Login / Signup

An Improved Approach to Automated Measurement of Body Condition Score in Dairy Cows Using a Three-Dimensional Camera System.

Rodrigo I AlbornozKhageswor GiriMurray C HannahWilliam J Wales
Published in: Animals : an open access journal from MDPI (2021)
Body condition scoring is a valuable tool used to assess the changes in subcutaneous tissue reserves of dairy cows throughout the lactation resulting from changes to management or nutritional interventions. A subjective visual method is typically used to assign a body condition score (BCS) to a cow following a standardized scale, but this method is subject to operator bias and is labor intensive, limiting the number of animals that can be scored and frequency of measurement. An automated three-dimensional body condition scoring camera system is commercially available (DeLaval Body Condition Scoring, BCS DeLaval International AB, Tumba, Sweden), but the reliability of the BCS data for research applications is still unknown, as the system's sensitivity to change in BCS over time within cows has yet to be investigated. The objective of this study was to evaluate the suitability of an automated body condition scoring system for dairy cows for research applications as an alternative to visual body condition scoring. Thirty-two multiparous Holstein-Friesian cows (9 ± 6.8 days in milk) were body condition scored visually by three trained staff weekly and automatically twice each day by the camera for at least 7 consecutive weeks. Measurements were performed in early lactation, when the greatest differences in BCS of a cow over the lactation are normally present, and changes in BCS occur rapidly compared with later stages, allowing for detectable changes in a short timeframe by each method. Two data sets were obtained from the automatic body condition scoring camera: (1) raw daily BCS camera values and (2) a refined data set obtained from the raw daily BCS camera data by fitting a robust smooth loess function to identify and remove outliers. Agreement, precision, and sensitivity properties of the three data sets (visual, raw, and refined camera BCS) were compared in terms of the weekly average for each cow. Sensitivity was estimated as the ratio of response to precision, providing an objective performance criterion for independent comparison of methods. The camera body condition scoring method, using raw or refined camera data, performed better on this criterion compared with the visual method. Sensitivities of the raw BCS camera method, the refined BCS camera method, and the visual BCS method for changes in weekly mean score were 3.6, 6.2, and 1.7, respectively. To detect a change in BCS of an animal, assuming a decline of about 0.2 BCS (1-8 scale) per month, as was observed on average in this experiment, it would take around 44 days with the visual method, 21 days with the raw camera method, or 12 days with the refined camera method. This represents an increased capacity of both camera methods to detect changes in BCS over time compared with the visual method, which improved further when raw camera data were refined as per our proposed method. We recommend the use of the proposed refinement of the camera's daily BCS data for research applications.
Keyphrases
  • dairy cows
  • high speed
  • convolutional neural network
  • electronic health record
  • big data
  • deep learning
  • physical activity
  • high throughput
  • preterm birth