Optimized YOLO-AC Model for Behavior Detection in Smart Classrooms Using GSConv and Attention Mechanisms

The rapid advancement of information technology has made behavior detection in smart classrooms a crucial research focus in modern education. However, existing methods face challenges in achieving high detection accuracy and robustness, particularly in complex scenarios such as target occlusion, dense environments and subtle behavioral variations. To address these challenges, we propose an enhanced YOLO-AC model that integrates the GSConv convolutional structure and the CBAM attention mechanism. These innovations improve feature extraction efficiency and enable the model to focus more effectively on key regions, thus enhancing both detection accuracy and robustness. Specifically, the GSConv structure refines the feature extraction process, enabling better capture of fine details in complex settings. The CBAM attention mechanism further strengthens the recognition of essential features, leading to improved model performance. Experimental results demonstrate that the YOLO-AC model significantly outperforms mainstream models such as YOLOv8n and YOLOv7-Tiny, achieving superior mAP50 and mAP50-95 metrics on the RoboFlow and SCB-Dataset3 datasets. Moreover, the proposed model reduces both parameter count and computational complexity. These findings confirm the efficiency and practical applicability of YOLO-AC for behavior detection in smart classrooms, providing a robust, deployable solution for dynamic environments.