EEG Acquisition And Analysis System Based On Time-frequency And Neural Network

Brain-Computer Interface(BCI)can build a way to transmit information between human brain and external devices to realize the interaction between brain activity and the external world,such as controlling the device or receiving signal stimulation,etc.It has been widely used in the fields of intelligent control,medical rescue,sports rehabilitation and consumer entertainment.Motor Imagery(MI)-based brain-computer interface refers to the direct control of external devices through the brain activity of the subject to imagine the motor intention of the limb control without the corresponding actual action,and through the local brain cell activity generated by the EEG(Electroencephalogram).The problems of weak signal amplitude,strong noise interference and poor signal resolution of motor imagery EEG signals affect the efficient operation of brain-computer interface systems;at the same time,existing EEG acquisition systems also have the problems of poor portability and low accuracy of device signal acquisition.To this end,this paper proposes a multi-scale convolutional neural network combined with time-frequency analysis,and designs and implements a hardware and software system for EEG signal acquisition and analysis equipped with this network model to improve the acquisition quality and processing capability of EEG signals.The main work is as follows:Aiming at the problem of poor time-frequency resolution of EEG signals,the improved generalized Stockwell transform is introduced to perform timefrequency analysis of them,and the improved generalized Stockwell transform,applied to the feature extraction method of EEG signals,is proposed and compared with the traditional time-frequency analysis method for experiments,and the results show that the time-frequency analysis method has better energy focus and time-frequency resolution.To address the problem of low accuracy of EEG signal classification and recognition,an improved classification method combining generalized Stockwell transform time-frequency spectrogram and multi-scale convolutional neural network is proposed.The feature-extracted time-frequency spectrogram is first segmented according to the EEG signals with different rhythms,then convolved in the spatial domain,and finally the temporal features are extracted by the variance layer and classified using the fully connected layer.The results show that the neural network model achieves a certain degree of classification accuracy improvement.To address the problem of poor quality of EEG signal acquisition,this paper designs and implements an EEG signal acquisition and processing system based on ESP32 main control chip and ADS1299 analog-to-digital conversion,including hardware circuit,driver and upper computer visualization platform.The system is used for data acquisition test and classification recognition test in the upper computer.From the test results,the system is able to acquire high quality EEG signals and retain the effective information to identify the corresponding features.