Scholarlink

The increasing threat of plant disease threatens world crop production, causing huge economic and food security losses, particularly in crops like corn, potato, rice, wheat, and sugarcane. This paper presents the LEAFNET (Lightweight Efficient Agricultural Feature Network) model, a new deep learning model that integrates Swin Transformer, MobileNetV3-Small, and ResNet18-SE to deliver accurate and explainable disease classification in 13,324 images of 17 classes. LEAFNET employs state-of-the-art models, pretrained weights, and Grad-CAM visualizations to deliver robust performance: Swin Transformer attained the highest validation accuracy of 96.96% (training accuracy 96.91%, validation loss 0.1029), ResNet18-SE attained 94.63% (training 93.45%, loss 0.1544), and the lightweight MobileNetV3-Small attained 93.70% (training 93.62%, loss 0.1561), which is deployable in real time on resource-constrained devices. ResNet18-SE attention mechanisms offer interpretability, with visual disease localization to maximize user trust. This dual emphasis on accuracy and usability assists to significantly support early disease detection, offering a possible 20–30% reduction of crop yield losses, and making sustainable agriculture possible. LEAFNET, in total, offers farmers, agronomists, and policymakers a powerful, scalable, and societally impactful tool that drives precision farming and international food security.

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

LEAFNET Model for Early Crop Disease Detection: Integrating CNNs and Vision Approach in Precision Agriculture

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ritu Chauhan,
  • Mehak Jena,
  • Aarushi Mishra,
  • Dhananjay Singh

摘要

The increasing threat of plant disease threatens world crop production, causing huge economic and food security losses, particularly in crops like corn, potato, rice, wheat, and sugarcane. This paper presents the LEAFNET (Lightweight Efficient Agricultural Feature Network) model, a new deep learning model that integrates Swin Transformer, MobileNetV3-Small, and ResNet18-SE to deliver accurate and explainable disease classification in 13,324 images of 17 classes. LEAFNET employs state-of-the-art models, pretrained weights, and Grad-CAM visualizations to deliver robust performance: Swin Transformer attained the highest validation accuracy of 96.96% (training accuracy 96.91%, validation loss 0.1029), ResNet18-SE attained 94.63% (training 93.45%, loss 0.1544), and the lightweight MobileNetV3-Small attained 93.70% (training 93.62%, loss 0.1561), which is deployable in real time on resource-constrained devices. ResNet18-SE attention mechanisms offer interpretability, with visual disease localization to maximize user trust. This dual emphasis on accuracy and usability assists to significantly support early disease detection, offering a possible 20–30% reduction of crop yield losses, and making sustainable agriculture possible. LEAFNET, in total, offers farmers, agronomists, and policymakers a powerful, scalable, and societally impactful tool that drives precision farming and international food security.