Steady-state visual evoked potential-based brain-computer interfaces(SSVEP-BCI) are very popular for assistive control applications because of high accuracy, high transmission rate and no training required. However, users need to spend a lot of energy focusing on visual stimuli to generate strong enough SSVEP. Users are very fatigued due to high luminance, frequent low-frequency stimuli and single task. In this thesis, a real-time fatigue detection system for the SSVEP-BCI is proposed. The system completes the hardware design and implementation of a wearable electroencephalogram(EEG) device, and the classification algorithm based on support vector machines(SVM). On this basis, this project investigates the entropy used for fatigue detection accuracy and finds that fuzzy entropy and approximate entropy are consistent in detection. The fuzzy entropy change is more prominent in the case of weak fatigue changes while the approximate entropy change is more significant in the case of significant fatigue changes. In addition, this paper compares prefrontal and occipital lobe signals and finds that prefrontal signals usually have higher classification accuracy than occipital lobe signals. And the composite accuracy is higher than the accuracy of either one when used alone.