基于自适应卷积和注意力融合的ResNet人脸表情识别方法  

ResNet-based Facial Expression Recognition Method Based on Adaptive Convolution and Attention Fusion

作　　者：严武军[1] 叶金霞 李建昌 YAN Wujun;YE Jinxia;LI Jianchang(College of Computer Science and Technology,Taiyuan Normal University,Jinzhong 030619,China)

机构地区：[1]太原师范学院计算机科学与技术学院,山西晋中030619

出　　处：《现代信息科技》2025年第5期39-44,50,共7页Modern Information Technology

摘　　要：在实际应用中,人脸图像受光线、遮挡和姿态等因素影响,表情识别准确率较低。为此,提出一种融合自适应卷积和注意力机制的表情识别方法。该方法基于ResNet34网络,引入自适应卷积(AKConv)模块以捕捉多尺度特征,并整合注意力混合(ACmix)机制提升分类精度。同时,采用滑动损失(SlideLoss)替代传统交叉熵损失函数,解决数据不平衡问题。实验结果表明,该模型在FER2013数据集上达到75.21%的准确率,验证了其有效性和优越性,为表情识别领域提供了新思路和方法。In practical applications,the accuracy of facial expression recognition is relatively low because facial images are affected by factors such as lighting,occlusion,and pose.To address this,an expression recognition method that integrates adaptive convolution and the Attention Mechanism is proposed.Based on the ResNet34 network,this method introduces an AKConv module to capture multi-scale features and integrates an ACmix mechanism to improve the classification accuracy.Meanwhile,the traditional cross-entropy loss function is replaced with SlideLoss to address the problem of data imbalance.Experimental results show that this model achieves an accuracy of 75.21%on the FER2013 dataset,verifying its effectiveness and superiority and providing new ideas and methods for the field of facial expression recognition.

关 键 词：ResNet 面部情绪识别 深度学习 注意力机制 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]