BBSNet: An Intelligent Grading Method for Pork Freshness Based on Few-Shot Learning.

Area of Science:

• Artificial Intelligence
• Food Science
• Computer Vision

Background:

• Deep learning for pork freshness grading demands extensive datasets, hindering practical, cost-effective applications.
• High data collection costs limit the widespread adoption of automated pork quality assessment.

Purpose of the Study:

• To develop a lightweight few-shot learning model (BBSNet) for accurate pork freshness classification with limited image data.
• To address the challenge of data scarcity in automated food quality monitoring systems.

Main Methods:

• Proposed BBSNet, a novel architecture incorporating Batch Channel Normalization (BCN) for improved feature distinctiveness.
• Utilized BiFormer for efficient and fine-grained feature extraction.
• Trained and evaluated on a dataset of 600 pork images, assessing freshness via microbial cell concentration.

Main Results:

• Achieved high average accuracy of 96.36% in a challenging 5-way 80-shot learning task.
• Demonstrated the model's effectiveness in maintaining accuracy despite limited training data.
• Validated BBSNet as a viable solution for cost-effective, real-time pork quality monitoring.

Conclusions:

• BBSNet significantly reduces data requirements for pork freshness grading, offering a practical solution for industrial applications.
• The developed framework bridges laboratory freshness indicators with industrial needs under data-scarce conditions.
• Future work includes extending BBSNet to other food types and optimizing for portable device deployment.