Abstract: In view of the problems that classroom teaching scenes with severe student occlusion and extreme posture, and the inapplicability of traditional facial emotion recognition algorithms to classroom scenes and the lack of public academic emotion data, a student classroom facial detection dataset and a classroom emotion dataset are constructed, and a classroom academic emotion recognition algorithm based on face correction recovery is proposed. A real-time classroom face detection model was constructed to obtain students’ facial information. A correction recovery algorithm was used to repair damaged faces. A two-way academic emotion recognition network was designed to identify academic emotions. The system module was integrated to achieve real-time recognition with sparse video sampling under limited classroom conditions.