基于关键点和局部注意力机制的面部表情识别方法研究  

The Research on Facial Expression Recognition Method Based on Key Points and Local Attention Mechanism

作　　者：谢梦瑶 Xie Mengyao(Changsha Social Work College,Changsha Hunan 410004,China;Hunan Intelligent Rehabilitation Robot and Auxiliary Equipment Engineering Technology Research Center,Changsha Hunan 410004,China)

机构地区：[1]长沙民政职业技术学院,湖南长沙410004 [2]湖南省智能康复机器人与辅具器械工程技术研究中心,湖南长沙410004

出　　处：《长沙民政职业技术学院学报》2025年第1期122-128,共7页Journal of Changsha Social Work College

基　　金：2024年度湖南省智能康复机器人与辅具器械工程技术研究中心基金项目“基于骨骼关键点的体态评估与交互研究”(2024KF303)。

摘　　要：随着计算机视觉和人工智能技术的进步,面部表情作为情感传达的重要手段,已成为机器智能研究的一个热门领域。面部表情识别通过分析视觉特征来识别情绪,广泛应用于人机交互、心理监测等领域。然而,在自然场景中,光照变化和遮挡等因素对识别效果构成了挑战。为此,本文提出了一种基于关键点和局部注意力机制的面部表情识别方法。该方法设计了空间注意力机制,以提取人脸关键点特征,并结合多分支局部注意力机制以提取图像深层特征,从而增强了模型在遮挡和姿态变化下的鲁棒性。实验结果表明,该方法在RAF-DB和AffectNet数据集上的准确率分别达到了87.58%和66.50%,进一步验证了该方法的有效性与优越性。With the advancement of computer vision and artificial intelligence technologies,facial expressions,as an important means of emotional communication,have become a popular research area in machine intelligence.Facial expression recognition identifies emotions by analyzing visual features and is widely used in fields such as human-computer interaction and psychological monitoring.However,factors such as lighting variations and occlusions pose challenges to recognition performance in natural scenes.To address this,this paper proposes a facial expression recognition method based on key points and local attention mechanism.The method designs a spatial attention mechanism to extract facial key point features and combines a multi-branch local attention mechanism to extract deep image features,thereby enhancing the model's robustness under occlusions and pose variations.Experimental results show that the proposed method achieves accuracy rates of 87.58%and 66.50%on the RAF-DB and AffectNet datasets,respectively,further validating the effectiveness and superiority of the method.

关 键 词：面部表情识别 关键点 注意力机制 TRANSFORMER 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]