Classroom Facial Emotion Recognition Based on Channel Attent

Abstract

In order to solve the problem of camera shooting distance, indoor illumination and other factors in classroom scene, and to solve the problem of unbalanced expression image and poor recognition effect caused by fuzzy facial images of students, a deep convolution' classroom facial analysis' method based on channel interaction is proposed. Firstly, there is an imbalance in the camera intake of students' facial expression images in the classroom state, and GANs are used for image expansion preprocessing. Secondly, the collected fuzzy facial images are enhanced by deep separable convolution and channel interaction ECA to avoid dimension reduction and appropriate cross-channel interaction to improve recognition rate. Finally, the self-built classroom expression data set is used for comparative experiments in the lightweight model. Experimental results show that data enhancement and model improvement have good effects on facial expression analysis.