机器视觉技术在大体型家畜无接触体尺测量中的研究进展  

Research progress on machine vision technology for non-contact body measurement of large livestock

作　　者：李振波[1,2,3,4] 孙浩翔 郭倩男 张涵钰 刘皓南 LI Zhenbo;SUN Haoxiang;GUO Qiannan;ZHANG Hanyu;LIU Haonan(College of Information and Electrical Engineering,China Agricultural University,Beijing 100083,China;Key Laboratory of Smart Animal Farming Technology,Ministry of Agriculture and Rural Affairs,Beijing 100083,China;Scientific Research Base for Integrated Technologies of Precision Agriculture(Animal Husbandry),Ministry of Agriculture and Rural Affairs,Beijing 100083,China;Key Laboratory of Agricultural Information Acquisition Technology,Ministry of Agriculture and Rural Affairs,Beijing 100083,China)

机构地区：[1]中国农业大学信息与电气工程学院,北京100083 [2]农业农村部智慧养殖重点实验室,北京100083 [3]农业农村部精准农业技术集成科学试验基地(畜牧业),北京100083 [4]农业农村部农业信息获取技术重点实验室,北京100083

出　　处：《农业工程学报》2025年第7期1-12,共12页Transactions of the Chinese Society of Agricultural Engineering

基　　金：北京市智慧农业创新团队项目(BAIC10-2024);科技创新2030-“新一代人工智能”重大项目-典型畜禽疫病诊断与主动防控智慧云平台。

摘　　要：家畜体尺能直接反映其生长发育状态,对育种和养殖过程管理具有重要意义。基于机器视觉的家畜无接触体尺测量技术可以解决传统人工接触式测量中耗时、费力和主观误差等问题,同时能够降低养殖人员的劳动强度,避免家畜产生应激反应。近年来,随着机器视觉技术的迅猛发展,家畜无接触体尺测量方法也取得了突破性的进步。该研究聚焦于牛、羊、马和猪4种常见大体型家畜,按照体尺测量任务流程,概述了常见的家畜图像采集场景、图像采集设备和设备部署方式。基于近5年机器视觉在家畜无接触体尺测量中的应用,阐述了目前家畜图像分割算法和家畜体尺测量算法的研究现状。当前研究的着重点主要在于加速体尺测量过程,提升测量结果精度,以及增强测量设备的便携性这3个核心方面。结合当前研究中存在的公开数据集不足、深度学习前沿方法应用较少、算法在实际生产中应用和部署困难等问题,提出了未来应围绕应用生成式模型扩充家畜图像数据集、加速深度学习方法的迁移,开发适用多种家畜的通用测量模型等方面展开研究,旨在为后续的研究及应用提供参考。Livestock breeding is often required for animal growth and development.Among them,the systematic evaluation of livestock body measurements can also be highlighted to represent the animal growth and developmental stages.Such measurements are of great importance for the decision-making on the overall breeding.Manual contact measurements have been used in traditional practices.However,manual contact is usually susceptible to subjective errors,due to the cumbersome,time-consuming,and labor-intensive tasks.It is very necessary for the accurate data of the correct decisions.Fortunately,machine vision has revolutionized the agricultural industry in recent years.The contactless body measurement can be expected to replace the manual contact measurements using machine vision.The potential stress reactions can also be prevented to reduce the labor intensity in livestock breeding.This study aims to review the research progress of the non-contact livestock body measurement using machined vision.Four commonly large-bodied livestock were selected,including cattle,sheep,horses,and pigs.Initially,the common acquisition of livestock images was outlined to evaluate the types of imaging devices and various deployments.All tasks were aligned with the body size measurement.Subsequently,machine vision was applied to the contactless body measurements of livestock over the past five years.The current research status of image segmentation was also summarized during livestock body measurements.The speed,accuracy,and portability of equipment were then concentrated mainly on the body measurement at present.Several challenges were proposed,including the limited supply of public datasets and deep learning in the deployment of the algorithms in real-world environments.As such,the generative models can be expected to augment the dataset of the livestock images.Deep learning can be promoted to develop the generalized measurement suitable for a wide range of livestock.The findings can also provide valuable insights and references for future resear

关 键 词：体尺测量 家畜 机器视觉 深度学习 图像采集 图像分割 数据集 

分 类 号：S126[农业科学—农业基础科学]