Scholarlink

<p>Pomegranate diseases, especially scab, dry rot, and anthracnose, threaten high-quality production and cause severe yield losses. In dynamic orchard environments, leaf–branch occlusion, similar lesions, and complex backgrounds hinder accurate disease detection. To address these challenges, an improved GAC-YOLO disease detection method based on a&#xa0;self-developed inspection platform was proposed in this study. First, to mitigate information loss caused by shadow occlusion, the Gated Multi-Scale Convolution (GMSConv) module was introduced into the YOLOv10 backbone to enhance the model’s multi-scale feature extraction and reconstruction capabilities. Second, the Attention-based Intra-scale Feature Interaction-Position-Adaptive Encoding (AIFI-PAE) module replaced the original Spatial Pyramid Pooling–Fast (SPPF) and Position-Sensitive Attention (PSA) structures, strengthening the representation of key lesion areas and effectively suppressing environmental noise. Finally, the Cross-Scale Feature Fusion Module (CCFM) was integrated into the neck structure to improve the efficiency of combining disease information at different feature levels. Experiments showed that on the self-constructed dataset, GAC-YOLO achieved a&#xa0;precision of 94.0%, recall of 90.3%, and F1 Score of 92.1%, representing improvements of 2.2%, 4.2%, and 2.2%, respectively, over the baseline YOLOv10 model. To further verify the practical applicability of the proposed method in real-world scenarios, GAC-YOLO was deployed on an orchard inspection robot. The deployed system demonstrates stable disease detection capability and reliable data transmission in unstructured orchard environments. These results indicate that the proposed method effectively addresses the challenges of accurate disease fruit identification in mobile inspection scenarios, providing strong technical support for unmanned and precision disease management in smart orchards.</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

GAC-YOLO: A Pomegranate Disease Rapid Detection Method with Multi-Scale Feature Enhancement for an Orchard Inspection Robot

  • 期刊论文

  • 发表日期：

    2026年7月23日

查看全文
  • Shenghui Fu,
  • Yongchang Deng,
  • Shuangxi Liu,
  • Rongwei Li,
  • Yuefeng Du,
  • Linlin Sun,
  • Hongjian Zhang,
  • Tao Xu,
  • Wen Zhang

摘要

Pomegranate diseases, especially scab, dry rot, and anthracnose, threaten high-quality production and cause severe yield losses. In dynamic orchard environments, leaf–branch occlusion, similar lesions, and complex backgrounds hinder accurate disease detection. To address these challenges, an improved GAC-YOLO disease detection method based on a self-developed inspection platform was proposed in this study. First, to mitigate information loss caused by shadow occlusion, the Gated Multi-Scale Convolution (GMSConv) module was introduced into the YOLOv10 backbone to enhance the model’s multi-scale feature extraction and reconstruction capabilities. Second, the Attention-based Intra-scale Feature Interaction-Position-Adaptive Encoding (AIFI-PAE) module replaced the original Spatial Pyramid Pooling–Fast (SPPF) and Position-Sensitive Attention (PSA) structures, strengthening the representation of key lesion areas and effectively suppressing environmental noise. Finally, the Cross-Scale Feature Fusion Module (CCFM) was integrated into the neck structure to improve the efficiency of combining disease information at different feature levels. Experiments showed that on the self-constructed dataset, GAC-YOLO achieved a precision of 94.0%, recall of 90.3%, and F1 Score of 92.1%, representing improvements of 2.2%, 4.2%, and 2.2%, respectively, over the baseline YOLOv10 model. To further verify the practical applicability of the proposed method in real-world scenarios, GAC-YOLO was deployed on an orchard inspection robot. The deployed system demonstrates stable disease detection capability and reliable data transmission in unstructured orchard environments. These results indicate that the proposed method effectively addresses the challenges of accurate disease fruit identification in mobile inspection scenarios, providing strong technical support for unmanned and precision disease management in smart orchards.