SwinKANet:An Improved Face Expression Recognition Network Based on KAN and Attention Mechanism

Facial expression recognition(FER) is an important task in affective computing and human-computer interaction. With the rapid development of deep learning, deep learning-based FER methods have gradually become a research hotspot. However, existing methods still face challenges in dealing with complex emotions, gesture changes, and illumination changes, especially the limitations in feature extraction and redundant information processing. To address these issues, this paper proposes SwinKANet, an improved face expression recognition network based on KAN(Knowledge-Aware Network) and attention mechanism. The network combines the window multi-head self-attention and sliding window multi-head self-attention mechanisms, and further enhances the comprehension and representation of facial expression features by introducing a KAN component. In order to improve the feature representation ability of the network, SwinKANet designs a feature redundancy encoder to reconstruct the spatial and channel features respectively, which improves the feature expression ability and stability. Experimental results show that SwinKANet achieves an accuracy of 96.94% on the KDEF dataset, surpassing six advanced expression recognition methods and verifying its superiority.