ACFSENet: an adaptive cross-frequency global sparse encoding network for end-to-end EEG emotion recognition

End-to-end EEG-based emotion recognition is attracting increasing attention due to its potential in human-computer interaction, mental health, and affective brain-computer interfaces (aBCIs). However, most existing methods overlook cross-frequency interactions in neural oscillations and suffer from high computational complexity, limiting their applicability in real-time or resource-constrained scenarios. To this end, we propose ACFSENet, a novel end-to-end neural architecture that integrates adaptive cross-frequency modeling with global sparse encoding. ACFSENet employs an adaptive frequency-aware mechanism to dynamically focus on subject- and task-specific local brain dynamics, thereby enhancing the flexibility of emotional representation. In addition, it incorporates a sparse attention mechanism with a temporal distillation structure to reduce computational complexity while preserving the ability to model long-range temporal dependencies. We evaluate ACFSENet using cross-block validation on three benchmark datasets: DEAP, SEED, and SEED-IV. Results demonstrate that ACFSENet outperforms state-of-the-art methods and achieves a favorable balance between recognition performance and computational efficiency.