您好，欢迎访问内蒙古农牧业科学院 机构知识库！
内蒙古农牧业科学院机构知识库

全部

  • 全部
  • 标题
  • 学者
  • 机构
  • 关键词

Sheep-YOLO: improved and lightweight YOLOv8n for precise and intelligent recognition of fattening lambs' behaviors and vitality statuses

文献类型： 外文期刊

作者： Zhang, Mengjie 1 ; Hong, Dan 1 ; Wu, Jiabao 1 ; Zhu, Yanfei 1 ; Zhao, Qinan 2 ; Zhang, Xiaoshuan 1 ; Luo, Hailing 1 ;

作者机构： 1.China Agr Univ, Beijing 100083, Peoples R China

2.Inner Mongolia Acad Agr & Anim Husb Sci, Hohhot 010000, Peoples R China

关键词： Fattening lambs; Computer vision; Deep learning; Improved and lightweight YOLOv8n

期刊名称：COMPUTERS AND ELECTRONICS IN AGRICULTURE （ 影响因子：8.9； 五年影响因子：9.3 ）

ISSN： 0168-1699

年卷期： 2025 年 236 卷

页码：

收录情况： SCI

摘要： Fattening lambs' behaviors and vitality statuses reflect health status and animal welfare directly. However, current detection methods for these aspects primarily rely on manual observation, which is inefficient. This paper aims to achieve precise and intelligent recognition of fattening lambs' behaviors and vitality statuses. Through the comprehensive literature review, field observations, and consultations with experts, four pivotal behaviors (running, sleeping, socialization, and wandering), alongside four vitality statuses (excitement, hungry, lethargy, and normal) are identified as crucial indicators that mirror the health status of fattening lambs. A visual system is established to collect and construct a dataset of fattening lambs, and a five-fold cross-validation method is employed to determine the basic model. Furthermore, an improved lightweight YOLOv8n model, named SheepYOLO, is developed in this paper. Sheep-YOLO utilizes the FasterNet network to reduce model complexity and incorporates the Mixed Local Channel Attention (MLCA) module, SIoU loss function, and Content-Aware ReAssembly of FEatures (CARAFE) to enhance the model's adaptability to varying breeding densities and lighting conditions. To validate the effectiveness of the optimization strategies, comparison experiments are conducted in this paper. The experimental results show that Sheep-YOLO achieved Params of 1.905 M and GFLOPs of 5.5 G, with a high mAP0.5 of 96.1 % on the test set and a detection speed of 79.317 FPS(12.61 ms/per image). Compared to the basic YOLOv8n, Sheep-YOLO achieves a 36.6 % reduction in Params and a 30 % decrease in GFLOPs, while maintaining higher mAP0.5 and faster detection speed. Besides, Sheep-YOLO surpasses widely used algorithms such as YOLOv3-tiny, YOLOv5n, YOLOv10n, RT-DETR, and Faster-RCNN regarding both lightweight performance and precision. Therefore, this study provides potential technical support for precise and intelligent recognition of the behaviors and vitality statuses of fattening lambs, contributing to the health monitoring and early disease prediction of sheep.

  • 相关文献
作者其他论文 更多>>
置顶
购物车
反馈

© 京ICP备09089781号-29 主办单位：内蒙古农牧业科学院

学术资源： 中国农业科学院 农业专业知识服务系统 农科联盟资源共建共享平台 中国农业科技文献与信息服务平台 中国农业期刊集成服务平台