Analyzing the rear shape of dairy cows in 3D to better assess body condition score
Résumé
Body condition is an important trait in dairy cow management, mainly
because it reflects the level and the use of body reserves and indirectly
reproductive and health performance. Body condition score (BCS),
which is done visually or by palpation, is the usual method on farm but
is subjective and not very sensitive. The aim was here to develop and
to validate 3DBCS which estimates BCS from 3D-shapes of dairy cows
rear, the body area commonly used to assess BCS. For the calibration, a
set of 57 3D-shapes from 56 Holstein cows with large BCS variability
(0.5 to 4.75 on a 0–5 scale) were transformed with a principal component
analysis (PCA). A multiple linear regression was fitted on the principal
components to assess BCS. Four anatomical landmarks were extracted
to normalize the 3D-shapes: the validation results of a manual labeling
proved the concept. Then an automated labeling method was developed
to extract them. Prior to the PCA, the 3D-shapes were either regularized
to fill in the holes or not regularized. External validation was evaluated
on 2 sets: one with cows used for calibration, but with a different
lactation stage (valididem) and one with cows not used for calibration
(validdiff). Repeatability was estimated with 6 cows scanned 8 times
each the same day. The automated method performed slightly better
than manual method for external validation (RMSE = 0.27 versus 0.34
for validdiff) and both were more repeatable than usual BCS (σ = 0.20
for 3DBCS and 0.28 for BCS). Surprisingly, regularizing the 3D-shapes
performed slightly less than without regularization. Nevertheless
regularization should be an interesting process before BCS assessing,
especially to avoid discarding too many 3D-shapes. The first results
of 3D-BCS monitoring in dairy cows with a fully automated method
show promising results in terms of phenotyping. The next step will try
to reduce scanning time to decrease the number of bad 3D-shapes due
to cow’s movement without losing too much resolution.
Domaines
Sciences du Vivant [q-bio]
Origine : Fichiers éditeurs autorisés sur une archive ouverte