Scholarlink

Rice plant diseases significantly impact agricultural productivity, making automated detection essential for effective crop management. This study explores deep learning-based approaches for rice disease classification using two publicly available Kaggle datasets. A comparative analysis is conducted between a custom CNN model and three transfer learning architectures: VGG16, ResNet50, and InceptionV3. Preprocessing and data augmentation techniques are applied to improve model generalization. Models are evaluated using classification accuracy, with InceptionV3 achieving the highest testing accuracy of 100% on Dataset 2 and 90.72% on Dataset 1. The custom CNN achieved 99.41 and 85.03%, while VGG16 followed with 99.3 and 81.81%. ResNet50 underperformed, with 62.44 and 40.72% accuracy. Despite promising results, challenges such as dataset limitations and the need for real-time implementation remain. Future work could focus on improving model robustness through additional training data, hyperparameter tuning, and deploying the models in real-world agricultural settings.

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

Harnessing Deep Learning for Efficient Rice Disease Detection with CNN and Advanced Transfer Learning Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • K. Ramani Teja Sri Deepa,
  • K. Ratna Kumari

摘要

Rice plant diseases significantly impact agricultural productivity, making automated detection essential for effective crop management. This study explores deep learning-based approaches for rice disease classification using two publicly available Kaggle datasets. A comparative analysis is conducted between a custom CNN model and three transfer learning architectures: VGG16, ResNet50, and InceptionV3. Preprocessing and data augmentation techniques are applied to improve model generalization. Models are evaluated using classification accuracy, with InceptionV3 achieving the highest testing accuracy of 100% on Dataset 2 and 90.72% on Dataset 1. The custom CNN achieved 99.41 and 85.03%, while VGG16 followed with 99.3 and 81.81%. ResNet50 underperformed, with 62.44 and 40.72% accuracy. Despite promising results, challenges such as dataset limitations and the need for real-time implementation remain. Future work could focus on improving model robustness through additional training data, hyperparameter tuning, and deploying the models in real-world agricultural settings.