Recognition of Different Turning Behaviors of Pilots Based on Flight Simulator and fNIRS Data

Abstract

Pilot operational behaviour directly affects aviation safety, and exploring pilot behaviour can help decode, model, and predict pilot behavioural decisions. Therefore, it is necessary to recognize pilots’ states and predict the behaviours responsible for state changes. The purpose of this study was to investigate the relationship between pilot operating behaviour and brain activity. Using a flight simulator (DA42) and noninvasive functional near-infrared spectroscopy (fNIRS), changes in physiological signals were measured in 25 pilot cadets during different turning tasks. Twenty-five pilot cadets completed left and right turning tasks in a flight simulator, and the changes in blood oxygen concentration during different turning behaviours were analysed and modelled using machine learning models. Feature samples with significant differences ( $\text{p}< 0.05$ ) were extracted from the changes in oxyhaemoglobin concentration and selected for input into the SVM-linear classification model, the best results were achieved for the recognition of the pilot’s turning behaviour. After stratification based on feature importance, the SVM-RBF classification model obtained the highest accuracy (92.6%) after training with inputs containing a subset of features (60%) with the highest importance. In addition, the different turning behaviours of the pilots were closely related to activity in Brodmann area 17 (BA 17), BA 18 and BA 46. These findings provide insight into the neuroscientific mechanisms of pilot behaviour and provide an important physiological reference for improving pilot training.

Keywords

• Aviation safety
• behavioural recognition
• flight simulator
• fNIRS
• machine learning