Scholarlink

<p>Due to wheat fungal leaf diseases (WD), the global rate of wheat production is decreasing by 3.6% annually. The WD decreases the grain yield production rate which affects the agriculture industry worldwide. Thus, identifying wheat diseases is essential for early intervention and effective management to maintain crop health. This paper presents a robust Class attention image transformer based You look only once with version 9 (CaiT-YOLOv9) model that recognizes and localizes the seven different wheat fungal leaf diseases including leaf rust, stripe rust, stem rust, powdery mildew, septoria leaf blotch, tan spot, and fusarium leaf blotch diseases. The CaiT-YOLOv9 model finds the disease symptoms by classifying and localizing each leaf disease by visually similar symptoms in densely distributed lesions. Firstly, the CaiT-YOLOv9 uses Convolutional neural networks (CNN) as a backbone for feature extraction but the CNN model doesn’t extract global features in distributed leaf lesions and spots. For extraction of global features in distributed leaf lesions, a new multiclass head class attention mechanism is added to the CaiT transformer. A total of 40,330 wheat leaf images were collected from primary as well as secondary sources and labeled for lesion localization. Experimental results show that the CaiT-YOLOv9 achieves a mean average precision (mAP) of 94.51%, outperforming RetinaNet, Faster Region-based CNN (Faster-RCNN), Mask Region-based CNN (Mask-RCNN), and other You look only once (YOLO) versions in disease classification and localization. This advanced class attention mechanism improves detection accuracy and offers a reliable solution for managing wheat leaf fungal diseases.</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

CaiT-YOLOv9: hybrid transformer model for wheat leaf fungal head prediction and diseases classification

  • 期刊论文

  • 发表日期：

    2025年2月13日

查看全文
  • Deepak Kumar,
  • Vinay Kukreja

摘要

Due to wheat fungal leaf diseases (WD), the global rate of wheat production is decreasing by 3.6% annually. The WD decreases the grain yield production rate which affects the agriculture industry worldwide. Thus, identifying wheat diseases is essential for early intervention and effective management to maintain crop health. This paper presents a robust Class attention image transformer based You look only once with version 9 (CaiT-YOLOv9) model that recognizes and localizes the seven different wheat fungal leaf diseases including leaf rust, stripe rust, stem rust, powdery mildew, septoria leaf blotch, tan spot, and fusarium leaf blotch diseases. The CaiT-YOLOv9 model finds the disease symptoms by classifying and localizing each leaf disease by visually similar symptoms in densely distributed lesions. Firstly, the CaiT-YOLOv9 uses Convolutional neural networks (CNN) as a backbone for feature extraction but the CNN model doesn’t extract global features in distributed leaf lesions and spots. For extraction of global features in distributed leaf lesions, a new multiclass head class attention mechanism is added to the CaiT transformer. A total of 40,330 wheat leaf images were collected from primary as well as secondary sources and labeled for lesion localization. Experimental results show that the CaiT-YOLOv9 achieves a mean average precision (mAP) of 94.51%, outperforming RetinaNet, Faster Region-based CNN (Faster-RCNN), Mask Region-based CNN (Mask-RCNN), and other You look only once (YOLO) versions in disease classification and localization. This advanced class attention mechanism improves detection accuracy and offers a reliable solution for managing wheat leaf fungal diseases.