Non-destructive detection of milk nutritional components based on hyperspectral imaging

Area of Science:

• Food Science and Technology
• Spectroscopy
• Machine Learning

Background:

• Growing consumer demand for safe, nutritious dairy products necessitates rapid quality assessment.
• Hyperspectral imaging (HSI) offers non-destructive analysis but faces challenges with high-dimensional data and multi-component analysis.
• Existing methods often focus on single nutrient prediction, limiting comprehensive milk quality evaluation.

Purpose of the Study:

• To develop a rapid, non-destructive method for simultaneous detection of fat, protein, and lactose in milk.
• To optimize hyperspectral data analysis for improved band selection and multi-target regression.
• To enhance the accuracy and reliability of milk nutritional component detection.

Main Methods:

• Integration of moving average smoothing and first derivative (MA-FD) preprocessing for HSI data.
• Application of an improved coati optimization algorithm (ICOA) for efficient band selection.
• Utilizing the CatBoost model for accurate multi-target regression of nutritional components.

Main Results:

• Achieved high prediction accuracy on calibration and prediction sets (MultiR² of 0.9992 and 0.9797, respectively).
• Demonstrated excellent individual component prediction (R² values for fat, protein, lactose: 0.9658, 0.9910, 0.9825).
• The proposed method showed robust predictive accuracy and reliability in milk quality assessment.

Conclusions:

• The combined MA-FD, ICOA, and CatBoost approach provides a reliable, non-destructive solution for milk quality assessment.
• This method enables simultaneous detection of key nutritional components, supporting dairy industry quality control.
• The technology holds substantial potential for broader applications in food quality assessment and consumer health protection.