基于改进YOLOv10n的轻量化番茄叶片病虫害检测方法  

Lightweight Tomato Leaf Disease and Pest Detection Method Based on Improved YOLOv10n

作　　者：吴六爱[1] 许雪珂 WU Liuai;XU Xueke(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学电子与信息工程学院,甘肃兰州730070

出　　处：《智慧农业(中英文)》2025年第1期146-155,共10页Smart Agriculture

基　　金：国家自然科学基金项目(51567014);甘肃省科技计划项目(22JR5RA797)。

摘　　要：[目的/意义]为解决番茄叶片病虫害检测中面临的环境复杂、目标小、精度低、参数冗余及计算复杂度高等问题,提出了一种新型轻量化、高精度、实时的检测模型——YOLOv10n-YS (You Only Look Once Version10-YS)。[方法]首先,采用C2f_RepViTBlock模块替换主干网络的C2f,减少了模型的计算量和参数量。其次,加入带切片操作的注意力机制SimAM,结合原有卷积形成Conv_SWS模块,提升了小目标的特征提取能力。另外,在颈部网络中使用DySample轻量动态上采样模块,使采样点集中在目标区域而不会关注背景部分,实现病虫害的有效识别。最后,将跨通道交互的高效率通道注意力(Efficient Channel Attention with Cross-Channel Interaction,EMCA)替换主干网络的金字塔空间注意力机制(Pyramid Spatial Attention, PSA),进一步提高了主干网络的特征提取能力。[结果与讨论]实验结果显示,YOLOv10n-YS模型在番茄病虫害数据集上展现出了卓越的性能。其平均识别精度、检测准确率和召回率分别达到了92.1%、89.2%和82.1%,相较于原模型,这些指标分别提升了3.8、3.3和4.2个百分点。同时,模型在参数量和计算量上也实现了显著的优化,分别减少了13.8%和8.5%。[结论]这些改进不仅提升了模型的性能,还保持了其轻量化特性,对番茄叶片病虫害的检测具有重要参考价值。[Objective]To address the challenges in detecting tomato leaf diseases and pests,such as complex environments,small goals,low precision,redundant parameters,and high computational complexity,a novel lightweight,high-precision,real-time detection model was proposed called YOLOv10n-YS.This model aims to accurately identify diseases and pests,thereby providing a solid scientific basis for their prevention and management strategies.[Methods]The dataset was collected using mobile phones to capture images from multiple angles under natural conditions,ensuring complete and clear leaf images.It included various weather conditions and covered nine types:Early blight,leaf mold,mosaic virus,septoria,spider mites damage,yellow leaf curl virus,late blight,leaf miner disease,and healthy leaves,with all images having a resolution of 640×640 pixels.In the proposed YOLOv10n-YS model,firstly,the C2f in the backbone network was replaced with C2f_RepViTBlock,thereby reducing the computational load and parameter volume and achieving a lightweight design.Secondly,through the introduction of a sliced operation SimAM attention mechanism,the Conv_SWS module was formed,which enhanced the extraction of small target features.Additionally,the DySample lightweight dynamic up sampling module was used to replace the up sampling module in the neck network,concentrating sampling points on target areas and ignoring backgrounds,thereby effectively identifying defects.Finally,the efficient channel attention(ECA)was improved by performing average pooling and max pooling on the input layer to aggregate features and then adding them together,which further enhanced global perspective information and features of different scales.The improved module,known as efficient channel attention with cross-channel interaction(EMCA)attention,was introduced,and the pyramid spatial attention(PSA)in the backbone network was replaced with the EMCA attention mechanism,thereby enhancing the feature extraction capability of the backbone network.[Results and Discussions]Af

关 键 词：番茄叶片 病虫害检测 YOLOv10n 注意力机制 轻量化 

分 类 号：S433[农业科学—农业昆虫与害虫防治] TP183[农业科学—植物保护] TP391.41[自动化与计算机技术—控制理论与控制工程]