Scholarlink

<p>Mung bean (<i>Vigna radiata</i>) is a highly valued pulse that plays a critical role in food security and sustainable agriculture due to its nutritional profile, cooking quality, and versatile culinary applications. It is consumed in multiple forms, including whole pulses, split pulses with or without husk, and sprouts obtained after soaking and germination. Despite its commercial significance and high market value, traditional quality assessment methods, including physical grading, manual inspection, and chemical analysis, are inherently subjective, time-consuming, labour-intensive, and potentially destructive. Therefore, there is an urgent need for fast, accurate, and non-destructive quality evaluation systems that can be integrated into pulse-processing workflows. In the present study, a non-destructive machine-vision-based image classification system was developed using five deep learning models (ResNet50, DenseNet121, MobileNetV2, GoogleNet, and YOLOv8s-cls) to assess the appearance quality of mung beans. A balanced dataset of 1530 images comprising four classes, that were whole mung bean (Class I), split mung bean with husk (Class II), split mung bean without husk (Class III), and husk alone (Class IV), was captured under the developed controlled illumination machine vision system. All five models were trained and compared on the acquired mung bean dataset under identical experimental conditions. On comparing the achieved classification accuracies of different applied models, ResNet50 (98.76%), DenseNet121 (98.76%), GoogleNet (97.53%), MobileNetV2 (97.53%), and YOLOv8s-cls achieved the highest classification accuracy with 98.90% among all. These results further confirm that the YOLOv8s-cls model can be used for quality assessment, achieving the highest accuracy with strong, consistent performance across all four selected mung bean classes. This promising approach may thus be applied for rapid, automated, and non-destructive appearance-based quality monitoring in the pulse processing and allied industries.</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

Non-destructive Machine Vision System for Quality Assessment of Mung Bean (Vigna radiata)

  • 期刊论文

  • 发表日期：

    2026年4月30日

查看全文
  • Komalpreet Kaur,
  • Rajni Goyal,
  • Shubhra Shekhar,
  • Amar Nath,
  • Kamlesh Prasad

摘要

Mung bean (Vigna radiata) is a highly valued pulse that plays a critical role in food security and sustainable agriculture due to its nutritional profile, cooking quality, and versatile culinary applications. It is consumed in multiple forms, including whole pulses, split pulses with or without husk, and sprouts obtained after soaking and germination. Despite its commercial significance and high market value, traditional quality assessment methods, including physical grading, manual inspection, and chemical analysis, are inherently subjective, time-consuming, labour-intensive, and potentially destructive. Therefore, there is an urgent need for fast, accurate, and non-destructive quality evaluation systems that can be integrated into pulse-processing workflows. In the present study, a non-destructive machine-vision-based image classification system was developed using five deep learning models (ResNet50, DenseNet121, MobileNetV2, GoogleNet, and YOLOv8s-cls) to assess the appearance quality of mung beans. A balanced dataset of 1530 images comprising four classes, that were whole mung bean (Class I), split mung bean with husk (Class II), split mung bean without husk (Class III), and husk alone (Class IV), was captured under the developed controlled illumination machine vision system. All five models were trained and compared on the acquired mung bean dataset under identical experimental conditions. On comparing the achieved classification accuracies of different applied models, ResNet50 (98.76%), DenseNet121 (98.76%), GoogleNet (97.53%), MobileNetV2 (97.53%), and YOLOv8s-cls achieved the highest classification accuracy with 98.90% among all. These results further confirm that the YOLOv8s-cls model can be used for quality assessment, achieving the highest accuracy with strong, consistent performance across all four selected mung bean classes. This promising approach may thus be applied for rapid, automated, and non-destructive appearance-based quality monitoring in the pulse processing and allied industries.