优化算法在人脸表情识别中的应用研究  

Application of optimization algorithm in facial expression recognition

作　　者：黄豪豪 李铭田 张富春[1] HUANG Haohao;LI Mingtian;ZHANG Fuchun(School of Physics and Electronic Information,Yan’an University,Yan’an 716000,China)

机构地区：[1]延安大学物理与电子信息学院,陕西延安716000

出　　处：《延安大学学报（自然科学版）》2022年第3期56-60,65,共6页Journal of Yan'an University：Natural Science Edition

基　　金：延安大学研究生教育创新计划项目(YCX2021066);延安大学研究生教育教学改革研究项目(YDYJG2019016)。

摘　　要：在人脸表情识别任务中,适用的优化算法可以有效地提高表情识别的效率。针对人脸表情识别任务中的优化算法选择问题,比较研究了SGD、Momentum以及Adagrad、Adadelta、Adam 3种自适应学习率方法在人脸表情识别任务上的表现。特别是为了检验结果的可靠性,采用相同方法在MNIST数据集上进行手写数字识别测试。实验结果显示,在人脸表情识别与其他任务中,自适应学习率方法和动量法性能优于SGD方法,且自适应学习率方法在提高模型准确率上更为突出,Adadelta在表情识别和手写数字识别任务上的准确率达到了96.12%和99%。研究表明,在人脸表情识别任务中,自适应学习率的优化算法具有明显优势。In facial expression recognition,the applicable optimization algorithm can effectively improve the efficiency.Aiming at the selection of optimization algorithm in facial expression recognition task,this paper compares the performance of SGD,Momentum and three adaptive learning rate methods(Adagrad,Adadelta and Adam).In order to test the reliability of the results,this paper uses the same method to test handwritten numeral recognition on Minist data set.The results show that in the tasks of expression recognition and handwritten numeral recognition,the performance of adaptive learning rate method and Momentum method is better than SGD method,and the adaptive learning rate method is more prominent in improving the accuracy of the model.The accuracy of Adadelta in the tasks of expression recognition and handwritten numeral recognition is 96.12%and 99%.The above research shows that the optimization algorithm of adaptive learning rate demonstrates obvious advantages in the task of facial expression recognition.

关 键 词：深度学习 表情识别 随机梯度下降 自适应学习率 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]