Detection of poultry egg freshness by dielectric spectroscopy and machine learning techniques

Abstract

The physicochemical changes that occur in poultry egg during storage, make a reduction in its quality. The present research investigates the possibility of the nondestructive classification and quality inspection of eggs using dielectric detection technique in the range of radio frequency. Several machine learning (ML) techniques were developed for freshness detection including artificial neural networks (ANN), Bayesian networks (BNs), decision trees (DTs) and support vector machines (SVMs). Among ANNs, the ANN with topology of 62-18-6 gave a perfect capability to predict the class of freshness for all samples with accuracy of 100%. Also all types of BNs represented excellent results with Kappa statistic of 1 and overall accuracy of 100%. From developed SVMs, the SVM with polynomial kernel function gave the best results with Kappa value of 1 and accuracy of 100%. Among DTs, LMT tree had the highest Kappa value (0.846) and the highest accuracy (87.5%) compared to other DT approaches. Different ML methods were used to predict air cell height. Among ANNs the 24-12-1 structure with R value of 0.817, among DTs the MSP tree with R value of 0.906 and among SVMs the RBF form with R value of 0.920 had the highest value of correlation coefficient and the lowest standard error of 0.452.