Pilot task monitoring using neural networks

Kohonen and backpropagation neural networks were used to process physiological data in order to identify pilot tasks and to cluster pilot tasks with similar workloads. Electrooculograms, electrocardiograms, and head motion data were collected from three pilots during a night close air support air-to-ground mission in a manned combat simulator. A backpropagation network and a Kohonen network were configured and trained. The backpropagation network was trained to a 99.9% recognition of the four flight modes within the training data. The network produced a 67% recognition with the testing data. Its performance was limited by the training data selection. The Kohonen network made four clusters which corresponded to four flight modes: navigation, threat evasion, target search, and target tracking. The Kohonen network has a 98% accuracy in determining the flight mode. A measure of pilot workload was also produced by the Kohonen network. Pilot tasks with similar workloads were clustered together