Automatic estimation of number of piglets in a pen during farrowing, using image analysis

Abstract

The objective was to develop a camera based monitoring system of the farrowing process in sows. The system, when used in practical farm conditions, should support the farm staff in reducing the problems of mortality in piglets due to perinatal asphyxia and crushing. The experiments took place in the research farm of the University of Veterinary Medicine Vienna, using a herd of 120 Large White sows. Data were collected from sows housed in farrowing pens with possibility of temporary crating. The farrowing process of 8 sows was video recorded and labelled for a period of 6 h before the start of farrowing until the end of farrowing. Timestamps of births of all piglets in the litter were labelled. Images obtained during the experiment were segmented with focus on piglet detection. Three parameters were extracted from segmented images: number of objects detected, area and perimeter of all objects. On the basis of the parameters, a Transfer Function (TF) model was estimated with output variable defined as number of piglets in the pen. The developed model explained 82% (R2) variability in the training set composed of 5 sows and 81% (R2) in the validation set composed of 3 sows. Number of piglets in the pen was estimated with a standard error of 1.73 piglets in the training set and 1.72 in the validation set. The potential application of the developed technique is monitoring of start of farrowing, perinatal asphyxia and crushing in piglets.