Scholarlink

<p>Early detection of citrus Huanglongbing (HLB) is crucial for effective disease control and enhanced citrus cultivation efficiency. To address the limitations of traditional manual inspection methods, such as low efficiency and subjectivity, this study proposes a&#xa0;real-time detection model for citrus HLB based on an improved YOLOv8n, referred to as HLB-YOLO. First, this model introduces an adaptive rotated convolution (ARConv) module to replace standard convolutions, significantly enhancing the model’s rotational invariance for multiangle lesion detection and improving sensitivity to subtle pathological features. Second, the study optimizes the feature fusion mechanism in the detection head using a&#xa0;dynamic head structure, substantially boosting the model’s discriminative performance in complex field environments. Finally, the incorporation of an extended intersection over union (EIoU) loss function further refines the accuracy of the bounding-box regression. The experimental results demonstrate that the proposed method achieves an&#xa0;mAP50 of 86.6% on the test dataset while reducing computational complexity by 19.5%. Compared to the original YOLOv8n model, our improved model exhibits significant performance advantages, with improvements of 2.7% and 2.5% in mAP<sub>50</sub> and mAP<sub>50–95</sub> metrics, respectively. This study successfully realizes efficient automated detection of early citrus HLB symptoms, providing a&#xa0;reliable technical solution for intelligent disease monitoring and precise prevention in smart agriculture. The proposed model holds substantial practical value for promoting the development of agricultural intelligence.</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

HLB-YOLO: A Real-Time Citrus Huanglongbing Detection Model with Dynamic Head

  • 期刊论文

  • 发表日期：

    2025年10月22日

查看全文
  • Jian Chen,
  • Guowu Yuan

摘要

Early detection of citrus Huanglongbing (HLB) is crucial for effective disease control and enhanced citrus cultivation efficiency. To address the limitations of traditional manual inspection methods, such as low efficiency and subjectivity, this study proposes a real-time detection model for citrus HLB based on an improved YOLOv8n, referred to as HLB-YOLO. First, this model introduces an adaptive rotated convolution (ARConv) module to replace standard convolutions, significantly enhancing the model’s rotational invariance for multiangle lesion detection and improving sensitivity to subtle pathological features. Second, the study optimizes the feature fusion mechanism in the detection head using a dynamic head structure, substantially boosting the model’s discriminative performance in complex field environments. Finally, the incorporation of an extended intersection over union (EIoU) loss function further refines the accuracy of the bounding-box regression. The experimental results demonstrate that the proposed method achieves an mAP50 of 86.6% on the test dataset while reducing computational complexity by 19.5%. Compared to the original YOLOv8n model, our improved model exhibits significant performance advantages, with improvements of 2.7% and 2.5% in mAP50 and mAP50–95 metrics, respectively. This study successfully realizes efficient automated detection of early citrus HLB symptoms, providing a reliable technical solution for intelligent disease monitoring and precise prevention in smart agriculture. The proposed model holds substantial practical value for promoting the development of agricultural intelligence.