Research On Motor Imagery Decoding Based On Portable EEG Recording Device

Brain Computer Interface(BCI)has a wide range of applications in medical,military,and gaming fields.Especially in the field of rehabilitation medicine,BCI can help patients with muscle or nerve damage to express their intentions or control devices,in which the acquisition and decoding of EEG signals is the technical difficulty of BCI system application.In order to accelerate the development of Brain Computer Interface in the field of rehabilitation,this dissertation firstly designs a 40-channel portable EEG acquisition device to acquire EEG signals.Then,a motor imagery decoding algorithm based on multi-frequency brain network and deep learning has proposed,which can effectively decode the motor intention of the subject.To address the issues of EEG is weak and hard to acquire,this dissertation has designed a portable EEG acquisition device containing 40 channels,and names it GS-AI-40.Meanwhile,an accompanying host computer software has designed.The software used the self-developed Brain Link communication protocol to transmit EEG data and can display real-time waveform.In order to evaluate the performance of GS-AI-40,this dissertation has done sufficient performance tests on it.The GS-AI-40 has an input impedance of about 800M(?),a common mode rejection ratio of more than 90 d B,a signal-to-noise ratio of about 60 d B,and a static noise of less than 2μV.In the process of hardware design,the possible influence of the external environment or static electricity on the device has fully considered.Therefore,8KV air discharge test,6KV contact discharge test,and radiation emission test were conducted.The test results all comply national standards.To address the issues of difficult decoding of motor imagery paradigm signals,a decoding algorithm based on multi-frequency brain network and deep learning has proposed in this dissertation.The algorithm applies the advantages of complex networks to decode motor imagery EEG signals.The electrodes are used as the nodes of the brain network,and the PLV coefficients between different channels are used as the connected edges to establish the multi-frequency brain network.A deep learning framework with the convolutional neural network as the core has constructed for the characteristics of the brain network.The FBCSP algorithm has also introduced to supplement the more refined frequency information.In this dissertation,the effectiveness of the algorithm was first evaluated using the dataset I based on the BCI competition IV 2a,and achieved the average classification accuracy of 83.83% and the kappa value of 0.784.Further,data from 10 subjects were collected using the GS-AI-40 device as dataset II,which was used to evaluate the performance of the device and the algorithm.The average accuracy of 80.96% with the kappa value of 0.746 was obtained on dataset II.All results indicate that the algorithm based on multi-frequency brain networks and deep learning can effectively extract features concealed in EEG signals and demonstrates the good performance of GS-AI-40 in practical applications.