Research on an Improved Model for Expression Recognition in Classrooms

Aiming at the real-time demand of expression recognition in online classroom application, this paper proposes an improved model to balance the real-time and accuracy of expression recognition. The optimization model can comprehensively consider the recognition time and recognition accuracy. As a result, it can better support the real-time understanding of students' class status in the process of online teaching. This paper mainly improves the model from the following aspects. Firstly, in the input stream structure, large convolution core and small pool core are used to reduce the amount of calculation of the training network. Secondly, in the intermediate stream structure, the depth separable convolution is used to make full use of the spatial correlation and channel correlation of the images in the video. Thirdly, in the intermediate flow structure, the deep separable convolution and residual network are fused into a hybrid module to reduce over fitting and enhance the generalization ability of the model. Finally, in the output stream structure, the global average pooling layer is used to replace the full connection layer to improve the efficiency of expression recognition. Through detailed experimental comparison and model analysis, the model improvement scheme proposed in this paper can occupy a good advantage in an ideal range, considering the training time, recognition time and recognition accuracy. The optimization model can be used to obtain students' expressions and analyze students' online class status through students' video data.