湖北农业科学 ›› 2026, Vol. 65 ›› Issue (11): 189-193.doi: 10.14088/j.cnki.issn0439-8114.2026.11.032

• 信息工程 • 上一篇    下一篇

基于改进VGG16(L-VGG16)的农作物病虫害智能识别系统研究

孙小广1, 万若楠1, 余光正2   

  1. 1.广州城市理工学院电子信息工程学院,广州 510800;
    2.华南理工大学物理与光电学院,广州 510640
  • 收稿日期:2026-06-01 发布日期:2026-11-17
  • 作者简介:孙小广(1983-),男,湖南武冈人,讲师,硕士,主要从事信息处理技术与超声技术研究,(电话)15989112452(电子信箱)86451264@qq.com
  • 基金资助:
    国家自然科学基金资助项目(12474465); 广东省教育科学规划项目(2024GXJK521); 广州城市理工学院科研项目(53-K0223007)

Research on an intelligent identification system for crop diseases and pests based on improved VGG16 (L-VGG16)

SUN Xiao-guang1, WAN Ruo-nan1, YU Guang-zheng2   

  1. 1. School of Electronic and Information Engineering, Guangzhou City University of Technology, Guangzhou 510800, China;
    2. School of Physics and Optoelectronics, South China University of Technology, Guangzhou 510640, China
  • Received:2026-06-01 Online:2026-11-17

摘要: 为了克服传统农作物病虫害识别方法耗时费力、主观性强且易误判的局限,提升识别的准确率,基于深度学习构建了一种智能化识别系统。首先,以PlantVillage数据集为基础,通过几何变换、颜色变换以及标准化与归一化处理,完成数据增强与预处理,以增强模型泛化能力。其次,对比LeNet、AlexNet、VGG16、GoogLeNet、ResNet 5种经典卷积神经网络模型在数据集上的表现,选择识别准确率最高的VGG16作为基线模型;针对ReLU激活函数易导致“神经元死亡”的缺陷,将其替换为LeakyReLU激活函数,构建L-VGG16模型。在MATLAB R2022a平台上进行训练与测试,结果显示,L-VGG16模型的识别准确率明显高于VGG16模型,达94.05%。系统仿真输出表明,该模型可同步识别作物种类、病害及严重程度,具备良好的多属性综合识别能力。

关键词: 改进VGG16, L-VGG16, 农作物, 病虫害, PlantVillage数据集, 卷积神经网络, 智能识别系统

Abstract: To overcome the limitations of traditional crop disease and pest identification methods, which were time-consuming, labor-intensive, highly subjective and prone to misjudgment, and to improve identification accuracy, an intelligent identification system was constructed based on deep learning. First, based on the PlantVillage dataset, data augmentation and preprocessing were completed through geometric transformation, color transformation, standardization and normalization, so as to enhance the generalization ability of the model.Second, the performance of five classic convolutional neural network models, namely LeNet, AlexNet, VGG16, GoogLeNet and ResNet, on the dataset was compared, and VGG16, which had the highest identification accuracy, was selected as the baseline model; in view of the defect that the ReLU activation function easily caused “neuron death”, it was replaced by the LeakyReLU activation function, and the L-VGG16 model was constructed. Training and testing were conducted on the MATLAB R2022a platform, and the results showed that the identification accuracy of the L-VGG16 model was significantly higher than that of the VGG16 model, reaching 94.05%. The system simulation output indicated that the model could simultaneously identify crop types, diseases and severity levels, and possessed good multi-attribute comprehensive identification ability.

Key words: improved VGG16, L-VGG16, crop, diseases and pests, PlantVillage dataset, convolutional neural network, intelligent identification system

中图分类号: