Article Details

Abstract

Facial expression is one of the most intuitive forms of expressing human psychological states. Therefore, accurately recognizing facial expressions in images using computers is of great significance. This study proposes an improved lightweight MobileNetV3 network and introduces a class-balanced loss function to alleviate the imbalance in the numbers of image samples among different categories in the dataset. Meanwhile, the Convolutional Block Attention Module (CBAM) is employed to enhance the network’s ability to extract key information in both the channel and spatial dimensions. In addition, a Squeeze-and-Excitation (SE) pooling module is proposed to enhance the model’s ability to identify important features during pooling. Finally, based on the RAF-DB dataset, the method distinguishes seven facial expression categories, namely anger, disgust, fear, happiness, neutral, sadness, and surprise, and the overall test accuracy reaches 85.23%. The method not only achieves high recognition accuracy and strong robustness but also requires fewer parameters and shorter processing time, making it suitable for portable mobile devices.

Keywords

facial expression recognition; MobileNetV3; class-balanced loss; CBAM; SE pooling

References

1. [1] Chen Qiaohong, Sun Jiajin, Lou Yangbo, etc Multi modal sentiment analysis model based on multitask learning and stacked Transformer [J]. Journal of Zhejiang University (Engineering Edition), 2023, 57 (12): 2421-2429
2. [2] Huang Yuanqin Research on Facial Expression Recognition Method Based on Improved Convolutional Neural Network [D]. Guilin: Guangxi Normal University, 2022
3. [3] Lu Haoyang Research and Application of Facial Expression Recognition Based on Lightweight Convolutional Neural Networks [D]. Xi'an: Xi'an University of Technology, 2025
4. [4] Howard A，Sandler M，Chu G，et al. Searching for MobileNetV3 [J].In Proceedings of the IEEE/CVF international conference on computer vision，2019：1314-1324.
5. [5] Wang Liangliang, Xue Mingming, Wang Xiaoling Facial expression recognition based on improved convolutional neural network [J]. Computer Age, 2025 (3): 16-20
6. [6] Li Meiyu, Ding Jifeng, Ao Yingjian, etc Inverted Residual Image Super Resolution Reconstruction Model Based on Attention Mechanism [J]. Intelligent Computers and Applications, 2025, 15 (12): 31-38
7. [7] Yang Lian, Shi Baofeng, Dong Yizhe A credit risk assessment model for non-equilibrium samples based on Class Balanced Loss corrected cross entropy [J]. Journal of Systems Management, 2022, 31 (2): 255-269
8. [8] Wang Xinyu Research on Facial Emotion Recognition Technology Based on Convolutional Neural Networks [D]. Dalian: Dalian Jiaotong University, 2025
9. [9] Wu Zhenrong, Qiu Weigen, Zhang Lichen Improved MobileNet Lightweight Facial Expression Recognition Method [J]. Information Technology, 2024 (8): 44-50, 58
10. [10] Li S，Deng W H. Reliable Crowdsourcing and Deep Locality-Preserving Learning for Unconstrained Facial Expression Recognition [J].IEEE Transactions on Image Processing，2019，28（1）：356-370.
11. [11] Tan M X，Le Q V. Efficientnetv2: Smaller Models and Faster Training [J/OL].arXiv:2104.00298 [cs.CV]. https://arxiv.org/abs/2104.00298.
12. [12] Ma N N，Zhang X Y，Zheng H T，et al. Shufflenet v2: Practical Guidelines for Efficient CNN Architecture Design [J/OL].arXiv:1807.11164 [cs.CV].https://arxiv.org/abs/1807.11164.
13. [13] Tang Y H， Han K，Guo J Y，et al. GhostNetv2: Enhance Cheap Operation with Long-Range Attention [J/OL].arXiv:2211.12905 [cs.CV].https://doi.org/10.48550/arXiv.2211.12905.
14. [14] Zhou D，Hou Q，Chen Y，et al. Rethinking bottleneck structure for efficient mobile network design [C]//Computer Vision–ECCV 2020:16th European Conference，2020：680-697.

Funding

Collaborative Education Project of the Ministry of Education (2024CXHZ0205)

About the authors

Li Jiaqi (1997-), female, Han ethnicity, from Wuhan, Hubei Province, full-time teacher, master's degree, research direction: big data analysis, applied spectroscopy; Li Juan (1977-), female, Han ethnicity, from Huai'an, Jiangsu Province, professor, PhD, research direction: computer application technology.