Scholarlink

In this research, a group of convolutional neural network (CNN) architectures for peach image classification task were compared. The models were trained to identify the categories of a set of images that could be found in a sorting machine, such as “Healthy” or “Damaged”, which correspond to the separation of peaches with or without damage. Specifically, five ResNet models (ResNet-18, ResNet-34, ResNet-50, ResNet-101, and ResNet-152) from the Torchvision library were evaluated for their ability to categorize peach images. The models were trained employing feature extraction in a transfer learning approach and evaluated using established metrics. The evaluation results indicate that ResNet-18 is the optimal CNN model for this task. The metrics for ResNet-18 are as follows: Accuracy – 95.68%, Precision – 95.74%, Recall – 95.68%, Specificity – 90.81%, F1-score – 95.61%, Geometric Mean – 93.07%, and Index of Balanced Accuracy – 87.04%. By using the AI-based grading system, peach producers could optimize their classification production and have each product evaluated according to international standards.

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

Convolutional Neural Network Models for Classifying of Peach (Prunus persica L)

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Flossi Puma-Ttito,
  • Carlos Guerrero-Mendez,
  • Daniela Lopez-Betancur,
  • Tonatiuh Saucedo-Anaya,
  • Rafael Castaneda-Diaz,
  • Luis Martinez-Ytuza

摘要

In this research, a group of convolutional neural network (CNN) architectures for peach image classification task were compared. The models were trained to identify the categories of a set of images that could be found in a sorting machine, such as “Healthy” or “Damaged”, which correspond to the separation of peaches with or without damage. Specifically, five ResNet models (ResNet-18, ResNet-34, ResNet-50, ResNet-101, and ResNet-152) from the Torchvision library were evaluated for their ability to categorize peach images. The models were trained employing feature extraction in a transfer learning approach and evaluated using established metrics. The evaluation results indicate that ResNet-18 is the optimal CNN model for this task. The metrics for ResNet-18 are as follows: Accuracy – 95.68%, Precision – 95.74%, Recall – 95.68%, Specificity – 90.81%, F1-score – 95.61%, Geometric Mean – 93.07%, and Index of Balanced Accuracy – 87.04%. By using the AI-based grading system, peach producers could optimize their classification production and have each product evaluated according to international standards.