目的 构建一种基于识别模型的包装磨损特征识别与归纳方法。方法 以卷积神经网络（CNN）为核心,针对包装磨损图像的局部特征提取需求,设计轻量级网络模型,采用交叉熵损失函数优化训练过程;基于自建的包装磨损图像数据集完成模型训练,并引入迁移学习策略强化网络泛化能力,提升训练速度与识别精度;同时简化硬件协同逻辑,仅保留与视觉数据传输适配的4B串口通信核心逻辑,保证方法的实用性。结果 所构建的卷积神经网络模型对包装磨损特征的识别准确率达93.5%,相较传统视觉模型,在参数规模相近的前提下,准确率提升了7.8%;补充的分类评价指标显示,模型对严重磨损、一般磨损、轻微磨损、未磨损4个分类的识别精确率分别为94.2%、93.8%、92.9%、95.1%,F1分数均高于92%。结论 基于卷积神经网络的包装磨损特征识别与归纳方法,可满足复杂包装磨损特征分析需求,实现高精度分类的同时,提升磨损包装识别效率,为包装运输领域的质量管控提供高效解决方案。
Abstract
The work aims to develop a packaging wear feature recognition and summarization method based on identification models. Centered on convolutional neural networks (CNNs), a lightweight network model was designed to address the local feature extraction requirements of packaging wear images, with the cross-entropy loss function employed to optimize the training process. Model training was completed based on a self-built packaging wear image dataset, and transfer learning strategies were introduced to enhance network generalization capabilities, improving training speed and recognition accuracy. Additionally, hardware coordination logic was simplified, retaining only the core 4B serial communication logic compatible with visual data transmission to ensure practicality. The constructed CNN model achieved a recognition accuracy of 93.5% for packaging wear features, representing a 7.8% improvement in accuracy compared with traditional visual models under similar parameter scales. Supplementary classification evaluation metrics revealed precision rates of 94.2%, 93.8%, 92.9%, and 95.1% for severe wear, moderate wear, minor wear, and no wear categories, respectively, with all F1 scores exceeding 92%. In conclusion, the CNN-based packaging wear feature recognition and summarization method can meet the analytical demands of complex packaging wear features, achieving high-precision classification while improving the efficiency of wear packaging identification. This provides an efficient solution for quality control in the packaging transportation sector.
关键词
包装磨损 /
卷积神经网络 /
系统架构
Key words
packaging wear /
convolutional neural networks /
system architecture
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] BITKINA O V, PARK J, RYU D H.Color Vision Deficiency Recognition Based on Eye-Tracking Metrics Using Machine Learning Approaches[J]. International Journal of Human-Computer Interaction, 2025, 41(14): 8928-8942.
[2] KAVANAGH S R.Identifying Split Vacancy Defects with Machine-Learned Foundation Models and Electrostatics[J]. Journal of Physics: Energy, 2025, 7(4): 045002.
[3] ANDANI M T, AMERI F.Automated Pipe Defect Identification in Underwater Robot Imagery with Deep Learning[J]. Journal of Marine Science and Application, 2026, 25(1): 197-215.
[4] LI X, LIU X L, JIANG Y Q, et al.Identifying Neuroimaging Biomarkers of Attention-Deficit Hyperactivity Disorder (ADHD) from Cortical Hemodynamic Responses during Go/NoGo Task Using Machine Learning Approaches[J]. Progress in Neuro-Psychopharmacology and Biological Psychiatry, 2025, 140: 111417.
[5] BESKOPYLNY A N, STEL’MAKH S A, SHCHERBAN’ E M, et al. Defects Identification and Crack Depth Determination in Porous Media on the Brick Masonry Example Using Ultrasonic Methods: Numerical Analysis and Machine Learning[J]. Journal of Composites Science, 2025, 9(6): 267.
[6] WU L J, WANG W, SHI Z W, et al.Rapid Identification of Defects in Doped Organic Crystalline Films via Machine Learning-Enhanced Hyperspectral Imaging[J]. Chemical Engineering Journal, 2025, 513: 162696.
[7] LIN S Q, YE H J, TAN D Y, et al.Identification of Defects in Underground Structures Using Machine Learning Aided Distributed Fiber Optic Sensing[J]. Journal of Rock Mechanics and Geotechnical Engineering, 2025, 17(4): 2194-2207.
[8] 王祺, 陈玉婷, 董国忠, 等. 机器学习在木材表面缺陷检测中的应用研究[J]. 中国人造板, 2024, 31(4): 16-24.
WANG Q, CHEN Y T, DONG G Z, et al.Application and Research of Machine Learning in Wood Surface Defect Detection[J]. China Wood-Based Panels, 2024, 31(4): 16-24.
[9] 姚浩, 夏桂然, 刘泽佳, 等. 基于机器学习的黏钢构件黏接层缺陷识别方法研究[J]. 应用数学和力学, 2024, 45(4): 429-442.
YAO H, XIA G R, LIU Z J, et al.A Defect Identification Method for Bonding Layers of Adhesive Steel Members Based on Machine Learning[J]. Applied Mathematics and Mechanics, 2024, 45(4): 429-442.
[10] 许啸振. 基于联邦学习的卷烟外包装缺陷识别算法研究[D]. 杭州: 浙江大学, 2024.
XU X Z.Research on Cigarette Packaging Defect Recognition Algorithm Based on Federated Learning[D]. Hangzhou: Zhejiang University, 2024.
[11] 李超华, 许再胜. 基于融合机器学习的管道焊缝缺陷识别方法研究[J]. 石油化工自动化, 2024, 60(1): 72-76.
LI C H, XU Z S.Study on Defect Identification Method of Pipeline Weld Based on Integrated Machine Learning[J]. Automation in Petro-Chemical Industry, 2024, 60(1): 72-76.
[12] 王杰, 张志芬, 秦锐, 等. 基于机器学习的不锈钢薄板MIG焊焊穿缺陷识别[J]. 电焊机, 2023, 53(9): 70-77.
WANG J, ZHANG Z F, QIN R, et al.Identification of MIG Welding Burn-through Defects in Stainless Steel Sheet Based on Machine Learning[J]. Electric Welding Machine, 2023, 53(9): 70-77.
[13] 廖嘉杰, 黄胜, 马保松, 等. 排水管道缺陷图像的智能识别分类技术综述[J]. 给水排水, 2023, 59(7): 148-156.
LIAO J J, HUANG S, MA B S, et al.A Review of Intelligent Identification and Classification Techniques for Drainage Pipe Defect Images[J]. Water & Wastewater Engineering, 2023, 59(7): 148-156.
[14] 尤伟军, 李聪, 张江雄. 利用YOLOv5机器学习模型识别混凝土排水管道接头缺陷试验研究[J]. 给水排水, 2023, 59(S1): 489-494.
YOU W J, LI C, ZHANG J X.Experimental Research on Concrete Pipe Joint Defect Inspection by Using YOLOv5machine Learning Model[J]. Water & Wastewater Engineering, 2023, 59(S1): 489-494.
[15] 赵治. 基于机器学习的管道缺陷识别量化方法研究[D]. 沈阳: 沈阳工业大学, 2023.
ZHAO Z.Research on Quantitative Method for Pipeline Defect Recognition Based on Machine Learning[D]. Shenyang: Shenyang University of Technology, 2023.
[16] 涂一枝. 基于机器学习的二维材料缺陷识别[D]. 南京: 东南大学, 2023.
TU Y Z.Defect Identification in Two-Dimensional Materials Based on Machine Learning[D]. Nanjing: Southeast University, 2023.
基金
河南省科技攻关项目（252102210220）; 河南省科技攻关项目（242102211036）
PDF(2246 KB)
渝公网安备 50010702501716号 渝ICP备15012534号-2
