Search for a command to run...
The objectives of this study were to develop and evaluate a novel algorithm for image extraction of structural conformation traits and estimate variance components among skeletal conformation, growth, and herd retention traits. An Intel RealSense D435i camera was used to obtain left-side-view RGB images on individual purebred Duroc pigs (n = 846) at 156 d of age. Frames were selected by a trained swine evaluator when either the left front leg (n = 1056), left back leg (n = 888), or both left legs (n = 728) were present in the field of view and the respective foot pads from toe to heel were in contact with the ground. Selected images were processed through Apple Inc's image segmentation algorithm to extract the pig from the background. Segmented pig images were then processed through a novel algorithm developed in this study. The algorithm identified the leg and estimated 21 skeletal conformation traits from each leg. Steps for user intervention were added to assist the algorithm in identifying which leg(s) were present and the general location of each leg to increase the accuracy of leg identification and trait acquisition. The algorithm correctly identified at least one front and one back leg from an image for 99.9% and 98.0% of the pigs, respectively. Heritability estimates ranged from 0.01 to 0.33 for all conformation traits with the quadratic term for the curvature of the anterior side of the front and the height of the back leg having the highest heritability for each location (h2 = 0.33 and 0.30, respectively). Genetic correlations among image feet and leg conformation traits and production traits (finishing average daily gain, weight per day of age, and finishing feed efficiency) ranged from -0.37 to 0.19. Boars that remained in the breeding herd for longer than 200 d tended (P = 0.08) to have greater curvature of the front leg and lower (P = 0.07) angularity between the midpoint of the foot and the anterior point of the pastern and had significantly (P = 0.03) shorter distance between the pastern and the top of the shoulder than those that were removed prior to 200 d. Gilts that remained in the breeding herd for longer than 200 d tended (P = 0.08) to have less curvature of the back leg. The current study presents an algorithm that extracts novel, objective structural conformation traits and reports corresponding genetic and phenotypic parameters.