YOLOv11-CoordAttention轻量化烟叶目标检测模型

doi:10.14088/j.cnki.issn0439-8114.2026.01.025

摘要/Abstract

摘要： 为提升YOLOv11模型在烟叶智能分级目标检测任务中的性能,解决资源受限环境下烟叶目标检测的准确性和时效性问题,提出一种轻量级的YOLOv11-CoordAttention烟叶目标检测模型。通过对比不同主干网络、卷积模块及注意力机制对模型精度与速度的影响,评估各组件的有效性。在此基础上设立消融试验,以探究优化组合的实际效果,从而全面揭示模型在实际应用中的性能表现。结果表明,YOLOv11-CoordAttention模型在烟叶目标检测任务中具有更优的综合性能,其精确率为100%,召回率为99.4%,F1分数为99.7%,mAP50为99.5%,模型大小为5.2 MB,参数量为2.3×10⁶,计算量为6.3×10⁹,帧率为198.2帧/s。相较于YOLOv11模型,YOLOv11-CoordAttention模型的精确率提升1.2个百分点,平均精度均值提升0.1个百分点。YOLOv11-CoordAttention模型的训练过程稳定有效,表现出色。训练集和验证集的各项损失均随训练轮次增加稳步下降并趋于收敛,表明模型学习过程充分且未出现过拟合。在性能指标方面,该模型的精确率与召回率均保持高位,实现高精度与低漏检率;mAP50与mAP50-95指标俱佳,表明其检测能力强大且鲁棒性高。YOLOv11-CoordAttention模型兼具轻量、高效与精准的优势,可在资源受限设备上稳定运行,胜任复杂场景下的烟叶检测任务。

关键词: YOLOv11-CoordAttention, 轻量化, 烟叶, 目标检测模型

Abstract: To enhance the performance of the YOLOv11 model in intelligent grading tasks for tobacco leaf object detection and to address the issues of accuracy and timeliness in tobacco leaf object detection within resource-constrained environments, a lightweight YOLOv11-CoordAttention tobacco leaf object detection model was proposed. The effectiveness of various components was evaluated by comparing the impact of different backbone networks, convolutional modules, and attention mechanisms on model accuracy and speed.Ablation experiments were set up on this basis to investigate the practical effects of optimized combinations, thereby comprehensively revealing the model’s performance in practical applications. The results indicated that the YOLOv11-Coord Attention model demonstrated superior comprehensive performance in the tobacco leaf object detection task, achieving a precision of 100%, recall of 99.4%, F1-score of 99.7%, mAP50 of 99.5%, with a model size of 5.2 MB, 2.3×10⁶ parameters, 6.3×10⁹ FLOPs, and a frame rate of 198.2 f/s. Compared to the YOLOv11 model, the YOLOv11-CoordAttention model improved precision by 1.2 percentage points and mean average precision by 0.1 percentage points. The training process of the YOLOv11-CoordAttention model was stable, effective, and exhibited outstanding performance. The losses for both the training and validation sets steadily decreased and converged as the training epochs increased, indicating a sufficient learning process without overfitting. In terms of performance metrics, the model maintained high precision and recall, achieving high accuracy and low missed detection rates. Its mAP50 and mAP50-95 metrics were both excellent, indicating powerful detection capability and high robustness. The YOLOv11-CoordAttention model combined the advantages of being lightweight, efficient, and accurate. It could run stably on resource-constrained devices and was competent for tobacco leaf detection tasks in complex scenarios.

Key words: YOLOv11-CoordAttention, lightweight, tobacco leaf, object detection model

中图分类号:

张千子, 朱云聪, 杜啟霞, 赵文军, 李丽华, 李学明, 邓邵文, 王剑松, 高云才, 曹静. YOLOv11-CoordAttention轻量化烟叶目标检测模型[J]. 湖北农业科学, 2026, 65(1): 152-158.

ZHANG Qian-zi, ZHU Yun-cong, DU Qi-xia, ZHAO Wen-jun, LI Li-hua, LI Xue-ming, DENG Shao-wen, WANG Jian-song, GAO Yun-cai, CAO Jing. A lightweight YOLOv11-CoordAttention model for tobacco leaf object detection[J]. HUBEI AGRICULTURAL SCIENCES, 2026, 65(1): 152-158.

参考文献

[1] XU G P, LIAO W T, ZHANG X, et al.Haar wavelet downsampling: A simple but effective downsampling module for semantic segmentation[J]. Pattern recognition, 2023, 143: 109819.
[2] CHEN G Y, KRZYZAK A.Face recognition via selective denoising, filter faces and hog features[J]. Signal, image and video processing, 2024, 18(1): 369-378.
[3] SINGH R P, DASH R, MOHAPATRA R K.LBP and CNN feature fusion for face anti-spoofing[J]. Pattern analysis and applications, 2023, 26(2): 773-782.
[4] MALGE A, DHADUK H M, M M S P. An approach to face detection and recognition using Viola Jones[J]. International journal of engineering and advanced technology, 2019, 8(5s): 52-56.
[5] 邹香玲. 目标检测算法研究述评[J].河南广播电视大学学报,2021,34(1):108-112.
[6] QIAO H L, YANG X N, LIANG Z Q, et al.A method for extracting joints on mountain tunnel faces based on mask R-CNN image segmentation algorithm[J]. Applied sciences, 2024, 14(15): 6403.
[7] Unleashing hidden canines: A novel fast R-CNN based technique for automatic auxiliary canine impaction[J].International journal of advanced technology and engineering exploration (IJATEE),2024,11(115):10102600.
[8] MALLAHI E I,RIFFI J,TAIRI H, et al.Efficient vehicle detection and classification algorithm using Faster R-CNN models[J].Journal of automation, mobile robotics and intelligent systems,2024, 18(4):86-93.
[9] JIAO L, ABDULLAH M I.YOLO series algorithms in object detection of unmanned aerial vehicles: A survey[J]. Service oriented computing and applications, 2024, 18(3): 269-298.
[10] YIN Z B, LIU F Y, GENG H, et al.A high-precision jujube disease spot detection based on SSD during the sorting process[J]. PLoS one, 2024, 19(1): e0296314.
[11] 肖恒树,李军营,梁虹,等.基于改进YOLOv8和无人机遥感影像的大田烟株数量检测[J].电子测量技术,2024,47(9):163-171.
[12] 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(6):98-105.
[13] 朱波,胡朋,刘宇晨,等.基于CSS-Cascade Mask R-CNN的有遮挡多片烟叶部位识别[J].农业工程学报,2024,40(9):271-280.
[14] 周运磊,董效杰,刘三军,等.基于改进YOLOv11n的轻量级电力设备过热故障红外图像检测算法[J].湖北民族大学学报(自然科学版),2025,43(1):114-118,140.
[15] 吴钟仁,周莲英,丁腊春.基于双向融合和特征增强的SSD小目标检测算法[J].计算机与数字工程,2025,53(1):115-118,175.
[16] QIN D F, LEICHNER C, DELAKIS M, et al.MobileNetV4: Universal models for the mobile ecosystem[A]. European conference on computer vision[C]. Cham: Springer nature switzerland, 2024.78-96.
[17] TANG Y, HAN K, GUO J, et al.GhostNetv2: Enhance cheap operation with long-range attention[J]. Advances in neural information processing systems, 2022, 35: 9969-9982.
[18] TAN M X, LE Q.EfficientNetV2: Smaller models and faster training[A]. International conference on machine learning[C]. PMLR, 2021.10096-10106.
[19] CHEN J R, KAO S H, HE H, et al.Run, don’t walk: Chasing higher FLOPS for faster neural networks[A].2023 IEEE/CVF conference on computer vision and pattern recognition(CVPR)[C]. Vancouver, BC, Canada: IEEE, 2023.12021-12031.
[20] ZHONG J C, CHEN J Y, MIAN A.DualConv: Dual convolutional kernels for lightweight deep neural networks[J]. IEEE transactions on neural networks and learning systems, 2023, 34(11): 9528-9535.
[21] SINGH P, VERMA V K, RAI P, et al.HetConv: Heterogeneous kernel-based convolutions for deep CNNs[A].2019 IEEE/CVF conference on computer vision and pattern recognition(CVPR)[C].Long Beach, CA, USA:IEEE, 2020.4830-4839.
[22] ZHANG J N, LI X T, LI J, et al.Rethinking mobile block for efficient attention-based models[A]. 2023 IEEE/CVF international conference on computer vision[C]. Piscataway, USA: IEEE, 2023. 1389-1400.
[23] HOU Q B, ZHOU D Q, FENG J S. Coordinate attention for efficient mobile network design[EB/OL].(2021-03-04). https://arxiv.org/abs/2103.02907.
[24] 周秀珊,文露婷,介百飞,等.改进YOLOv11的水面膨化饲料颗粒图像实时检测算法[J].智慧农业(中英文),2024,6(6):155-167.
[25] 张永强,李胜男,张子强,等.R-YOLO轨道人员目标检测模型[J].河北科技大学学报,2023,44(6):580-588.