Files
Download Full Text (1.2 MB)
Date of Graduation
5-2026
Description
Beef is one of the most important sources of animal protein and plays a crucial role in daily diets to support muscle growth, energy production and overall health. Historically, each beef carcass was evaluated by 3-5 skilled United States Department of Agriculture (USDA) meat graders, which are still considered as the gold standard for grading. Since 2009, electronic imaging grading systems have been approved by the USDA and have been gradually adopted in the grading facilities. Accurate and consistent beef grading plays a critical role in maintaining the quality, market value, and consumer trust in U.S. beef products. The USDA beef grading system, primarily based on fat marbling characteristics, has long served as the industry standard for classifying beef quality. However, traditional grading methods relying on human visual inspection have been prone to variability among graders and inconsistencies across processing plants. To address these challenges, this study developed a deep learning based solution to predict USDA beef grades by analyzing fat marbling patterns in carcass images collected from multiple processing facilities. The dataset provided by the USDA, which includes expert-graded images of beef carcasses from seven processing plants, provides numerical scores on a range of 0-1500, which was used to train convolutional neural networks (CNNs) such as ResNet, VGG, and Inception. Additionally, this study proposes a unified unsupervised domain adaptation regression method, which was employed to enhance the models’ ability to generalize across different environments. The dataset reflected real-world variability, such as differences in lighting, equipment, and operational practices, ensuring the models could adapt to diverse plant conditions. By correlating image-based features with USDA grades, the models successfully identified intricate marbling patterns and reduced variability compared to human graders. Statistical analyses showed that the deep learning based approach provided more uniform results across plants, addressing the inconsistencies of traditional methods. In comparison with commonly utilized generic CNN deep learning models, the newly proposed unsupervised domain adaptation model achieves a significantly better marbling score prediction performance in different processing facilities. This study not only automated the beef grading process but also introduced a scalable and efficient solution for the industry. The integration of domain adaptation ensured robust performance in varying plant environments, making the system practical for widespread adoption. Beyond reducing labor costs and improving grading accuracy, this research contributed to a fairer system for producers by further standardizing grading criteria across facilities. By bridging advanced technology with traditional practices, this study provides a framework for policy makers to regulate and approve deep learning based beef grading protocol into commonplace beef grading practices.
Publication Date
2025
Document Type
Book
Degree Name
Bachelor of Science in Biological Engineering
Degree Level
Undergraduate
Department
Biological and Agricultural Engineering
Advisor/Mentor
Wang, Dongyi
Disciplines
Engineering
Keywords
Engineering
Citation
Vinson, S., & Wang, D. (2025). Cross-Facility Reliable Deep Learning Based Beef Marbling Assessment Via Unsupervised Domain Adaptation Regression. 2025 Research Poster Competition. Retrieved from https://scholarworks.uark.edu/hnrcsturpc25/19
