Scholarlink

<p>With the growing population of dairy goats, traditional management methods have become insufficient to meet the demands of modern farming. Precision livestock farming (PLF) utilizing advanced technologies holds promise for efficient monitoring and management. Among these, instance segmentation of dairy goat images plays a crucial role in processing image data for intelligent farming. However, existing instance segmentation methods for livestock and poultry often struggle to balance speed and accuracy, limiting their practical application. To address this challenge, we propose DG-DETR, a multi-scale transformer-based instance segmentation model designed specifically for dairy goat images. DG-DETR consists of a Res2Net backbone for feature extraction, an improved transformer module with deformable attention, and a unified query representation (UQR) module for end-to-end segmentation. We construct a dedicated dairy goat dataset comprising 18,600 high-quality images. Experimental results demonstrate that DG-DETR achieves superior segmentation performance, attaining an AP score of 69.20%. Notably, this study introduces deformable attention to dairy goat instance segmentation, significantly reducing labor costs and showing great potential for production monitoring. Here, we show how our approach not only introduces a novel method, but also contributes to the advancement of precise and large-scale goat farming. The code and datasets are available at <a href="https://github.com/tiana-tang/DG-DETR.git">https://github.com/tiana-tang/DG-DETR.git</a> and <a href="https://github.com/tiana-tang/DiaryGoatMVT.git">https://github.com/tiana-tang/DiaryGoatMVT.git</a>.</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

Enhanced dairy goat instance segmentation via multi-scale deformable transformer

  • 期刊论文

  • 发表日期：

    2025年9月5日

查看全文
  • Zihan Bai,
  • Jinglei Tang,
  • Xue Zhang,
  • Hao Rong,
  • Xianglong Pei,
  • Yawei Ding

摘要

With the growing population of dairy goats, traditional management methods have become insufficient to meet the demands of modern farming. Precision livestock farming (PLF) utilizing advanced technologies holds promise for efficient monitoring and management. Among these, instance segmentation of dairy goat images plays a crucial role in processing image data for intelligent farming. However, existing instance segmentation methods for livestock and poultry often struggle to balance speed and accuracy, limiting their practical application. To address this challenge, we propose DG-DETR, a multi-scale transformer-based instance segmentation model designed specifically for dairy goat images. DG-DETR consists of a Res2Net backbone for feature extraction, an improved transformer module with deformable attention, and a unified query representation (UQR) module for end-to-end segmentation. We construct a dedicated dairy goat dataset comprising 18,600 high-quality images. Experimental results demonstrate that DG-DETR achieves superior segmentation performance, attaining an AP score of 69.20%. Notably, this study introduces deformable attention to dairy goat instance segmentation, significantly reducing labor costs and showing great potential for production monitoring. Here, we show how our approach not only introduces a novel method, but also contributes to the advancement of precise and large-scale goat farming. The code and datasets are available at https://github.com/tiana-tang/DG-DETR.git and https://github.com/tiana-tang/DiaryGoatMVT.git.