Facial Expression Detection Algorithm Based on YOLOv8

DOI: https://doi.org/10.62517/jbdc.202401328

Author(s)

Zhengjun Wang

Affiliation(s)

University of Shanghai for Science and Technology, Shanghai, China,

Abstract

Facial expression detection is pivotal for the development of affective computing and human-computer interaction, but existing algorithms often fall short in real-time performance, accuracy, and complexity. This paper presents YOLOv8CRGD, a lightweight facial expression detection algorithm based on YOLOv8. The algorithm features a lightweight cross-scale feature fusion module (CCFM) to enhance the model's adaptability to scale variations and a SENetv2 module to improve feature representation, thereby increasing detection accuracy. To address efficiency in complex visual tasks, an improved visual Transformer structure with a dynamic attention mechanism (DAT) is adopted. Furthermore, GhostConv replaces traditional convolution operations to achieve a lightweight model design. Experimental results show that YOLOv8CRGD achieves a 91.3% accuracy and an mAP50 of 94.4% in facial expression detection tasks, while reducing parameters by 15% compared to the YOLOv8n model. With a frame rate (FPS) of 57.9, the algorithm not only maintains high detection accuracy but also excels in real-time performance, making it a compelling candidate for real-time facial expression analysis applications.

Keywords

YOLOv8; Facial Expression Detection; Lightweight; Dynamic Attention Mechanism; Cross-Scale Feature Fusion

References

[1] Hu Haixia, Xiang Cenyang, Fan Weiqin. Research on Tray Recognition Method Based on Improved YOLOv8n Model [J/OL]. Journal of Chongqing Technology and Business University (Natural Science Edition), 1-11 [2400-09-23] http://kns.cnki.net/kcms/detail/50.1155.N.20240914.1645.007.html. [2] Hu Jiusong, Liu Zhangchi, Yu Qian, et al. YOLOv8 Smoke Recognition Algorithm Incorporating GhostNet and CBAM [J/OL]. Journal of Electronic Measurement and Instrumentation, 1-8 [2400-09-23] http://kns.cnki.net/kcms/detail/11.2488.TN.20240906.1038.003.html. [3] Wang Baocheng, Yuan Hao, Han Feng, et al. Track surface damage detection method based on deep vision algorithm [J/OL]. Experimental Technology and Management, 1-12 [2400-09-23] http://kns.cnki.net/kcms/detail/11.2034.T.20240903.0944.004.html. [4] Cheng Shun, Li Jianrong, Wang Zhiqian, et al. A lightweight underwater optical image recognition algorithm based on YOLOv8 [J/OL]. Advances in Laser and Optoelectronics, 1-20 [2022-09-23] http://kns.cnki.net/kcms/detail/31.1690.tn.20240809.1638.010.html. [5] Yang Feng, Yao Xiaotong. A lightweight model for detecting wheat leaf diseases and pests based on improved YOLOv8 [J]. Smart Agriculture (Chinese and English), 2024, 6 (01): 147-157.