Predicting carcass cut yields in cattle from digital images using artificial intelligence.

Area of Science:

• Agricultural Science
• Computer Science
• Animal Science

Background:

• Deep Learning (DL) excels in image classification but hasn't been widely applied to carcass analysis.
• Accurate prediction of carcass cut yields is crucial for the meat industry.

Purpose of the Study:

• To train DL models for predicting carcass cut yields.
• To compare DL model performance against standard machine learning (ML) methods.
• To evaluate different data inputs, including phenotypic data, carcass images, and carcass dimensions.

Main Methods:

• Three approaches were used: ML with phenotypic data, DL (Convolutional Neural Networks) with carcass images, and ML with carcass dimensions plus phenotypic data.
• The study utilized large datasets of 54,598 and 69,246 animals for predicting Grilling and Roasting cuts, respectively.

Main Results:

• Deep Learning models can be successfully trained to predict carcass cut yields.
• An approach utilizing carcass dimensions as features in ML algorithms demonstrated slightly superior predictive performance in absolute terms compared to DL models using only images.

Conclusions:

• While DL shows potential for carcass yield prediction from images, integrating directly measured carcass dimensions into ML models currently offers a marginal improvement in accuracy.
• Further research could explore hybrid approaches combining image-derived features with dimensional data for enhanced prediction accuracy.