Table 1.
Input data information for each particular step.
Fig 1.
The mental fatigue detection framework.
Fig 2.
68 special landmarks from face image.
Fig 3.
The pseudocode of the RFE algorithm.
Fig 4.
The DRN-RF model.
Fig 5.
The structure of residual block.
Table 2.
The steps of applying RF algorithm.
Fig 6.
The workflow of the RF algorithm (the size of the training sampleS is N; the sample has M features and the forest has k trees).
Table 3.
The original Karolinska Sleepiness Scale.
Table 4.
The modified KSS and experiment sample size.
Fig 7.
The overall workflow of the experiment.
Table 5.
Cross- validation result in threshold selection.
Table 6.
MFD related features selected by using RFECV. (the ‘~’ represents that all features in the interval are retained; x represents horizontal coordinate, y presents vertical coordinate).
Fig 8.
A total of 83 selected features.
Table 7.
Parameters of GBM.
Table 8.
Parameters of KNN.
Table 9.
Parameters of RF.
Table 10.
The test results of different MFD comparison.
Table 11.
Comparison of difference between DRN-RF and the others.
Table 12.
The test results of different MFD comparison in Sleepy & Extremely Sleepy.
Table 13.
Comparison of difference between DRN-RF and the others in Sleepy & Extremely Sleepy.
Table 14.
The performance of different MFD models comparison in Sleepy & Extremely Sleepy.
Fig 9.
The confusion matrices of different MFD models comparison.
Fig 10.
Characteristic correlation thermal diagram.
(Feature correlation coefficients among three classes, including eye features, nose features, and mouth features. Red color represents strong correlation, and blue color represents poor correlation).