Deep learning-based image segmentation for predicting hot carcass weight in tropical beef cattle.

The application of computer vision and deep learning in the meat processing industry enables automated carcass evaluation. This study aimed to develop and validate a deep learning-based pipeline for automatic carcass segmentation and prediction of hot carcass weight (HCW) in tropical beef cattle. A total of 598 RGB images of bovine half-carcasses were collected under commercial slaughterhouse conditions and manually annotated to delineate carcass boundaries. For segmentation, a YOLOv11 model was trained. From the segmented images, geometric and shape descriptors were extracted and subsequently used in a LASSO regression model to predict HCW. A strong segmentation performance was achieved, with an Intersection over Union (IoU) of 0.92 and a Precision of 0.98. For HCW prediction, the model achieved R² = 0.84 and MAPE = 5.77%. The integration of deep learning–based segmentation with regularized regression provides a practical and scalable approach for carcass evaluation. The combination of computer vision and statistical learning enables real-time, accurate prediction of beef carcass weight.