CALVING   MONITORING  WITH VISION CAMERA SYSTEM USING DEEP LEARNING ALGORITHMS

Abstract: The calving processIn dairy cattleIs a crldcal sltJJatlon for thedairy farmers economlcally and for the cows In their production cycle [1].The failurein assist ngthe cow bythe personnel when needed can cause sevel"S illnass to bodi mother cow and calf.Thel"Sfore,pl"Sdict ngthe calv ng event time can benefit the farmers in deciding     an appropriate time ofIntervention during the process. The objective of this projectis to mon tor the calving process in dairy cattle witha vision camera system using Convolut onal Neural Network (CNN) based deep learning methods.The data set comprises different calving events recordedIn the Delaval test farm using vision cameras. Based on thecalving stages discussed laterin subsection 1.1.2, the following     three st.ages areidentified to be important - the appearance of an amniotic sac (water bag) atthe vulva, hooves after the amniotic sac: cracks, andlastly the birthof the calf [2].Inthis project, objectdetectionis used as a methodinidentifying these stages by detecting  the pr ncipal object classes namely the amniot c sac, hooves. and calf from the recorded calving events. This method avoids the use of any devices to be in close contact with the cows and requires only the vision cameras to be placed above the calving area. In this project we use state-of-the-art one-stage obje<:t detectionmethods such as YOLOvS [3] and RetinaNet [4] with backbonesCSPDarkNet-5] [5] and ResNet-50 [6], with pretralned COCO and lmagenet weights. Tne two-stage object detection method,FasterR-CNN [7] with backbone R.esNet-SO is also evaluatedin this task. Annotations for thebounding boxes are clone usingthe Computer Vision Annotation Tool(CVAT) for the object classes -amnotic sac, hooves, and calf. The detection results obtained in this project, using dfferent object detection models, promise that the calving process can be efficiently monitored and tracked with the help of vision cameras using deep learning algorithms in real-time.This can in tum avoid the need forthe fanners to be present constantly in monitoring the cows. Also, the identification of the calving stagesin real-time can help the farmersto takeImmediate action Md InterveneIn the process when needed. This can help the farmersin avoiding the cause of serious injuries to mother cow and prevent any calf losses.