基于改进轻量卷积神经网络MobileNetV3的人脸表情识别  

Facial expression recognition based on improved lightweight convolutional neural network MobileNetV3

作　　者：雷晓鹏 Lei Xiaopeng(School of Computer Science and Technology,Shanghai University of Electric Power,Shanghai 200000,China)

机构地区：[1]上海电力大学计算机科学与技术学院,上海200000

出　　处：《现代计算机》2024年第10期29-34,共6页Modern Computer

摘　　要：人脸表情识别在授课中应用的及时检测可有效提升教育质量和学生参与度。为实现人脸表情识别在授课中的实时检测,该研究基于卷积神经网络MobileNetV3进行学习,对SE和卷积层进行了改进,以构建人脸表情识别模型,可识别八种不同的表情类别。研究空洞卷积的位置对模型性能的影响,发现将空洞卷积放在网络的前部对性能有积极影响,而放在后部则会导致性能下降。同时,通过引入SSE(space squeeze-and-excitation)模块并优化其位置和结构,进一步提高了模型性能。最终提出的MobileNetV3改进版本在参数数量和模型文件大小上有显著减小,但精度下降了1%左右。对模型进行了多次随机试验,鲁棒性良好。该研究可为人脸表情识别在授课中的实时应用提供理论基础和技术支持,未来将致力于开发可在移动端应用的人脸表情识别系统。The timely detection of facial expression recognition in teaching can effectively improve the quality of education and student engagement.To achieve real-time detection of facial expression recognition in teaching,this study is based on the con-volutional neural network MobileNetV3 for learning,and improves SE and convolutional layers to construct a facial expression rec-ognition model that can recognize 8 different expression categories.Studying the impact of the position of void convolutions on model performance and found that placing void convolutions at the front of the network has a positive impact on performance,while placing them at the back can lead to performance degradation.At the same time,the performance of the model was further im-proved by introducing the SSE(Space Squeeze and Extraction)module and optimizing its position and structure.The final proposed improved version of MobileNetV3 significantly reduced the number of parameters and model file size,but the accuracy decreased by about 1%.And multiple random experiments were conducted on the model,which showed good robustness.This study can pro-vide theoretical basis and technical support for the real-time application of facial expression recognition in teaching.In the future,

关 键 词：人脸表情识别 卷积神经网络 MobileNetV3 空洞卷积 SSE模块 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]