Abstract:

In order to identify the fatigue state of flight alert, a fatigue detection method of flight alert was proposed based on multi-physiological signals and SVM. Firstly, the evaluation and detection methods of fatigue were studied, and the experimental platform was built by combining the self-developed flight alert test system with multichannel biofeedback instrument and eye movement instrument. Physiological signals such as electrocardiogram, respiration, eye movement and subjective fatigue evaluation values were collected. The sensitive physiological indexes were extracted by nonparametric test of paired samples as feature vectors. With these feature vectors, fatigue detection models based on multi-physiological signals and SVM were constructed through machine learning training. Lastly, the effects of models were compared and analyzed based on ROC curve and model accuracy. The results show that many physiological indexes of the operators change significantly in the state of flight alert fatigue. Multi-physiological signal fusion has better detection effect than single signal. Its ROC curve area is 0.802. The training and prediction accuracy of fatigue detection model based on Gaussian radial basis function (RBF) can reach 93% and 87.50%. The state detection of flight alert fatigue can be realized based on multi-physiological signals and SVM.

Key words: multi-physiological signals, flight alert fatigue, eye movement index, support vector machine (SVM), receiver operating characteristic (ROC) curve