EEG Sensor Based Semi-Supervised Inattention Prediction Framework For Unmanned Aerial Vehicles

摘要

With the advance in sensor devices, electroencephalog-raphy (EEG) can be unobtrusively collected enabling the inattention prediction of unmanned aerial vehicle (UAV) operators, which is one solution for reducing the high accident rate of UAVs. Several studies using statistical learning methods on EEG data have shown satisfactory results. However, it is almost impossible to obtain accurate training data containing attention status labels due to the absence of standardized measure for the attention status. Therefore, in this paper, we propose a semi-supervised inattention prediction framework which does not require training data nor any prior information by utilizing the fact that operators keep their attention at the beginning of a task and adopting a cumulative sum algorithm to detect the duration. Moreover, weighted dissimilarity measures are applied to enhance the prediction performance of the proposed framework. From experiments conducted on real-world datasets, the proposed framework showed promising results.