Scholarlink

<p>Broiler chicken production is a cornerstone of global food security, yet monitoring animals in commercial farms remains challenging due to high stocking densities and variable environmental conditions. Progress in automated monitoring is hindered by the scarcity of domain-specific, publicly available datasets. To address this gap, we present Poultry Images for Object detection (PIO), a dataset designed to support the development and evaluation of computer vision models for poultry farming. PIO comprises 1,487 manually annotated images containing 327,289 instances of broiler chickens, collected from both commercial and prototype poultry houses across different growth stages. The dataset reflects realistic conditions such as variations in morphology, lighting and bird density. Annotations were generated using the LabelImg tool, with bounding boxes normalized to image dimensions for compatibility with state-of-the-art detection frameworks. To illustrate its utility, three YOLOv10 variants were trained and evaluated on PIO, demonstrating its suitability for benchmarking object detection models in precision livestock farming contexts, as shown in figure 1.</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

PIO, A Large-Scale Dataset for Broiler Chicken Detection under Real Poultry Farming Conditions

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月9日

查看全文
  • Keyla Boniche,
  • Edmanuel Cruz,
  • José Carlos Rangel,
  • Miguel Hidalgo-Rodríguez,
  • Francisco Gómez-Donoso

摘要

Broiler chicken production is a cornerstone of global food security, yet monitoring animals in commercial farms remains challenging due to high stocking densities and variable environmental conditions. Progress in automated monitoring is hindered by the scarcity of domain-specific, publicly available datasets. To address this gap, we present Poultry Images for Object detection (PIO), a dataset designed to support the development and evaluation of computer vision models for poultry farming. PIO comprises 1,487 manually annotated images containing 327,289 instances of broiler chickens, collected from both commercial and prototype poultry houses across different growth stages. The dataset reflects realistic conditions such as variations in morphology, lighting and bird density. Annotations were generated using the LabelImg tool, with bounding boxes normalized to image dimensions for compatibility with state-of-the-art detection frameworks. To illustrate its utility, three YOLOv10 variants were trained and evaluated on PIO, demonstrating its suitability for benchmarking object detection models in precision livestock farming contexts, as shown in figure 1.