Figure 1. Schematic illustration of the proposed model.

EEG signals from PSG recordings were separated into 30 s segments and the size was equalized after obtaining the spectrograms. The whole data was divided into training, testing, and validation and first applied as input to the developed ResNet64 model. The same spectrograms were also classified using YOLOv5 and YOLOv8 models respectively and their performance metrics were compared.