Leveraging pre-trained computer vision models for accurate classification of meat freshness

Area of Science:

• Food Science
• Computer Science
• Artificial Intelligence

Background:

• Growing consumer demand for safe and high-quality food necessitates reliable methods for assessing meat freshness.
• Traditional methods for meat freshness assessment are often time-consuming, destructive, or subjective.
• Deep learning offers potential for rapid, non-destructive analysis of food properties.

Purpose of the Study:

• To develop and evaluate a novel, efficient, and non-destructive method for classifying meat freshness using image analysis.
• To leverage pre-trained deep convolutional neural networks (DCNNs) and random encoding of aggregated deep activation maps (RADAM) for feature extraction.
• To assess the performance of machine learning (ML) classifiers trained on these extracted features.

Main Methods:

• Utilized pre-trained DCNNs to extract deep features from meat images.
• Applied RADAM for encoding these deep features.
• Trained traditional ML classifiers using the RADAM-encoded features.
• Validated the approach on three distinct datasets for beef and chicken meat freshness.

Main Results:

• Achieved state-of-the-art classification performance, with accuracy metrics ranging from 93% to 100%.
• Demonstrated superior performance compared to existing methods reported in the literature.
• The methodology proved to be simpler and more efficient than current approaches.

Conclusions:

• The proposed image-based approach using DCNNs and RADAM is a highly effective and accurate method for meat freshness classification.
• This technique offers a practical and efficient solution for real-world applications in the food industry.
• The findings support the potential for widespread industry deployment to enhance food quality and safety monitoring.