摘 要：针对深度卷积神经网络随着卷积层数增加而导致网络模型难以训练和性能退化等问题，提出了一种基于深度残差网络的人脸表情识别方法。采用改进后的 ResNet18 模型，结合数据增强、mixup、label smoothing 等辅助策略对 FER2013 训练集进行 300 个 epoch 的训练，利用最优的权重，在 FER2013 的验证数据集上达到了 72.09% 的准确率；并结合 YOLOv5Face预训练权重，实现了人脸检测和表情识别。
关键词：人脸表情识别；人脸检测；深度学习；ResNet18
DOI:10.19850/j.cnki.2096-4706.2022.20.022
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2022）20-0090-05
Research on Facial Expression Recognition System Based on Deep Learning
FAN Wenjie, TIAN Xiuyun
(School of Electronics and Information Engineering, Guangdong Ocean University, Zhanjiang 524088, China)
Abstract: Aiming at the problems of network model training difficulty and performance degradation due to the increase of convolutional layers in deep convolutional neural network, a facial expression recognition method based on deep residual network is proposed. By using the improved ResNet18 model and combining auxiliary strategies such as data enhancement, mixup and label Smoothing, FER2013 training sets are trained for 300 epoch. Using the optimal weights, the accuracy of FER2013 validation data set reaches 72.09%.Combined with the weight of YOLOv5Face pre-training, face detection and expression recognition are realized.
Keywords: facial expression recognition; face detection; deep learning; ResNet18
参考文献：
[1] REDMON J，DIVVALA S，GIRSHICK R. You Only Look Once：Unified，Real-Time Object Detection [J/OL].（2022-06-09）. https：//arxiv.org/abs/1506.02640.
[2] BOCHKOVSKIY A，WANG C Y，LIAO H Y. YOLOv4： Optimal Speed and Accuracy of Object Detection [J/OL].（2022-06-09）. https：//arxiv.org/abs/2004.10934.
[3] HOWARD A，SANDLER M，CHU G，et al. Searching for MobileNetV3 [J/OL].（2022-06-09）.https：//arxiv.org/abs/1905.02244.
[4] HE K M，ZHANG X Y，REN S Q，et al. Deep Residual Learning for Image Recognition [J/OL].（2022-06-09）.https：//arxiv. org/abs/1512.03385.
[5] GOODFELLOW I J，ERHAN D，CARRIER P L， et al. Challenges in Representation Learning：A report on three machine learning contests [J/OL].（2022-06-09）.https：//arxiv.org/ abs/1307.0414.
[6] ZHANG H，CISSE M，DAUPHIN Y N. mixup：Beyond Empirical Risk Minimization [J/OL].（2022-06-09）.https：//arxiv.org/ abs/1710.09412.
[7] MÜLLER R，KORNBLITH S，HINTON G. When Does Label Smoothing Help? [J/OL].（2022-06-09）.https：//arxiv.org/ abs/1906.02629.
作者简介：范文杰（2000—），男，汉族，广东清远人，本科在读，研究方向：区块链及深度学习；通讯作者：田秀云（1974—），女，汉族，广东湛江人，讲师，硕士，研究方向：光电技术。