Interpretive Summary: Comparison of data analytics strategies in computer vision systems to predict pig body composition traits from 3D images
By Anne Zinn
Computer vision systems have been shown to be a helpful tool for the measurement of live pig body weight without causing the animal stress, and, as precision farming continues to advance, it is now possible to evaluate growth performance of individual pigs much more accurately. Despite these advances, some traits, such as muscle and fat deposition, can still only be evaluated using ultrasound, typically computed tomography or through dual-energy x-ray absorptiometry. Therefore, a recent study published in the Journal of Animal Science aimed to develop a computer vision system for prediction of live body weight, muscle depth, and back fat from top view 3D images of finishing pigs and to compare the predictive ability of different approaches and machine learning techniques, such as deep learning. The datasets evaluated, supplied by the Pig Improvement Company, contained over 12,000 images from 557 finishing pigs and were split into training and testing sets using a 5-fold cross-validation technique.
The study concluded that it is possible to successfully predict body weight, muscle depth, and back fat via computer vision systems on a fully automated setting using 3D images collected in farm conditions without the need for preprocessing images. Specifically, it was demonstrated that a deep learning model using the raw 3D images as input provided higher prediction accuracy compared with the other methods evaluated. These findings will play a key role in optimizing data analysis workflows in computer vision systems, but there is still room for improvement regarding the predictions of muscle depth and back fat and additional research in this area is warranted.
The full paper will be available on the Journal of Animal Science website.