Scholarlink

The appearance of strawberries is a crucial factor for both consumers and the fruit processing industry. The visual quality of strawberries is directly related to their degree of ripeness. With the advancement of deep learning, the analysis of strawberry appearance has become more accurate, though these methods still require significant time and computational power. In this article, we analyze the efficiency of hundreds of different combinations of convolutional neural network (CNN) models and supervised classifiers to evaluate the quality of strawberries. We utilized seventy-one CNN models to extract features from strawberry images and applied ten different classifiers to perform the classification. The best results were obtained with CNNs from the ConvNeXt family (ConvNeXtBase, ConvNeXtSmall, and ConvNeXtTiny) and VGG models (VGG16 and VGG19) in combination with Gradient Boosting, Histogram-Based Gradient Boosting, and SVM classifiers, achieving accuracies up to 78% and F1-scores up to 85%. The objective of our study is to help farmers accurately classify the appearance of strawberries in real-world situations. The methods used can facilitate the future development of intelligent strawberry classification systems.

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

Classification of Strawberry Maturity Level from Images Using Supervised Classifiers and Convolutional Neural Networks as Feature Extractors

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Érika Kayoko Hamaguti,
  • Fabricio Aparecido Breve

摘要

The appearance of strawberries is a crucial factor for both consumers and the fruit processing industry. The visual quality of strawberries is directly related to their degree of ripeness. With the advancement of deep learning, the analysis of strawberry appearance has become more accurate, though these methods still require significant time and computational power. In this article, we analyze the efficiency of hundreds of different combinations of convolutional neural network (CNN) models and supervised classifiers to evaluate the quality of strawberries. We utilized seventy-one CNN models to extract features from strawberry images and applied ten different classifiers to perform the classification. The best results were obtained with CNNs from the ConvNeXt family (ConvNeXtBase, ConvNeXtSmall, and ConvNeXtTiny) and VGG models (VGG16 and VGG19) in combination with Gradient Boosting, Histogram-Based Gradient Boosting, and SVM classifiers, achieving accuracies up to 78% and F1-scores up to 85%. The objective of our study is to help farmers accurately classify the appearance of strawberries in real-world situations. The methods used can facilitate the future development of intelligent strawberry classification systems.