基于改进YOLOv8s算法的饮品瓶盖缺陷检测  

Beverage bottle cap defect detection based on improved YOLOv8s algorithm

作　　者：孙铭见 刘国洋 李春伟[1] SUN Mingjian;LIU Guoyang;LI Chunwei(College of Home and Art Design,Northeast Forestry University,Harbin 150040,China)

机构地区：[1]东北林业大学家居与艺术设计学院,哈尔滨150040

出　　处：《包装与食品机械》2025年第2期20-29,共10页Packaging and Food Machinery

基　　金：黑龙江省自然科学基金项目(LH2019E001);东北林业大学院级大学生创新训练项目(202310225649)。

摘　　要：针对饮品瓶盖缺陷检测技术存在特征提取能力不足、检测精度低、收敛速度慢等问题,提出一种基于改进YOLOv8s算法的饮品瓶盖缺陷检测模型(YOLOv8s-GCW)。以YOLOv8s算法为基准模型,在主干网络中SPPF模块的上一层加入Global Context注意力模块;使用WIoU(v3)替换原有的CIoU损失函数;构建饮品瓶盖缺陷数据集,并采用数据增强技术扩充。试验结果表明,与原模型相比,在不增加计算成本的基础上,可实现更高的检测精度,平均检测精度达到91.1%,精确率达到88.8%,召回率达到86.1%,分别提升2.5,2.5,3.0个百分点,检测速度达到96.15帧/s。研究为饮品瓶盖的质量检测提供理论支持。To address the limitations of insufficient feature extraction capability,low detection accuracy,and slow convergence in beverage bottle cap defect detection,this study proposes an improved YOLOv8s-based model(YOLOv8s-GCW).Building upon the baseline YOLOv8s architecture,we integrated a Global Context attention module into the backbone network above the SPPF layer,replaced the original CIoU loss function with WIoU(v3),and constructed a dedicated beverage bottle cap defect dataset augmented with data enhancement techniques.Experimental results demonstrate that,without increasing computational costs,the enhanced model achieves superior detection performance:mean average precision(mAP)reaches 91.1%,with precision and recall rates of 88.8%and 86.1%,representing improvements of 2.5,2.5,and 3.0 percentage points over the original model,respectively.The detection speed remains at 96.15 FPS.This research provides theoretical support for quality inspection of beverage bottle caps.

关 键 词：缺陷检测 YOLOv8 注意力模块 损失函数 

分 类 号：TB487[一般工业技术—包装工程] TH165.4[机械工程—机械制造及自动化]