Efficient Hardware and Software Co-design for EEG Signal Classification based on Extreme Learning Machine

ElectroEncephaloGram (EEG) is associated with multiple functions, including communications with neurons, organic monitoring, and interactions with external stimuli. By decoding EEG signals, certain human activities such as sleeping, brain diseases, motor imagery, movement of limbs, and others can be observed and controlled through the brain-computer interface (BCI). Therefore, it is vital to efficiently process EEG signals with a robust and accurate system to build BCI systems with powerful applications. However, as a weak biosignal, EEG demands a fast-reaction system signal processing with high accuracy and sensitivity. In this work, a hardware/software co-design network based on Extreme Learning Machine (ELM) is introduced for the classification of certain actions, motor imagery of the human brain. This system is based on an optimized Hierarchical Extreme Learning Machine (H-ELM) on the software layer. The proposed method has advantages over previous designs with an accuracy of 90.3%. It also improves the training speed by around 25X compared to conventional methods. The software model is also translated into efficient FPGA hardware design to maintain high computation efficiency and reduce power consumption for biomedical applications.