基于YOLOv7模型的孕羊产前行为识别方法  

Prenatal behavior recognition method of pregnant ewes based on YOLOv7 model

作　　者：李少涵 袁洪波[1] 蔡振江[1] 程曼[1] LI Shaohan;YUAN Hongbo;CAI Zhenjiang;CHENG Man(College of Mechanical and Electrical Engineering,Hebei Agricultural University,Baoding 071000,China)

机构地区：[1]河北农业大学机电工程学院,河北保定071000

出　　处：《河北农业大学学报》2025年第2期80-87,共8页Journal of Hebei Agricultural University

基　　金：河北省重点研发计划项目(21327402D).

摘　　要：分娩对于母羊和羔羊来说都是一个关键时期,若在这一时期忽视母羊的接产和助产,可能会导致母羊难产、羔羊死亡等,这将直接影响到养殖场的经济效益。怀孕母羊的某些特定的生理表现或者行为变化可能是即将分娩的指示信号,在孕羊产前对此进行检测有助于预测分娩行为的开始。本文提出了一种基于YOLOv7的深度学习网络对视频中孕羊的产前站立、普通躺卧、进食、饮水和分娩躺卧5种行为进行识别。试验结果表明该方法能够准确监测和识别出与分娩事件相关的个体孕羊行为,对于产前5种行为的识别精确度和召回率都在97%以上,全类别平均精度达到98.77%。Labor is a critical period for both ewes and lambs,and neglecting the delivery and assistance of ewes during this period may lead to ewes’difficult births and lamb deaths,which will negatively impact the farm’s economic efficiency.Some specific physiological or behavioral changes in pregnant ewes may be indicative of impending labor.Detecting these changes before parturition can help predict the onset of labor behavior in pregnant ewes.This paper proposes a deep learning model based on YOLOv7 to identify five behaviors of pregnant ewes in surveillance video,namely standing,lying,feeding,drinking,and lambing.The experimental results showed that the proposed approach can accurately monitor and recognize the individual pregnant ewe’s behaviors related to the birthing event.The recognition precision and recall of all five prenatal behaviors were above 97%,and the mean average precision reached 98.77%.

关 键 词：孕羊 深度学习 目标检测 行为识别 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]