In the field of computer vision, human action recognition has received a lot of interest, particularly with regard to education environments. The classroom camera can flexibly automatically detect violations of classroom rules or students’ expressed opinions without the intervention of the teacher. Therefore, this paper presents a student behavior recognition system in the classroom based on deep learning. The three main processing functions in this system are skeletal data collection, feature extraction, and behavior classification. Initially, the OpenPose model is used to collect skeletal data. Subsequently, the system performs feature extraction to generate descriptive feature vectors that encapsulate key attributes of human postures. Finally, behavior categorization is conducted to identify and classify specific student behaviors based on the extracted features. The model’s training and validation accuracy are highly accurate, with mAP50 scores of 0.9994 and 0.9991, respectively. The experimental results demonstrate the effectiveness of the proposed system.