目的 针对烟草行业一线应用场景中烟包图像识别存在的小目标检测难、密集排列易漏检等问题,提出一种面向边缘部署的轻量化烟包识别解决方案,以提升识别效率与实用性。方法 结合实际陈列场景中烟包尺寸小、排列密集的特点,优化检测模型的损失函数,增强其对小尺度与高密度目标的特征学习能力;在此基础上,引入模型量化技术对训练收敛后的网络进行压缩,构建低参数量、低计算开销的轻量化模型。结果 实验表明,相较于使用原始损失函数的基准模型,优化后模型的烟包图像识别准确率提升了8.05个百分点,达到88.16%;经量化处理后,模型体积压缩约68.83%,在嵌入式设备上的推理耗时较全精度模型降低47.82%以上,显著提升了运行效率。结论 本研究实现了一种高效、轻量化的烟包图像识别方法,所构建的模型具备良好的识别性能与资源适应性,可有效部署于算力受限的终端设备,提升了卷烟识别在实际业务场景中的可行性、实用性和操作便利性。
Abstract
To address the challenges of cigarette pack image recognition in frontline scenarios of the tobacco industry, specifically the difficulty in detecting small targets and high risk of missed detections in densely packed arrangements, the work aims to propose a lightweight recognition solution tailored for edge deployment, aiming to enhance recognition efficiency and practical applicability. Given the characteristics of in-store display scenarios, such as the small size and dense arrangement of cigarette packs, the loss function of the detection model was optimized to strengthen its ability to learn features from small-scale and high-density targets. Furthermore, model quantization techniques were applied to compress the converged training network, yielding a lightweight model with fewer parameters and lower computational overhead. Experimental results showed that, compared with the baseline model using the original loss function, the optimized model achieved an 8.05 percentage point improvement in cigarette pack image recognition accuracy, reaching 88.16%. After quantization, the model size was compressed by approximately 68.83%, and inference latency on embedded devices was reduced by more than 47.82% relative to the full-precision model, thus significantly enhancing runtime efficiency. In conclusion, this study develops an efficient lightweight method for cigarette pack image recognition. The constructed model exhibits robust recognition performance and excellent resource adaptability, enabling effective deployment on computationally constrained terminal devices. This improves the feasibility, practicality, and operational convenience of automated cigarette pack recognition in real-world business scenarios.
关键词
烟包识别 /
模型量化 /
YOLOv10
Key words
cigarette packaging recognition /
model quantization /
YOLOv10
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 郑远攀, 李广阳, 李晔. 深度学习在图像识别中的应用研究综述[J]. 计算机工程与应用, 2019, 55(12): 20-36.
ZHENG Y P, LI G Y, LI Y.Survey of Application of Deep Learning in Image Recognition[J]. Computer Engineering and Applications, 2019, 55(12): 20-36.
[2] 王俊杰. 县域经济中现代零售终端卷烟陈列策略探讨[C]// 浙江省烟草学会学术年会, 2013. WANG J J. Exploration of Modern Retail Terminal Cigarette Display Strategies in County Economy[C]// Zhejiang Tobacco Society Academic Annual Meeting, 2013.
[3] 陈哲凡, 郑建彬, 詹恩奇, 等. 基于聚类SURF特征的商品识别算法[J]. 计算机应用研究, 2017, 34(11): 3490-3493.
CHEN Z F, ZHENG J B, ZHAN E Q, et al.Product Recognition Algorithm Based on Clustered SURF Features[J]. Application Research of Computers, 2017, 34(11): 3490-3493.
[4] 李春国, 刘杨, 杨哲, 等. 一种复杂场景下香烟品牌识别方法: China, 109961049B[P].2022-04-26.
LI C G, LIU Y, YANG Z. A Method for Cigarette Brand Recognition in Complex Scenes: China, 201910236775.1[P].2019-07-02.
[5] 单宇翔, 龙涛, 楼卫东, 等. 基于深度学习的复杂场景中卷烟烟盒检测与识别方法[J]. 中国烟草学报, 2021, 27(5): 71-80.
SHAN Y X, LONG T, LOU W D, et al.A Detection and Recognition Method of Cigarette Cases in Complex Scene Based on Deep Learning[J]. Acta Tabacaria Sinica, 2021, 27(5): 71-80.
[6] 罗会兰, 陈鸿坤. 基于深度学习的目标检测研究综述[J]. 电子学报, 2020, 48(6): 1230-1239.
LUO H L, CHEN H K.Survey of Object Detection Based on Deep Learning[J]. Acta Electronica Sinica, 2020, 48(6): 1230-1239.
[7] 郑银环, 林晓琛, 吴飞, 等. 基于改进YOLO v4的轻量化烟梗识别方法[J]. 合肥工业大学学报(自然科学版), 2023, 46(9): 1196-1202.
ZHENG Y H, LIN X C, WU F, et al.Lightweight Tobacco Stem Identification Method Based on Improved YOLO V4[J]. Journal of Hefei University of Technology (Natural Science), 2023, 46(9): 1196-1202.
[8] 唐兵, 罗钢印, 邓胜勇, 等. 一种烟草输送设备用杂物检测方法和系统: CN116563778A[P].2023-08-08.
TANG B, LUO G Y, DENG S Y, et al. A Method and System for Detecting Debris in Tobacco Conveying Equipment: CN116563778A[P].2023-08-08.
[9] CHEN H, CHEN K, DING G G, et al.YOLOv10: Real-Time End-to-End Object Detection[C]// Advances in Neural Information Processing Systems 37. Vancouver, BC, Canada. Neural Information Processing Systems Foundation, Inc.(NeurIPS), 2024: 107984-108011.
[10] REDMON J, DIVVALA S, GIRSHICK R, et al.You Only Look Once: Unified, Real-Time Object Detection[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA. IEEE, 2016: 779-788.
[11] REN S Q, HE K M, GIRSHICK R, et al.Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
[12] BOCHKOVSKIY A, WANG C Y, LIAO H M. YOLOv4: Optimal Speed and Accuracy of Object Detection[EB/OL].2020: arXiv: 2004.10934. https://arxiv.org/abs/2004.10934
[13] LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature Pyramid Networks for Object Detection[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, HI, USA. IEEE, 2017: 936-944.
[14] WANG J W, XU C, YANG W, et al. A Normalized Gaussian Wasserstein Distance for Tiny Object Detection[EB/OL].2021: arXiv: 2110.13389. https://arxiv. org/abs/2110.13389
[15] 李江昀, 赵义凯, 薛卓尔, 等. 深度神经网络模型压缩综述[J]. 工程科学学报, 2019, 41(10): 1229-1239.
LI J Y, ZHAO Y K, XUE Z E, et al.A Survey of Model Compression for Deep Neural Networks[J]. Chinese Journal of Engineering, 2019, 41(10): 1229-1239.
[16] 杨春, 张睿尧, 黄泷, 等. 深度神经网络模型量化方法综述[J]. 工程科学学报, 2023, 45(10): 1613-1629.
YANG C, ZHANG R Y, HUANG L, et al.A Survey of Quantization Methods for Deep Neural Networks[J]. Chinese Journal of Engineering, 2023, 45(10): 1613-1629.
基金
阳江市烟草专卖局（公司）科技项目（阳烟科项202401）,该项目由首批烟草行业青年科技托举人才计划（国烟人〔2023〕3号）
PDF(2907 KB)
渝公网安备 50010702501716号 渝ICP备15012534号-2
