Research On Denoising And Classification Algorithms Of Motor Imagination EEG Signals And Acquisition System Design

The 21st century is considered by the scientific community to be the era of biological sciences and brain sciences.Driven by the global brain research plan,my country has put forward the strategic deployment of "brain science and brain-like research".Among them,motor imaging has attracted much attention and research in the field of brain-computer interface,and it has important development value in the fields of medical-industry integration and intelligent control.As EEG signals are microvolt-level bioelectric signals,they are susceptible to environmental interference,which makes it very difficult to extract effective information from them.At present,many algorithms have been successfully applied to the denoising and classification of EEG signals,such as using wavelet transform or independent component analysis to remove noise;using empirical mode decomposition or co-space mode to extract EEG features.However,these traditional methods all have the problem of unsatisfactory denoising effect and low recognition accuracy.Therefore,the research object of this article is motor imaging EEG signals,focusing on the unsatisfactory denoising effect of EEG signals,and classification and recognition accuracy.To improve the denoising and classification and recognition capabilities of the EEG signal acquisition system,the research on the denoising and classification algorithms of EEG signals and the design of the signal acquisition system are carried out to improve the denoising and classification and recognition capabilities of the EEG signal acquisition system.The main work of the paper is as follows:?.Research on Denoising Algorithms for EEG SignalsIn order to solve the problem that traditional EEG denoising methods tend to filter the useful information contained in the high-frequency part of the signal as noise,resulting in the distortion of the de-noising EEG signal,a complementary total empirical mode decomposition(CEEMD)denoising method based on cosine similarity is proposed.Firstly,the EEG signal is decomposed into multiple intrinsic mode functions(IMF)with different scales by CEEMD method.The cosine similarity method is used to calculate the similarity between each IMF and the original signal.The IMF component after the first minimum value in the similarity curve is selected as the dividing point between the dominant mode of signal and the dominant mode of noise.Then wavelet packet transform is used to extract the useful information of the dominant mode of noise,and finally the denoised signal is reconstructed with the rest of IMF.The experimental results show that the proposed algorithm can effectively retain the useful information in the high frequency mode,and the denoising performance is better than the traditional algorithm under different noise intensities.?.Research on Classification Algorithms of EEG SignalsIn this paper,two effective EEG classification algorithms are proposed.a)Aiming at the problems of poor adaptability and low recognition rate of single feature recognition of EEG signals,a feature extraction method based on multi feature fusion of dual tree complex wavelet(DTCWT)is proposed.First,the best time-frequency band is extracted by DTCWT transform.Then,the Hilbert transform and Lempel Ziv complexity calculation are performed on the extracted signal frequency band to obtain the time-frequency-domain nonlinear features.Finally,the linear discriminant analysis(LDA)is used for classification.The experimental results show that this method can effectively classify EEG signals,and the accuracy is significantly improved,up to 89.84%.b)In order to solve the problem that the recognition rate is low because the machine learning can't get the effective information completely,a classification recognition method based on convolutional neural network is proposed based on the ability of deep learning to automatically extract features.By designing different number of convolution pool blocks,three convolutional network structures of different depth are constructed.The three convolutional network structures are fused in parallel to form feature extractors.Finally,the classifier is connected to form a multi-level fusion network model.Experiments show that the method can classify EEG signals effectively,and the recognition accuracy is 93.3%.?.Design and implementation of EEG signal acquisition systemA EEG signal acquisition system based on ADS 1299 is designed.ADS 1299 analog front-end is used to realize the amplification and analog-to-digital conversion of EEG signal.The data is transmitted to STM32 main control chip through SPI communication,and then STM32 transmits the data to the upper computer through Bluetooth,so as to display the EEG signal in real time and save the data for further development.Finally,the function of the system is tested.The test results show that the system meets the design requirements and can collect EEG data and display them in real time.To sum up,this paper completed the design of a portable wireless EEG acquisition system based on ADS 1299,and focused on the EEG signal denoising and classification algorithm.The prototype of the system has been tested in the subject laboratory,and the power consumption,wireless transmission distance and other important indicators basically meet the expected design requirements,which has a certain reference value for BCI application research.