改进YOLOv9的非机动车头盔佩戴检测算法

doi:10.16039/j.cnki.cn22-1249.2026.01.003

下载:

PDF (5245KB)
输出: BibTeX | EndNote (RIS)

摘要针对在高密度复杂骑行条件下，非机动车头盔佩戴检测精度低、漏检率高等问题，提出了一种基于改进YOLOv9的非机动车头盔佩戴检测算法。首先，在分类预测损失上，将Slide Loss损失函数添加到YOLOv9中，提高模型对小目标和遮挡物体的检测性能；其次，在主干网络中引入深度可分离卷积模块（Depthwise Separable Convolution，DSConv），减少了模型的参数量；最后，在模型的Head网络中添加DASI小目标检测模块，提高了对小目标的检测能力。与YOLOv9算法相比，改进后算法的精确度（Precision）、召回率（Recall）、平均精度（mean Average Precision）分别提升了2.3%、1.9%和2.8%，且参数量相较于原模型下降了9.1%，能够实现对非机动车头盔佩戴的高精度检测。

	服务
	把本文推荐给朋友
	加入引用管理器
	E-mail Alert
	RSS
	（）
	作者相关文章
	刘麒
	张朔
	常广良
	王影

关键词: 头盔检测 YOLOv9 Slide Loss DSConv DASI小目标检测

Abstract: This study proposes an improved YOLOv9-based helmet wearing detection algorithm for non-motor vehicle riders, aiming to solve the problems of low accuracy and high missed detection rate under high-density and complex riding scenarios.Firstly, the Slide Loss loss function was incorporated into YOLOv9 to enhance the model's detection performance on small targets and occluded objects. Secondly, Depthwise Separable Convolution (DSConv) modules were introduced into the backbone network, reducing the model's parameter count. Lastly, a DASI small target detection module was added to the model's Head network, improving the detection capability for small targets. Compared to the YOLOv9 algorithm, the improved algorithm achieved increases of 2.3% in Precision, 1.9% in Recall, and 2.8% in mean Average Precision, while the parameter count was reduced by 9.1% compared to the original model, enabling high-precision detection of non-motor vehicle helmet wearing.

Key words: helmet detection YOLOv9 Slide Loss DSConv DASI small target detection

出版日期: 2026-01-25 发布日期: 2026-07-11 整期出版日期: 2026-01-25

ZTFLH:

TP 391.41

引用本文:

刘麒, 张朔, 常广良, 王影. 改进YOLOv9的非机动车头盔佩戴检测算法[J]. 吉林化工学院学报, 2026, 43(1): 16-22.
LIU Qi, ZHANG Shuo , CHANG Guangliang, WANG Ying. Improve the Helmet Wearing Detection Algorithm for Non-motorized Vehicles Based on YOLOv9. Journal of Jilin Institute of Chemical Technology, 2026, 43(1): 16-22.

链接本文:

https://xuebao.jlict.edu.cn/CN/10.16039/j.cnki.cn22-1249.2026.01.003 或 https://xuebao.jlict.edu.cn/CN/Y2026/V43/I1/16