Image-Based Machine Learning for Predicting Acceptability Limits in Frozen Pizza Shelf Life

Area of Science:

• Food Science
• Computer Science
• Sensory Science

Background:

• Consumer acceptability of frozen foods is crucial for shelf life determination.
• Quantifying the effects of storage duration and temperature fluctuations on food perception is complex.
• Tomato sauce degradation is a key visual indicator of frozen pizza quality decline.

Purpose of the Study:

• To develop a non-destructive, image-based method for estimating frozen pizza acceptability.
• To utilize machine learning to predict product quality decay based on visual cues.
• To identify tomato sauce saturation as a reliable indicator for assessing frozen food quality.

Main Methods:

• An image processing pipeline was created to isolate tomato sauce regions.
• Color extraction, specifically saturation in the HSV color space, was performed on sauce samples.
• A polynomial regression model tracked saturation trends, and a logistic regression classifier predicted consumer acceptability using saturation and storage duration.

Main Results:

• The regression model showed good performance with R² of 0.68 and RMSE of 12.8.
• The logistic regression classifier achieved high accuracy (88.2%) and AUC (0.93) in predicting acceptability.
• Tomato sauce saturation was confirmed as a primary driver of visual rejection by consumers (90% feedback).

Conclusions:

• The developed framework offers an early, non-invasive estimation of frozen food acceptability.
• This approach has significant potential for practical application in the frozen food industry's shelf life studies.
• Image-based analysis combined with machine learning provides a robust tool for quality assessment of frozen products.