基于端到端表情识别方法的课堂教学分析

Classroom teaching analysis based on end-to-end facial expression recognition

为了更好地掌握及分析课堂教学情况,采用端到端的方法搭建了一种表情识别模型,用于评价学生的上课抬头率和活跃度等。模型使用Opencv库中的Haar特征分类器进行人脸检测,基于卷积神经网络方法进行表情识别。该卷积神经网络在Tensorflow框架下搭建,模型在输入层之后加入1*1的卷积核来保证输入的非线性,再使用逐渐减小的卷积核逐步提取更加精细的人脸表情特征,并将人脸划分为7种不同的表情,采用FER2013数据集进行训练。研究结果显示,此模型在FER2013数据集的测试集上取得了较好的识别率且网络参数和计算量均较小,便于实时检测。

In order to better grasp and analyze the classroom teaching situation,an end-to-end method is adopted to build an expression recognition model to evaluate students′head-up rate and activity in class.The model uses the Haar feature classifier in the Opencv library for face detection and facial expression recognition based on the convolutional neural network method.The network is built under the Tensorflow framework.The model adds a 1*1 convolution kernel after the input layer to ensure the nonlinearity of the input,and then uses the gradually reduced convolution kernel to extract more refined facial expression features step by step,and students′faces are divided into 7 different expressions,which are trained by the application of the FER2013 data set.This model has achieved a good recognition rate on the test set of the FER2013 data set,and has fewer network parameters and a small amount of calculation,which is more suitable for real-time detection.