Easy-to-use EEG data collection platform for motor imagery

Abstract

Brain-Computer Interface (BCI), as a novel form of human-computer interaction, holds great potential to revolutionize the world. Among various BCI paradigms, the Motor Imagery-based Electroencephalogram (MI-EEG) BCI stands out as a non-invasive method that allows for the control of external devices such as robots solely through the power of imagination, without the need for external stimulation.

However, the development of MI-EEG BCI faces challenges due to various factors, such as the limited availability of large-scale databases specifically designed for MI-EEG and the selection of suitable classification algorithms. Additionally, the requirement to have all data collected before training poses limitations in scenarios where incremental learning or transfer learning is necessary.

In the dissertation, A large-scale database named EPOCX-DATABASE is obtained using animated guidance to mitigate the distractions caused by subjects' inattention or fatigue during the MI-EEG data collection process. The EEG data is collected using lightweight EEG equipment with saline-based electrodes. In order to simultaneous model training and data acquisition across different hosts, a socket API has been developed to facilitate communication between Unity and PyCharm.

Finally, to evaluate the effectiveness of the proposed data collection platform, we employed two MI-EEG classification algorithms: FBCSP and ATCNet, as performance indicators. A higher classification accuracy is considered indicative of a superior data collection platform. Regrettably, our EEG collection equipment malfunctioned. As a result, we could only evaluate the feasibility of the data evaluation method on the publicly available BCI Competition IV 2a dataset, postponing the assessment of our dataset to future research.