Development of a non-invasive system for monitoring dairy cattle sleep

Tuesday, July 22, 2014
Exhibit Hall AB (Kansas City Convention Center)
Jenna M Klefot , University of Kentucky, Lexington, KY
Jordan L Murphy , University of Kentucky, Lexington, KY
Kevin D Donohue , University of Kentucky, Lexington, KY
Bruce F O'Hara , University of Kentucky, Lexington, KY
Mike E Lhamon , University of Kentucky, Lexington, KY
Jeffrey M Bewley , University of Kentucky, Lexington, KY
Abstract Text:

Lack of sleep in dairy cattle may indicate shortcomings in housing, environment, or increased physiological disturbances. Little research has been conducted to assess sleep in production livestock, primarily because of limitations with monitoring abilities.  Consequently, biological understanding of the production circumstances and facility options that affect sleep is limited.  The objective of this study was to test a non-invasive system using a three-axis accelerometer monitor to measure head position of the cow in order to classify sleep, and wake behaviors. The duration of the study consisted of two 24-hour periods of observing 4 Holstein dairy cows in September 2013 at the University of Kentucky Coldstream Dairy. The three-axis accelerometers were attached to a harness on the side of each cow’s neck to determine head and body movement.  Human observation of the animals noted the times of active behaviors and very low activity, or sleep behaviors. Wake behaviors were classified as standing and alert. Sleep was classified with the behaviors of lying with no movement and eyes closed with head rested on the ground or flank. The radial signal was extracted from the xyz components of the accelerometer to obtain a motion signal independent of direction.  Radial signal features were examined for maximizing the performance of detecting sleep behavior using a Fishers linear discriminant analysis (LDA) classifier. This study included a total of 652 minutes of high activity behaviors, and 107 minutes of sleep behavior recorded from 2 cows with usable data.  Results from a bootstrapping analysis show an agreement between human observation and the LDA classifier of 93.7 ± 0.7% for wake behavior and 92.2 ± 0.8% for sleep behavior, with a 95% confidence interval.  This monitor may be used to help understand options for monitoring sleep in research and production settings.

Keywords: behavior, sleep, accelerometer