Scholarlink

<p>In this study, deep learning-based object detection models were developed using various versions of the YOLOv5 architecture (YOLOv5s, m, l, x) to detect visual symptoms of Grapevine fanleaf virus (GFLV) on grapevine leaves, which is a pathogen responsible for significant economic losses in viticulture. A custom high-resolution image dataset was created with 399 images and was split into training (70%), validation (20%), and testing (10%) subsets. All models were trained for 100 epochs and evaluated based on key performance metrics such as Precision, Recall, and Mean Average Precision (mAP@0.5 and mAP@0.5–0.95). The comparative evaluation of YOLOv5 variants showed that performance metrics were distributed across models rather than concentrated in a single architecture. YOLOv5l achieved the highest precision (0.91), YOLOv5s demonstrated the highest recall (0.88) and F1 score (0.88), while YOLOv5x delivered the strongest localization performance with high mAP@0.5 (0.89) and mAP@0.5–0.95 (0.53) values. Although YOLOv5l and YOLOv5x produced slightly higher precision and localization scores, YOLOv5m showed a well-balanced performance with a precision of 0.86, recall of 0.85, F1 score of 0.86, and mAP@0.5 of 0.85, while requiring lower computational resources than the larger variants. Considering the balance between detection accuracy and computational efficiency, YOLOv5m appears to provide a practical and efficient solution for real-time vineyard disease monitoring applications.</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

Precision viticulture with AI: Comparing YOLOv5 models for grapevine fanleaf virus diagnosis

  • 期刊论文

  • 发表日期：

    2026年5月1日

查看全文
  • Bahadır Şin,
  • Lerzan Öztürk

摘要

In this study, deep learning-based object detection models were developed using various versions of the YOLOv5 architecture (YOLOv5s, m, l, x) to detect visual symptoms of Grapevine fanleaf virus (GFLV) on grapevine leaves, which is a pathogen responsible for significant economic losses in viticulture. A custom high-resolution image dataset was created with 399 images and was split into training (70%), validation (20%), and testing (10%) subsets. All models were trained for 100 epochs and evaluated based on key performance metrics such as Precision, Recall, and Mean Average Precision (mAP@0.5 and mAP@0.5–0.95). The comparative evaluation of YOLOv5 variants showed that performance metrics were distributed across models rather than concentrated in a single architecture. YOLOv5l achieved the highest precision (0.91), YOLOv5s demonstrated the highest recall (0.88) and F1 score (0.88), while YOLOv5x delivered the strongest localization performance with high mAP@0.5 (0.89) and mAP@0.5–0.95 (0.53) values. Although YOLOv5l and YOLOv5x produced slightly higher precision and localization scores, YOLOv5m showed a well-balanced performance with a precision of 0.86, recall of 0.85, F1 score of 0.86, and mAP@0.5 of 0.85, while requiring lower computational resources than the larger variants. Considering the balance between detection accuracy and computational efficiency, YOLOv5m appears to provide a practical and efficient solution for real-time vineyard disease monitoring applications.