Nondestructive Determinations of Texture and Quality of Preserved Egg Gel by Hyperspectral Method
Chen Yuan-zhe,Wang Qiao-hua,Tian Wen-qiang,Xu Bu-yun,Hu Jian-chao
DOI: https://doi.org/10.3964/j.issn.1000-0593(2022)06-1985-08
2022-01-01
Spectroscopy and spectral analysis
Abstract:As an important quality parameter, texture can significantly affect the gelatin quality of preserved eggs. There is no effectively rapid detection method at present. In this study, hyperspectral imaging technology was used to predict preserved eggs' textural characteristics and classify different qualities. Hyperspectral data of high-quality eggs, qualified eggs and unqualified eggs were collected. The original spectra were transformed by single and combined transformation to analyze the correction between one-dimensional spectral and textural hardness and springiness. It was found that the spectral reflectance after CR-FD transformation was most correlated with the hardness and springiness of the gel texture, and the maximum values were 0. 882, 0. 86 5 at 683, 715 nm, respectively; The hardness and springiness were taken as disturbance factors to explore the optimal research area of the hardness and springiness, the results showed that : When hardness was used as disturbance factor, autocorrelation peaks existed at 476, 539, 647, 672, 728 and 851 nm, Spectral signals at 483, 572, 657, 739 and 826 nm were more sensitive to springiness value. Therefore, two sensitive bands, 476 similar to 851 and 483 similar to 826 nm, were finally selected as the study regions for gel hardness and springiness, respectively. Comparing five different variable selection methods(SPA, CARS, GA, PSO and UVE) , it was found that PSO-PLSR model had the highest detection accuracy : theR(p)(2) and RMSEP for predicted hardness were 0. 826 and 0. 874 with an RPD of 2. TheR(p)(2) and RMSEP for predicted springiness were 0. 886 and 0. 402 with an RPD of 1. 9. Three different classifiers were used to predict preserved eggs, and the classification accuracy of high-quality eggs, qualified eggs and unqualified eggs reached 97% , 92% and 100% , respectively. The accuracy and generalization ability of the PLS-DA model were better than the BP and RF model based on the confusion matrix and ROC curves of prediction results. In conclusion, the hyperspectral technique can be used to predict preserved eggs' textural characteristics and classify different quality of preserved eggs.