基于MobileNetV3的人脸微表情识别系统研究

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）24-0077-06

Research on Facial Micro-Expression Recognition System Based on MobileNetV3

CENChengrui,LI Haixia (CollegeofMechanicalandControlEngineering,Guilin UniversityofTechnology,Guilin 541oo6,China)

Abstract: Facial MEs are highly challnging to recognize due to their transient, subtle yet emotionally genuine characteristics.Adressng the prevalent isues ofbulkymodels orinsuffcient performanceoflightweight models in existing neural networks forMErecognition,this study proposes and implementsalightweight solution.BasedontheMobileNetV3 model,the solution performs fine-tuning in combination with ME characteristics and selects the Fer2013 and CK+ datasets for trainingandoptimization. Ultimately,thisstudydesignsand deploysalightweightMErecognitionsystemintegrated withaGUI. Tests coducted on the AfectNet dataset demonstrate thatthe system not only achieves high recognition accuracyand strong robustnessbutaomeetsreal-tmeequirements inprocessingspeedbeingcapableofompletingecogitionwitinthical duration ofMEs. This provides a valuable reference for practical applications.

KeyWords: Micro-Expresion Recognition; Deep Learning; Lightweight Convolutional Neural Networks;model deployment; MobileNetV3

0 引言

人脸微表情（ME）作为情感的真实流露，具有自发性和短暂性（持续1/25至1/5秒），其精确识别不仅在基础情感研究中意义重大，更在安防领域的测谎、临床心理诊断辅助、人机交互中的情感理解以及智能驾驶员的疲劳监测等方面具有重要的实际应用价值[]。（剩余11402字）