Electroencephalography (EEG) is extensively employed in emotion recognition and cognitive workload decoding. However, signal characteristics and inter-subject variability pose significant challenges for deep learning models, particularly due to data scarcity and limited generalization. Although data augmentation (DA) is a critical approach to addressing data scarcity, a notable paucity of systematic reviews exists within deep learning frameworks focused exclusively on these two tasks. Through a systematic review of relevant literature, we summarize commonly used public EEG datasets, input representations, and deep learning classifiers. Subsequently, we focus on analyzing the specific applications and effectiveness of seven categories of DA methods in emotion recognition and cognitive workload decoding. The investigation identifies current challenges in this field, explores future research directions, and provides valuable references for researchers seeking to select and apply DA techniques to enhance model performance.
Radically different