Leveraging Wavelets and Deep CNN for Sleep Pattern Recognition in Road Safety: An EEG Study

A major cause of fatal traffic accidents and injuries worldwide is drowsy driving. The fatal aftereffects can be considerably reduced, and road safety can rise, with earlier and more efficient diagnosis. In this study, electroencephalography (EEG) biosignals-based brain-computer interface (BCI) is established for early identification of human inattentiveness during driving activities. EEG signals are recorded from six EEG electrode positions viz Fp1, Fp2, F7, F8, O1, and O2. Spectral information of the alpha and beta frequency bands is extracted in terms of continuous wavelet transform (CWT) with a temporal correlation for complete experiments. CWT components are processed as scalograms which are color images used for visual inspection of brain states. The scalogram images of 10 s epochs are stored as the image dataset. The SqueezeNet deep convolutional neural network (CNN) is fine-tuned with transfer learning for effective drowsiness detection from CWT scalograms. Transfer-learned CNN classification model achieved the best results with 89.8% accuracy, 88.0% precision, 96.1% recall, 89.9% F \(_1\) -score, 87.7% specificity, and 79.7% Matthew’s correlation coefficient. The statistical significance of the achieved results is validated with a p-value of less than 0.05. Using wavelet information of EEG biosignals, the suggested BCI methodology offers a viable method for quicker and more accurate driving drowsiness detection.