Big data and Graphic Processing Unit (GPU) accelerated data science for high-throughput carcass and pork quality phenotyping

Project 21-06 - Dr. Stephanie Lam

Dr. Stephanie Lam, Agriculture and Agri-Food Canada - Lacombe

Current methods used for measuring pork meat quality and carcass traits are disadvantaged due to their high-cost, large footprint, invasive sampling, low-throughput, or susceptibility to human error suggesting a need for alternative tools to perform accurate measures of desirable traits in the pork industry. The application of Graphic Processing Unit (GPU) accelerated data science has led to improved methods for data analysis in multiple fields, leading to innovative solutions that were not possible a few years ago. Data analytics includes ‘machine learning’, which uses a variety of techniques to construct models on data. Machine learning techniques include ‘deep learning’, which automates this process; however, it can require significant computing power (GPUs) and large amounts of data to train deep neural networks to effectively perform human-level tasks and surpass human-level accuracies. Using this technology and existing large pork phenomics databases and image sets, a computer vision system will be developed and tested to measure pork primal cut quality traits, including meat colour and fat and lean content at a high throughput level. The system and data analysis models will be tested and adjusted to optimize performance and use for research and potentially in-line production. This project will expand the current pork phenomics database and identify relationships between desirable meat quality traits, serving as a platform for precision farming. This will result in innovative solutions to improve pork meat quality research and production efficiency, bringing the Canadian Pork industry to the forefront of research and sustainability.