基于RI-YOLO的学生行为检测算法

Student behavior detection algorithm based on RI-YOLO

牛泽刚 ¹赵玉兰²

扫码查看

作者信息

1. 吉林化工学院,吉林吉林 132022
2. 吉林化工学院,吉林吉林 132022;吉林农业科技学院,吉林吉林 132101
折叠

摘要

针对学生行为检测算法准确率不高、易出现漏检误检问题,文章提出了一种改进的RI-YOLO学生行为检测算法.该算法通过引入感受野注意力卷积(Receptive-Field Attention Convolution,RFAConv)对C3 模块进行优化,提出新型的RFAC3 模块,可以更精确地捕捉细微的局部特征,提升网络特征提取能力.此外,采用基于辅助边框的交并比(Intersection over Union,IoU)损失函数Inner-IoU替代传统损失函数,加速模型的收敛速度.在学生课堂行为数据集 SCB-Dataset3 上验证表明,RI-YOLO平均精度mAP50 较YOLOv5 提升了1.5%,mAP50:95 提升了1.2%,与其他主流检测模型对比,展示出了优异检测效果.

Abstract

To address the issues of low accuracy in student behavior detection algorithms,which often lead to missed detections and false positives,the article proposes an improved student behavior detection algorithm based on YOLOv5s called RI-YOLO.The algorithm optimizes the C3 module by introducing Receptive-Field Attention Convolution(RFAConv),proposing a new RFAC3 module that can more accurately capture subtle local features,thereby enhancing the network's feature extraction capabilities.Additionally,it adopts an Inner-IoU loss function based on auxiliary bounding boxes to replace traditional loss functions,accelerating the convergence speed of the model.Testing on the student classroom behavior dataset SCB-Dataset3 shows that RI-YOLO improves mean average precision(mAP50)by 1.5%compared to YOLOv5,and mAP50:95 by 1.2%,demonstrating superior detection performance when compared with other mainstream detection models.

关键词

目标检测/学生行为/RFAC3/辅助边框/平均精度

Key words

object detection/student behavior/RFAC3/auxiliary bounding box/mean average precision

引用本文复制引用

出版年

2024

无线互联科技

江苏省科学技术情报研究所

无线互联科技

影响因子：0.263

ISSN：1672-6944

段落导航