2025/11/07 by Magadray, Juliette, Saintilan, Romain, Brenaut, Pauline
#Apprentissage profond #Deep Learning #classification #imagerie #imaging #maturité néonatale #neonatal maturity #selection #survie #survival #sélection
paper · doi:10.17180/ciag-2025-vol107-art22
The improvement made on sow prolificity led to a rise in litter heterogeneity. Immature piglets are born with a delay in intra-uterine growth which makes them more susceptible to death by hypothermia, starving or crushing. The Pic’Let study aimed at developing an unbiased maturity phenotyping tool that can be implemented easily in the routine procedure of the breeder. To that end, a phenotyping stand taking pictures of the head (front and side) of the piglet was made and tested during birth weighing. With these prototypes 9,620 piglets representing three maturity levels were photographed. Two algorithms were trained based on 1 509 pictures of the front and side. The piglet maturity level was deduced from the results of both models. When tested on a set of pictures from a new farm, the right level was found in 76 % of the cases. From all the phenotypes collected, the heritability of the litter proportion of immatures piglets was estimated in Landrace (0,25 ± 0,11) and Large White (0,13 ± 0,07). These results could lead to genetic selection based on an automatically acquired trait. However, more data needs to be collected.