基于改进YOLOX的教师行为检测方法研究  

Research on Teacher Behavior Detection Method Based on Improved YOLOX

作　　者：王朋涛 叶黎伟 WANG Peng-tao;YE Li-wei(North China University of Water Resources and Electric Power,Zhengzhou 450046,China)

机构地区：[1]华北水利水电大学信息工程学院,河南郑州450046

出　　处：《电脑与信息技术》2023年第5期58-61,共4页Computer and Information Technology

摘　　要：目前对线下课堂教学行为的评价还停留在原始的随堂专家评审的阶段,教师的课堂教学行为检测存在着的费时费力以及评审结果不准确等问题。现有研究中,针对教师的课堂行为数据较少且大多数实验数据不规范。根据以上问题,文章提出一种规范化的教师课堂行为检测数据集,该数据集包含二千多张高清图片,共有四千多个标签,涵盖八类教学行为。文章还提出了基于改进YOLOX的实时检测模型。该模型在YOLOX中引入高效通道注意力模块(ECA),在教师课堂行为检测数据集上mAP提高了10.99%达到了95.29%。然后在Neck部分加入自适应空间特征融合(ASFF),使mAP再次提高了2.19%,达到了97.48%。并且推理速度也得到了明显提升。At present,the evaluation of offline classroom teaching behavior is still in the original stage of in-class expert evaluation,and there are some problems such as time-consuming and laborious detection of teachers'classroom teaching behavior and inaccurate evaluation results.In the existing researches,the classroom behavior data of teachers are few and most experimental data are not standard.According to the above problems,this paper proposes a standardized classroom behavior detection dataset of teachers,which contains more than 2,000 high-definition images and more than 4,000 labels,covering eight types of teaching behaviors.This paper also proposes a real-time detection model based on improved YOLOX.In this model,the efficient channel attention module(ECA)is introduced into YOLOX,and the mAP is improved by 10.99%to 95.29%on the teacher's classroom behavior detection dataset.Then adaptive spatial feature fusion(ASFF)is added to the Neck part,which increases the mAP by 2.19%again to 97.48%.And the reasoning speed has also been significantly improved.

关 键 词：教师课堂行为 YOLOX 目标检测 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术]