Scholarlink

Purpose <p>This study addresses the challenge of low detection accuracy for plant leaf diseases in complex farmland environments by developing an enhanced, real-time object detection algorithm. The primary research question is whether integrating dynamic convolution, bionic structural modeling, and global attention mechanisms can substantially improve small-spot detection and overall model performance.</p> Methods <p>We propose ADG-YOLO, a three-stage optimization framework built upon YOLOv11. First, an Adaptive Kernel Convolution (AKConv) module reconstructs the backbone network to reinforce multi-scale feature extraction, enabling accurate representation of both small lesions and larger disease areas. Second, we design a Dynamic Snake Convolution (DySnakeConv) architecture that synergistically captures local fine‐grained features and global morphological cues through bionic modeling, augmented via cross-layer connections for enhanced sensitivity to small targets. Third, we integrate a GAM global attention mechanism to strengthen discriminative feature representation while preserving computational efficiency, ensuring suitability for low-power embedded platforms.</p> Results <p>On a benchmark plant disease dataset, ADG-YOLO achieves a mean Average Precision (mAP) of 95.8%, representing a 3.9% improvement over the baseline YOLOv11 model. The single-class mAP for small targets increases by 4.3%, indicating notable gains in fine-grained lesion detection. The model maintains an overall computational complexity of 6.3 GFLOPs, enabling real-time inference on resource-constrained edge platforms.</p> Conclusion <p>The ADG-YOLO algorithm delivers high-accuracy, multi-scale leaf disease detection in challenging farmland scenarios. Its compact architecture and reduced computational demand make it well-suited for deployment on UAVs, portable agricultural diagnostic tools, and other edge devices, and offer reliable technical support for the development of intelligent plant-protection systems in precision agriculture.</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

Research on Plant Leaf Disease Detection Method Based on Improved YOLOv11

  • 期刊论文

  • 发表日期：

    2025年9月24日

查看全文
  • Yan Zhang,
  • Weizhong Cheng,
  • Yinghui Wang,
  • Zhensheng Lu,
  • Miao Cheng

摘要

Purpose

This study addresses the challenge of low detection accuracy for plant leaf diseases in complex farmland environments by developing an enhanced, real-time object detection algorithm. The primary research question is whether integrating dynamic convolution, bionic structural modeling, and global attention mechanisms can substantially improve small-spot detection and overall model performance.

Methods

We propose ADG-YOLO, a three-stage optimization framework built upon YOLOv11. First, an Adaptive Kernel Convolution (AKConv) module reconstructs the backbone network to reinforce multi-scale feature extraction, enabling accurate representation of both small lesions and larger disease areas. Second, we design a Dynamic Snake Convolution (DySnakeConv) architecture that synergistically captures local fine‐grained features and global morphological cues through bionic modeling, augmented via cross-layer connections for enhanced sensitivity to small targets. Third, we integrate a GAM global attention mechanism to strengthen discriminative feature representation while preserving computational efficiency, ensuring suitability for low-power embedded platforms.

Results

On a benchmark plant disease dataset, ADG-YOLO achieves a mean Average Precision (mAP) of 95.8%, representing a 3.9% improvement over the baseline YOLOv11 model. The single-class mAP for small targets increases by 4.3%, indicating notable gains in fine-grained lesion detection. The model maintains an overall computational complexity of 6.3 GFLOPs, enabling real-time inference on resource-constrained edge platforms.

Conclusion

The ADG-YOLO algorithm delivers high-accuracy, multi-scale leaf disease detection in challenging farmland scenarios. Its compact architecture and reduced computational demand make it well-suited for deployment on UAVs, portable agricultural diagnostic tools, and other edge devices, and offer reliable technical support for the development of intelligent plant-protection systems in precision agriculture.