Scholarlink

<p>As a highly valuable agricultural product, soybean plants experience major leaf problems that decrease output while burdening farmers with significant financial losses. Our research introduces a combined model that merges Swin Transformer (ST) and ResNet50 for extracting features and applies Random Forest (RF) with bagging and boosting updates for accurate disease prediction. The Swin Transformer (ST) detects how different parts of images share information across distances to create long-range views while ResNet50 reads fine image details to give complete information about plant diseases. The system trains on PlantVillage soybean leaf disease dataset and tests it with multiple image alterations to achieve better results across all conditions. Our model brings new performance heights in this field by producing 99.12% accuracy, 99.34% precision, 99.12% recall, and 99.23% F1-score in experimental results. Our proposal stands out because it links multiple attention mechanisms together with deep feature merging and team-based selection in order to build an effective and expandable system for agricultural disease detection.</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

ST-ResNet50-RF: a hybrid model for robust classification of soybean leaf disease using Swin transformer, ResNet50 and random forest

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • Anjali Chaudhary,
  • Neeraj Raheja

摘要

As a highly valuable agricultural product, soybean plants experience major leaf problems that decrease output while burdening farmers with significant financial losses. Our research introduces a combined model that merges Swin Transformer (ST) and ResNet50 for extracting features and applies Random Forest (RF) with bagging and boosting updates for accurate disease prediction. The Swin Transformer (ST) detects how different parts of images share information across distances to create long-range views while ResNet50 reads fine image details to give complete information about plant diseases. The system trains on PlantVillage soybean leaf disease dataset and tests it with multiple image alterations to achieve better results across all conditions. Our model brings new performance heights in this field by producing 99.12% accuracy, 99.34% precision, 99.12% recall, and 99.23% F1-score in experimental results. Our proposal stands out because it links multiple attention mechanisms together with deep feature merging and team-based selection in order to build an effective and expandable system for agricultural disease detection.