目的 针对高端包装印刷中二维码模切及印刷误差的亚毫米级精度要求,以及工业场景下高光干扰样本稀缺导致的模型泛化能力不足问题,构建了一套基于物理感知数据生成与级联视觉测量的高精度亚像素测量框架。方法 构建融合物理感知合成数据与级联视觉测量的检测框架。基于Blinn-Phong光照理论设计了一种物理感知的样本合成机制,模拟了复杂包装材质的高反光光学特性,强化检测模型的提取特征能力。此外,设计“粗定位-精测量”级联视觉测量模型,即采用YOLO11-Seg网络实现目标掩膜提取。结合最小二乘法展开亚像素边缘拟合,结合透视矫正,抑制深度网络的离散化误差。结果 实验结果表明,该系统的边界框精度达到98%,能够精准定位检测目标区域。单帧检测总耗时在100 ms以内。结论 本方法将深度学习的语义理解能力与精密测量相结合,使得绝对测量误差被控制在0.04 mm量级,满足了小于0.8 mm的高精度公差指标,适用于复杂光学材质表面的微小定位偏差的量化分析并兼顾鲁棒性、可解释性与现场部署可行性。
Abstract
The work aims to propose a high-precision sub-pixel measurement framework integrating physics-aware data generation with cascaded visual measurement to address sub-millimeter precision requirements for QR code die-cutting and printing errors in high-end packaging, and to mitigate the lack of model generalization caused by the scarcity of samples with highlight interference in industrial scenarios. A detection framework fusing physics-aware synthetic data with cascaded visual measurement was developed and a physics-aware sample synthetic mechenicam was designed based on the the Blinn-Phong illumination theory to simulate the highly reflective properties of complex packaging materials to enhance model feature extraction. Subsequently, a "coarse-to-fine" cascaded visual measurement model was deployed: with YOLO11-Seg used to extract target masks, followed by least-squares sub-pixel edge fitting and homography-guided perspective correction to eliminate discretization artifacts. Experiments demonstrated that the system achieved 98% bounding box accuracy, accurately localizing the target regions. The single-frame processing time was within 100 ms. In conclusion, by integrating semantic perception with geometric precision, the method restricts absolute measurement errors to 0.04 mm, operating well within the 0.8 mm tolerance requirement. It is applicable for quantifying micro-positioning deviations on complex optical surfaces, balancing robustness, interpretability, and on-site deployability.
关键词
机器视觉 /
印刷误差 /
亚像素测量 /
实例分割 /
YOLO
Key words
machine vision /
overprinting error /
sub-pixel measurement /
instance segmentation /
YOLO
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] LI P, YANG J, JIMÉNEZ-CARVELO A M, et al. Applications of Food Packaging Quick Response Codes in Information Transmission toward Food Supply Chain Integrity[J]. Trends in Food Science & Technology, 2024, 146: 104384.
[2] 国家新闻出版署(国家版权局). 平版装潢印刷品: GB/T 7705—2008[S]. 2008.
National Press and Publication Administration (National Copyright Administration of the People's Republic of China). The offset lithographic prints for decorating: GB/T 7705—2008[S]. Beijing: Standards Press of China, 2008.
[3] 国家标准委. 防伪标识通用技术条件: GB/T 22258— 2008[S]. 2008.
Standardization Administration of the People's Republic of China. General technical specifications for anti-counterfeiting marks: GB/T 22258—2008[S]. Beijing: Standards Press of China, 2008.
[4] 门超. 包装印刷检测的新篇章——机器视觉[J]. 印刷杂志, 2017(8): 59-61.
MEN C.A new chapter in packaging and printing inspection: Machine vision[J]. Printing Field, 2017(8): 59-61.
[5] SEE J E, DRURY C G, SPEED A, et al.The Role of Visual Inspection in the 21st Century[J]. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 2017, 61(1): 262-266.
[6] SWAIN A D, GUTTMANN H E. Handbook of Human-Reliability Analysis with Emphasis on Nuclear Power Plant Applications. Final Report: NUREG/CR-1278; SAND-80-0200[R]. Sandia National Labs., Albuquerque, NM(USA), 1983.
[7] ZHOU Q B, CHEN R, HUANG B, et al.DeepInspection: Deep Learning Based Hierarchical Network for Specular Surface Inspection[J]. Measurement, 2020, 160: 107834.
[8] TONG K, WU Y, ZHOU F.Recent Advances in Small Object Detection Based on Deep Learning: A Review[J]. Image and Vision Computing, 2020, 97: 103910.
[9] LIU Y, SUN P, WERGELES N, et al.A Survey and Performance Evaluation of Deep Learning Methods for Small Object Detection[J]. Expert Systems with Applications, 2021, 172: 114602.
[10] WEI W, CHENG Y, HE J, et al.A Review of Small Object Detection Based on Deep Learning[J]. Neural Computing and Applications, 2024, 36(12): 6283-6303.
[11] SHORTEN C, KHOSHGOFTAAR T M.A Survey on Image Data Augmentation for Deep Learning[J]. Journal of Big Data, 2019, 6(1): 60.
[12] GHIASI G, CUI Y, SRINIVAS A, et al.Simple Copy-Paste Is a Strong Data Augmentation Method for Instance Segmentation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021: 2918-2928.
[13] MOHAMMED ELTOUM M A, IQBAL E, ZWEIRI Y, et al. BladeSynth: A High-Quality Rendering-Based Synthetic Dataset for Aero Engine Blade Defect Inspection[J]. Scientific Data, 2025, 12(1): 1268.
[14] FULIR J, BOSNAR L, HAGEN H, et al.Synthetic Data for Defect Segmentation on Complex Metal Surfaces[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops.[S.l.]: IEEE, 2023: 4424-4434.
[15] AKENINE-MÖLLER T, HAINES E, HOFFMAN N. Real-Time Rendering, Fourth Edition[M]. Boca Raton London New York: A K Peters/CRC Press, 2019.
[16] TOBIN J, FONG R, RAY A, et al.Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real World[C]//2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). 2017: 23-30.
[17] ZHONG Z, ZHENG L, KANG G, et al.Random Erasing Data Augmentation[C]//Proceedings of the AAAI Conference on Artificial Intelligence.[S. l.]: AAAI Press, 2020, 34(7): 13001-13008.
[18] LI D, PAN X, FU Z Z, et al.Real-Time Accurate Deep Learning-Based Edge Detection for 3-D Pantograph Pose Status Inspection[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-12.
[19] GE Z, LIU S T, WANG F, et al. YOLOX: Exceeding YOLO Series in2021[EB/OL]. (2021-07-18)[2026-04-21]. https://arxiv.org/abs/2107.08430.
[20] LIU H J, LIU F Q, FAN X Y, et al.Polarized Self-Attention: Towards High-quality Pixel-wise Regression[J]. Neurocomputing, 2022, 506: 158-167.
[21] HARTLEY R, ZISSERMAN A.Multiple View Geometry in Computer Vision[M]. Cambridge University Press, 2003.
PDF(2054 KB)
渝公网安备 50010702501716号 渝ICP备15012534号-2
