Assessment of a non-invasive accelerometer for detecting cattle urination and defecation events

Abstract

Nitrogen (N) excreted in the urine from cattle is the primary source of N leaching loss from grazed pasture systems. The objective of this study was to use non-invasive accelerometer sensors to detect the time and duration of urination and defecation events from grazing cattle. Two accelerometer sensors were attached to cows to detect the back arching of cows during urination and defecation events. Trials were conducted under outdoor grazing conditions in autumn and summer with a total of 160 urination recorded events. Changes in tilt angle of 8–22.5 degrees were observed during most urination and defecation events. Ensemble machine learning models to predict urination and defecation events were developed based on 18 selected features of a cow back arching event. The accelerometer urination event classifier had an F1 statistic of 0.80 (harmonic mean of precision (0.82) and sensitivity (0.78)) and the defecation event classifier had an F1 statistic of 0.73 (based on 408 h of data obtained 9a.m. to 3p.m. from 19 cows). Models to predict urination event duration were significant but had low performance partly due to the moderate association between the duration of back arching and the duration of the urination event (R2 of 0.29 and 0.47 for the autumn and summer trials, respectively). This demonstrates the potential for non-invasive accelerometer sensors to identify the urination characteristics of individual cows and provide individual farm information that could be used as an input parameter in farm decision support tools (e.g. urine-N excretion) and farm management decisions (e.g. culling/breeding).