针对高原茶园病害人工检测效率低、光照变化下小尺度病斑检测性能受限等问题,本研究提出轻量化YOLOv8n-GCW模型。基于云南大叶种茶园构建典型病害数据集,通过三重优化提升性能：（1）GhostNet重构特征提取网络适配边缘计算;（2）集成坐标注意力（Coordinate attention,CA）机制增强复杂光照下的鲁棒性;（3）动态损失函数（WIoU）优化重叠病斑定位精度。研究结果表明,改进型模型YOLOv8n-GCW较原先的YOLOv8n模型精确率提升5.6%,召回率提高4.0%,F1分数增长5.3%,平均精度均值（mAP@0.5）上升4.5%,参数量减少42.7%。验证集上关键损失函数下降9.87%~10.74%,推理速度每秒帧数（Frames per second,FPS）达53.84,能满足便携式农业设备轻量化部署需求。消融实验与梯度加权类激活映射（Gradient-weighted class activation mapping,Grad-CAM）可视化进一步验证该模型的有效性。本研究为高原茶园病害智能监测提供了轻量化解决方案,结合农业物联网技术（Agricultural internet of things,Agri-IoT）,未来有望扩展至茶园全域智能防控系统,推动智慧茶园的可持续发展。
Abstract
This study proposed a lightweight YOLOv8n-GCW model to address the challenges of low efficiency in manual disease detection and limited performance in small-scale lesion identification under varying illumination conditions in plateau tea plantations. Based on a dataset of typical diseases in Yunnan large-leaf tea gardens, the model was optimized in three key areas to improve performance: (1) the feature extraction network was reconstructed using GhostNet for edge computing compatibility. (2) The Coordinate attention mechanism was integrated to enhance robustness under complex lighting. (3) The overlapping lesion localization accuracy was optimized through a dynamic Wise-IoUloss function. Experimental results demonstrate that the improved model outperformed the original YOLOv8n, with a 5.6% increase in precision, 4.0% in recall, 5.3 % in F1-score, and 4.5% in mean average precision (mAP@0.5), while reducing the number of parameters by 42.7%. On the validation set, key loss functions decreased by 9.87%-10.74%, and the inference speed reached 53.84FPS, meeting the lightweight deployment requirements of portable agricultural devices. Ablation studies and Grad-CAM visualizations further validated the model's effectiveness. This study provided a lightweight solution for intelligent disease monitoring in plateau tea plantations. Combined with agricultural IoT technology, it held potential for extension to comprehensive smart disease control systems across tea plantations, promoting sustainable development in smart tea cultivation.
关键词
YOLOv8n-GCW /
茶树病害 /
轻量化检测 /
注意力机制 /
WIoU损失函数
Key words
YOLOv8n-GCW /
tea disease /
lightweight detection /
attention mechanism /
WIoU loss function
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 陈雪芬. 我国茶树病害的发生趋势与绿色防控[J]. 中国茶叶, 2022, 44(6): 7-14.
Chen X F.Trends and green control of tea plant diseases in China[J]. China Tea, 2022, 44(6): 7-14.
[2] 赵永田, 马悦, 胡德禹, 等. 我国茶树叶部主要真菌病害绿色防控现状与展望[J]. 植物保护, 2023, 49(5): 133-144, 166.
Zhao Y T, Ma Y, Hu D Y, et al.Current status and prospects of green control for major fungal diseases in tea leaves in China[J]. Plant Protection, 2023, 49(5): 133-144, 166.
[3] 宋攀. 贵州安顺市山地茶园茶饼病的发生特点及防控对策[J]. 现代园艺, 2025, 48(8): 77-79.
Song P.Occurrence characteristics and control strategies of tea blister blight in mountain tea plantations of Anshun, Guizhou[J]. Modern Horticulture, 2025, 48(8): 77-79.
[4] 边磊, 何旭栋, 季慧华, 等. 基于机器视觉的小贯小绿叶蝉智能识别的研究与应用[J]. 茶叶科学, 2022, 42(3): 376-386.
Bian L, He X D, Ji H H, et al.Research and application of intelligent recognition for Empoasca onukii based on machine vision[J]. Journal of Tea Science, 2022, 42(3): 376-386.
[5] Xue Z Y, Xu R J, Bai D, et al. YOLO-Tea: a tea disease detection model improved by YOLOv5 [J]. Forests, 2023, 14: 415. https://doi.org/10.13390/f14020415.
[6] Li H X, Yuan W X, Xia Y X, et al. YOLOv8n-WSE-Pest: a lightweight deep learning model based on YOLOv8n for pest identification in tea gardens [J]. Applied Sciences, 2024, 14(19): 8748. https://doi.org/10.13390/app14198748.
[7] Wang Z J, Zhang S H, Chen L J, et al. Microscopic insect pest detection in tea plantations: improved YOLOv8 model based on deep learning [J]. Agriculture, 2024, 14(10): 1739. https://doi.org/10.13390/agriculture14101739.
[8] Alam B M, Ahammed F, Kibria G, et al. TeaLeafBD: a comprehensive image dataset to classify the diseased tea leaf to automate the leaf selection process in Bangladesh [J]. Data in Brief, 2025,61, 111769. https://doi.org/10.11016/j.dib.2025.111769.
[9] 余文森. 复杂背景茶树病害图像识别[J]. 武夷学院学报, 2022, 41(3): 33-40.
Yu W S.Image recognition of tea plant diseases in complex backgrounds[J]. Journal of Wuyi University, 2022, 41(3): 33-40.
[10] 孙道宗, 刘欢, 刘锦源, 等. 基于改进YOLOv4模型的茶叶病害识别[J]. 西北农林科技大学学报(自然科学版), 2023, 51(9): 145-154.
Sun D Z, Liu H, Liu J Y, et al.Tea disease recognition based on improved YOLOv4 model[J]. Journal of Northwest A & F University (Natural Science Edition), 2023, 51(9): 145-154.
[11] 梁俊杰, 王东霞. 基于改进EfficientNet-Lite适用于复杂背景下茶树病害识别轻量级移动模型[J]. 农业技术与装备, 2024, 45(9): 23-25, 28.
Liang J J, Wang D X.A lightweight mobile model for tea disease recognition in complex backgrounds based on improved EfficientNet-Lite[J]. Agricultural Technology & Equipment, 2024, 45(9): 23-25, 28.
[12] 俞淑燕, 杜晓晨, 冯海林, 等. TTLD-YOLOv7: 非结构化环境下茶树病害的检测算法[J]. 茶叶科学, 2024, 44(3): 453-468.
Yu S Y, Du X C, Feng H L, et al.TTLD-YOLOv7: a detection algorithm for tea plant diseases in unstructured environments[J]. Journal of Tea Science, 2024, 44(3): 453-468.
[13] 肖双喜, 姚彤彤, 李灿. 基于改进MobileNetV2的轻量化茶叶病害检测方法[J]. 华南农业大学学报, 2025, 46(6): 801-809.
Xiao S X, Yao T T, Li C.A lightweight tea disease detection method based on improved MobileNetV2[J]. Journal of South China Agricultural University, 2025, 46(6): 801-809.
[14] Ye R, Shao G Q, He Y, et al. YOLOv8-RMDA: lightweight YOLOv8 Network for early detection of small target diseases in tea [J]. Sensors, 2024, 24(9): 2896. https://doi.org/10.13390/s24092896.
[15] Lin Y X, Xiao X, Lin H F.YOLOv8-FDA: lightweight wheat ear detection and counting in drone images based on improved YOLOv8[J]. Frontiers in Plant Science, 2025, 16: 1682243. https://doi.org/10.13389/fpls.2025.1682243.
[16] 陈禹, 吴雪梅, 张珍, 等. 基于改进YOLOv5s的自然环境下茶叶病害识别方法[J]. 农业工程学报, 2023, 39(24): 185-194.
Chen Y, Wu X M, Zhang Z, et al.A tea disease recognition method in natural environments based on improved YOLOv5s[J]. Transactions of the Chinese Society of Agricultural Engineering, 2023, 39(24): 185-194.
[17] Li X T, Zhang T H, Yu M, et al. A YOLOv8-based method for detecting tea disease in natural environments [J]. Agronomy Journal, 2025, 117(2): 43. https://doi.org/10.11002/agj2.70043.
[18] Su B, Zhu Y Y, Lin, Y F. Pest-YOLOv8: enhanced detection for small and complex agricultural pests using triple attention and wise-IoU [J]. Journal of Plant Diseases and Protection, 2025, 132(5): 1134. https://doi.org/10.11007/s41348-025-01134-w.
[19] Wu Z X, Lü J, Pan J F, et al. REB-Tea: An intelligent detection model for tea buds with clarity and multi-scale feature enhancement [J]. Agriculture, 2026, 16(12): 134 https://doi.org/10.13390/agriculture1612134.
[20] 余京豪, 陈炫宇, 韩章, 等. 基于SAM实例分割的茶树叶片病害严重度估计方法[J]. 中国农机化学报, 2026, 47(4): 139-146.
Yu J H, Chen X Y, Han Z, et al.A SAM-based instance segmentation method for estimating severity of tea leaf diseases[J]. Journal of Chinese Agricultural Mechanization, 2026, 47(4): 139-146.
[21] Ai J C, Li Y D, Gao S X, et al. Tea disease detection method based on improved YOLOv8 in complex background [J]. Sensors, 2025, 25(13): 4129. https://doi.org/10.13390/s25134129.
[22] 李伟豪, 詹炜, 周婉, 等. 轻量型Yolov7-TSA网络在茶叶病害检测识别中的研究与应用[J]. 河南农业科学, 2023, 52(5): 162-169.
Li W H, Zhan W, Zhou W, et al.Research and application of lightweight Yolov7-TSA network in tea disease detection[J]. Journal of Henan Agricultural Sciences, 2023, 52(5): 162-169.
[23] 余盈潭, 袁琳, 聂臣巍,等. 基于无人机遥感的茶园多胁迫分层监测方法研究[J]. 茶叶科学, 2026, 46(2): 292-310
Yu Y T, Yuan L, Nie C W, et al.A hierarchical monitoring method for multiple stresses in tea plantations based on UAV remote sensing[J]. Journal of Tea Science, 2026, 46(2): 292-310.
[24] Chen Z L, Guo Y., He J, et al.Research on lightweight detection and recognition of tobacco disease based on RT-YOLOv10 and UAV remote sensing images[J]. Smart Agricultural Technology, 2025, 12: 101415. https://doi.org/10.11016/j.atech.2025.101415.
[25] Wu X J, Liang J Z, Yang Y Y, et al. SAW-YOLO: a multi-scale YOLO for small target citrus pests detection [J]. Agronomy, 2024, 14(7): 1571. https://doi.org/10.13390/agronomy14071571.
[26] Phan Q H, Setyawan B, Duong T P, et al. Enhanced detection of algal leaf spot, tea brown blight,tea grey blight diseases using YOLOv5 Bi-HIC model with instance and context information [J]. Plants, 2025, 14(20): 3219. https://doi.org/10.13390/plants14203219.
基金
云南省科技厅农业联合专项（202401BD070001-053）; 云南省茶叶产业人工智能与大数据应用创新团队（202405AS350025）
PDF(3936 KB)
浙公网安备 33019902000101号