针对智慧课堂行为检测中因存在学生人数密集、相互遮挡及目标尺度差异大而导致检测精度较低的问题,提出了一种基于改进YOLO11的课堂行为检测模型ADU-YOLO11。首先,将部分卷积下采样层替换为Adown下采样模块,减少模型复杂度及关键特征信息损失;其次,采用动态检测头DyHead（Dynamic Head）,利用其尺度、空间、任务感知注意力增强检测能力;最后,采用UIoU（Unified-IoU）损失函数优化训练收敛速度并提高预测框的回归精度。实验结果表明,ADU-YOLO11在STBD-08学生教师行为数据集上,平均精度均值mAP50及mAP50-95、精确率、召回率相较原始YOLO11n分别提升了1.1%、1.6%、2.7%和0.6%,且优于其他目标检测算法,证明了其在智慧课堂行为检测中的有效性与优越性。
Abstract
To address the issue of low detection accuracy in smart classroom behavior detection due to dense student populations, mutual occlusion, and large differences in target scales, an improved YOLO11-based classroom behavior detection model, ADU-YOLO11, is proposed. Firstly, the partial convolutional downsampling layers are replaced with the Adown downsampling module to reduce model complexity and minimize the loss of key feature information. Secondly, the dynamic detection head DyHead (Dynamic Head) is adopted, which enhances detection capabilities through scale, spatial, and task-aware attention. Finally, the UIoU (Unified-IoU) loss function is used to optimize the training convergence speed and improve the regression accuracy of the predicted bounding boxes. Experimental results show that ADU-YOLO11 achieves mAP50 and mAP50-95 improvement of 1.1% and 1.6% respectively, along with a precision increase of 2.7% and a recall increase of 0.6% compared to the original YOLO11n on the STBD-08 student-teacher behavior dataset. Moreover, it outperforms other object detection algorithms, demonstrating its effectiveness and superiority in smart classroom behavior detection.
关键词
课堂行为检测 /
YOLO11 /
Adown /
DyHead /
UIoU
Key words
classroom behavior detection /
YOLO11 /
Adown /
DyHead /
UIoU
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 中共中央、国务院印发《教育强国建设规划纲要(2024—2035年)》[N].人民日报,2025-01-20(006).
[2] 周宇航,张海宇,王永彬,等.基于AlphaPose的课堂行为识别算法[J].信息技术与信息化,2023(12):204-208.
[3] 吕文超. 基于改进SSD的高校学生课堂行为智能识别技术研究[D].廊坊:防灾科技学院,2023.
[4] Zhang X,Nie J,Wei S,et al.A study of classroom behavior recognition incorporating super-resolution and target detection[J].Sensors,2024,24(17):5640.
[5] 王鑫钰,高东怀,宁玉文,等.基于改进YOLO算法的学生行为检测方法研究[J/OL].计算机科学,1-15[2025-07-18].http://kns.cnki.net/kcms/detail/50.1075.tp.20250317.1846.032.html.
[6] Khanam,R.,& Hussain,M.(2024).Yolov11:An overview of the key architectural enhancements.arXiv preprint arXiv:2410.17725.
[7] Wang C Y,Yeh I H,Mark Liao H Y.Yolov9:Learning what you want to learn using programmable gradient information[C]//European conference on computer vision.Cham:Springer Nature Switzerland,2024:1-21.
[8] Dai X,Chen Y,Xiao B,et al.Dynamic head:Unifying object detection heads with attentions[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2021:7373-7382.
[9] Luo X,Cai Z,Shao B,et al.Unified-IoU:For High-Quality Object Detection[J].arXiv preprint arXiv:2408.06636, 2024.
[10] Zhao J,Zhu H.Cbph-net:A small object detector for behavior recognition in classroom scenarios[J].IEEE transactions on instrumentation and measurement,2023,72:1-12.
