Scholarlink

<p>This study aims to evaluate the effectiveness of deep learning techniques in predicting grape number and weight. Specifically, grape detection and weight estimation were performed using machine vision processes with the object recognition algorithm You Only Look Once Version&#xa0;8 (YOLOv8). Tests performed with different model sizes (nano, small, medium, large, xlarge) showed that the YOLOv8L model had the highest accuracy rate. The dataset was augmented with 120 photos taken in the field and 540 images obtained from the internet, divided into 80% for training and 20% for testing. The data augmentation techniques applied improved the generalization ability of the model and minimized the problem of overfitting. The results obtained show that the YOLOv8L model performs better than other models on metrics such as mAP50-95, precision and recall. In addition, the model developed for grape weight prediction has the potential to be used in agriculture for object recognition and weight estimation. This study aims to support sustainable production processes by providing data-driven decision-making mechanisms to overcome challenges in modern 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

Predicting Grain Count and Weight of Grape Clusters by Image Processing with Deep Learning

  • 期刊论文

  • 发表日期：

    2025年5月12日

查看全文
  • Erhan Kahya

摘要

This study aims to evaluate the effectiveness of deep learning techniques in predicting grape number and weight. Specifically, grape detection and weight estimation were performed using machine vision processes with the object recognition algorithm You Only Look Once Version 8 (YOLOv8). Tests performed with different model sizes (nano, small, medium, large, xlarge) showed that the YOLOv8L model had the highest accuracy rate. The dataset was augmented with 120 photos taken in the field and 540 images obtained from the internet, divided into 80% for training and 20% for testing. The data augmentation techniques applied improved the generalization ability of the model and minimized the problem of overfitting. The results obtained show that the YOLOv8L model performs better than other models on metrics such as mAP50-95, precision and recall. In addition, the model developed for grape weight prediction has the potential to be used in agriculture for object recognition and weight estimation. This study aims to support sustainable production processes by providing data-driven decision-making mechanisms to overcome challenges in modern viticulture.