Scholarlink

<p>In this study, real-time transformer-based object detection (RF-DETR) is utilized for the detection of grapevine fanleaf virus (GFLV), a&#xa0;serious viral disease that significantly reduces grape yield and quality. A&#xa0;dataset of high-resolution grapevine leaf images showing typical GFLV symptoms was used to train four RF-DETR versions (Nano, Small, Medium, and Base). Their performance was evaluated using standard object detection metrics. All four RF-DETR models demonstrated high detection performance for GFLV symptoms, with mean average precision (mAP)@50 scores ranging from 91.4% to 94.1%. The Small and Base variants reached the highest mAP values (94.1%), the Medium model achieved the highest recall (93.5%), and the Small model obtained the highest precision (92.1%). Although the Nano model showed the lowest recall, it still maintained high precision. Overall, the Medium and Base models exhibited the most balanced detection performance. These findings demonstrate that RF-DETR can reliably detect GFLV symptoms even under changing light conditions, varied backgrounds and natural differences in leaf shape. As a&#xa0;result, RF-DETR is confirmed as a&#xa0;powerful and practical tool for early disease detection in vineyards. It reduces the need for manual field inspections and supports faster, data-based decisions in precision viticulture.</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

Real-Time Transformer-Based Object Detection for Advanced Plant Disease Recognition: The Grapevine Fanleaf Virus Example

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • Lerzan Öztürk,
  • Bahadır Şin

摘要

In this study, real-time transformer-based object detection (RF-DETR) is utilized for the detection of grapevine fanleaf virus (GFLV), a serious viral disease that significantly reduces grape yield and quality. A dataset of high-resolution grapevine leaf images showing typical GFLV symptoms was used to train four RF-DETR versions (Nano, Small, Medium, and Base). Their performance was evaluated using standard object detection metrics. All four RF-DETR models demonstrated high detection performance for GFLV symptoms, with mean average precision (mAP)@50 scores ranging from 91.4% to 94.1%. The Small and Base variants reached the highest mAP values (94.1%), the Medium model achieved the highest recall (93.5%), and the Small model obtained the highest precision (92.1%). Although the Nano model showed the lowest recall, it still maintained high precision. Overall, the Medium and Base models exhibited the most balanced detection performance. These findings demonstrate that RF-DETR can reliably detect GFLV symptoms even under changing light conditions, varied backgrounds and natural differences in leaf shape. As a result, RF-DETR is confirmed as a powerful and practical tool for early disease detection in vineyards. It reduces the need for manual field inspections and supports faster, data-based decisions in precision viticulture.