Frondelius, L. | Search bioRxiv

bioRxiv2018

A Hidden Markov Model to Estimate the Feeding Time of Dairy Cows Based on Indoor Positioning Data

The feeding time of dairy cows is linked with the health status of the animal and can be used to estimate daily feed intake together with other measurements. The aim of this study was to develop a model to measure the time a dairy cow spends at a feed bunk using an Ultra wide-band indoor positioning system.\n\nWe measured the feeding behavior of 50 dairy cows during 7 days using Ubisense indoor positioning system and Insentec roughage feeders. We calculated the feeding (presence at the feeder) probability of the cow using logistic regression model with the distance to feed barrier as input and used the Viterbi algorithm to calculate the most likely state (feeding or not feeding) given state transition probabilities. The model was able to predict whether the cow was at the feeding trough or not with the accuracy of 97.6%, sensitivity 95.3% and specificity 97.9%. The model was also able to estimate the mean bout duration and the number of feeding bouts.

animal behavior and cognition↗

bioRxiv2017

Filtering Methods to Improve the Accuracy of Indoor Positioning Data for Dairy Cows

Several indoor positioning systems for livestock buildings have been developed to be used as a tool in automated animal welfare monitoring. In many environments the measurements from positioning systems still contain unwanted noise and the quality of the measurement data can be enhanced using filters.\n\nThe aim of this study was to develop an efficient filter for positioning data measured from dairy cows with UWB-based indoor positioning system in a free stall barn. We developed and tested a heuristic jump filter combined with median filter and extended Kalman filter. The performance of the filters were compared against reference data collected from Insentec Roughage intake feeders and scan sampling of animal presence in a specific lying stall with over 1500 reference observations from both methods.\n\nWe were able to improve the quality of the positioning data significantly using filtering. The 9th order median filter provided best estimates for cow position when the cows were not moving with median 100% of measurements located in correct stall and 84% in correct feeding trough when compared to the reference observations and measurements. The extended Kalman filter also improved the positioning accuracy significantly when compared to raw data and provides better of estimates of the trajectory of moving cows.

animal behavior and cognition↗