Non-destructive measurement of eggshell strength using NIR spectroscopy and explainable artificial intelligence

Area of Science:

• Agricultural Science
• Spectroscopy
• Artificial Intelligence

Background:

• Eggshell strength is vital for egg quality and consumer satisfaction.
• Traditional eggshell strength testing is destructive, slow, and impractical for large-scale use.
• Developing non-destructive methods is crucial for the egg industry.

Purpose of the Study:

• To evaluate near-infrared (NIR) spectroscopy combined with explainable artificial intelligence (AI) for non-destructive eggshell strength prediction.
• To explore multivariate analysis techniques for improving prediction accuracy.
• To assess the interpretability of the AI model for practical application.

Main Methods:

• Near-infrared (NIR) spectroscopy was used to collect spectral data from eggshells.
• Various multivariate analysis techniques, including principal component analysis (PCA) and partial least squares discriminant analysis (PLSDA), were applied.
• Regression models such as random forest (RF) and gradient boosting were developed and validated.
• Shapley additive explanation (SHAP) was employed for model interpretability.

Main Results:

• NIR spectroscopy effectively classified eggs based on shell strength using PCA and PLSDA.
• The random forest (RF) model achieved high prediction accuracy (Rp² = 0.83) using only 14 spectral variables.
• The model demonstrated a low prediction error (RMSEP = 1.49 N) and a good prediction-to-deviation ratio (RPD = 2.44).
• SHAP analysis provided insights into the key spectral variables influencing eggshell strength predictions.

Conclusions:

• NIR spectroscopy integrated with explainable AI provides a robust, non-destructive method for eggshell strength prediction.
• This innovative approach is environmentally sustainable and suitable for industrial quality control.
• The findings support optimizing resource use and enhancing quality assurance in the egg industry.