Scholarlink

<p>Yak husbandry is crucial for communities in high-altitude regions. Traditional animal identification methods are often invasive, costly, and inefficient. To address these challenges, this paper presents an automated, non-invasive system for real- time yak identification, re-recognition, and monitoring utilizing a multi-stage deep learning pipeline. The proposed framework integrates a lightweight YOLOv11 s model for initial object detection, a fine-tuned MobileNetV2 architecture for extracting discriminative appearance features, and the DeepSORT algorithm for robust multi-object tracking. The models were trained and validated on a custom dataset generated from video footage captured in a yak shed of Arunachal, India. Experimental results demonstrate the system’s high efficacy. The MobileNetV2 feature extractor showed excellent discriminative capability enabling consistent and accurate re-identification. This study validates the feasibility of using an end-to-end deep learning approach for efficient and scalable livestock monitoring, offering a significant advancement in precision agriculture and animal welfare.</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

Computer Vision Based Deep Learning Framework for Detection, Re-identification and Tracking of Yaks in High-Altitude Regions

  • 期刊论文

  • 发表日期：

    2026年1月9日

查看全文
  • Udayaditya Sankar Das,
  • Kishor Kumar Baruah,
  • Vijay Paul,
  • Rupesh Mandal,
  • Mokhtar Hussain,
  • Mihir Sarkar,
  • Nupur Choudhury

摘要

Yak husbandry is crucial for communities in high-altitude regions. Traditional animal identification methods are often invasive, costly, and inefficient. To address these challenges, this paper presents an automated, non-invasive system for real- time yak identification, re-recognition, and monitoring utilizing a multi-stage deep learning pipeline. The proposed framework integrates a lightweight YOLOv11 s model for initial object detection, a fine-tuned MobileNetV2 architecture for extracting discriminative appearance features, and the DeepSORT algorithm for robust multi-object tracking. The models were trained and validated on a custom dataset generated from video footage captured in a yak shed of Arunachal, India. Experimental results demonstrate the system’s high efficacy. The MobileNetV2 feature extractor showed excellent discriminative capability enabling consistent and accurate re-identification. This study validates the feasibility of using an end-to-end deep learning approach for efficient and scalable livestock monitoring, offering a significant advancement in precision agriculture and animal welfare.