Scholarlink

<p>Smart technologies in modern poultry farms enable precise monitoring of broiler performance through computer vision and artificial intelligence, supporting data driven management and improved productivity. This study investigates three deep learning models Mask R-CNN with MobileNetv2, YOLOv8-large, and SAM for broiler body segmentation. A dataset of 1122 top-view images of Arian broilers were collected over 13 days using a custom-built image acquisition platform. Models specific adjustments were made to optimize training and evaluation. Among the tested models, YOLOv8-large achieved the highest segmentation accuracy (99.5%) and real-time processing speed of 33 frames per second within 50 epochs, making it suitable for embedded applications. Mask R-CNN showed fast convergence but was limited by the lightweight MobileNetv2 backbone, while SAM produced accurate and smooth segmentation using a region of interest (RoI) approach but required high computational resources. Overall, YOLOv8-large demonstrated the best trade off between accuracy, speed, and efficiency, highlighting the potential of deep learning-based segmentation to improve automation and precision in poultry production.</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

A comparative study of deep learning-based models for broiler body segmentation

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Hossein Akhtari,
  • Hossein Navid,
  • Ali Ghaffarnezhad,
  • Redmond R. Shamshiri

摘要

Smart technologies in modern poultry farms enable precise monitoring of broiler performance through computer vision and artificial intelligence, supporting data driven management and improved productivity. This study investigates three deep learning models Mask R-CNN with MobileNetv2, YOLOv8-large, and SAM for broiler body segmentation. A dataset of 1122 top-view images of Arian broilers were collected over 13 days using a custom-built image acquisition platform. Models specific adjustments were made to optimize training and evaluation. Among the tested models, YOLOv8-large achieved the highest segmentation accuracy (99.5%) and real-time processing speed of 33 frames per second within 50 epochs, making it suitable for embedded applications. Mask R-CNN showed fast convergence but was limited by the lightweight MobileNetv2 backbone, while SAM produced accurate and smooth segmentation using a region of interest (RoI) approach but required high computational resources. Overall, YOLOv8-large demonstrated the best trade off between accuracy, speed, and efficiency, highlighting the potential of deep learning-based segmentation to improve automation and precision in poultry production.