基于SE-MobileNetV2的葡萄叶部病害识别方法  

Grape Leaf Disease Recognition Method Based on SE-MobileNetV2

作　　者：程浈浈 张必详 程一帆 缪百灵 龚守富 CHENG Zhenzhen;ZHANG Bixiang;CHENG Yifan;MIAO Bailing;GONG Shoufu(College of Horticulture,Xinyang Agriculture and Forestry College,Xinyang,Henan 464399;College of Horticulture,Huazhong Agricultural University,Wuhan,Hubei 430070;College of Optoelectronic Information Engineering,Huazhong University of Science and Technology,Wuhan,Hubei 430074)

机构地区：[1]信阳农林学院园艺学院,河南信阳464399 [2]华中农业大学园艺学院,湖北武汉430070 [3]华中科技大学光电信息工程学院,湖北武汉430074

出　　处：《北方园艺》2025年第5期131-140,共10页Northern Horticulture

基　　金：河南省重点研发与推广专项(科技攻关)资助项目(232102111118)。

摘　　要：以PlantVillage公开数据集的4种葡萄叶部病害为试材,提出了一种基于改进MobileNetV2模型的轻量化识别方法,该方法以轻量级MobileNetV2模型为基础,通过在模型瓶颈层中引入SE注意力机制,增强模型对关键特征的关注能力,从而进一步优化识别性能和减少模型参数数量,以期为实现病害的高精度诊断,同时有效降低计算资源需求提供参考依据。结果表明:改进后的模型在测试集上的识别准确率达97.5%,较原始MobileNetV2提升4.5%;与ResNet50、ResNet34和ShuffleNetV2模型相比,平均准确率分别提高10.2、18.7、28.2个百分点,且模型大小仅为20.7 MB,实现了模型运行成本和精确度的平衡,为葡萄叶部病害识别问题提供了解决方案。Taking four grape leaf diseases in the open data set of Plant Village as the test materials,a lightweight recognition method was proposed based on an improved MobileNetV2 model,aiming to achieve high-accuracy disease diagnosis while effectively reducing computational resource requirements.The method utilizes the lightweight MobileNetV2 framework and integrates the SE attention mechanism into the bottleneck layers,enhancing the model′s ability to focus on key features and further optimizing recognition performance while reducing the number of model parameters,in order to provide reference for the realization of high precision disease diagnosis,while effectively reducing the need for computing resourees.The results showed that the improved model achieved a recognition accuracy of 97.5%on the test set,which was 4.5%higher than the original MobileNetV2.Compared to ResNet50,ResNet34,and ShuffleNetV2,the proposed method improved the average accuracy by 10.2,18.7,and 28.2 percentage points,respectively,with a model size of only 20.7 MB.This study successfully balances computational cost and recognition accuracy,providing an effective solution for the challenge of grape leaf disease recognition.

关 键 词：葡萄 叶部病害 注意力机制 图像分类 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术] S431.9[自动化与计算机技术—计算机科学与技术]