摘要
【目的】为解决雪茄烟支人工清点工序繁琐、耗时较长等问题。【方法】以不同形状成品雪茄烟支为对象,通过电荷耦合器件获取雪茄烟支端面图像,利用卷积神经网络的表征学习功能提取雪茄烟支图像中的高水平特征,并人工标注图像中的雪茄烟支,经过大量样本图像训练,建立雪茄烟支图像识别与计数的实例分割模型。【结果】雪茄烟支图像识别计数模型能有效识别各种形状烟支和特殊用途烟支,识别计数准确率达到99.04%,图像识别计数较人工目数计数效率提升169.53%。【结论】本研究建立的模型及方法在雪茄烟支识别入库、计数核验方面具有较好的实用性。
[Background]This study aims to solve the problems of cumbersome and time-consuming manual counting of cigars.[Methods]Using different shaped finished cigars as subjects,cigar end images were obtained through charge-coupled devices.The high-level features of the cigar images were extracted using the representation learning function of convolutional neural networks.The cigars in the images were manually annotated,and an instance segmentation model for cigar image recognition and counting was established through training with a large number of sample images.[Results]The cigar image recognition and counting model can effectively identify various shapes of cigars and cigars for special purposes,with an accuracy rate of 99.04%.The efficiency of image recognition and counting is 169.53%higher than that of manual counting.[Conclusion]The model and method established in this study have good practicability in cigar recognition for warehousing and counting verification.
作者
谭再钰
王剑
潘勇
贾梦珠
吴创
吴英乔
郭霜
蔡尧
王靖渊
施友志
TAN Zaiyu;WANG Jian;PAN Yong;JIA Mengzhu;WU Chuang;WU Yingqiao;GUO Shuang;CAI Yao;WANG Jingyuan;SHI Youzhi(China Tobacco Hubei Industry Co.,Ltd.,Yichang Hubei,443100,China)
出处
《中国烟草学报》
CAS
CSCD
北大核心
2024年第6期110-118,共9页
Acta Tabacaria Sinica
关键词
雪茄烟支
电荷耦合器件
卷积神经网络
图像识别技术
图像识别计数模型
cigars
charge-coupled device
convolutional neural network
image recognition technology
image recognition counting model
