基于改进YOLOv8的绿豆叶片主要病害检测研究

doi:10.13560/j.cnki.biotech.bull.1985.2025-1295

摘要/Abstract

摘要：

目的为解决自然环境中绿豆叶片病害检测场景复杂、检测精度不足、小目标检测能力较弱等问题，提出一种YOLOv8-EMA-CARAFE-Layer的深度学习算法。方法在主干网络中引入了高效多尺度注意力机制（efficient multi-scale attention, EMA），提升模型对病害特征的提取与筛选能力；在颈部网络中采用CARAFE（Content-Aware ReAssembly of FEatures）上采样模块，通过动态生成上采样核增强全局信息聚合能力，从而提高特征重建精度；在Neck和Head网络中添加一个小目标检测层，提升对细小病害目标的检测性能。结果改进后的模型在mAP@50、精确度和召回率等关键指标上分别达到了91.0%、86.9%、83.4%，且与主流目标检测模型YOLOv3、YOLOv5、YOLOv6和YOLOv11相比，平均精度mAP@50分别提升了6.8%、0.3%、13.5%和1.6%，在包含复杂田间背景的混合数据集上，表现出较强的绿豆叶片病害检测性能；此外，模型通过轻量化设计保持了较低的参数量，仅为5.95 Mb。结论 YOLOv8-EMA-CARAFE-Layer模型能够满足无人机、智能农机等移动终端的部署需求，在现有的田间混合数据集上，对提高绿豆叶片主要病害的快速、高效、精准检测水平具有实际应用借鉴价值。

关键词: 图像识别, 深度学习, YOLOv8, 绿豆叶片病害, 高效多尺度注意力机制

Abstract:

Objective To address issues such as complex detection scenarios, insufficient detection accuracy, and weak small-target detection capability for mung bean leaf diseases in natural environments, this study proposes a deep learning algorithm named YOLOv8-EMA-CARAFE-Layer. Method The EMA (efficient multi-scale attention) mechanism was introduced into the backbone network to enhance the model’s ability to extract and screen disease features. The CARAFE (Content-Aware ReAssembly of FEatures) upsampling module was adopted in the neck network, which enhanced global information aggregation capability by dynamically generating upsampling kernels, thereby improving feature reconstruction accuracy. A small-target detection layer was added to the neck and head networks to improve the detection effectiveness for minute disease targets. Result The improved model achieved a mAP@50 of 91.0%, a precision of 86.9%, and a recall of 83.4%. Compared with mainstream object detection models YOLOv3, YOLOv5, YOLOv6, and YOLOv11, the mAP@50 improved by 6.8%, 0.3%, 13.5%, and 1.6% respectively, demonstrating superior performance in mung bean leaf disease detection. Furthermore, through lightweight design, the model maintained a low parameter count of only 5.95 MB. Conclusion YOLOv8-EMA-CARAFE-Layer model enables deployment on mobile terminals such as drones and smart agricultural machinery, offering practical application value for enhancing the rapid, efficient, and accurate detection of major diseases in mung bean leaves.

Key words: image recognition, deep learning, YOLOv8, mung bean leaf diseases, efficient multi-scale attention mechanism

王超, 杨钰超, 郭思宇, 贾瑶镟, 陈艳霞, 赵钰, 乔星星, 肖璐洁, 冯美臣, 杨武德. 基于改进YOLOv8的绿豆叶片主要病害检测研究[J]. 生物技术通报, 2026, 42(10): 282-292.

WANG Chao, YANG Yu-chao, GUO Si-yu, JIA Yao-xuan, CHEN Yan-xia, ZHAO Yu, QIAO Xing-xing, XIAO Lu-jie, FENG Mei-chen, YANG Wu-de. Research on Detection of Major Diseases in Mung Bean Leaves Based on an Improved YOLOv8[J]. Biotechnology Bulletin, 2026, 42(10): 282-292.

图/表 13

参考文献 30

[1]	戴高星. 发展绿豆生产大有可为 [J]. 四川农业科技, 2011(6): 18-19.
	Dai GX. Developing mung bean production is promising [J]. Sichuan Agric Sci Technol, 2011(6): 18-19.
[2]	Maiti D, Majumdar M. Impact of bioprocessing on phenolic content and antioxidant activity of soy seed to improve hypoglycemic functionality [J]. Asian J Plant Sci Res, 2012, 2(2): 924-931.
[3]	孙瑞卿, 秦曙, 李晋栋, 等. 我国绿豆病虫害化学防治现状及质量安全问题分析 [J]. 农产品质量与安全, 2017(3): 91-94.
	Sun RQ, Qin S, Li JD, et al. Current situation of chemical control of mung bean diseases and insect pests in China and analysis of quality and safety problems [J]. Qual Saf Agro Prod, 2017(3): 91-94.
[4]	赵玉梅. 农业信息化在蔬菜病虫害防治技术上的应用分析 [J]. 农业工程技术, 2022, 42(27): 31-32.
	Zhao YM. Analysis on the application of agricultural informatization in vegetable pest control technology [J]. Agric Eng Technol, 2022, 42(27): 31-32.
[5]	梁斯佳, 李雷, 顾月, 等. 基于改进YOLO v8n的甘蔗叶片病害识别算法 [J]. 农业机械学报, 2025, 56(10): 567-574.
	Liang SJ, Li L, Gu Y, et al. Sugarcane leaf disease recognition algorithm based on improved YOLO v8n [J]. Trans Chin Soc Agric Mach, 2025, 56(10): 567-574.
[6]	Kumar Sahu S, Pandey M. An optimal hybrid multiclass SVM for plant leaf disease detection using spatial Fuzzy C-Means model [J]. Expert Syst Appl, 2023, 214: 118989.
[7]	黄林生, 江静, 黄文江, 等. Sentinel-2影像和BP神经网络结合的小麦条锈病监测方法 [J]. 农业工程学报, 2019, 35(17): 178-185.
	Huang LS, Jiang J, Huang WJ, et al. Wheat yellow rust monitoring method based on Sentinel-2 image and BPNN model [J]. Trans Chin Soc Agric Eng, 2019, 35(17): 178-185.
[8]	朱锐, 张家瑜, 黄继超, 等. 基于卷积神经网络的农作物叶片病害检测研究进展 [J]. 农业工程学报, 2025, 41(17): 15-28.
	Zhu R, Zhang JY, Huang JC, et al. Research progress of crop leaf disease detection based on convolutional neural network [J]. Trans Chin Soc Agric Eng, 2025, 41(17): 15-28.
[9]	徐洪丽, 安凤, 史宇轩. 基于改进YOLO v8的苹果叶片病害检测研究 [J]. 农业机械学报, 2025, 56(10): 530-538.
	Xu HL, An F, Shi YX. Apple leaf disease detection method based on improved YOLO v8 [J]. Trans Chin Soc Agric Mach, 2025, 56(10): 530-538.
[10]	苗荣慧, 李志伟, 武锦龙. 基于改进YOLO v7的轻量化樱桃番茄成熟度检测方法 [J]. 农业机械学报, 2023, 54(10): 225-233.
	Miao RH, Li ZW, Wu JL. Lightweight maturity detection of cherry tomato based on improved YOLO v7 [J]. Trans Chin Soc Agric Mach, 2023, 54(10): 225-233.
[11]	郑俊键, 兰玉彬, 熊万杰, 等. 基于YOLOv5s改进模型的小白菜虫害识别方法 [J]. 农业工程学报, 2024, 40(13): 124-133.
	Zheng JJ, Lan YB, Xiong WJ, et al. Method for identification of Pak choi pests and diseases based on improved YOLOv5s [J]. Trans Chin Soc Agric Eng, 2024, 40(13): 124-133.
[12]	祝诗平, 张越, 唐茂杰, 等. 基于改进YOLO v10n的百香果成熟果实检测方法 [J]. 农业机械学报, 2025, 56(10): 549-557, 595.
	Zhu SP, Zhang Y, Tang MJ, et al. Ripe fruit detection model of passion fruit based on improved YOLO v10n [J]. Trans Chin Soc Agric Mach, 2025, 56(10): 549-557, 595.
[13]	Khalid M, Sarfraz M, Iqbal U, et al. Real-time plant health detection using deep convolutional neural networks [J]. Agriculture, 2023, 13(2): 510.
[14]	公徐路, 张淑娟. 基于改进YOLOv5s的苹果叶片小目标病害轻量化检测方法 [J]. 农业工程学报, 2023, 39(19): 175-184.
	Gong XL, Zhang SJ. Lightweight detection of small target diseases in apple leaf using improved YOLOv5s [J]. Trans Chin Soc Agric Eng, 2023, 39(19): 175-184.
[15]	袁杰, 谢霖伟, 郭旭, 等. 基于改进YOLOv7的苹果叶片病害检测方法 [J]. 农业机械学报, 2024, 55(11): 68-74.
	Yuan J, Xie LW, Guo X, et al. Apple leaf disease detection method based on improved YOLOv7 [J]. Trans Chin Soc Agric Mach, 2024, 55(11): 68-74.
[16]	毛锐, 张宇晨, 王泽玺, 等. 利用改进Faster-RCNN识别小麦条锈病和黄矮病 [J]. 农业工程学报, 2022, 38(17): 176-185.
	Mao R, Zhang YC, Wang ZX, et al. Recognizing stripe rust and yellow dwarf of wheat using improved Faster-RCNN [J]. Trans Chin Soc Agric Eng, 2022, 38(17): 176-185.
[17]	王会征, 孙良晨, 李新龙, 等. 基于改进YOLOv7-tiny的番茄叶片病虫害检测方法 [J]. 农业工程学报, 2024, 40(10): 194-202.
	Wang HZ, Sun LC, Li XL, et al. Detecting tomato leaf pests and diseases using improved YOLOv7-tiny [J]. Trans Chin Soc Agric Eng, 2024, 40(10): 194-202.
[18]	甘雨, 郭庆文, 王春桃, 等. 基于改进EfficientNet模型的作物害虫识别 [J]. 农业工程学报, 2022, 38(1): 203-211.
	Gan Y, Guo QW, Wang CT, et al. Recognizing crop pests using an improved EfficientNet model [J]. Trans Chin Soc Agric Eng, 2022, 38(1): 203-211.
[19]	刘诗怡, 胡滨, 赵春. 基于改进YOLOv7的黄瓜叶片病虫害检测与识别 [J]. 农业工程学报, 2023, 39(15): 163-171.
	Liu SY, Hu B, Zhao C. Detection and identification of cucumber leaf diseases based improved YOLOv7 [J]. Trans Chin Soc Agric Eng, 2023, 39(15): 163-171.
[20]	孙丰刚, 王云露, 兰鹏, 等. 基于改进YOLOv5s和迁移学习的苹果果实病害识别方法 [J]. 农业工程学报, 2022, 38(11): 171-179.
	Sun FG, Wang YL, Lan P, et al. Identification of apple fruit diseases using improved YOLOv5s and transfer learning [J]. Trans Chin Soc Agric Eng, 2022, 38(11): 171-179.
[21]	Arnal Barbedo JG. Plant disease identification from individual lesions and spots using deep learning [J]. Biosyst Eng, 2019, 180: 96-107.
[22]	罗志聪, 何陈涛, 陈登捷, 等. 基于轻量化YOLOv8s-GD的自然环境下百香果快速检测模型 [J]. 农业机械学报, 2024, 55(8): 291-300.
	Luo ZC, He CT, Chen DJ, et al. Passion fruit rapid detection model based on lightweight YOLOv8s-GD [J]. Trans Chin Soc Agric Mach, 2024, 55(8): 291-300.
[23]	李嘉诚, 陈中举, 许浩然. 基于改进YOLOv8n的真实场景下草莓病害检测方法 [J]. 中国农机化学报, 2024, 45(12): 267-274.
	Li JC, Chen ZJ, Xu HR. A method for strawberry disease detection in real scenarios based on improved YOLOv8n [J]. J Chin Agric Mech, 2024, 45(12): 267-274.
[24]	Zhang YP, Liu J, Li SZ, et al. ESM-YOLOv11: a lightweight deep learning framework for real-time peanut leaf spot disease detection and precision severity quantification in field conditions [J]. Comput Electron Agric, 2025, 238: 110801.
[25]	Chen SY, Li YX, Zhang YD, et al. Soft X-ray image recognition and classification of maize seed cracks based on image enhancement and optimized YOLOv8 model [J]. Comput Electron Agric, 2024, 216: 108475.
[26]	Liu S, Qi L, Qin HF, et al. Path aggregation network for instance segmentation [C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. June 18-23, 2018, Salt Lake City, UT, USA. Piscataway, NJ: IEEE, 2018: 8759-8768.
[27]	于营, 王春平, 付强, 等. 语义分割评价指标和评价方法综述 [J]. 计算机工程与应用, 2023, 59(6): 57-69.
	Yu Y, Wang CP, Fu Q, et al. Survey of evaluation metrics and methods for semantic segmentation [J]. Comput Eng Appl, 2023, 59(6): 57-69.
[28]	Deng H, Chen YY, Xu YL. ALD-YOLO: a lightweight attention detection model for apple leaf diseases [J]. Front Plant Sci, 2025, 16: 1616224.
[29]	Liu F, Wang SD, Pang SC, et al. Detection and recognition of tea buds by integrating deep learning and image-processing algorithm [J]. Food Measure, 2024, 18(4): 2744-2761.
[30]	杨锋, 姚晓通. 基于改进YOLOv8的小麦叶片病虫害检测轻量化模型 [J]. 智慧农业, 2024, 6(1): 147-157.
	Yang F, Yao XT. Lightweighted wheat leaf diseases and pests detection model based on improved YOLOv8 [J]. Smart Agric, 2024, 6(1): 147-157.

病害名称 Disease types	病害特征 Disease characters
细菌性疫病 Bacterial blight	初期叶片表面出现小的水浸斑，变成淡黄色到棕褐色；后期病斑在叶片上扩展，病斑呈黑色
轮纹病 Ring spot	初期叶片产生小的、褐色圆形病斑；后期发展成大的具有同心环的圆形或不规则的深褐斑
叶斑病 Leaf spot	初期叶片上出现小型不规则或圆形褐色斑点，边缘清晰；后期病斑扩展，形成大的不规则坏死区域
健康 Health	叶片呈深绿色，叶形卵状，叶脉清晰可见，质地柔软薄嫩，表面光滑

病害名称 Disease types	病害特征 Disease characters
细菌性疫病 Bacterial blight	初期叶片表面出现小的水浸斑，变成淡黄色到棕褐色；后期病斑在叶片上扩展，病斑呈黑色
轮纹病 Ring spot	初期叶片产生小的、褐色圆形病斑；后期发展成大的具有同心环的圆形或不规则的深褐斑
叶斑病 Leaf spot	初期叶片上出现小型不规则或圆形褐色斑点，边缘清晰；后期病斑扩展，形成大的不规则坏死区域
健康 Health	叶片呈深绿色，叶形卵状，叶脉清晰可见，质地柔软薄嫩，表面光滑

模型	精确度 Precision（%）	召回率 Recall（%）	mAP@50（%）	mAP@50-95（%）	参数量 Params（MB）
YOLOv8n	85.5	81.1	86.2	62.8	5.73
YOLOv8-EMA-CARAFE-Layer	86.9	83.4	91.0	67.5	5.95

模型	精确度 Precision（%）	召回率 Recall（%）	mAP@50（%）	mAP@50-95（%）	参数量 Params（MB）
YOLOv8n	85.5	81.1	86.2	62.8	5.73
YOLOv8-EMA-CARAFE-Layer	86.9	83.4	91.0	67.5	5.95

模型 Model	精确度 Precision（%）	召回率 Recall（%）	mAP@50（%）	mAP@50-95（%）	参数量 Params （MB）
YOLOv8_CBAM	84.5	78.6	86.8	64.3	5.86
YOLOv8_CA	87.4	80.3	86.7	64.0	5.75
YOLOv8_EMA	83.1	81.2	87.4	64.0	5.73