A Spatial Mamba Model with Kolmogorov-Arnold Networks Classifier for Facial Expression Recognition

With the growing ubiquity of human-computer interactions (HCI), facial expression recognition (FER) has become essential for machines to understand human emotions. However, current research on FER is affected by class imbalance within databases, leading to a decrease in the overall recognition performance of the FER model. To address this problem, this study proposes a spatial Mamba model architecture containing a Kolmogorov-Arnold Networks (KAN) classifier for FER. This architecture not only significantly reduces the computational complexity of the model, but also enhances its ability to extract long-range dependency (LRD) features from facial expression images. Experimental results indicate that the proposed model, evaluating on the CK+ database with a class imbalance challenge, has achieved 98.94%, 98.41 %, and 98.61 % in recognition performance, evaluated by the recall, precision, and F1-score metrics, respectively. Compared to existing FER methods, the model proposed in this study demonstrates the superior ability to extract critical features from class-imbalanced databases.