[1]唐秀英,孙中清,杨琳琳,等.基于改进YOLO v8n轻量化的番茄叶霉病发病程度分级检测[J].江苏农业学报,2025,(10):1985-1996.[doi:doi:10.3969/j.issn.1000-4440.2025.10.012]
　TANG Xiuying,SUN Zhongqing,YANG Linlin,et al.Grading detection for tomato leaf mold severity based on improved lightweight YOLO v8n[J].,2025,(10):1985-1996.[doi:doi:10.3969/j.issn.1000-4440.2025.10.012]
点击复制

基于改进YOLO v8n轻量化的番茄叶霉病发病程度分级检测()

江苏农业学报[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2025年10期
页码:
1985-1996
栏目:
农业信息工程
出版日期:
2025-10-31

文章信息/Info

Title:
Grading detection for tomato leaf mold severity based on improved lightweight YOLO v8n
作者:
唐秀英; 孙中清; 杨琳琳; 余静; 刘正林; 王佩; 施杰
(云南农业大学机电工程学院,云南昆明650201)
Author(s):
TANG Xiuying; SUN Zhongqing; YANG Linlin; YU Jing; LIU Zhenglin; WANG Pei; SHI Jie
(Faculty of Mechanical & Electrical Engineering, Yunnan Agricultural University, Kunming 650201, China)
关键词:
YOLO v8n模型; 番茄叶霉病; 发病程度分级
Keywords:
YOLO v8n model; tomato leaf mold; severity grading
分类号:
S126；TP391.4
DOI:
doi:10.3969/j.issn.1000-4440.2025.10.012
文献标志码:
A
摘要:
为进一步提高番茄叶霉病发病程度分级识别的精度和效率,降低检测模型的复杂度和权重,便于部署在移动端,本研究对YOLO v8n模型进行了改进,提出了一种轻量化病害发病程度分级检测方法。引入MobileNetV4中的UIB模块,替换YOLO v8n模型中的C2f卷积层,降低模型的计算量和参数量,满足移动端轻量化部署要求；在主干网络的最高维度后引入级联群体注意力机制模块（CGA),同时引入位置偏置，最后将模型的检测头由解耦检测头替换为双重注意力增强的目标检测头,实现对叶霉病症状特征的精确定位。研究结果表明,级联群体注意力机制模块（CGA)对模型性能的提升效果最为明显。相比YOLO v8n模型,YOLO v8n-UC-DAE模型的P、R、mAP50和mAP50-95分别提高了2.0个百分点、7.7个百分点、3.8个百分点和2.9个百分点；同时,计算量和权重分别降低了43.33%和32.86%。相较于其他主流模型,本研究构建的YOLO v8n-UC-DAE模型能够满足番茄叶霉病发病程度分级检测的需求,并解决了移动端部署的问题。
Abstract:
To further improve the accuracy and efficiency of tomato leaf mold severity grading recognition, reduce the complexity and weight of the detection model, and facilitate deployment on mobile devices, this study improved the YOLO v8n model and proposed a lightweight disease severity grading detection method. The UIB module from MobileNetV4 was introduced to replace the C2f convolutional layer in the YOLO v8n model, reducing computational load and the number of parameters to meet the requirements for lightweight deployment on mobile devices. The cascaded group attention (CGA) module was incorporated into the highest-dimensional layer of the backbone network, along with the introduction of positional bias. Additionally, the original decoupled detection head was replaced with a dual attention-enhanced detection head. These modifications enabled the model to achieve precise localization of leaf mold symptoms. The results demonstrated that the cascaded group attention (CGA) module yielded the most significant improvement in model performance. Compared to the YOLO v8n model, the YOLO v8n-UC-DAE model achieved increases of 2.0 percentage points in P, 7.7 percentage points in R, 3.8 percentage points in mAP50, and 2.9 percentage points in mAP50-95. Meanwhile, the computational load and the number of parameters were reduced by 43.33% and 32.86%, respectively. Compared with other mainstream models, the YOLO v8n-UC-DAE model developed in this study is capable of meeting the requirements for grading the severity of tomato leaf mold and has effectively addressed the challenges associated with mobile deployment.

参考文献/References:

[1]李江峰. 番茄叶霉病的诊断与综合防治技术[J]. 特种经济动植物,2023,26(7):119-121.
[2]张雪辉. 番茄叶霉病的发生与防治[J]. 现代园艺,2012(13):68.
[3]吴华瑞. 基于深度残差网络的番茄叶片病害识别方法[J]. 智慧农业,2019,1(4):42-49.
[4]王少波. 基于光谱信息和图像信息的番茄叶片病害识别[D]. 昆明:云南师范大学,2024.
[5]ZHANG D, HUANG Y, WU C, et al. Detecting tomato disease types and degrees using multi-branch and destruction learning[J]. Computers and Electronics in Agriculture,2023,213:108244.
[6]胡玲艳,周婷,刘艳,等. 基于轻量级网络自适应特征提取的番茄病害识别[J]. 江苏农业学报,2022,38(3):696-705.
[7]郭小清,范涛杰,舒欣. 基于改进Multi-Scale AlexNet的番茄叶部病害图像识别[J]. 农业工程学报,2019,35(13):162-169.
[8]储鑫,李祥,罗斌,等. 基于改进YOLOv4算法的番茄叶部病害识别方法[J]. 江苏农业学报,2023,39(5):1199-1208.
[9]李仁杰,宋涛,高婕,等. 基于改进YOLOv5的自然环境下番茄患病叶片检测模型[J]. 江苏农业学报,2024,40(6):1028-1037.
[10]周巧黎,马丽,曹丽英,等. 基于改进轻量级卷积神经网络MobileNetV3的番茄叶片病害识别[J]. 智慧农业,2022,4(1):47-56.
[11]孟凯鑫. 基于高光谱成像技术的番茄叶斑病与枯萎病早期检测方法研究[D]. 泰安:山东农业大学,2024.
[12]YAN Z, HUA-RUI W, HUA-JI Z. Hyperspectral latent period diagnosis of tomato Gray Mold based on TLBO-ELM model[J]. Spectroscopy and Spectral Analysis,2022,42(9):2969-2975.
[13]崔江南,付芸,赵森,等. 基于高光谱成像的番茄叶霉病的无损检测[J]. 长春理工大学学报(自然科学版),2022,45(4):65-71.
[14]周占坤. 基于THz-NIR高光谱相融合的番茄病虫害检测研究[D]. 镇江:江苏大学,2022.
[15]QIN D, LEICHNER C, DELAKIS M, et al. MobileNetV4:universal models for the mobile ecosystem[C]. Berlin:Springer,2024.
[16]LIU X, PENG H, ZHENG N,et al. EfficientViT: memory efficient vision transformer with cascaded group attention[EB/OL].
[2023-05-12]（2024-10-22). https://arxiv.org/abs/2305.07027.
[17]中华人民共和国农业农村部. 番茄主要病害抗病性鉴定技术规程:NY/T 1858.2-2010[S]. 北京:中国标准出版社,2010.
[18]李国花,王晓敏,赵宇飞,等. 180份番茄种质资源对3种主要叶部病害的田间抗病性评价[J]. 新疆农业科学,2021,58(4):741-755.
[19]XIE S, SUN H. Tea-YOLO v8s:a tea bud detection model based on deep learning and computer vision[J]. Sensors,2023,23(14):6576.
[20]LOU H, DUAN X, GUO J, et al. DC-YOLO v8: small-size object detection algorithm based on camera sensor[J]. Electronics,2023,12(10):2323.
[21]WU C, WU F, GE S, et al. Neural news recommendation with multi-head self-attention[C]. Hong Kong:Association for Computational Linguistics,2019.
[22]FU J, LIU J, TIAN H, et al. Dual attention network for scene segmentation[C]. Long Beach:IEEE,2019.
[23]WOO S, PARK J, LEE J Y, et al. CBAM:convolutional block attention module[C]. Munich:Springer,2018.
[24]SELVARAJU R R, COGSWELL M, DAS A, et al.Grad-CAM: visual explanations from deep networks via gradient-based localization[J]. International Journal of Computer Vision,2020,128(2):336-359.

备注/Memo

备注/Memo:
收稿日期：2024-12-23基金项目：云南省重大科技专项(202302AE090020)；云南省农业基础研究联合专项(202401BD070001-069)；云南省作物生产与智慧农业重点实验室开放课题作者简介：唐秀英(1981-),女,四川南充人,硕士,副教授,主要从事智能农业装备、植物保护喷雾减药增效研究。(E-mail)xitb05@126.com通讯作者：施杰,(E-mail)km_shijie@126.com
更新日期/Last Update: 2025-11-17