PSI-11 Testing a Computer Vision System for Automatically Detecting Piglet Interbirth Interval

Abstract

Abstract An apparent opportunity for improvement in the swine industry is preweaning piglet mortality. Preweaning piglet mortality results in economic loss, decreasing productivity per sow, and food waste. One aspect of preweaning mortality, stillborn piglets, can be reduced with timely caretaker interventions to reduce interbirth interval. However, a single caretaker is often in charge of managing many farrowing events concurrently. To address this labor need, piglet interbirth interval could be monitored using thermal cameras. In the present study the performance of a custom image capturing and processing system was evaluated for its ability to identify piglet births. Thermal images of 4 sows in farrowing stalls were collected once every 2 seconds to compare the image processing algorithm’s ability to identify piglet interbirth interval with human observations. There were 66 piglets born in total. Piglets born less than 2 minutes apart were classified as one birthing event, resulting in 49 birthing events total. When the algorithm detected piglet birthing events, they were considered correct if they were within 2 minutes of the human timestamp. The algorithm correctly identified 27 out of 49 birthing events. The algorithm resulted in 22 false negatives and 27 false positives. Results show that 55% of birthing events were accurately detected. With further refinement, the algorithm has the potential to increase accuracy of piglet birthing event identification. The present study demonstrates that computer vision systems can be implemented to monitor piglet birthing events in real-time, allowing caretakers to target their efforts on the most at-risk animals in the farrowing room. This continuous monitoring system has the potential to change the view of farrowing in the swine industry.