Scholarlink

<p>Inventory management is a key component in ensuring seamless business operations; yet, many Small and Medium-sized Enterprises (SMEs) still rely on manual counting, which is prone to error and inefficiency. While deep learning models such as You Only Look Once (YOLO) have shown promise for object detection, their application in real-world warehouse environments often faces practical challenges, such as the indistinct texture of white fabrics and the low resolution of Closed-Circuit Television (CCTV) cameras. In this paper, we present a fine-tuned YOLOv11-based system designed to detect and count inventory items in real time. The system is tailored to handle the complexities of SME warehouses, where products are frequently scattered across limited spaces. Our approach involves constructing a diverse dataset from CCTV footage, consisting of 335 images representing three product types: white fabric rolls, blue fabric rolls, and large pillow bags. To improve detection accuracy, we employed image processing techniques, including Canny Edge Detection and Contour Detection, to augment and refine the dataset. By leveraging transfer learning, we fine-tuned YOLOv11 to detect and count objects under various warehouse conditions. The model achieved a detection accuracy of 90% based on the mean Average Precision (mAP) at an Intersection over Union (IoU) threshold of 0.5 (mAP50) and an inventory counting accuracy of 97%, as verified through five experimental rounds. These findings highlight the potential of combining deep learning with traditional computer vision methods to develop a practical, scalable inventory management solution. This system provides SMEs with a cost-effective alternative to manual counting, addressing key limitations in real-world inventory tracking.</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

Real-time object detection and counting for inventory management using fine-tuned YOLOv11

  • 期刊论文

  • 发表日期：

    2026年3月19日

查看全文
  • Peaysararn Rapinrangchang,
  • Krit Jamkachornkiat,
  • Panupong Khamruen,
  • Thitirat Siriborvornratanakul

摘要

Inventory management is a key component in ensuring seamless business operations; yet, many Small and Medium-sized Enterprises (SMEs) still rely on manual counting, which is prone to error and inefficiency. While deep learning models such as You Only Look Once (YOLO) have shown promise for object detection, their application in real-world warehouse environments often faces practical challenges, such as the indistinct texture of white fabrics and the low resolution of Closed-Circuit Television (CCTV) cameras. In this paper, we present a fine-tuned YOLOv11-based system designed to detect and count inventory items in real time. The system is tailored to handle the complexities of SME warehouses, where products are frequently scattered across limited spaces. Our approach involves constructing a diverse dataset from CCTV footage, consisting of 335 images representing three product types: white fabric rolls, blue fabric rolls, and large pillow bags. To improve detection accuracy, we employed image processing techniques, including Canny Edge Detection and Contour Detection, to augment and refine the dataset. By leveraging transfer learning, we fine-tuned YOLOv11 to detect and count objects under various warehouse conditions. The model achieved a detection accuracy of 90% based on the mean Average Precision (mAP) at an Intersection over Union (IoU) threshold of 0.5 (mAP50) and an inventory counting accuracy of 97%, as verified through five experimental rounds. These findings highlight the potential of combining deep learning with traditional computer vision methods to develop a practical, scalable inventory management solution. This system provides SMEs with a cost-effective alternative to manual counting, addressing key limitations in real-world inventory tracking.