Application of intelligent classroom management system based on deep learning in smart classrooms

In the prevailing educational paradigm, classroom management predominantly depends on teachers' subjective judgments and static assessments as the primary evaluative approach. This conventional model lacks real-time and dynamic feedback mechanisms, resulting in inherent inefficiencies. To address this problem, this paper develops an intelligent classroom management monitoring system based on deep learning. The system consists of three parts: image acquisition module, image processing module and real-time feedback module. The system detects the behavioral information of students in the classroom through sensor technology and image recognition technology. In the image processing module section, we propose an improved YOLO11 model, the YOLO11-DP model, which effectively achieves real-time and efficient detection of student behavior in the classroom. The YOLO11-DP model significantly enhances detection performance through three key innovations: first, the introduction of a Diverse Branch Block (DBB) module in the C3k2 module effectively overcomes the representation capability bottleneck of the original YOLO11 during the feature fusion stage. Second, the traditional SPPF module is replaced with Poly-Scale Convolution (PSConv), which uses its dynamically varying dilation rate to achieve adaptive multi-scale feature extraction. Finally, the innovative mixed local channel attention mechanism (MLCA) is added, further optimizing the model's detection accuracy and computational efficiency in multi-object scenarios. Results show that the YOLO11-DP-MLCA model demonstrates excellent detection accuracy on the STBD-08 dataset, achieving a mAP@0.5 of 94.2%, and also exhibits good recognition performance in detecting student behavior in real classroom scenarios.