基于改进YOLOv8n模型的烟梗识别与检测

doi:10.3969/j.issn.1004-1524.20240508

浙江农业学报 ›› 2025, Vol. 37 ›› Issue (11): 2395-2407.DOI: 10.3969/j.issn.1004-1524.20240508

基于改进YOLOv8n模型的烟梗识别与检测

冯永新¹(), 姬远鹏², 崔英¹, 白力¹, 王健¹, 丁佳¹, 张晓东¹, 张伟伟², 李萌³, 张卫正²^,^*()

1.河北中烟工业有限责任公司,河北石家庄 050051
2.郑州轻工业大学计算机科学与技术学院,河南郑州 450001
3.郑州轻工业大学烟草科学与工程学院,河南郑州 450001

收稿日期:2024-06-11 出版日期:2025-11-25 发布日期:2025-12-08
作者简介:冯永新(1977—),男,安徽全椒人,学士,工程师,主要从事烟叶原料研究。E-mail:fyxman@163.com
通讯作者: ^*张卫正,E-mail: weizheng008@vip.126.com
基金资助:
河南省科技攻关项目(242102110334);河南省高等学校重点科研项目(24B520039);河北中烟工业有限责任公司科技计划项目(2023130000340016)

The recognition and detection of tobacco stems based on the improved YOLOv8n model

FENG Yongxin¹(), JI Yuanpeng², CUI Ying¹, BAI Li¹, WANG Jian¹, DING Jia¹, ZHANG Xiaodong¹, ZHANG Weiwei², LI Meng³, ZHANG Weizheng²^,^*()

1. Hebei Tobacco Industry Co., Ltd., Shijiazhuang 050051, China
2. School of Computer Science and Technology, Zhengzhou University of Light Industry, Zhengzhou 450001, China
3. School of Tobacco Science and Engineering, Zhengzhou University of Light Industry, Zhengzhou 450001, China

Received:2024-06-11 Online:2025-11-25 Published:2025-12-08

摘要/Abstract

摘要： 对烟梗结构进行实时、准确的识别是提高烟梗筛选质量和效率的关键,为了解决人工筛选烟梗效率低和质量差的问题,该研究提出了一种改进的YOLOv8n模型用于识别烟梗,以实现烟梗结构的自动化检测。以复烤后的烟梗为研究对象,在YOLOv8n模型基础上以可变核卷积代替标准卷积,增强了网络模型对烟梗的特征提取,减少了模型参数量。将三重注意力模块嵌入骨干网络中,通过跨维度交互增强网络对烟梗位置信息的关注。结果表明,改进后模型的全类平均精度(mAP)为95.40%,参数量为2.52×10⁶,每秒浮点运算数为7.50 G,检测速度为每秒62.38帧。与YOLOv8n模型相比,改进后的模型mAP提高了4.55百分点,参数量和每秒浮点运算数分别减少了0.49×10⁶和0.70 G。与Faster R-CNN、SSD、YOLOv5n、YOLOv6n、YOLOv7tiny模型相比,改进后的模型具有更高的准确性和更快的检测速度,展示出优越的综合性能。同时,将改进后模型嵌入到计算机设备中,在实际应用中测试模型性能,该模型展现出了较高的准确性和实时性,有效提高了烟梗识别与分拣效率。

关键词: 烟梗, 改进YOLOv8n模型, 识别, 检测, 分拣

Abstract:

Real-time and accurate recognition of tobacco stems is essential for enhancing the quality and efficiency of tobacco stem sorting. To tackle the challenges of low efficiency and subpar quality in manual tobacco stem sorting, this paper proposed a tobacco stem recognition method based on the improved YOLOv8n model, aiming to achieve automated detection of tobacco stem structures. Taking the tobacco stems after re-roasting as the research object, variable kernel convolution was used instead of standard convolution based on the YOLOv8n model to enhance the feature extraction of tobacco stems by the network model and reduce parameters of the model. A triple attention module is embedded into the backbone network to enhance the network’s focus on tobacco stem positional information through cross-dimensional interactions. The results demonstrated that the improved model achieved an mAP of 95.40% with a parameter of 2.52×10⁶, floating-point operations per second (FLOPs) of 7.50 G, and detection speed of 62.38 frames per second. Compared to the YOLOv8n model, the improved model achieved a 4.55 percentage points increase in mAP while reducing parameters and FLOPs by 0.49×10⁶ and 0.70 G, respectively. Compared with Faster R-CNN, SSD, YOLOv5n, YOLOv6n, and YOLOv7tiny, the improved model had superior accuracy and faster detection speed, demonstrating superior-comprehensive performance. Furthermore, embedding the improved model into computing devices and testing its performance in real-world applications. The model demonstrated high accuracy and real-time capabilities, effectively improving the efficiency of recognition and sorting of tobacco stems.

Key words: tobacco stem, improved YOLOv8n model, recognition, detection, sorting

中图分类号:

S572

冯永新, 姬远鹏, 崔英, 白力, 王健, 丁佳, 张晓东, 张伟伟, 李萌, 张卫正. 基于改进YOLOv8n模型的烟梗识别与检测[J]. 浙江农业学报, 2025, 37(11): 2395-2407.

FENG Yongxin, JI Yuanpeng, CUI Ying, BAI Li, WANG Jian, DING Jia, ZHANG Xiaodong, ZHANG Weiwei, LI Meng, ZHANG Weizheng. The recognition and detection of tobacco stems based on the improved YOLOv8n model[J]. Acta Agriculturae Zhejiangensis, 2025, 37(11): 2395-2407.

图/表 11

图1 烟梗识别与分类平台

Fig.1 The platform for recognition and classification of tobacco stems

图2 烟梗图像

Fig.2 Images of tobacco stems

图3 改进的YOLOv8n模型结构 YGshort,短烟梗;YGlong,长烟梗。后面的数字表示判定为该类烟梗的概率。

Fig.3 Architecture of the improved YOLOv8n model YGshort, short tobacco stems; YGlong, long tobacco stems. The subsequent numbers indicating the probability of being classified into each category.

图4 AKConv结构 C表示通道数,H和W分别表示烟梗图像的高度和宽度,图5同。N表示卷积核的大小。

Fig.4 Structure of alterable kernel convolution C represents the number of channels, H and W represent the height and width of the tobacco stem images, respectively, the same as Fig.5. N represents the size of the convolution kernel.

图5 三重注意力模块结构

Fig.5 Structure of triplet attention module

表1 改进的YOLOv8n模型与6个不同模型在烟梗测试集上的结果

Table 1 Results of improved YOLOv8n model and six different models on the test set of the tobacco stem

模型 Model	AP_long/%	AP_short/%	AP_broken/%	mAP/%	参数量 Parameter/10⁶	FLOPs/G	FPS
Faster R-CNN	76.09	86.12	57.23	73.15	136.69	200.84	13.22
SSD	79.49	91.74	92.45	87.89	23.61	136.59	29.58
YOLOv5n	90.50	83.93	85.37	86.60	1.90	4.50	63.16
YOLOv6n	89.25	83.46	82.95	85.22	4.70	11.40	52.43
YOLOv7tiny	84.32	83.51	82.97	83.60	6.01	13.00	50.40
YOLOv8n	91.21	89.63	91.72	90.85	3.01	8.20	56.65
改进的YOLOv8n	95.60	94.71	95.89	95.40	2.52	7.50	62.38
Improved YOLOv8n

图6 七个模型的综合性能对比分析

Fig.6 Comprehensive performance comparison and analysis of seven models APlong, Average precision of long tobacco stem; APshort, Average precision of short tobacco stem; APbroken, Average precision of broken tobacco stem; FLOPs, Floating-point operation per second; FPS, Frames per second.

图7 七个模型检测效果对比黄色虚线框表示误检。YGbroken,碎烟梗;YGshort,短烟梗;YGlong,长烟梗。后面的数字表示判定为该类烟梗的概率。

Fig.7 Comparison of detection performance of seven models The yellow dashed box indicates false detection. YGbroken, Broken tobacco stems; YGshort, Short tobacco stems, YGlong, Long tobacco stems. The subsequent numbersindicating the probability of being classified into each category.

表2 不同优化模块的消融实验结果

Table 2 Results of ablation experiments on different optimization modules

YOLOv8n模型 YOLOv8n model	AKConv	TAM	AP_long/%	AP_short/%	AP_broken/%	mAP/%	参数量 Parameter/10⁶	FLOP_S/G	FPS
√			91.21	89.63	91.72	90.85	3.01	8.20	56.65
√	√		92.06	91.53	92.83	92.14	2.52	7.40	64.10
√	√	√	95.60	94.71	95.89	95.40	2.52	7.50	62.38

图8 烟梗识别与分拣流程

Fig.8 Workflow of recognition and sorting of tobacco stems

表3 改进的YOLOv8n模型与人工分选对比

Table 3 Comparison between improved YOLOv8n model and manual sorting

方法 Method	长烟梗质量 Quality of long tobacco stem/g	短烟梗质量 Quality of short tobacco stem/g	碎烟梗质量 Quality of broken tobacco stem/g	长梗率 Long tobacco stem rate/%	碎梗率 Broken tobacco stem rate/%	长烟梗识别误差 Identification error of long tobacco stem/%	短烟梗识别误差 Identification error of short tobacco stem/%	碎烟梗识别误差 Identification error of broken tobacco stem/%
人工分选	986.2	413.5	100.3	65.75	6.69	0	0	0
Manual sorting
改进的YOLOv8n模型	982.7	420.4	96.9	65.51	6.46	0.35	1.82	3.40
Improved YOLOv8n model

参考文献 19

[1]	ZOU X D, BK A, ABU-IZNEID T, et al. Current advances of functional phytochemicals in Nicotiana plant and related potential value of tobacco processing waste: a review[J]. Biomedicine & Pharmacotherapy, 2021, 143: 112191.
[2]	YANG X Y, LIU Z C, LIU J S, et al. Two-step strategy for the comprehensive utilization of tobacco stem[J]. Chemical Papers, 2023, 77(4): 1797-1808.
[3]	GUO D Y, HU Q. Design of multi-indicator integrated testing system for tobacco intelligent silk production line[J]. Journal of Intelligent & Fuzzy Systems, 2022, 42(3): 2615-2627.
[4]	WANG W, ZHANG L L, WANG J, et al. Locating defects and image preprocessing: deep learning in automated tobacco production[J]. Journal of Sensors, 2022, 2022(1): 6797207.
[5]	朱文魁, 刘斌, 毛伟俊, 等. 基于低能X射线透射成像的打叶片烟中烟梗在线检测[J]. 烟草科技, 2015, 48(2): 69-74.
	ZHU W K, LIU B, MAO W J, et al. A method for on-line detection of stem in strips based on low-energy X-ray transmission imaging[J]. Tobacco Science & Technology, 2015, 48(2): 69-74.
[6]	席建平, 易浩, 刘斌, 等. 基于FPGA的烟梗在线检测系统设计[J]. 中国烟草学报, 2016, 22(5): 50-54.
	XI J P, YI H, LIU B, et al. Design of online tobacco stems detection system based on FPGA[J]. Acta Tabacaria Sinica, 2016, 22(5): 50-54.
[7]	崔云月, 管一弘, 孙娜, 等. BP神经网络在烟梗长短梗率检测中的应用[J]. 软件导刊, 2021, 20(2): 63-67.
	CUI Y Y, GUAN Y H, SUN N, et al. The application of BP neural network in the determination of the stalk length and stem rate[J]. Software Guide, 2021, 20(2): 63-67.
[8]	杨耀伟, 张月华, 崔廷, 等. 基于机器视觉和深度学习的烟梗识别方法[J]. 计算机应用, 2022, 42(S1):118-122.
	YANG Y W, ZHANG Y H, CUI T, et al. Tobacco stem recognition method based on machine vision and deep learning[J]. Journal of Computer Applications, 2022, 42(S1):118-122.
[9]	刘新宇, 郝同盟, 张红涛, 等. 基于改进YOLOv3网络的烟梗识别定位方法[J]. 食品与机械, 2022, 38(3): 103-109.
	LIU X Y, HAO T M, ZHANG H T, et al. Cigarette stem identification and location method based on improved YOLOv3 network[J]. Food & Machinery, 2022, 38(3): 103-109.
[10]	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(6): 65-74.
	XIAO L Y, WANG S, LIU Y G, et al. Classification and identification of tobacco stem morphology based on deep learning technology[J]. Tobacco Science & Technology, 2021, 54(6): 65-74.
[11]	郑银环, 林晓琛, 吴飞, 等. 基于改进YOLOv4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, 1253.
	ZHENG Y H, LIN X C, WU F, et al. Lightweight tobacco stem identification method based on improved YOLOv4[J]. Journal of Hefei University of Technology (Natural Science), 2023, 46(9): 1196-1202, 1253.
[12]	MISRA D, NALAMADA T, ARASANIPALAI A U, et al. Rotate to attend: convolutional triplet attention module[C]//2021 IEEE Winter Conference on Applications of Computer Vision (WACV). January 3-8, 2021, Waikoloa, HI, USA. IEEE, 2021: 3138-3147.
[13]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]// Computer Vision-ECCV 2018. Cham: Springer, 2018: 3-19.
[14]	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.
[15]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[C]// Computer Vision-ECCV 2016. Cham: Springer, 2016: 21-37.
[16]	MA L, YU Q W, YU H L, et al. Maize leaf disease identification based on YOLOv5n algorithm incorporating attention mechanism[J]. Agronomy, 2023, 13(2): 521.
[17]	LI C Y, LI L L, JIANG H L, et al. YOLOv6:a single-stage object detection framework for industrial applications[EB/OL]. (2022-09-07)[2024-06-08]. https://arxiv.org/abs/2209.02976.
[18]	WANG C Y, BOCHKOVSKIY A, LIAO H M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 17-24, 2023, Vancouver, BC, Canada. IEEE, 2023: 7464-7475.
[19]	周雅宁. 烟梗加工处理技术与设备研究进展[J]. 中国烟草学报, 2019, 25(2): 121-129.
	ZHOU Y N. Research progress in tobacco stem processing technology and equipment[J]. Acta Tabacaria Sinica, 2019, 25(2): 121-129.

基于改进YOLOv8n模型的烟梗识别与检测

The recognition and detection of tobacco stems based on the improved YOLOv8n model

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 19

相关文章 15

编辑推荐

Metrics

本文评价

[1]	刘睿, 王丽娟, 王秋皓, 林旭东, 郭启航, 许多霖, 李文妍. 基于改进YOLOv8s的茶叶病虫害检测[J]. 浙江农业学报, 2025, 37(9): 1933-1942.
[2]	吕胤春, 段恩泽, 朱一星, 郑霞, 柏宗春. 基于YOLOv8-Swin Transformer模型的翻覆肉鸭实时检测[J]. 浙江农业学报, 2025, 37(7): 1556-1566.
[3]	陈伟, 姚洁, 冯明峰, 李帅, 蒋西子, 蒋磊, 江彤. 蔬菜病毒不同检测技术灵敏度的比较研究[J]. 浙江农业学报, 2025, 37(5): 1072-1081.
[4]	沈岚, 杨肖芳, 张国芳. 草莓根腐病病原菌的鉴定及其对杀菌剂的敏感性[J]. 浙江农业学报, 2025, 37(2): 417-425.
[5]	汤奥冉, 金秀, 王坦, 饶元, 李佳佳, 张武. 基于弯曲大豆植株主茎骨架重构的生理株高测量方法[J]. 浙江农业学报, 2025, 37(2): 466-479.
[6]	孙硕, 刘昭华, 王可, 郑纪业, 邢凡彬, 宋现雪, 王建英, 孟宪锋, 杨景晁, 张霞. 基于SNP与机器学习的羊品种识别算法研究[J]. 浙江农业学报, 2025, 37(11): 2387-2394.
[7]	余红伟, 邓惠丹, 杨华, 肖英平, 丁向英, 吉小凤. 动物源性食品中兽药残留检测技术及兽药残留污染状况分析[J]. 浙江农业学报, 2025, 37(1): 203-216.
[8]	谷瑞, 宋翠玲, 钱春花. 融合沙漏结构与改进坐标注意力的轻量级番茄叶片病害识别模型[J]. 浙江农业学报, 2025, 37(1): 217-230.
[9]	郭秀明, 王大伟, 刘升平, 诸叶平, 刘晓辉, 林克剑, 王佳宇, 李非. 基于深度学习的近地面草原鼠洞识别计数关键问题研究与应用[J]. 浙江农业学报, 2024, 36(9): 2146-2154.
[10]	王健霖, 田兴苗, 王景松, 戴莎莎, 郭亚男, 何生虎, 李继东. 牛支原体可视化重组酶聚合酶扩增检测方法的建立[J]. 浙江农业学报, 2024, 36(8): 1811-1819.
[11]	朱铭敏, 张国平, 谭建军, 孙玲姣, 朱黎, 焦洁. 基于YOLOv5s的轻量级茶叶嫩芽终端检测模型[J]. 浙江农业学报, 2024, 36(6): 1413-1424.
[12]	陆胜民, 黄子馨, 李小琼, 郑美瑜, 韩永斌. 热加工食品中晚期糖基化终末产物和5-羟甲基糠醛的形成、检测与控制[J]. 浙江农业学报, 2024, 36(6): 1458-1468.
[13]	汤永华, 石非凡, 林森, 张志鹏, 孟妍君, 刘兴通. 基于融合非局部操作的YOLOv5s高密度锦鲤鱼苗检测方法[J]. 浙江农业学报, 2024, 36(4): 952-967.
[14]	牛钰, 李晶, 王俊文, 李瑞瑞, 田强, 武玥, 郁继华. 高等植物花青素生物合成、调控、生物活性及其检测的研究进展[J]. 浙江农业学报, 2024, 36(4): 978-996.
[15]	郭伟娜, 陶晶, 何梦婷, 王紫苇, 马佰贺, 赵磊. 鸡源鼠伤寒沙门菌的分离鉴定、药敏试验与毒力基因检测[J]. 浙江农业学报, 2024, 36(2): 284-294.