Rapid assessment of egg freshness by handheld NIR spectrometer coupled with Random Forest algorithm

Ha, Thi Thanh Nga and Mac, Xuan Hoa and Pham, Tung Thanh and Nguyen, Lien and Baranyai, László and Friedrich, László Ferenc and Tóth, Adrienn and Németh, Csaba and Nguyen, Duc Quang (2025) Rapid assessment of egg freshness by handheld NIR spectrometer coupled with Random Forest algorithm. PROGRESS IN AGRICULTURAL ENGINEERING SCIENCES, 21 (1). pp. 377-391. ISSN 1786-335X

Preview

Text 446-article-p377.pdf - Published Version Available under License Creative Commons Attribution. Download (1MB) | Preview

Abstract

This study evaluated the potential of a handheld near-infrared (NIR) spectrometer for rapid estimation of freshness of egg. Random Forest (RF) and Partial Least Squares (PLS) models were compared for Regression (R) and Classification (C) using various spectral pre-processing methods such as Savitzky-Golay smoothing (SGS), multiplicative scatter correction (MSC), first derivative (1D), and second derivative (2D). Pre-processing the spectra with multiplicative scatter correction (MSC) yielded the best results. RF-R outperformed PLS-R for predicting Haugh unit. MSC-RF-R with selected wavelengths showed a slight reduction in performance but offered an eight-fold decrease in computational time compared to full-spectrum model. Balancing accuracy and efficiency, MSC-RF-R based on selected wavelengths ( R 2 = 0.943, RMSE = 2.600) is recommended. For classifying eggs based on storage time, MSC-RF-C achieved higher performance (F1-score = 99.31%) than MSC-PLS-DA (F1-score = 95.90%). In classifying eggs by grade, MSC-RF-C (F1-score = 98.75%) performed comparably to MSC-PLS-DA (F1-score = 98.83%). These findings demonstrate that handheld NIR spectrometer combined with RF models is an effective tool for assessing egg freshness.

Actions (login required)

View Item