Scholarlink

Existing cotton pests and diseases detection algorithms face challenges in adapting to variations in target sizes and complex scenarios, with inherent trade-offs between detection accuracy and speed during algorithmic optimization. To address these limitations, we propose GCDN-YOLOv11, an enhanced algorithm based on YOLOv11, which introduces three critical innovations: Global Guided Contextual Attention (GGCA) strengthens feature representation of key pest characteristics while improving adaptability to dynamic and complex environments, thereby enhancing detection precision; DCNv4 module enables precise spatial deformation estimation of pest targets, boosting the net- work’s accuracy and responsiveness to critical pest features; and ADown module enhances multi-scale and subtle feature recognition while significantly reducing computational complexity. Evaluated on the CottonInsect dataset, our GCDN model achieves a remarkable 97.1% mAP (3.4% improvement) and 2.6% higher accuracy than mainstream alternatives, while compressing parameters to 2.2 MB and reducing inference time to 1.0 ms, about 30% speed enhancement over the original model.

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

GCDN: A Novel YOLOv11-Based Approach for Cotton Pest and Disease Detection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xinkang Li,
  • Liejun Wang,
  • Shaochen Jiang

摘要

Existing cotton pests and diseases detection algorithms face challenges in adapting to variations in target sizes and complex scenarios, with inherent trade-offs between detection accuracy and speed during algorithmic optimization. To address these limitations, we propose GCDN-YOLOv11, an enhanced algorithm based on YOLOv11, which introduces three critical innovations: Global Guided Contextual Attention (GGCA) strengthens feature representation of key pest characteristics while improving adaptability to dynamic and complex environments, thereby enhancing detection precision; DCNv4 module enables precise spatial deformation estimation of pest targets, boosting the net- work’s accuracy and responsiveness to critical pest features; and ADown module enhances multi-scale and subtle feature recognition while significantly reducing computational complexity. Evaluated on the CottonInsect dataset, our GCDN model achieves a remarkable 97.1% mAP (3.4% improvement) and 2.6% higher accuracy than mainstream alternatives, while compressing parameters to 2.2 MB and reducing inference time to 1.0 ms, about 30% speed enhancement over the original model.