Mental fatigue can reduce work performance and cause safety accidents in humanmachine systems. Therefore, it is important to detect fatigue in real time. A great deal of work has focused on this problem, but there are still no standards for the physiological index. Multiphysiological measurement becomes a trend, at the same time, the increasing complexity of instruments for multiphysiological measurement brings challenges due to the complicacy of mental fatigue. Functional NearInfrared Spectroscopy (fNIRS) can measure cerebral hemoglobin and reflect cognitive function indirectly. However, cardiac and respiratory signals in the fNIRS signal are sensitive to physiological activity, which have always been removed as interference in previous studies. To increase the information capacity and establish a multiphysiological fatigue detection model using fNIRS, this paper extracts the cardiac and respiratory features from the fNIRS signal as new sensitive feature. A fatigue detection model is proposed based on the support vector machine (SVM) by combining cardiac and respiratory features with common features, such as the mean value and slope. We use a verbal 2back task for a total of 60 minutes to induce mental fatigue. The fNIRS signals from 10 channels in the prefrontal cortex (PFC) are measured from 15 healthy subjects. The results show that the new cardiac and respiratory features are significantly sensitive to the fatigue state and increase the classification accuracy compared with a common fatigue model based on fNIRS (84%→90%). Our findings can detect mental fatigue effectively and reduce the complexity of equipment significantly for multiphysiological fatigue detection.