unfavorable to the future application of BCI system in embedded direction.The model introduces a lightweight Mobile Net network based on the CBCNN-Bi LSTM hybrid model,which can improve the classification accuracy and reduce the model volume at the same time.The BCI Competition IV public dataset is used for the experiments.By comparing the experimental results with the CBCNN-Bi LSTM model,the average accuracy on both data is higher than that of the comparison model,and the classification accuracy of EEG signal recognition is improved and the model volume is reduced.By comparing the experimental results with the classical deep learning model,the average accuracy of the experimental results on both data are 76.23% and77.72%,which are higher than the average accuracy of other models.The experiments prove that the introduction of Mobile Net network can improve the classification accuracy. |
