Thermal Imaging Technology and Computer Vision Models for Rapid in-Vivo Evaluation of Carcass Composition in Growing-Finishing Pigs

The Canadian hog market pays producers based on pork carcass merit (i.e., carcass leanness percentage), which is assessed postmortem. However, selecting market hogs is an antemortem process based on animal weight and conformation, increasing the likelihood of selecting excessively lean or overly conditioned animals for slaughter. This study aimed to utilize thermal imaging with computer vision models (CVM) to predict carcass traits and composition in live animals. Crossbred Large White × Landrace barrows and gilts (n = 243; body weight ≈ 122 kg) were used in the experiment, and 1 dorsal image per pig (640 × 512 pixels) was captured with a thermal camera (between 5–15 μm; long-wavelength infrared) 3 d before slaughter. After slaughtering, hot-carcass evaluations included lean and fat depth measurements and leanness percentage by using an electronic grading probe (Destron) on the left side carcasses. At 24 h postmortem, chilled carcasses were assessed for leanness using dual-energy x-ray absorptiometry (DEXA) technology. Images were segmented, and then CVM were trained using data augmentation. The models performed poorly (R2 ≤ 0.09) in predicting individual traits (i.e., total lean). However, when a model was trained to classify the images based on lean-grade index (> 109 scores; carcasses between 57.7–64.2% of leanness and 80–105 kg of hot-carcass weight), a moderate performance was obtained based on DEXA lean yield (F1 score = 0.73). These results suggest that thermal imaging could enable producers to select and market their hogs based on the best grid grade.