Innovation Series: Advanced Science (ISSN 2938-9933, CNKI Indexed)

Research on Multi-Class Facial Expression Recognition Method Based on YOLOv8

Zhiyuan Lu, Qiannan Wei*

School of Electronics and Information Engineering, University of Science and Technology Liaoning, Anshan, 114051, China

Corresponding Author: Qiannan Wei

Abstract: Facial expression recognition is an important research direction in emotion computing, human-computer interaction, intelligent monitoring, and psychological state analysis. To address the shortcomings of traditional expression recognition methods in handling complex postures, changes in lighting, fine-grained differences in expressions, and multi-category recognition capabilities, this paper proposes a nine-category facial expression recognition method based on YOLOv8. The study uses a self-built multi-category facial expression dataset, with a data size of approximately 60,000 to 70,000 images, including nine categories of expressions: angry, contempt, disgust, fear, happy, natural, sad, sleepy, and surprised. This paper models facial expression recognition as a task combining object detection and category discrimination, utilizing the C2f feature extraction structure, SPPF spatial pyramid pooling module, PAN-FPN multi-scale feature fusion structure, and decoupled detection head of YOLOv8 to achieve end-to-end recognition of facial regions and their categories. Experimental results show that the constructed model achieved 92.8% Precision, 91.6% Recall, 93.4% mAP@0.5, and 78.9% mAP@0.5:0.95 on the test set. The recognition effects of the happy, natural, and surprised categories are better, while the fine-grained categories such as disgust, contempt, and fear still have some confusion. The results indicate that YOLOv8 can better adapt to multi-category facial expression recognition tasks, achieving a better balance between recognition accuracy, inference speed, and deployment convenience. This provides an effective technical foundation for subsequent applications in real-time emotion perception systems.

Keywords: YOLOv8; Facial Expression Recognition; Deep Learning; Multi-class Detection; Emotional Computing

References