| 509 | 0 | 6 |
| 下载次数 | 被引频次 | 阅读次数 |
【目的】切丝质量的在线检测是卷烟制丝加工中的关键质量指标，直接影响成品质量。针对异常形态烟丝检测中存在与背景颜色相似、目标尺寸不规则、小目标易受背景干扰、以及模型计算量大等情况，提出一种基于CBS-YOLOv5s的异常烟丝检测方法。【方法】针对复杂背景下不同尺度的异常烟丝，在颈部网络中引入BiFormer注意力机制，以增强复杂背景下小目标特征的提取能力。其次，采用部分卷积与点卷积结合的C3-Faster模块，在保证模型精度的前提下，降低了计算复杂度和参数量。最后，引入Shape-IoU损失函数，进一步提高回归的准确性。【结果】本研究建立的模型在目标检测中的平均精确率达到了96.4%，相比于原模型提高2.5%，与Faster R-CNN、YOLOv4-tiny、YOLOv8s等模型相比分别提高14.8%、25.1%、1.58%；在计数任务中，能更好分析异常烟丝的波动状况。【结论】本研究为切丝质量稳定控制与优化提供了在线检测方法，有助于推动制丝工艺的现代化进程。
Abstract：[Objective] On-line detection of cut tobacco quality is a key quality index in cigarette processing, which directly affects the quality of finished products. Aiming at the problems of similar background color, irregular target size, small target susceptible to background interference, and large model calculation in abnormal shape cut tobacco detection, an abnormal cut tobacco detection method based on CBS-YOLOv5s is proposed. [Methods] Aiming at the abnormal cut tobacco of different scales in complex background, the BiFormer attention mechanism is introduced into the neck network to enhance the extraction ability of small target features in complex background. Secondly, the C3-Faster module, which combines partial convolution and point convolution, is used to reduce computational complexity and the number of parameters while ensuring the model's accuracy. Finally, the Shape-IoU function is introduced to further improve the regression accuracy. [Results] The average accuracy of the model established in this study in target detection reached 96.4 %,which was 2.5 % higher than that of the original model. Compared with Faster R-CNN, YOLOv4-tiny, YOLOv8s, and other models, it increased by 14.8 %, 25.1 % and 1.58 %, respectively. In the counting task, it can more effectively analyze the fluctuation of abnormal tobacco. [Conclusion] This study provides an on-line detection method for the stability control and optimization of cutting quality, which is beneficial for promoting the modernization of the silk-making process.
[1]唐军，何邦华，易斌，等.国内卷烟加工过程烟草异物除杂技术研究进展[J].轻工学报，2022,37(2):94-101.TANG Jun,HE Banghua,YI Bin,et al.Research progress of tobacco foreign material eliminating technology for cigarette processing in china[J].Journal of Light Industry,2022,37(2):94-101.
[2]曾静，李斌，冯志斌，等.卷烟机剔除梗签物中含丝量的检测[J].烟草科技，2012(8):5-7.ZENG Jing,LI Bin,FENG Zhibin,et al.Determination of tobacco shred content in rejects from cigarette maker[J].Tobacco Science&Technology,2012(8):5-7.
[3]张雷，李金学，堵劲松，等.基于DGRU网络的烘丝机筒壁温度动态预测[J].轻工学报，2022,37(06):85-91,100.ZHANG Lei,LI Jinhua,DU Jinsong,et al.Dynamic predietion of cylinder wall temperature for drum dryer based on DGRUnetwork[J].Journal of Light Industry,2022,37(06):85-91,100.
[4]李善莲，安佳敏，刘朝贤，等.基于自编码器和PCA的滚筒烘丝机多块过程监测方法[J].轻工学报，202338(6):110-117.LI Shanlian,AN Jiamin,LIU Chaoxian,et al.Multi block process monitoring method of drum dryer based on auto encoder and PCA[J].Journal of Light Industry,2023,38(6):110-117.
[5]张鹏飞，陈姣文，郭洪滨，等.雪茄烟烟支外观质量检测方法[J].烟草科技，2024,57(01):91-97.ZHANG Pengfei,CHEN Jiaowen,GUO Hongbin,et al.An inspection method for cigar appearance quality[J].Tobacco Science&Technology,2024,57(01):91-97.
[6]肖雷雨，王澍，刘渊根，等.基于深度学习技术的烟梗形态分类与识别[J].烟草科技，2021,54(06):65-74.XIAO Leiyu,WANG Shu,LIU Yuangen,et al.Classification and identification of tobacco stem morphology based on deep learning technology[J].Tobacco Science&Technology,2021,54(06):65-74.
[7]洪金华，忻惠琴，陆海华，等.基于YOLOV3模型的卷烟厂烟虫识别方法[J].烟草科技，2020,53(09):77-84.HONG Jinhua,XIN Huiqin,LU Haihua,et al.Tobacco insect recognition in cigarette factory using YOLOV3 model[J].Tobacco Science&Technology,2020,53(09):77-84.
[8]杨超，李佳田，张泽龙，等.YOLOv4-tiny算法的融合模块在卷烟小包外观缺失检测中的应用[J].中国烟草学报，2022,28(02):59-64.YANG Chao,LI Jiatian,ZHANG Zelong,et al.Application of fusion module of YOLOv4-tiny algorithm in cigarette packet appearance missing detection[J].Acta Tabacaria Sinica,2022,28(02):59-64.
[9]刘军，辛秋生，李文灿，等.基于加权K-means与Ghost Net融合改进YOLOv4的卷烟厂物料输送巷道异物视觉检测方法[J].中国烟草学报，2023,29(05):42-48.LIU Jun,XIN Qiusheng,LI Wencan,et al.Improved YOLOv4visual detection method of foreign bodies in material delivery roadway of cigarette factory based on the fusion of weighted K-means and Ghost Net[J].Acta Tabacaria Sinica,2023,29(05):42-48.
[10]汪睿琪，张炳辉，顾钢，等.基于YOLOv5的鲜烟叶成熟度识别模型研究[J].中国烟草学报，2023,29 (02):46-55.WANG Ruiqi,ZHANG Binghui,GU Gang,et al.Recognition model of tobacco fresh leaf maturity based on YOLOv5[J].Acta Tabacaria Sinica,2023,29(2):46-55.
[11]杨心安，朱文魁，张明建，等.盒装机制雪茄烟缺陷分类视觉检测方法[J].烟草科技，2023,56 (10):95-101.YANG Xinan,ZHU Wenkui,ZHANG Mingjian,et al.A visual inspection method for integrity of machine-made cigars in packets[J].Tobacco Science&Technology,2023,56(10):95-101.
[12]Zhu L,Wang X,Ke Z,et al.Bi Former:Vision Transformer with Bi-Level Routing Attention[C]//Proceedings of the IEEE/CVFConference on Computer Vision and Pattern Recognition.2023:10323-10333.
[13]Chen J,Kao S,He H,et al.Run,don’t walk:chasing higher FLOPS for faster neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2023:12021-12031.
[14]Zhang H,Zhang S.Shape-Io U:More accurate metric considering bounding box shape and scale[J].ar Xiv preprint ar Xiv:2312.17663,2023.
[15]余红霞，罗瑞林，云利军，等.基于YOLOv5的烤烟烟叶散把程度检测算法研究[J].烟草科技，2022,55(06):98-105.YU Hongxia,LUO Ruilin,YUN Lijun,et al.YOLOv5-based detection algorithm for evaluating degree of bundle-loosening of flue-cured tobacco[J].Tobacco Science&Technology,2022,55(06):98-105.
[16]Wang Q,Yang L,Zhou B,et al.YOLO-SS-Large:A lightweight and high-performance model for defect detection in substations[J].Sensors,2023,23(19):8080.
[17]Cui C,Gao T,Wei S,et al.PP-LCNet:A lightweight CPUconvolutional neural network[J].ar Xiv preprint ar Xiv:2109.15099,2021.
[18]Howard A G,Zhu M,Chen B,et al.Mobilenets:Efficient convolutional neural networks for mobile vision applications[J].ar Xiv preprint ar Xiv:1704.04861,2017.
[19]Han K,Wang Y,Tian Q,et al.Ghostnet:More features from cheap operations[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2020:1580-1589.
[20]Girshick R,Donahue J,Darrell T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.2014:580-587.
基本信息:
DOI：10.16472/j.chinatobacco.2024.T0192
中图分类号:TS47;TP183;TP391.41
引用信息:
[1]胡东辉,刘振宇,林苗俏等.基于改进YOLOv5s的异常烟丝识别检测轻量化算法[J].中国烟草学报,2025,31(03):78-87.DOI:10.16472/j.chinatobacco.2024.T0192.
基金信息:
中国烟草总公司科技项目“梗签形成机制及大工艺协同控制技术研究与应用”(110202202010)