Based On The Movement Of The FPGA Imagine EEG Classification Research

The Brain Computer Interface(Brain Computer Interface,BCI)establishes a channel for direct transmission of information between the brain and the outside world by interpreting human physiological information in the process of thinking activities,which has broad prospects in applications such as health monitoring and advanced human-computer interaction.The brain-computer interface based on motor imagination is a very important type of brain-computer interaction strategy,which is characterized by detecting and classifying the brain signals related to the user's thinking "imagination"to control external devices such as neural prostheses and wheelchairs to complete functions that have lost their functions.The peripheral nerves should have the functions of movement and walking.Aiming at the left and right hand movement imagining EEG signals,this paper adopts a joint classification method based on wavelet transform feature extraction,principal component analysis(PCA)and Support Vector Machine(SVM).The multi-resolution characteristics of wavelet transform are very suitable for extracting the characteristics of random,non-stationary EEG signals.In practical applications,the BCI Competition ? 2003 EEG data training samples are pre-processed and convolved with the db4 wavelet filter coefficients,and then the low-frequency coefficients sampled at the even-numbered positions of the convolution results are used as the next layer input signal and filter coefficient Operation,repeating the above steps can realize four-layer decomposition of Mallat and use low-frequency coefficients as feature vectors.Due to the high dimension of the extracted signal features,the PCA dimensionality reduction and SVM classification algorithm can avoid the dimensionality disaster and overfitting caused by wavelet transform.The dimensionality reduction process of PCA can be regarded as the difference between the feature vector and the sample mean,and the PCA projection matrix is multiplied to achieve dimensionality reduction.Use training samples to train the SVM,find the weight coefficient vector and bias that have the best classification effect,and finally get the decision function,and then send the dimensionality-reduced feature vector to the decision function to get the classification result.In FPGA design of classification algorithm,first,the EEG signal passes through the Mallat four-layer decomposition module,and the resulting dimension is 98 as the feature vector;Secondly,in the PCA_SVM module,the product of the training sample mean matrix,PCA projection matrix and SVM optimal weight coefficient vector obtained in matlab is solidified into the ROM.The PCA_SVM module essentially reduces the ca4 dimension to 15 feature vectors and multiplies the SVM optimal weight coefficient vector;Finally,the timing control module and the output effective bit control module ensure that the PCA_SVM module outputs correctly and effectively to the output selection module,and the output selection module classifies left and right hand movement imaging EEG signals.The simulation results show that the EEG signal classification system designed this time has a system response time of 9.1 us,a recognition rate of 93.33%and a Kappa coefficient of0.867,which has certain stability and real-time performance.Figure[52]Table[4]References[63]...