Classification of Sleep EEG Signals for Individual Specialization by 1D Deep Convolutional Network and Data Augmentation

Sleep plays a crucial role in maintaining good health, and high-quality sleep helps people maintain and promote overall health and well-being. In modern clinical medicine, analyzing EEG during sleep is one of the most important ways to study sleep. However, the data volume of sleep EEG is very large that manual recognition consumes a lot of human resources, at the same time, there is obvious subject-independence between different people's sleep EEG, making it difficult for artificial intelligence recognition models to achieve decent high accuracy. In this paper, a data augmentation technique that can generate a significant amount of artificial data based on Discrete Cosine Transform from a small amount of real EEG data is introduced, then we mixed the data augmentation with the publicly available dataset as input and trained the 1D Deep Convolutional Network to obtain a model with very high classification accuracy for a specific individual. Our experimental results demonstrated that we could train a customized high-accuracy classification model relying on only a small amount of labeled data from a specific individual, thus bypassing the subject-independence problem.