Scholarlink

<p>This study compares the classification performance of different optimization algorithms using transformer-based deep learning models for the early diagnosis of olive fruit disease and pest damage (particularly pest damage caused by olive fly and fungal diseases). Experiments were conducted on a&#xa0;balanced and labeled field image dataset comprising 1644 olive samples collected from the Balıkesir-Bandırma/Edincik region. The dataset was tested using current transformer architectures, such as ViT, BEiT, DeiT, LeViT, MaxViT, and MobileViT, along with ten optimization algorithms including SGD, Adam, AdamW, RAdam, NAdam, and Adafactor. Model performance was evaluated using metrics such as accuracy, macro F1 score, precision, and recall. The results showed that the MobileViT and MaxViT models achieved the highest classification performance when combined with modern adaptive optimization algorithms such as NAdam and AdamW, reaching an accuracy rate of 99.60%. RMSprop, on the other hand, showed the lowest performance overall and failed in class separation. The findings indicate that not only the architecture but also the optimization strategy is decisive in classification performance. This study contributes to the development of artificial intelligence-supported systems that provide high accuracy, class balance, and learning stability for the automatic diagnosis of olive diseases, thereby establishing a&#xa0;scientific foundation for sustainable and digital agriculture applications.</p>

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

Impact of Optimizers on Transformer Models for Classification of Olive Fruit Disease

  • 期刊论文

  • 发表日期：

    2026年2月10日

查看全文
  • Cağla Toprak Erdurak,
  • Serhat Kiliçarslan

摘要

This study compares the classification performance of different optimization algorithms using transformer-based deep learning models for the early diagnosis of olive fruit disease and pest damage (particularly pest damage caused by olive fly and fungal diseases). Experiments were conducted on a balanced and labeled field image dataset comprising 1644 olive samples collected from the Balıkesir-Bandırma/Edincik region. The dataset was tested using current transformer architectures, such as ViT, BEiT, DeiT, LeViT, MaxViT, and MobileViT, along with ten optimization algorithms including SGD, Adam, AdamW, RAdam, NAdam, and Adafactor. Model performance was evaluated using metrics such as accuracy, macro F1 score, precision, and recall. The results showed that the MobileViT and MaxViT models achieved the highest classification performance when combined with modern adaptive optimization algorithms such as NAdam and AdamW, reaching an accuracy rate of 99.60%. RMSprop, on the other hand, showed the lowest performance overall and failed in class separation. The findings indicate that not only the architecture but also the optimization strategy is decisive in classification performance. This study contributes to the development of artificial intelligence-supported systems that provide high accuracy, class balance, and learning stability for the automatic diagnosis of olive diseases, thereby establishing a scientific foundation for sustainable and digital agriculture applications.