Discrimination of origin of eggs using visible-near-infrared spectroscopy and random forest

In order to study the feasibility of rapid and nondestructive identification of the origin of eggs.Visible-near-infrared spectroscopy was used to obtain the spectral transmittance of eggs ( 500⁹00 nm) , which were the origin of four areas of Hubei.Combining with centering, normalization, standard normal variables, Savitzky-Golay smoothing filter, multiple scatter correction and the direct orthogonal signal correction method were to preprocessing the spectral data, the t-distributed stochastic neighbor embedding ( t-SNE) and PCA were to reduce the dimension of the preprocessed data.The processed data was transmitted as the input of extreme learning machine and random forest.The models were established to discriminate the origin of eggs.Compared the two methods, the RF model based on DOSC and t-SNE obtained the best results. The correct rates of calibration and prediction were 100%, 98.33% respectively.The results showed that the visible-near-infrared spectroscopy technology could be competent for the discrimination of the origin of eggs, which would provide a technical support for further research and development of portable devices for the discrimination of the origin of eggs.