杨学先
(科达制造股份有限公司,广东 佛山 528313)
摘要:瓷砖表面缺陷不仅影响外观,还可能缩短使用寿命并带来装修安全隐患。针对YOLOv8深度模型进行瓷砖表面缺陷检测时,需构建有效的训练数据集以保证模型的稳定性,提出基于深度学习的瓷砖表面缺陷检测数据增强方法。首先,通过高分辨率线阵相机采集瓷砖图像,并结合公共的纹理瓷砖数据集,构建瓷砖数据集;然后,利用Copy-Paste算法对瓷砖图像的缺陷目标进行分割、变换并粘贴到新的背景图像中,以提高YOLOv8深度模型的表面缺陷检测性能。实验结果表明,该方法构建并增强的瓷砖数据集可有效提高YOLOv8深度模型的瓷砖表面缺陷检测能力。
关键词:瓷砖表面缺陷检测;深度学习;数据增强;Copy-Paste算法;YOLOv8深度模型
中图分类号:TP27 文献标志码:A 文章编号:1674-2605(2024)06-0009-05
DOI:10.3969/j.issn.1674-2605.2024.06.009 开放获取
Data Enhancement Method for Ceramic Tile Surface Defect Detection Based on Deep Learning
YANG Xuexian
(Keda Industrial Group Co., Ltd., Foshan 528313, China)
Abstract: Surface defects on ceramic tiles not only affect their appearance, but may also shorten their service life and pose safety hazards to decoration. When using YOLOv8 deep model for tile surface defect detection, it is necessary to construct an effective training dataset to ensure the robustness of the model. A deep learning based data augmentation method for tile surface defect detection is proposed. Firstly, a high-resolution linear array camera is used to capture images of ceramic tiles, and combined with a public texture ceramic tile dataset, a ceramic tile dataset is constructed; Then, the Copy Paste algorithm is used to segment, transform, and paste the defect targets of the tile image into a new background image to improve the surface defect detection performance of the YOLOv8 depth model. The experimental results show that the method constructed and enhanced the tile dataset can effectively improve the tile surface defect detection ability of YOLOv8 depth model.
Keywords: ceramic tile surface defect detection; deep learning; data enhancement; Copy-Paste algorithm; YOLOv8 deep model