Scholarlink

Red Rot, a severe disease in sugarcane caused by the fungal pathogen Colletotrichum falcatum, results in significant yield losses and deterioration in crop quality. Prompt and precise identification of disease severity plays a vital role in enabling timely interventions and ensuring effective management of crops. This study employs transfer learning for multi-class classification of Red Rot severity in sugarcane leaves into three distinct categories -normal, mild, and severe, by comparing the performance of three well-known deep learning architectures: VGG19, ResNet50, and InceptionV3, trained under uniform conditions. Among the analyzed models, VGG19 exhibited the greatest classification accuracy of 92.75%, followed by ResNet50 with 92.39% and InceptionV3 with 88.77%. These findings demonstrate the effectiveness of deep CNN-based transfer learning methods, particularly VGG19 and ResNet50, for accurate severity classification. A basic CNN was also evaluated as a lightweight alternative, though with lower accuracy, to highlight trade-offs between performance and deployment feasibility in real-world scenarios. The proposed approach holds significant potential for integration into real-time agricultural decision-support systems, offering a robust and scalable solution for early disease diagnosis and precision crop management in sugarcane cultivation.

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

Multi-class Classification of Red Rot Disease Severity in Sugarcane Using Transfer Learning Approaches

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ashwini Kanade,
  • Priyanka Paygude,
  • Rohini Jadhav

摘要

Red Rot, a severe disease in sugarcane caused by the fungal pathogen Colletotrichum falcatum, results in significant yield losses and deterioration in crop quality. Prompt and precise identification of disease severity plays a vital role in enabling timely interventions and ensuring effective management of crops. This study employs transfer learning for multi-class classification of Red Rot severity in sugarcane leaves into three distinct categories -normal, mild, and severe, by comparing the performance of three well-known deep learning architectures: VGG19, ResNet50, and InceptionV3, trained under uniform conditions. Among the analyzed models, VGG19 exhibited the greatest classification accuracy of 92.75%, followed by ResNet50 with 92.39% and InceptionV3 with 88.77%. These findings demonstrate the effectiveness of deep CNN-based transfer learning methods, particularly VGG19 and ResNet50, for accurate severity classification. A basic CNN was also evaluated as a lightweight alternative, though with lower accuracy, to highlight trade-offs between performance and deployment feasibility in real-world scenarios. The proposed approach holds significant potential for integration into real-time agricultural decision-support systems, offering a robust and scalable solution for early disease diagnosis and precision crop management in sugarcane cultivation.