FA-YOLOv9: Improved YOLOv9 Based on Feature Attention Block

In educational environments, controlling students’ behaviors and gestures is a significant challenge. With the advancements in deep learning technology, utilizing videos and images for detecting behaviors, events, and objects has gained considerable attention. This research introduces a novel method to enhance the detection capability of the YOLOv9 model by integrating the Feature Extraction (FA) Block for detecting student behaviors in educational settings. With the incorporation of FA, an attention mechanism, the model can fine-tune feature representations and prioritize crucial multi-spatial and channel features. Our results highlight the effectiveness of employing attention mechanisms within deep learning frameworks for recognizing subtle behaviors. This research not only contributes to the advancement of computer vision techniques but also holds promise for practical applications in educational settings, facilitating efficient and comprehensive monitoring of student engagement and behavior. Through detailed experiments on benchmark datasets, we demonstrate the effectiveness of our fine-tuned FA-YOLOv9 model in accurately detecting diverse student behaviors. Specifically, we achieved SOTA results of 77.8% mAP50 and 74.3% Recall compared to other YOLO models.