基于改进YOLOX_Nano的番茄叶片病害识别研究  

Research on Tomato Leaf Disease Recognition Based on Improved YOLOX-Nano

作　　者：方晓捷 严李强 张福豪 宋沛琳 FANG Xiao-jie;YAN Li-qiang;ZHANG Fu-hao(School of Information Science and Technology,Tibet University,Lhasa,Xizang 850000)

机构地区：[1]西藏大学信息科学技术学院,西藏拉萨850000

出　　处：《安徽农业科学》2025年第6期238-242,共5页Journal of Anhui Agricultural Sciences

基　　金：2021年中央引导地方科技发展资金项目(XZ202101YD-0014C);西藏大学研究生“高水平人才培养计划”项目(2022-GSP-S106)。

摘　　要：近年来,番茄遭受的病害种类越来越多,这些病害对番茄产量和果实品质产生巨大影响,及时高效识别病害并采取有效措施成为当前番茄生产的迫切需求。针对现有模型番茄病害识别率较低以及模型较大较复杂的问题,提出一种基于改进YOLOX_Nano的病害识别模型。通过引入全局注意力机制以增强特征图的全局信息捕捉能力,改进特征金字塔网络中的上采样模块和路径聚合网络中的下采样模块,以提升特征的表达能力和融合效果。试验结果表明,该方法对番茄叶片病害识别的mAP达到89.16%。优化后的模型不仅在番茄叶片病害识别上表现出高准确率和快速检测性能,而且参数量和计算量较少,便于部署于手机等移动设备。该方法可为番茄叶片病害轻量化快速高效识别提供参考。In recent years,tomatoes have suffered from an increasing number of diseases,which have a huge impact on tomato yield and fruit quality.Timely and efficient identification of diseases and taking effective measures have become an urgent need for tomato production.In response to the low recognition rate of tomato diseases in existing models and the problem of large and complex models,this paper proposes a disease recognition model based on improved YOLOX-Nano.By introducing a global attention mechanism to enhance the global information capture ability of feature maps,improving the upsampling module in the feature pyramid network and the downsampling module in the path aggregation network,the expression ability and fusion effect of features can be improved.The experimental results showed that the mAP of this method for identifying tomato leaf diseases reached 89.16%.The optimized model not only exhibits high accuracy and fast detection performance in tomato leaf disease recognition,but also has fewer parameters and calculations,making it easy to deploy on mobile devices such as smartphones.This method can provide a reference for lightweight,fast,and efficient identification of tomato leaf diseases.

关 键 词：YOLOX_Nano网络 GAM注意力机制 番茄病害识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]