Search papers, labs, and topics across Lattice.
This paper introduces ALC-YOLOv8s, an improved YOLOv8s model, to address challenges in student behavior recognition within complex classroom settings, specifically targeting dense, small, and occluded objects with imbalanced class distributions. The model incorporates SPPF-LSKA for enhanced contextual feature extraction, CFC-CRB and SFC-G2 for optimized multi-scale feature fusion, and ATFLoss to improve learning for minority classes. Experiments demonstrate that ALC-YOLOv8s achieves a 1.8% increase in mAP50 and a 2.1% increase in mAP50-95 compared to the baseline, outperforming other mainstream detection methods.
Overcome the chaos of classroom behavior recognition with ALC-YOLOv8s, achieving state-of-the-art detection of dense, occluded, and imbalanced student actions.
In classroom teaching, student behavior can reflect their learning state and classroom participation, which is of great significance for teaching quality analysis. To address the problems of dense student targets, numerous small objects, frequent occlusions, and imbalanced class distribution in real classroom scenes, this paper proposes an improved student classroom behavior recognition model named ALC-YOLOv8s based on YOLOv8s. The model introduces SPPF-LSKA to enhance contextual feature extraction, employs CFC-CRB and SFC-G2 to optimize multi-scale feature fusion, and incorporates ATFLoss to improve the learning ability for minority classes and hard samples. Experimental results show that compared with the baseline model, the improved model achieves increases of 1.8% in mAP50 and 2.1% in mAP50-95. Compared with several mainstream detection methods, the proposed model can well meet the requirements of automatic student behavior recognition in complex classroom scenarios.
