Scholarlink

<p>Ensuring Personal Protective Equipment (PPE) compliance in the food industry is vital for maintaining hygiene standards, yet manual monitoring remains error-prone. This study introduces an automated PPE detection system using the YOLOv7 algorithm, specifically targeting aprons, hairnets, and footwear within the Malaysian food sector. A dataset of 1,500 images was expanded to 2,859 through mosaic augmentation to enhance model generalization. Among the variants tested, YOLOv7-d6 emerged as the most reliable and balanced configuration, achieving an mAP (0.5) of 0.918. While YOLOv8-l achieved a slightly higher mAP (0.5) of 0.923, it exhibited a critical imbalance between precision and recall, posing a risk of missed violations in safety-sensitive environments. In contrast, YOLOv7-d6 maintained a balanced precision (0.934) and recall (0.930), ensuring high detection reliability. Real-time validation further confirmed the model’s efficacy, with detection accuracies of 88.24% for aprons, 90.91% for footwear, and 83.33% for hairnets. These results demonstrate that the YOLOv7-d6 framework serves as a robust automated monitoring system for strengthening food safety compliance.</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

Robust baseline evaluation of scalable YOLO architectures for PPE compliance monitoring: a foundational study for lightweight DNN development in the food industry

  • 期刊论文

  • 发表日期：

    2026年6月27日

查看全文
  • Nurbaity Sabri,
  • Hafizul Ikram Azmi,
  • Raihah Aminuddin,
  • Shafaf Ibrahim,
  • Budi Sunarko

摘要

Ensuring Personal Protective Equipment (PPE) compliance in the food industry is vital for maintaining hygiene standards, yet manual monitoring remains error-prone. This study introduces an automated PPE detection system using the YOLOv7 algorithm, specifically targeting aprons, hairnets, and footwear within the Malaysian food sector. A dataset of 1,500 images was expanded to 2,859 through mosaic augmentation to enhance model generalization. Among the variants tested, YOLOv7-d6 emerged as the most reliable and balanced configuration, achieving an mAP (0.5) of 0.918. While YOLOv8-l achieved a slightly higher mAP (0.5) of 0.923, it exhibited a critical imbalance between precision and recall, posing a risk of missed violations in safety-sensitive environments. In contrast, YOLOv7-d6 maintained a balanced precision (0.934) and recall (0.930), ensuring high detection reliability. Real-time validation further confirmed the model’s efficacy, with detection accuracies of 88.24% for aprons, 90.91% for footwear, and 83.33% for hairnets. These results demonstrate that the YOLOv7-d6 framework serves as a robust automated monitoring system for strengthening food safety compliance.