Assessment of a non-invasive acoustic sensor for detecting cattle urination events

Abstract

Excretion of urinary nitrogen (N) from cattle is the main source of N leaching loss from grazed pasture systems. The purpose of this study was to evaluate the suitability of non-invasive acoustic sensors to detect the time and duration of urination events from grazing cattle. The acoustic sensor identified the spectral acoustic patterns with a urination event like spectral signature and had an F1 = 0.88 in the validation study (the F1 statistic denotes the harmonic mean of precision (0.94) and sensitivity (0.83)). The acoustic prediction of urination event duration had an RMSE = 1.64 s in the validation study. The ensemble of audio recordings could also be assessed by an expert observer to provide the urination event time, event duration and the cow that generated the event with similar accuracy to the visual observation method. The urination surface type (pasture, concrete) could be detected based on the acoustic recordings (F1 = 0.93), which provides information on the environmental impact of events. The between-cow variation in urination frequency and duration has a coefficient of variation of 15–25%, which demonstrates the opportunity to identify and manage cows with increased urination frequency, lower volume per urination event, lower N load per event and reduced N leaching. • Acoustic sensors were attached to cows to detect urination events. • The acoustic urination classifier had good performance with an F1 = 0.88. • The acoustic prediction of urination event duration had an RMSE = 1.64 s. • Urination surface (pasture, concrete) can be detected from the acoustic recordings. • Therewas a between cow variation in urination frequency and duration of 15–25%.