Nondestructive egg freshness assessment using hyperspectral imaging and deep learning with distance correlation...

Area of Science:

• Food Science
• Spectroscopy
• Machine Learning

Background:

• Traditional egg freshness assessment using Haugh units is destructive and time-consuming.
• There is a need for rapid, non-destructive methods for evaluating egg quality.

Purpose of the Study:

• To investigate the efficacy of hyperspectral imaging (450-1100 nm) for nondestructive egg freshness evaluation.
• To identify optimal wavelengths for freshness prediction using distance correlation.
• To compare the performance of various regression models for egg freshness assessment.

Main Methods:

• Hyperspectral imaging data acquisition (450-1100 nm).
• Preprocessing using standard normal variates (SNV).
• Wavelength selection using distance correlation.
• Development and comparison of regression models: CNN, GBT, MLR, PLSR, SVR.

Main Results:

• The convolutional neural network (CNN) model combined with distance correlation yielded the highest accuracy (R=0.9056, RMSE=4.4152).
• This approach outperformed other models and common wavelength selection techniques.
• Generated pseudocolor maps effectively visualized egg freshness distribution.

Conclusions:

• Hyperspectral imaging, particularly with CNN and distance correlation, provides an effective non-destructive method for egg freshness assessment.
• This technique offers a promising alternative to conventional destructive methods.
• The study highlights the potential of distance correlation in hyperspectral data analysis for food quality evaluation.