Development of an algorithm to detect feed bunk replacement events in growing cattle from feeding event data acquired by an electronic feed intake measurement system

Abstract

Visual observations of competitive feeding behavior in cattle have been associated with animal productivity, health, and social hierarchy. However, human visual observations are time consuming, limiting the number of animals that can be evaluated. The objective of this study was to develop and assess an algorithm to detect disruptive feed bunk replacement events using feeding event data acquired from the Vytelle SENSE feed intake measurement system. Crossbred beef steers (N = 20) fed a grower diet were housed in one of two pens each equipped with three electronic feed bunks and a video recorder. A trained video observer recorded all feed bunk replacement events (when an actor animal displaced a reactor animal from the feed bunk and occupied the same feed bunk within a specified time interval, termed as the replacement criterion) and other agonistic activities at the feed bunk during a 4-d period. The Vytelle SENSE feed intake measurement system recorded the start and end timestamps of individual bunk visit (BV) events of animals. An algorithm was developed to determine replacement events based on the start and end times of consecutive BV events. Using video observation as the gold standard, the recall, precision, and F-score of the algorithm was calculated corresponding to time intervals from 1 to 60 s. The optimum replacement criterion was determined to be ≤ 18 s and ≤ 22 s for pens 1 and 2 respectively. The recall, precision, and F-score of the algorithm using these replacement criterions were high (> 0.75). A competition index was computed for each steer, calculated as the number of actor-initiated events divided by the number of actor and reactor events at the feed bunk. Spearman’s rank correlations (r_s) between total visually observed agonistic interactions at the feed bunk and the replacement events detected from the electronic feeding event data using the algorithm and replacement criterion of the respective pen was determined. There were strong correlations (r_s ≥ 0.71; P < 0.05) between the observed and electronic values for the number of actor and reactor events, and the competition index. These results demonstrate the potential to quantify disruptive feed bunk replacement events in confined beef cattle using electronically collected feeding event data from the Vytelle SENSE feed intake measurement system.