[1]ZHANG Rui,ZHANG Yanchun,ZHANG Jiaqi,et al.Class-wise distribution-aware semi-supervised student behavior detection in crowded classrooms[J].CAAI Transactions on Intelligent Systems,2026,21(4):1013-1029.[doi:10.11992/tis.202510035]
Copy

Class-wise distribution-aware semi-supervised student behavior detection in crowded classrooms

CAAI Transactions on Intelligent Systems[ISSN 1673-4785/CN 23-1538/TP] Volume: 21 Number of periods: 2026 4 Page number: 1013-1029 Column: 学术论文—智能系统 Public date: 2026-07-05
Title:
Class-wise distribution-aware semi-supervised student behavior detection in crowded classrooms
Author(s):
ZHANG Rui1;  ZHANG Yanchun1;  ZHANG Jiaqi1;  BAI Qiaofeng2;  FAN Guangrui1;  KOU Xupeng1
1. College of Computer Science and Technology, Taiyuan University of Science and Technology, Taiyuan 030024, China;
2. School of Mechanical Engineering, Taiyuan University of Science and Technology, Taiyuan 030024, China
Keywords:
deep learning; computer vision; semisupervised learning; object detection; neural networks; feature extraction; pseudolabel; dynamic threshold
CLC:
TP391.4
DOI:
10.11992/tis.202510035
Abstract:
A semi-supervised student behavior detection model for dense classroom environments is proposed to address missed and false detections caused by high seat density, low individual pixel proportion, large intra-class variation, imbalanced class distribution, and limited availability of high-quality labeled data. The proposed model integrates a dual-stream interactive fusion module and a granular multi-level perception module to extract individual and interactive behavioral features and mitigate the suppression of behavioral interaction information. A dynamic interclass distribution loss and detection head are introduced to reduce the bias of the network’s bias toward frequent categories. To address inconsistencies in pseudo-label confidence distributions, a class-wise distribution-aware semi-supervised learning strategy is employed, using dynamic thresholding for pseudolabel selection to suppress noisy samples. Experiments on a self-constructed classroom dataset demonstrated that, under 10% labeled data conditions, the proposed model improves mAP50 by 3.2% over YOLOv11m, and the semi-supervised strategy further increases performance by 5.0%. Validation on public datasets confirms the effectiveness and generalization capability of the model. These findings provide robust technical support for student behavior recognition and intelligent classroom learning status analysis in smart classroom scenarios.
References:
[1] NWAKOBY C, ILOKA P. Regulatory challenges in implementing digital education in anambra state for sustainable development[J]. Unizik journal of educational laws and leadership studies, 2025, 1(1): 224-236
[2] 张炜, 刘怡. 中国高等学校分类的背景与思路——基于实施《教育强国建设规划纲要》的思考[J]. 高等工程教育研究, 2025(5): 1-8, 45 ZHANG Wei, LIU Yi. Backgroundand approachto classification of Chinese higher education institutions—based on the implementation of master planon building China into a leading country in education[J]. Research in higher education of engineering, 2025(5): 1-8, 45
[3] SHI Yuxuan, HUANG Wen, SANG Yijing. A narrative review of utilizing generative artificial intelligence in classroom instructions[C]//2024 4th International Conference on Educational Technology. Wuhan: IEEE, 2024.
[4] KUANG Yuanyuan, TANG Ying, XIE Tao. Objectives, methods, and challenges of applying intelligent assessment in education: a systematic review[C]//2024 International Symposium on Educational Technology. Macao: IEEE, 2024.
[5] SHEHZADI T, HASHMI K A, LIWICKI M, et al. Object detection with transformers: a review[J]. Sensors, 2025, 25(19): 6025
[6] 广田薰, 周政, MERSHA Bemnet Wondimagegnehu, 等. 教室环境下的教学质量AI机器评估研究进展[J]. 北京理工大学学报, 2024, 44(10): 1077-1088 HIROTA Kaoru1, ZHOU Zheng, MERSHA Bemnet Wondimagegnehu, et al. AI machine evaluation level on classroom teaching quality[J]. Transactions of beijing institute of technology, 2024, 44(10): 1077-1088
[7] 谭苏燕,王祖煊,何高大,等. 基于CA-YOLOv9网络的实时全景多尺度课堂行为识别[J]. 现代教育技术, 2024, 34(7): 123-130 TAN Suyan, WANG Zuxuan, HE Gaoda. Real-time panoramic multi-scale classroom behaviors recognition based on CA-YOLOv9 network[J]. Modern educational technology, 2024, 34(7): 123-130
[8] 张浩鹏, 施铮, 刘峰, 等. CPViG-Net: 基于局部跨阶段视觉图卷积的学生课堂行为识别[J]. 计算机科学, 2026, 53(2): 57-66 ZHANG Haopeng, SHI Zheng, LIU Feng. CPViG-Net: students’ classroom behavior recognition based on cross-stage visual graph convolution[J]. Computer science, 2026, 53(2): 57-66
[9] 胡锦林,齐永锋,王佳颖. 基于时空图卷积网络的学生在线课堂行为识别[J]. 光电子·激光, 2022, 33(2): 149-156 HU Jinlin, QI Yongfeng, WANG Jiaying. Recognition of Students’ Online Classroom Action Based on Spatio-temporal Graph Convolutional Network[J]. Optoelectronics & laser, 2022, 33(2): 149-156
[10] 周珍玉, 秦学, 蔡芳, 等. 基于人物交互的学生课堂行为识别研究[J]. 现代教育技术, 2024, 34(2): 53-61 ZHOU Zhenyu, QIN Xue, CAI Fang, et al. Research on classroom behavior recognition of students based on human-object interaction[J]. Modern educational technology, 2024, 34(2): 53-61
[11] LU Sheng, GUO Yangming, LONG Jiang, et al. Dense small target detection algorithm for UAV aerial imagery[J]. Image and vision computing, 2025, 156: 105485
[12] CHEN Yuming, YUAN Xinbin, WANG Jiabao, et al. YOLO-MS: rethinking multi-scale representation learning for real-time object detection[J]. IEEE transactions on pattern analysis and machine intelligence, 2025, 47(6): 4240-4252
[13] 刘建伟, 刘媛, 罗雄麟. 半监督学习方法[J]. 计算机学报, 2015, 38(8): 1592-1617 LIU Jianwei, LIU Yuan, LUO Xionglin. Semi-supervised learning methods[J]. Chinese journal of computers, 2015, 38(8): 1592-1617
[14] XU Bowen, CHEN Mingtao, GUAN Wenlong, et al. Efficient teacher: semi-supervised object detection for YOLOv5[EB/OL]. (2023-02-15)[2025-05-20]. https://arxiv.org/abs/2302.07577.
[15] MI Peng, LIN Jianghang, ZHOU Yiyi, et al. Active teacher for semi-supervised object detection[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans: IEEE, 2022.
[16] LI Jiaming, LIN Xiangru, ZHANG Wei, et al. Gradient-based sampling for class imbalanced semi-supervised object detection[C]//2023 IEEE/CVF International Conference on Computer Vision. Paris: IEEE, 2023.
[17] LIU Yen-chen, MA Chih-yao, KIRA Z. Unbiased teacher v2: semi-supervised object detection for anchor-free and anchor-based detectors[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans: IEEE, 2022.
[18] 徐彦威, 李军, 董元方, 等. YOLO系列目标检测算法综述[J]. 计算机科学与探索, 2024, 18(9): 2221-2238 XU Yanwei, LI Jun, DONG Yuanfang, et al. Survey of development of YOLO object detection algorithms[J]. Journal of frontiers of computer science and technology, 2024, 18(9): 2221-2238
[19] ALHASHMI S, AL-AZAWI A. A review of the single-stage vs. two-stage detectors algorithm: comprehensive insights into object detection[J]. International journal of environmental sciences, 2025, 11(3s): 775-787
[20] KHANAM R, HUSSAIN M. YOLOv11: an overview of the key architectural enhancements[EB/OL]. (2024-10-23)[2025-08-20]. https://arxiv.org/abs/2410.17725.
[21] ZAIDI S S A, ANSARI M S, ASLAM A, et al. A survey of modern deep learning based object detection models[J]. Digital signal processing, 2022, 126: 103514
[22] SUN Kailai, WANG Xinwei, MIAO Xi, et al. A review of AI edge devices and lightweight CNN and LLM deployment[J]. Neurocomputing, 2025, 614: 128791
[23] ALONSO J, ORUE-ECHEVARRIA L, CASOLA V, et al. Understanding the challenges and novel architectural models of multi-cloud native applications–a systematic literature review[J]. Journal of cloud computing, 2023, 12(1): 6
[24] ABNAR S, SHAH H, BUSBRIDGE D, et al. Parameters vs FLOPs: scaling laws for optimal sparsity for mixture-of-experts language models[EB/OL]. (2025-01-21)[2025-08-20]. https://arxiv.org/abs/2501.12370.
[25] MAHMOUD E, BAWANEH A K. Best practices of effective classroom management strategies supported by digital ICT in higher education[J]. International journal of evaluation and research in education, 2025, 14(3): 2337
[26] QU Feilong, ALI KHAN N, JHANJHI N Z, et al. Improved YOLOv5 lane line real time segmentation system integrating seg head network[C]//The 7th International Global Conference Series on ICT Integration in Technical Education & Smart Society. Aizuwakamatsu City: MDPI, 2025.
[27] CAI Boyu, WANG Huibing, YAO Mingze, et al. Focus more on what? guiding multi-task training for end-to-end person search[J]. IEEE transactions on circuits and systems for video technology, 2025, 35(7): 7266-7278
[28] KLINKER F. Exponential moving average versus moving exponentialaverage[J]. Mathematische semesterberichte, 2011, 58(1): 97-107
[29] DEVERNAY F. A non-maxima suppression method for edge detection with sub-pixel accuracy[D]. Rocquencourt: INRIA, 1995.
[30] 国家标准化管理委员会. 中小学、幼儿园安全防范要求[S]. 北京: 中国标准出版社, 2022. Standardization Administration of China. Safety requirements for primary, secondary schools and kindergartens[S]. Beijing: China standards press, 2022.
[31] LIU Yen-cheng, MA Chih-yao, HE Zijian, et al. Unbiased teacher for semi-supervised object detection[EB/OL]. (2023-10-04)[2025-08-20]. https://arxiv.org/abs/2102.09480.
[32] YANG Fan, WANG Tao. SCB-Dataset3: a benchmark for detecting student classroom behavior[EB/OL]. (2023-10-04)[2025-08-20]. https://arxiv.org/abs/2310.02522.
Similar References:

Memo

-

Last Update: 1900-01-01

Copyright © CAAI Transactions on Intelligent Systems