基于畜牧養(yǎng)殖監(jiān)控視頻的圈養(yǎng)母羊分娩時(shí)間精準(zhǔn)預(yù)測

打開文本圖片集

中圖分類號：TP399 文獻(xiàn)標(biāo)志碼：A

Accurate Prediction of Parturition Time for Confined Ewes Based on Livestock Farming Surveillance Videos

WANG Miaomiao，JI Yuhao，LIU Cheng，YUN Xun （SchoolofInformationScienceandEnginering，ShandongAgriculturalUniversity，Taian271OoO，Shandong，China）

Abstract：Addressing the loss of newborn lambs due toaccidents after parturition in confinedewes，basedonreal-time surveillance videos inanimal husbandry，alimitedarea detection algorithm wasfirstlydesigned using the binarysegmentation method.Byapplying masking techniques outsidethe target detection area，the delineation of designated sheepfold regionswas achieved.A tracking technique forthe wandering trajectories of pregnant ewes was adopted based on DeepSORT algorithm，YOLOV5model wasutilized toidentify pregnant ewe targetsand obtain real-time locations of heir activitiessuchas feding and wandering within the targeted sheepfold.Byanalyzingthe walking and stopping rhythm paterns of pregnant ewes before parturition，precise predictions of ewe parturition times were made.The results show that the real-time tracking recognition rate during ewe wandering can reach 92.72% . When the density of ewe increases，the phenomenon of mutualobstruction among individuals intensifies，which afects thepositioning accuracyof the detection model toacertain extent，however，theconfidence level stillremains above O.8 in theend.Asparturition approaches， thefrequencyof dynamicand static behavioral changes inpregnant ewes exhibits distinct characteristic paterns，allowing for precise predictions ofparturition times，whichcanensure timelyasistance during lambing and nursing，thereby improving the survival rate of newborn lambs.

Keywords： multi-target recognition； deep learning; data mining; smart animal husbandry

現(xiàn)代化養(yǎng)殖企業(yè)已經(jīng)可以對畜禽養(yǎng)殖生態(tài)系統(tǒng)實(shí)施完整、全場景的監(jiān)測[1-4]。（剩余9791字）