EEG-Based Driving Fatigue Recognition Using Hybrid Deep Transfer Learning Approach

One of the main causes of traffic accidents is driving while tired. World Traffic Safety Organization statistics indicate that driving while tired leads to 35-45% of road accidents and directly causes 1550 deaths, 71000 injuries, and 12.5 billion dollars in economic damage every year. To detect fatigue over time during driving, it is crucial to design an accurate and efficient analysis architecture. Physiological signals can significantly reduce the subjectivity and individuality of fatigue in drivers compared to methods like face analysis or questionnaire design. EEG signals contain valuable information about fatigue. Fatigue detection using brain signals is still a challenge due to the significant differences in EEG signals among people and the difficulty in collecting enough signal samples during fatigue. To reduce the level of classification error, hybrid deep learning was used in this paper as an efficient and fast method. By combining dilated-ResNet and Inception module, the approach is faster than other learning methods and provides satisfactory accuracy for fatigue detection. Windowing is used to receive signals from a limited number of channels. Windows are selected based on different lengths and then their spectrogram is created using short-time Fourier transform (STFT). More than 99% accuracy was achieved by the suggested approach. In order to recognize people's fatigue, the method has used a small number of channels that can be generalized and interpreted.