摘要
印刷缺陷检测是确保印刷产品质量至关重要的步骤,其准确性直接影响到最终产品的市场表现。然而,由于缺陷样本稀缺、图像分辨率高、对比度低等挑战,自动化检测任务变得尤为困难。为了提高印刷缺陷检测的精度,本文提出了一种基于UNet的多尺度特征融合网络。引入了多尺度特征融合模块(MSF Block)和注意力机制,以增强模型对不同尺寸缺陷的感知能力。在印刷品缺陷数据集上进行的实验结果表明,我们的方法在F1-score、IoU等关键指标上均优于现有的UNet、UNet++、DeepLabV3+和SegNet等方法。
Print defect detection is a crucial step in ensuring the quality of printed products,with its accuracy directly impacting the market performance of the final product.However,challenges such as the scarcity of defect samples,high image resolution,and low contrast have made automated detection tasks especially difficult.To enhance the accuracy of print defect detection,this articale proposes a multi-scale feature fusion network based on UNet.This method introduces a Multi-Scale Feature Fusion Block(MSF Block)and a Spatial Attention Mechanism(SAM)to improve the network's capability to perceive defects of different sizes.Experimental results conducted on our self-constructed dataset of printed product defects indicate that our method surpasses existing approaches such as UNet,UNet++,DeepLabV3+,and SegNet in key metrics including F1-score and IoU.
作者
张自主
李桐
ZHANG Zizhu;LI Tong(Beijing Institute of Graphic Communication,Beijing 102600,China)
出处
《北京印刷学院学报》
2026年第3期21-28,共8页
Journal of Beijing Institute of Graphic Communication
关键词
印刷缺陷检测
多尺度特征融合
注意力机制
UNet
语义分割
printing defect detection
multi-scale feature fusion
attention mechanism
UNet
semantic segmentation
