Scholarlink

Sunflowers are a popular oil seed crop in Bangladesh, second only to soybeans. They have the capacity to generate high-quality edible oil, but because of their susceptibility to parasites, insects, and illnesses, crop yields may be reduced. To effectively tackle this problem, the implementation of digital image processing and computer vision techniques is necessary to identify diseases in sunflower leaves, fruits, and flowers. However, the specific sunflower leaf diseases found in Bangladesh have not been extensively studied. To fill the gap in knowledge, we created seven machine learning models, including four popular deep learning models (ResNet, VGG16, DenseNet, and Inception v3), and three advanced Vision Transformers (ViT, DeiT, and Swin). Among them, the ResNet and Inception v3 models showed good performance, but the Swin transformer surpassed them all with an accuracy score of 96.50%. According to a comparison of precision and recall as well as the F1 score, the Swin transformer outperforms all other models.

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

TTL: Transformer and Transfer Learning Approach to Detect Sunflower Disease

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Afsana Kabir Sinthia,
  • Nisat Islam Mozumder,
  • Md. Tariqul Islam Islam,
  • Md. Humaion Kabir Mehedi,
  • Annajiat Alim Rasel,
  • Munima Haque

摘要

Sunflowers are a popular oil seed crop in Bangladesh, second only to soybeans. They have the capacity to generate high-quality edible oil, but because of their susceptibility to parasites, insects, and illnesses, crop yields may be reduced. To effectively tackle this problem, the implementation of digital image processing and computer vision techniques is necessary to identify diseases in sunflower leaves, fruits, and flowers. However, the specific sunflower leaf diseases found in Bangladesh have not been extensively studied. To fill the gap in knowledge, we created seven machine learning models, including four popular deep learning models (ResNet, VGG16, DenseNet, and Inception v3), and three advanced Vision Transformers (ViT, DeiT, and Swin). Among them, the ResNet and Inception v3 models showed good performance, but the Swin transformer surpassed them all with an accuracy score of 96.50%. According to a comparison of precision and recall as well as the F1 score, the Swin transformer outperforms all other models.