Deep Learning with Noise Data Augmentation and Detrended Fluctuation Analysis for Physical Action Classification by Brain-Computer Interface

Deep neural networks (DNNs) demonstrate their high potential for investigation of time sequences in many domains, including analysis of electroencephalography (EEG) signals gathered by brain-computer interface (BCI). This work is targeted on physical actions classification problem based on analysis of EEG data collected by BCI. The relatively small DNN with components of fully-connected networks (FCN) and convolutional neural network (CNN) was considered to classify physical activities (namely, hand manipulations) from the grasp-and-lift (GAL) dataset. The noise data augmentation (NDA) allowed us to estimate reliability of the results obtained in the view of the possible influence from environment. The detrended fluctuation analysis (DFA) was applied to investigate the fluctuation properties and calculate the correspondent Hurst exponents. The results of DNN training, validation and testing with and without NDA along with DFA demonstrated that some physical actions can be divided in separate groups of actions that can be characterized by complexity and the feasibility of their classification: the easiest (HandStart), medium (LiftOff, Replace, and BothReleased), and hardest (BothStartLoadPhase and FirstDigitTouch) classification. As far as the statistically reliable results were obtained by the relatively small DNN with the low resource requirements, this approach can be used for porting such models to Edge Computing infrastructures on devices with the very limited computational resources.