Improving Traceability and Quality Control in the Red-Meat Industry Through Computer Vision-Driven Physical Meat Feature Tracking
This paper explores the application of CMOS imaging and computer vision technologies to enhance traceability and quality control in the red meat industry. We propose a novel computer vision-based traceability pipeline using internal, immutable meat features, achieving an F-1 score of 0.9942 in sample-to-sample tracing and 0.9479 in sample-to-database tracing, significantly increasing resistance to fraud. Based on the EfficientNet model, our method enables highly confident verification of provenance details, such as diet (90.90% accuracy) and breed (91.23% accuracy). It also improves product quality assessment through objective measures like marbling score prediction, achieving 96.24% top-1±1 and 99.57% top-1±2 accuracy. To support this research, we introduce a new comprehensive image dataset containing 38,528 high-resolution beef images taken from 602 steaks, specifically designed for traceability and quality control of beef. This innovative approach offers robust protection against product substitution, empowers end-users to authenticate product credentials, and enhances objective quality reporting. Ultimately, the proposed technology and dataset aim to strengthen consumer trust and integrity within the red-meat supply chain.
This repository will contain the source code and documentation for the computer vision-based traceability pipeline described in the paper. It will include:
- Implementation of the EfficientNet-based model
- Scripts for sample-to-sample and sample-to-database tracing
- Algorithms for provenance verification (diet and breed identification)
- Marbling score prediction model
- Dataset preprocessing and augmentation scripts
- Evaluation metrics and results
The source code and additional documentation for this project will be uploaded shortly. Please check back for updates.
For any questions or inquiries about this research, please contact Qiyu Liao at qiyu.liao@csiro.au.