Combining Feature Selection Techniques and Neurofuzzy Systems for the Prediction of Total Viable Counts in Beef...

Area of Science:

• Food Science
• Microbiology
• Image Processing
• Machine Learning

Background:

• Food quality and safety are critical for consumer health.
• Non-invasive sensory techniques are increasingly important for rapid quality assessment.
• Multispectral imaging is a promising technology for food product inspection.

Purpose of the Study:

• To develop a stacking-based ensemble prediction system for total viable counts of microorganisms in beef fillet.
• To utilize multispectral imaging data for predicting meat spoilage.
• To explore a feature fusion approach combining wavelengths from various feature selection techniques.

Main Methods:

• Development of a stacking-based ensemble prediction system.
• Utilizing multispectral imaging for data acquisition.
• Feature fusion approach combining wavelengths from multiple feature selection techniques.
• Employing clustering-based neuro-fuzzy network prediction models (average reflectance and standard deviation of pixel intensity).
• Comparison with regression algorithms: multilayer perceptron, support vector machines, and partial least squares.

Main Results:

• The proposed system successfully predicted total viable counts of microorganisms in beef fillet.
• The feature fusion approach enhanced prediction accuracy.
• Neuro-fuzzy models demonstrated strong performance in assessing microbiological quality.
• The ensemble system outperformed traditional regression algorithms in this application.

Conclusions:

• A combination of feature selection methods and neuro-fuzzy models is effective for assessing meat microbiological quality.
• Multispectral imaging, coupled with advanced machine learning, offers a viable non-invasive method for food safety monitoring.
• The developed system provides a valuable tool for the food industry to ensure product quality and safety.