Mini-SCXception: Balancing Accuracy and Efficiency in Real-Time Facial Expression Recognition

In Facial Expression Recognition (FER), existing models often encounter challenges such as high parameter counts and significant computational demands, making real-time application on resource-constrained devices difficult. To address this issue, this paper proposes an improved Mini-SCXception model based on the Mini-Xception architecture. By integrating Spatial and Channel Reconstruction Convolution (SCConv), the proposed model reduces redundant computations while preserving the efficiency of Mini-Xception and enhancing feature extraction capabilities. On the FER2013 dataset, Mini-SCXception achieved an accuracy of 64.50%, representing a 1.05% improvement over the baseline model's 63.83%. Experimental results demonstrate that Mini-SCXception balances accuracy and computational efficiency, providing a robust approach for real-time FER.