Fusion vs. Isolation: Evaluating the Performance of Multi-Sensor Integration for Meat Spoilage Prediction.

Area of Science:

• Food Science
• Analytical Chemistry
• Sensor Technology

Background:

• High-throughput, portable sensors offer rapid, non-invasive food quality and safety monitoring.
• Predictive models using sensor data and machine learning show variable accuracy based on food type and conditions.

Purpose of the Study:

• To explore multi-sensor fusion approaches for enhancing predictive accuracy in food spoilage detection.
• To integrate infrared spectroscopy and multispectral imaging for improved food quality assessment.

Main Methods:

• Utilized a fusion approach combining infrared spectroscopy and multispectral imaging data.
• Developed and validated predictive models for chicken and beef spoilage estimation (bacterial counts).
• Employed repeated nested cross-validation for robust out-of-sample performance evaluation.

Main Results:

• At least one fusion methodology outperformed single-sensor models in prediction accuracy across various food types and storage conditions.
• Observed performance improvements of up to 15% in aerobic, vacuum, and mixed aerobic/vacuum chicken spoilage scenarios.
• Demonstrated enhanced model robustness and cross-batch performance through the multi-sensor fusion approach.

Conclusions:

• Multi-sensor fusion significantly improves the accuracy and robustness of predictive models for food spoilage monitoring.
• This approach offers potential for enhanced real-world, minimally invasive food safety and quality assessment.
• The study highlights avenues for advancing sensor technology in food production and distribution.