Comparative study between scan sampling behavioral observations and an automatic monitoring image system on a commercial fattening pig farm

Abstract

Automation is an important element in modern livestock farming. Image computer analysis is the automation technology aiming to monitor farm animals by recording continuous images. Further analysis can be carried out to assess more precisely and effectively farm animals’ welfare. The aim of this study was to determine the applicability of the commercial multi-sensor device (Peek Analytics) developed by Copeeks SAS (France), in comparison to human-based observations used to assess behaviors in pigs, including posture (standing/resting), area within the pen (feeding/drinking/resting/enrichment), and activity level (number of active pigs). Two Peek Analytics (Peek 3 and Peek 4) were installed on a commercial fattening pig farm in Murcia (Spain). Each device recorded data of two pens (39 pigs in four pens in total). Scan sampling was the human-based behavioral observation method used in this study. Data was collected for five consecutive days, in the following intervals: 09:00-11:00, 13:00-15:00, and 16:00-18:00 (30 hours of observation in total). Every pig was observed six times per hour and hence the information analyzed includes 7020 observations (180 observations/pig). The comparison between data from human observation and Peek Analytics was performed by using Pearson correlation tests. Posture, areas of interest, and activity level were analyzed separately, as well as data recorded by Peek 3 and 4. Results indicated that Peek Analytics showed a better agreement with human observation, when recording posture(r=0.77, P<0.01) and area within the pen (r=0.77, P<0.01), than when recording activity level (r=0.35, P<0.01). Two devices performed differently in general, with Peek 3 having better agreement than Peek 4 with human observation, regardless of posture, area within the pen, and activity level. The better agreement in Peek 3 may be attributed to the smaller number of pigs in Peek 3 (18) compared to Peek 4 (22). We can conclude from the study that image computer analysis may be reliable in assessing posture and area within the pen of pigs. On the other hand, a moderate agreement in activity level between human observation and computer vision can be due to different methodologies of recording the activity, rather than due to low accuracy of Peek Analytics.