Research On Motor Imaging EEG Signal Recognition Method Improved Based On SCNN

The brain is the "headquarters" of the human body,taking charge of various parts of the body activities,the corresponding areas of the brain are activated automatically when people imagine movement.By studying the EEG signals collected from various regions of the brain to judge the user’s intention,the motor imagination technology can realize the communication between the brain and the external machine,so that the external equipment can work with the imagination of the human.At present,some achievements have been made in the research of motor imagery,but there is still room for improvement.For example,the traditional classification model of motor imagery fails to fully consider the time-frequency characteristics and spatial characteristics of EEG signals,resulting in insufficient classification accuracy.The main research work of this paper is as follows:Spatial convolutional neural network(SCNN)model fails to consider the timing property of EEG signals.In this paper,the temporal convolutional network(TCN)model is improved by taking advantage of its large scale of time perception.Although the spatial convolutional neural network model can extract certain time-frequency features from EEG signals,the integrated temporal convolutional network can make full use of the temporal features of EEG signals.The mixed model of spatial convolutional neural network model integrated with the temporal convolutional network model can extract the temporal features of EEG signal from two perspectives.The mixed model is applied to the classification test of BCI IV-2a data set,and the experimental results show that the optimal classification accuracy reaches90.47%,which is 4.16% higher than SCNN.In this paper,the channel attention mechanism(ECA)model was combined to improve the spatial convolutional neural network model.Channel attention mechanism has the ability to extract inter-channel relations locally across channels,and the features that can be extracted by the spatial convolutional neural network model after its integration include the features of inter-channel relations that were not previously available.The spatial convolutional neural network model integrated with ECA was applied to the classification test of BCI IV-2a data sets,and the experimental results showed that the optimal classification accuracy reached 90.87%,which was 4.56%higher than that of the spatial convolutional neural network model.In addition,this paper combines spatial convolutional neural network model,sequential convolutional network model and channel attention mechanism,but the classification accuracy is not as high as the first two improved methods in the classification test of BCI IV-2a data set.