基于Res2Net的人脸表情识别方法  

Facial expression recognition method based on improved Res2Net

作　　者：唐宏伟[1] 丁祥 邓嘉鑫 高方坤 罗佳强 王军权 TANG Hongwei;DING Xiang;DENG Jiaxin;GAO Fangkun;LUO Jiaqiang;WANG Junquan(Hunan Provincial Key Laboratory of Grids Operation and Control on Muti-Power Sources Area,Shaoyang University,Shaoyang 422000,China)

机构地区：[1]邵阳学院多电源地区电网运行与控制湖南省重点实验室,湖南邵阳422000

出　　处：《邵阳学院学报（自然科学版）》2024年第2期28-35,共8页Journal of Shaoyang University：Natural Science Edition

基　　金：湖南省自科基金(2022JJ50205);湖南省科技计划项目(2016TP1023);湖南省研究生科研创新项目(CX20221314);邵阳学院研究生科研创新项目(CX2022SY005);邵阳学院研究生科研创新项目(CX2022SY023)。

摘　　要：为解决自然条件下人脸表情识别易受角度、光线、遮挡物的影响以及人脸表情数据集各类表情数量不均衡等问题,提出基于Res2Net的人脸表情识别方法。使用Res2Net50作为特征提取的主干网络,在预处理阶段对图像随机翻转、缩放、裁剪进行数据增强,提升模型的泛化性。引入广义平均池化(generalized mean pooling, GeM)方式,关注图像中比较显著的区域,增强模型的鲁棒性;选用Focal Loss损失函数,针对表情类别不平衡和错误分类问题,提高较难识别表情的识别率。该方法在FER2013数据集上准确率达到了70.41%,相较于原Res2Net50网络提高了1.53%。结果表明,在自然条件下对人脸表情识别具有更好的准确性。In order to solve the problem that facial expression recognition was easily affected by angle,light and occlusion and the facial expression dataset suffers from class imbalance under non-laboratory conditions,a facial expression recognition method based on Res2Net was proposed.The proposed method used Res2Net50 as the backbone network for feature extraction.In the pre-processing stage,the image underwent random operations such as flipping,zooming,cropping to enhance the data and improve the generalization of the model.The generalized mean pooling method(GeM)was used to focus on the more significant regions in the images to enhance the robustness of the model.Focal Loss function was used to solve the problem of expression class imbalance and misclassification,and the recognition rate of difficult to recognize expressions was improved.The accuracy of the proposed method reaches 70.41%on the FER2013 public dataset,which is 1.53%higher than that of the original Res2Net network.The experimental results show that this method has better accuracy in face recognition under unnatural conditions.

关 键 词：表情识别 Focal Loss函数 广义平均池化模块 Res2Net50 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]