中文 | English

J. Hefei Univ. Tech. (Nat. Sci.)

合肥工业大学学报(自然科学版)

JOURNAL OF HEFEI UNIVERSITY OF TECHNOLOGY (NATURAL SCIENCE)

Competent Authorities: Ministry of Education of the People's Republic of China

Sponsor by: Hefei University of Technology

ISSN 1000-1646, CN 34-1001/N

Small target sundries detection method for tobacco yarn production line based on improved YOLOv5s

Download Full Text

基于改进 YOLOv5s 的烟丝制丝生产线小目标杂物检测方法

Journal Information

J. Hefei Univ. Tech. (Nat. Sci.), Sept.2025, Vol.48 No.9: 1183-1191

DOI: 10.3969/j.issn.1003-5060.2025.09.005

Authors

ZHENG Yinhuan, CHEN Enjie, WU Fei, ZHANG Shuaibin

(School of Mechanical and Electrical Engineering, Wuhan University of Technology, Wuhan 430070, China)

Abstract and Keywords

Abstract: A small target detection algorithm based on the improved YOLOv5s is proposed. Firstly, based on YOLOv5s, Focal Loss and EIoU Loss are introduced into the model to optimize the original BCE Loss and CIoU Loss to accelerate the convergence speed of the model. Secondly, a target detection head is added to improve the detection accuracy of small target sundries. Finally, the influence of different types of attention modules on the model is compared and analyzed, and coordinate attention is introduced into the model neck to strengthen the extraction of key features of the target and improve the learning ability of the model. The model is trained based on the self-made sundries dataset, and the experimental results show that compared with YOLOv5s, the accuracy of the improved model on the test set is improved by 4.9%, the recall rate increases by 5.5%, and the mean average precision (mAP) value increases by 7.3%. The model has better recognition effect, which meets the accuracy and real-time requirements in actual production.

Keywords: small target detection; YOLOv5s algorithm; attention mechanism; detection head; loss function improvement

作者信息

郑银环，陈恩杰，吴飞，张帅彬

（武汉理工大学机电工程学院，湖北武汉430070）

摘要和关键词

摘要: 文章提出一种基于改进 YOLOv5s 的小目标检测算法，以 YOLOv5s 算法为基础模型，首先将 Focal Loss 和 EIoU Loss 引入模型优化原有的 BCE Loss 和 CIoU Loss，加快模型的收敛速度；其次添加一个目标检测头，提高对小目标杂物的检测精度；最后对比分析不同类型注意力模块对模型的影响，并将坐标注意力引入模型颈部，加强模型对目标关键特征的提取，提高模型的学习能力。基于自制的杂物数据集对模型进行训练，实验结果表明，相较 YOLOv5s 算法，改进后的模型在测试集上的精确率、召回率、平均精度均值（mean average precision，mAP）值分别提高 4.9%、5.5%、7.3%，识别效果更好，满足实际生产中精确性和实时性要求。

关键词: 小目标检测；YOLOv5s算法；注意力机制；检测头；损失函数改进

Funding

国家自然科学基金资助项目（52275505）

Previous Next

Back to Contents