| An EEG acquisition device is an instrument used to record human EEG signals,which has a wide range of applications and research values in the fields of medicine,psychology,neuroscience,bioengineering and human-computer interaction.With the development of Brain Computer Interface(BCI)technology and wearable devices,high performance portable EEG acquisition devices have become one of the key technologies for BCI.However,the lack of low-cost,high-performance EEG acquisition devices on the market has been a limiting factor to the widespread application of EEG-related technologies.In the actual EEG acquisition process,due to the weakness of the EEG signals,the signals collected by the device will be masked by other physiological electrical artifact signals in the human body,which makes the extraction and analysis of EEG signals difficult.Removal of artifact signals can improve the reliability and accuracy of EEG signals.Therefore,the study of how to accurately and effectively eliminate artifact signals from EEG signals is crucial for the development of EEG technology.This thesis explores two areas of EEG acquisition systems and artifact removal methods for the implementation of portable EEG acquisition and EEG signal denoising application requirements:(1)Research on multi-channel EEG synchronous acquisition system: A multichannel EEG acquisition system was designed,including an EEG acquisition device and a computer GUI program.The EEG acquisition device supports multi-channel EEG acquisition with high accuracy,high sampling rate and high dynamic input range.A wireless transmission protocol is designed to meet the data loss problem in wireless transmission,and 74 Hamming error correction codes and CRC16 check codes are introduced to achieve reliable wireless data transmission.The computer GUI program implements data reception from the Bluetooth serial port,EEG acquisition device parameter setting,data frame parsing,filtering processing,data storage and real-time waveform display.To verify the reliability of the system,input noise experiment,signal simulator experiment,human external stimulation experiment and spontaneous EEG signal experiment were conducted.After experiments,it is proved that the EEG acquisition system meets the design specifications.(2)Artifact removal algorithm research: To solve the shortcomings of the traditional artifact removal methods that require human intervention,an automated artifact removal neural network model based on the Inception-RSBU block is proposed.For the feature extraction of the signal,the nonlinear expression capability of the model is improved by introducing the Inception module instead of the traditional one dimensional convolution module.The Gelu activation function is used to replace part of the Re LU activation function in order to improve the richness of the transmission weights and contribute to the model feature extraction capability.Then,an EEG artifact separation module based on channel thresholding residual shrinkage unit is proposed,which combines soft thresholding mechanism and attention mechanism to separate EEG features and artifact features,so as to realize the reconstruction of EEG signals and artifact signals respectively according to different signal features.In order to evaluate the artifact removal performance of the model,the proposed model is validated by using contaminated EEG signals with different signal-to-noise ratios,and the results are compared with the results of other models.Relative to other models,the performance of this model was improved by 9%,4%,0.6% and 44%,35%,7% in the OA and MA artifact removal tasks,respectively. |
