基于神经网络的面部表情鉴别系统的研究与实现.docx

PAGE PAGE 31 基于神经网络的面部表情鉴别系统 摘 要 随着人工智能时代的发展，以人为本的研究理念开始成为主要研究方向。为了能够让计算机理解人类的意愿和内心需求，面部表情鉴别成为计算机了解人类情绪变化的关键技术。它通过对人脸面部表情特征进行特征提取和分类，从而辨别面部表情。目前为止，在众多行业例如人机交互、网络在线教育、安全驾驶等均能看到该技术的应用价值。针对面部表情鉴别技术的研究方法众多，而本文主要研究的是基于神经网络的面部表情鉴别系统。 本次论文选择Fer2013数据集作为本次模型训练的数据集并对其进行预处理，得到六种表情的数据集。其次运用卷积神经网络原理采用Tensorflow 2.0深度学习框架设计一种11层网络结构的模型，包括4层卷积层、3层池化层和4层全连接层，并对其训练结果进行分析。根据最后的测试，该模型的面部表情鉴别精度为62%。将利用摄像头实时采集人脸图像并对其进行鉴别，而对于Normal和happy的鉴别效果较好，其余表情的鉴别次之。 针对以上结果，本文提出该方案的可行性以及不足点。 关键词： 面部表情鉴别技术，数据集预处理，卷积神经网络，Tensorflow框架 Facial expression identification system based on Neural Network Abstract With the development of the era of artificial intelligence, the people-oriented research concept has become the main research direction. In order to allow computers to understand human wishes and inner needs, facial expression identification has become a key technology for computers to understand changes in human emotions. It distinguishes facial expressions by extracting and classifying features of facial expressions.So far, the application value of this technology can be reflected in many industries such as human-computer interaction, online education and safe driving.There are many research methods for facial expression identification, and this paper mainly studies the facial expression identification system based on neural network. In this paper, the Fer2013 data set is selected as the data set for this model training and pre-processed to obtain a data set of six expressions. Then, an 11-layer network structure model based on the principle of convolutional neural network is designed, including 4 convolutional layers, 3 pooling layers and 4 fully connected layers, and the training results are analyzed. According to the final test, the model's facial expression discrimination accuracy is 62%. The camera will be used to collect face images in real time and identify them, and the identification effect for Normal and happy is better, followed by the identification of the remaining expression