Scholarlink

It is essential to detect the plant leaf diseases to reduce the crop loss and to improve the productivity. This paper presents a multi-class leaf disease classification using an enhanced version of MobileNetV3Large model, referred as the MobileNetV3Shuffle. This model is specifically trained on four variety of plant leaves: corn, tea, mulberry, and wheat. These four crops comprise of 17 distinct classes. The base model MobileNetV3Large is modified with two additional components: a Channel Shuffle mechanism, inspired from ShuffleNet, for enhanced feature diversity and a Squeeze-and-Excitation (SE) block to refine the features. In addition to these components, LIME XAI techniques are incorporated to provide trust in the model decisions and improving the transparency. Trained on a dataset of 11,269 images, an overall accuracy of 98% was achieved by the proposed model and for most of the classes both precision and recall exceeded 97%. The model performed well when compared to the existing state-of-the-art methods by balancing accuracy and computational efficiency. It is more suitable for real-time disease detection due to lightweight architecture and can be deployed on mobile and edge devices. These findings state the MobileNetV3Shuffle as a robust solution for leaf disease detection across multiple plant species and contribute the timely and precise agricultural disease management.

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

MobileNetV3Shuffle for Leaf Disease Detection: Integrating LIME for Explainable AI in Multi-class Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Devaprakash Umapathy,
  • Chintalacheruvu Chaitanya Deep,
  • K. G. Jyotheeshwar,
  • S. Rohini

摘要

It is essential to detect the plant leaf diseases to reduce the crop loss and to improve the productivity. This paper presents a multi-class leaf disease classification using an enhanced version of MobileNetV3Large model, referred as the MobileNetV3Shuffle. This model is specifically trained on four variety of plant leaves: corn, tea, mulberry, and wheat. These four crops comprise of 17 distinct classes. The base model MobileNetV3Large is modified with two additional components: a Channel Shuffle mechanism, inspired from ShuffleNet, for enhanced feature diversity and a Squeeze-and-Excitation (SE) block to refine the features. In addition to these components, LIME XAI techniques are incorporated to provide trust in the model decisions and improving the transparency. Trained on a dataset of 11,269 images, an overall accuracy of 98% was achieved by the proposed model and for most of the classes both precision and recall exceeded 97%. The model performed well when compared to the existing state-of-the-art methods by balancing accuracy and computational efficiency. It is more suitable for real-time disease detection due to lightweight architecture and can be deployed on mobile and edge devices. These findings state the MobileNetV3Shuffle as a robust solution for leaf disease detection across multiple plant species and contribute the timely and precise agricultural disease management.