【摘要】 人脸情绪识别一直是机器学习研究中重要的课题，发现准确率高、表现优秀网络的特点可以为情绪识别提供优化设计思路。利用卷积神经网络对情绪数据集进行学习，通过VGG16与ViT、MobileNet、ResNet50等模型在优化RAFDB数据集上的表现进行对比，可以探究各模型在情绪识别过程中的处理方法和测试结果，总结准确识别网络的特点。通过实验数据发现，VGG16在测试数据集上具有更显著的准确性，其综合识别准确度可以达到95.1%。在恐惧、伤心等特殊强烈情绪的准确率上，VGG16相较于其他卷积神经网络具有较明显的优势。对VGG16网络仍可能改进的方向进行分析，提出了未来VGG16可进一步在表情识别领域发展的方向。利用VGG16网络在人脸识别系统中引入情绪识别，能够为安防、医疗、监护、机器人制造等领域提供更精确有效的支持。更多还原

【Abstract】 Facial emotion recognition has always been an important topic in machine learning research. Finding the characteristics of high accuracy and excellent performance network can provide an optimal design idea for emotion recog-nition. The convolutional neural network is used to learn the emotion dataset. By comparing the performance of VGG16 with Vit, Mobilenet, Res Net50 and other models on the optimized RAF-DB dataset, the processing methods and test results of each model in the emotion recognition process can be explored, and the characteristics of the accurate recognition network can be summarized. Through the experimental data, it is found that VGG16 has more significant accuracy on the test dataset, and its comprehensive recognition accuracy can reach 95.1%. Compared with other convolutional neural networks,VGG16 has obvious advantages in the accuracy of special strong emotions such as fear and sadness. The possible improvement direction of VGG16 network is analyzed, and the further development direction of VGG16 in the field of expression recognition in the future is proposed in this paper.更多还原