何永伦1　张冲1,2,3　陈儒1,2　梁佳楠1,2

（1.华南智能机器人创新研究院，广东 佛山 528399

2.广东省科学院智能制造研究所，广东 广州 510070

3.五邑大学机械与自动化工程学院，广东 江门 529020）

摘要：瓶胚缺陷检测是保障PET瓶成型质量的关键环节。为了将缺陷检测模型部署到工业应用场景中实现在线检测，并提高瓶胚缺陷检测精度，提出一种基于改进YOLOv8的瓶胚缺陷检测模型——YOLOv8-FEMA模型。首先，将FasterNet Block引入YOLOv8模型的C2f模块中，以减少模型的参数量；然后，引入EMA机制，使网络更聚焦于有用的特征信息，以提升模型的检测精度。实验结果表明，该模型相较于YOLOv8n模型，参数量、浮点运算量分别减少了27%和26%，检测精度提升了0.03。该模型部署在瓶胚缺陷检测软件中，可有效检测出瓶胚缺陷。

关键词：瓶胚检测；缺陷检测；YOLOv8；轻量化

中图分类号：TP391            文献标志码：A          文章编号：1674-2605(2024)06-0007-06

DOI：10.3969/j.issn.1674-2605.2024.06.007                    开放获取

Bottle Preforms Defect Detection Model Based on Improved YOLOv8

HE Yonglun1  ZHANG Chong1,2,3  CHEN Ru1,2  LIANG Jianan1,2

(1.South China Robotics Innovation Research Institute, Foshan 528399, China

2.Institute of Intelligent Manufacturing, Guangdong Academy of Sciences, Guangzhou 510070, China

3.School of Mechanical and Automation Engineering, Wuyi University, Jiangmen 529020, China)

Abstract: Defect detection of bottle preforms is a crucial step in ensuring the quality of PET bottle molding. In order to deploy defect detection models to industrial application scenarios for online detection and improve the accuracy of preform defect detection, a bottle preform defect detection model based on improved YOLOv8, YOLOv8-FEMA model, is proposed. Firstly, embed the FasterNet Block into the C2f module of the YOLOv8 model to reduce the number of model parameters; Then, the EMA mechanism is introduced to make the network more focused on useful feature information and improve the detection accuracy of the model. The experimental results show that compared to the YOLOv8n model, this model reduces the number of parameters and floating-point operations by 27% and 26%, respectively, and improves detection accuracy by 0.03. This model is deployed in bottle embryo defect detection software and can effectively detect bottle preforms defects.

Keywords: bottle preforms detection; defect detection; YOLOv8; lightweighting