The development of curriculum culture is a profound reflection on the evolution of school curriculum in the present era. The article proposes a Retina Student network to detect students’ classroom behaviors and then incorporates the implementation of a classroom behavior coding system using computer vision. Based on this, we established a student detection dataset for the classroom attention recognition task, utilizing the Retina Studen student detection algorithm to achieve optimal recognition accuracy of 0.92 and 0.93, and recall of 0.91 and 0.94 for both gaze and non-gaze states. Then, the visualization information of students’ learning behaviors was targeted, and the classroom status was discussed. The experimental results demonstrate that this paper’s algorithm outperforms other algorithms in identifying student learning behaviors. Finally, through the coding analysis of the teaching video of the cultural education smart classroom, we found that students’ behaviors in the calligraphy, painting, and opera smart classroom are rich, and the traditional way of “explaining-questioning-answering-feedback” is still the main behavior in the classroom.