Scholarlink

<p>The tomato is one of the most important economic crop and food sources in the world. However, leaf diseases and pests have a major impact on yield and quality. To improve the identification accuracy of tomato leaf diseases, we proposed an integrated deep learning system that combines a residual network (ResNet), a squeeze and excitation network (SENet), and a capsule network (CapsNet) and called it RSCNet. First, we built two parallel and lightweight feature extraction branches to improve training efficiency and achieve lightweight training parameters. Then, we designed a compression module to retain more important feature information that enabled the previous network to fully learn the image information. Finally, the Euclidean distance was employed for completing the classification results of the categories corresponding to routing capsules. The experimental results demonstrate that our model has certain advantages over ShuffleNet-V2, MobileNet-V3, and Efficient-B7, including a quicker processing time, greater recognition accuracy, and more robust adaptability. The greatest test accuracy was 99.82%, while the total classification accuracy was 98.82%. The suggested technique offers some credible references for increasing informed detection of plant diseases and insect pests.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

RSCNet: an integrated deep learning classification model for tomato leaf diseases

  • 期刊论文

  • 发表日期：

    2025年7月31日

查看全文
  • Laixiang Xu,
  • Bei Li,
  • Jingfeng Su,
  • Yongfeng Fan,
  • Junmin Zhao

摘要

The tomato is one of the most important economic crop and food sources in the world. However, leaf diseases and pests have a major impact on yield and quality. To improve the identification accuracy of tomato leaf diseases, we proposed an integrated deep learning system that combines a residual network (ResNet), a squeeze and excitation network (SENet), and a capsule network (CapsNet) and called it RSCNet. First, we built two parallel and lightweight feature extraction branches to improve training efficiency and achieve lightweight training parameters. Then, we designed a compression module to retain more important feature information that enabled the previous network to fully learn the image information. Finally, the Euclidean distance was employed for completing the classification results of the categories corresponding to routing capsules. The experimental results demonstrate that our model has certain advantages over ShuffleNet-V2, MobileNet-V3, and Efficient-B7, including a quicker processing time, greater recognition accuracy, and more robust adaptability. The greatest test accuracy was 99.82%, while the total classification accuracy was 98.82%. The suggested technique offers some credible references for increasing informed detection of plant diseases and insect pests.