Asphyxia occurrence detection in sows during the farrowing phase by inter-birth interval evaluation

Abstract

The occurrence of asphyxia in sows during parturition often results in stillbirths and low vitality piglets, thus significantly affecting pig production and animal welfare in terms of piglet mortality. The losses during farrowing can account to a significant loss in a litter, a whole litter or in extreme cases, the loss of the sow. The objective of this study was to develop an infrared depth image sensor-based monitoring system as a surveillance support system for stockmen during the farrowing phase of sows, in an effort to reduce piglet mortality associated with asphyxiation of piglets. Experiments were performed at Jurong Research farm of Nanjing Agricultural University, using a herd of 105 Meishan sows. Data was collected by monitoring the farrowing process of 15 sows housed in farrowing pens. The sows were monitored by video recording for labeling and depth images taken at 10 fpm for automatic system development. Data labeling was based on the timestamps of births of all the piglets in a litter and each piglet vitality score based on the number of times a piglet attempts to stand within the 1st minute after birth. Data labeling was performed 1 h before the start of parturition until 0.5 h after the end of parturition. The depth images were processed to segment the piglets, and a developed parametrized ellipse-fitting algorithm performed piglet detection by multi-ellipse fitting. Piglet count and count tracking were established by the number of fitted ellipses on the image-object shape. Each sequential increment in the piglet count was used to compute the inter-birth interval. The developed model attained a detection accuracy of 0.832 for no piglets and 0.801 for 14 piglets in a range of 0 to 14 piglets in a litter. In piglet counting and count tracking, the model achieved an average accuracy of 0.918 (R2) at an RMSE of 1.225 piglets. For the classification of an asphyxia event, the model indicated an accuracy of 0.863, specificity of 0.921, precision of 0.791, and a sensitivity of 0.723. The developed system can serve as part of Precision Livestock Farming automatic farrowing monitoring system with the aim of detecting the start of parturition, piglet count tracking, and asphyxia occurrence.