吴仕莲¹，胡 涛²，汪增福¹，郑志刚¹，赵丙坤^2*，祝燕林²，杨 洋²

（1.中国科学技术大学信息科学技术学院，安徽合肥 230000；

2.泸州老窖股份有限公司，四川泸州 646000）

摘要：近年来，深度学习技术在不同计算机视觉任务中都取得了重大突破，基于深度学习的视觉检测方法具有精度高、鲁棒性强的优势。本文针对酒瓶包装缺陷检测领域进行研究，设计了一种基于网格的多尺度缺陷检测模型，通过对原始图像划分网格，将缺陷映射到对应网格位置以解决缺陷形状多样、难以定义的问题；并在不同尺度的特征上检测缺陷，解决酒瓶缺陷尺度变化大的问题。

关键词：缺陷检测；深度学习；多尺度；全卷积网络

中图分类号：TS206.1   文献标识码：A   文章编号：1674-506X（2023）05-0104-0005

Multi-scale Bottle Defect Detection Based on Grid Segmentation

WU Shilian¹，HU Tao¹，WANG Zengfu¹，ZHEN Zhigang¹，ZHAO Bingkun^2*，ZHU Yanlin²，YANG Yang²

（1.School of Information Science and Technology, University of Science and Technology of China,

Hefei Anhui 230000, China;

2.Luzhou Laojiao Co., Ltd., Luzhou Sichuan 646000, China）

Abstract：In recent years, deep learning technologies have made significant breakthroughs in various computer vision tasks. Deep learning- based visual inspection methods offer advantages such as high accuracy and strong robustness. This paper focuses on the field of bottle defect detection and presents a research study that designs a multi-scale defect detection model based on grids. By dividing the original image into grids, defects are mapped to their respective grid positions to address the issue of diverse and hard-to-define defect shapes. Additionally, defects are detected on features at different scales, addressing the challenge of varying defect scales in glass bottle defects.

Keywords：defect detection; deep learning; multi-scale; fully convolutional network

doi：10.3969/j.issn.1674-506X.2023.05-018

收稿日期：2023-30-12

作者简介：吴仕莲（1994-）男，博士，工程师。研究方向：缺陷检测。

*通信作者：赵丙坤（1979-）男，本科，高级工程师。研究方向：包材采购、包装设备开发。

引用格式：吴仕莲,胡涛,汪增福,等.基于网格分割的多尺度酒瓶缺陷检测研究[J].食品与发酵科技,2023,59(5):104-108.