The Research On Motor Imagery EEG Signals Recognition Algorithm Based On Simplified Convolution Neural Network

Brain computer interface(BCI)is a communication and control system established between human brain and electronic equipment,which has very important applications in many fields.Motor Imagery Electroencephalogram(MI-EEG)analysis is one of the hotspots of brain computer interface research.At present,the comprehensive performance of deep learning algorithms to classify motor imagery EEG signals needs to be improved.This article intends to improve the classification accuracy of two types of motor imaging EEG signals and enhance the comprehensive performance of standard convolutional neural networks.Combined with the characteristics of motor imagery EEG signals,the shortcomings of traditional preprocessing methods and standard convolutional neural networks are analyzed in detail through experiments and improved methods are proposed.The main work of this article is as follows:First,introduces the composition,research status and problems of the brain-computer interface system,introduces the physiological mechanism,composition,characteristics and classification of EEG signals,and explains the data of motor imaging EEG signals used for testing algorithms.Second,when using traditional methods such as common spatial pattern(CSP)and fast Fourier transform(FFT)to preprocess the EEG signals of motor imagery,and use convolutional neural networks(CNN)for classification,the traditional preprocessing method can not balance the overall and local characteristics of the motor imaging brain signal well,resulting in unsatisfactory classification results.In view of the above problems,this paper proposes a method based on continuous wavelet transform(CWT),which maps electroencephalogram signals of motor imagery to time-frequency images,which makes the overall and local characteristics of the electroencephalogram signals clear and can better extract effective features.Finally,although the pooling layer in CNN can simplify image features,it may also lose some effective features.For this problem,this paper proposes a simplified convolutional neural network(SCNN)to classify motor imagery EEG signals methods.The pooling layer is removed from the standard CNN to optimize the network structure and prevent the loss of effective features.Then use two one-dimensional convolution kernels to extract the time domain and frequency domain features to reduce the network parameters and improve the training rate.Finally,experiments prove that the improved scheme proposed in this paper not only simplifies the network structure and reduces network parameters,but also effectively prevents the loss of effective features.Experiments show that the classification accuracy rate is better than traditional or deep learning algorithms.