Masked Facial Expression Recognition with Attention Mechanism based on Mini_Xception Network

Abstract

Facial Expression Recognition (FER) plays a crucial role in applications such as human-computer interaction and health monitoring. However, traditional FER methods often suffer significant performance degradation when occlusions are present, especially when the mouth region is covered, as commonly observed during the COVID-19 pandemic. To address this challenge, we propose a novel FER model, Mini-Xception+CBAM, which combines a lightweight network architecture with an attention mechanism to enhance  performance under occlusion scenarios. The proposed approach utilizes the Mini-Xception architecture and integrates the Convolutional Block Attention Module (CBAM), enabling the model to dynamically focus on critical facial regions, such as the eyes and eyebrows, which are less likely to be occluded. Extensive experiments on the FER2013, SMFER2013, and MFER2013 datasets demonstrate that our model outperforms several state-of-the-art baseline models, including Mini-Xception, MobileNetV2, ResNet-18, and Vision Transformer (ViT), achieving 90.4% accuracy on FER2013, 89.2% on SMFER2013, and 91.5% on MFER2013, along with superior precision, recall, and F1 score. Furthermore, the model's efficient design, with only 0.68M parameters and 30.1M FLOPs, makes it highly suitable for deployment on resource-constrained devices.