Scholarlink

<p>Mango ripeness grading during the packaging phase involves categorizing fruits based on the duration they can remain packed before consumption. This process is essential for managing the long-distance transport of fruits. Fruit grading is a&#xa0;more challenging problem than binary categorization into fresh and rotted, since most fruits at the packaging stage are fresh and of high quality, with only minor differences that determine their shelf life. We define four categories for packaging based on the maturity level of the fruit: ‘ripe’ (optimal for packaging), ‘unripe’, ‘damaged’ and ‘overripe’. We propose a&#xa0;fusion method that combines deep features extracted from red–green–blue (RGB) images using three pre-trained convolution neural networks (CNNs), VGG13, VGG16, and VGG19. A&#xa0;support vector machine (SVM) is then trained on the obtained features for the classification. Additionally, we created a&#xa0;dataset of ‘Chaunsa’ mangoes, the most popular variety in South Asia, by acquiring 2000 images directly from the farms prior to the packaging. These images were then labeled into the four predefined categories by experts. A&#xa0;comprehensive evaluation shows that the proposed approach achieves an accuracy of 87.78% on the gathered dataset. The dataset is publicly available for further evaluation by the research community.</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

Ripeness Grading for Mango Packaging by Fusion of Deep Features

  • 期刊论文

  • 发表日期：

    2025年10月2日

查看全文
  • Muhammad Imran,
  • Syed Fahad Tahir,
  • Labiba Gillani Fahad

摘要

Mango ripeness grading during the packaging phase involves categorizing fruits based on the duration they can remain packed before consumption. This process is essential for managing the long-distance transport of fruits. Fruit grading is a more challenging problem than binary categorization into fresh and rotted, since most fruits at the packaging stage are fresh and of high quality, with only minor differences that determine their shelf life. We define four categories for packaging based on the maturity level of the fruit: ‘ripe’ (optimal for packaging), ‘unripe’, ‘damaged’ and ‘overripe’. We propose a fusion method that combines deep features extracted from red–green–blue (RGB) images using three pre-trained convolution neural networks (CNNs), VGG13, VGG16, and VGG19. A support vector machine (SVM) is then trained on the obtained features for the classification. Additionally, we created a dataset of ‘Chaunsa’ mangoes, the most popular variety in South Asia, by acquiring 2000 images directly from the farms prior to the packaging. These images were then labeled into the four predefined categories by experts. A comprehensive evaluation shows that the proposed approach achieves an accuracy of 87.78% on the gathered dataset. The dataset is publicly available for further evaluation by the research community.