Scholarlink

Detecting tomato ripeness with accuracy and speed plays a key role in improving harvest timing, limiting crop loss, and maintaining product standards. This study introduces a compact, interpretable model built on the YOLOv8n architecture to classify tomatoes into full ripe, half ripe, and green stages. Training was conducted on a custom-labeled dataset of 804 images, annotated using bounding boxes in CVAT. The model achieved a Precision of 82.68%, Recall of 78.94%, and a mean Average Precision (mAP@0,5) of 85.86%, reflecting consistent detection across ripeness levels. Beyond numerical evaluation, the integration of Gaussian heatmaps offers visual clarity by highlighting regions the model uses for prediction. This added layer of interpretation helps growers and field experts understand and trust the system’s decisions. The framework is suitable for practical use in environments such as greenhouses, sorting lines, or robotic harvesting systems, where both accuracy and explainability are essential.

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

Deep Learning for Tomato Ripeness Detection: A YOLO-Based Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Rameez Ahsen,
  • Pierpaolo Di Bitonto,
  • Michele Magarelli,
  • Donato Romano,
  • Pierfrancesco Novielli,
  • Sabina Tangaro

摘要

Detecting tomato ripeness with accuracy and speed plays a key role in improving harvest timing, limiting crop loss, and maintaining product standards. This study introduces a compact, interpretable model built on the YOLOv8n architecture to classify tomatoes into full ripe, half ripe, and green stages. Training was conducted on a custom-labeled dataset of 804 images, annotated using bounding boxes in CVAT. The model achieved a Precision of 82.68%, Recall of 78.94%, and a mean Average Precision (mAP@0,5) of 85.86%, reflecting consistent detection across ripeness levels. Beyond numerical evaluation, the integration of Gaussian heatmaps offers visual clarity by highlighting regions the model uses for prediction. This added layer of interpretation helps growers and field experts understand and trust the system’s decisions. The framework is suitable for practical use in environments such as greenhouses, sorting lines, or robotic harvesting systems, where both accuracy and explainability are essential.