Drowsiness Detection using Forehead Electrophysiological Signals and CNN-MultiGRU Networks

As drowsiness usually leads to performance degratation on tasks where long-term concentration is needed, developing an efficient wearable drowsiness monitoring equipment is of great significance. In this paper, to balance the contradiction between monitoring accuracy and wearing comfort in a wearable drowsiness monitoring equipment, we employ the electrophysiological signals acquired on the forehead, which are the compounds of electroencephalogram (EEG) and two types of electrooculogram (EOG) signals. With the acquired forehead electrophysiological signals as inputs, a one-dimensional convolutional neural network (1D-CNN) and a multi-layer gated recurrent unit (MultiGRU) were built to extract the intra-epoch and the inter-epoch features of signals, respectively, hence realizing automatic awake/drowsiness state detection. By evaluation on the data set with 30 healthy subjects recruited, the proposed method is shown to outperform the single-modal-signal-based drowsiness monitoring methods.