Features and Window Size Selection for Quadcopter Control Using Functional Near-Infrared Spectroscopy

In this paper, different feature combinations and time window sizes for the online/real-time control of a quadcopter using functional near-infrared spectroscopy (fNIRS) is investigated. fNIRS signals are acquired from the prefrontal cortex during the mental arithmetic and puzzle solving tasks. Six most frequently used features of oxy-hemoglobin (ΔHbO) including mean, peak, minimum, slope, skewness, and kurtosis are calculated in five different window sizes (i.e., 0~2, 0~4, 0~6, 0~8, and 0~10s). All possible 2- and 3-dimensional combinations of features are then formed and classify using linear discriminant analysis. The generated command is used to move quadcopter in a fixed path. The results showed that all combinations of 2- and 3-features containing mean, peak, and minimum values of fNIRS signals yielded significantly higher (p <; 0.05) classification accuracies. Mean, max, and minimum values in 0~2 s give 78% averaged classification accuracy that helps to generate a reliable brain command in a reduced time window for the online control of a quadcopter.