Research On The Quality Inspection Of Whole Egg Powder Based On Near Infrared Spectroscopy

Compared with fresh eggs,egg powder has the advantages of long storage period,convenient use and easy packaging and transportation,and has been widely used in food industry,livestock and poultry production,cosmetic and medical fields.However,the domestic egg powder standard system is not yet perfect,and egg powder quality detection technology is not yet mature.In order to obtain higher profits,some egg powder producers use cheap substances instead of high-quality egg powder.This not only disturbs the order of the egg powder market,but also brings great potential danger to food safety.Therefore,researching egg powder quality detection technology is of great significance for ensuring egg powder quality and safety.In this study,different brands of pure egg powders were selected as research subjects.Near-infrared diffuse reflectance spectroscopy data of whole egg powder samples in the4000 cm^-1-10000 cm^-1（1000nm-2500nm）wavelength range were collected.By analyzing the raw spectral data,comparing multiple pre-processing methods and variables selection methods,respectively,identification models for whole egg powder,concentration prediction models,and major physicochemical index prediction models were established.Based on these models,a quality detection software for whole egg powder was developed.The main findings of this study are as follows:（1）The model for authenticity identification was established.Soybean isolate protein,maltodextrin and wheat starch were used as adulterants,and egg powder samples were prepared with adulteration concentrations ranging from 0.1%to 60%.The NIR spectral data of adulterated egg powder and pure egg powder were collected,and after comparing different classification models,different processing methods and feature wavelength screening methods,a qualitative discrimination model of adulterated whole egg powder was established.The results showed that the optimal discrimination model was Ensemble Learning（EL）,the optimal pre-processing method was the first derivative,the optimal feature screening method was CARS.The overall discrimination accuracy of the optimal model for adulterated egg powder reached 98.18%,the adulteration identification rate reached 98.67%.The test time for a single whole egg powder sample was 0.48s.The detection accuracy of one-component adulteration,two-component adulteration and three-component adulteration egg powder was 98.33%,96.67%and 97.33%.The detection rate of adulteration was 100.00%,97.78%and 98.33%.The study shows that the FD-CARS-EL model can achieve qualitative discrimination of whole egg powder adulteration and meet the demand of online detection.（2）The prediction model of whole egg powder adulteration concentration was established.Based on the qualitative identification of egg powder adulteration,multiple preprocessing methods and variable screening methods were compared,and the PLSR regression prediction model（Partial Least Square Regression,PLSR）was developed.The results showed that the optimal pretreatment method was MSC and the optimal feature screening method was CARS.The Rp of one-component adulteration,two-component adulteration,three-component adulteration and all adulterated samples detected by the optimal model were 0.9585,0.9312,0.9456 and 0.9558.The RMSEP was 4.6891,5.8134and RMSEP was 4.6891,5.8134,4.6041 and 3.8029.The RPD was 3.5076,2.7434,3.0738 and 3.4012.The prediction model can better predict the adulteration concentration of whole egg powder.（3）The prediction model of the main physicochemical indexes were established.After comparing different classification models,different pretreatment methods and characteristic wavelength screening methods,quantitative prediction models for the main physicochemical indexes（protein,fat and moisture）were established.The results showed that the best prediction model for all three physicochemical indexes was PLSR,the best prediction treatment for the protein prediction model was first order derivative,the best feature variable screening method was CARS,the test set correlation coefficient Rp of PLSR model was 0.9915,the root mean square error RMSEP was 0.7499,and the RPD was 7.6860;the best prediction treatment for the fat prediction model The best prediction processing method for the fat prediction model was the standard normal transformation,and the best feature variable screening method was CARS,with the PLSR model test set correlation coefficient Rp was 0.9947,root mean square error RMSEP was 1.0533,and RPD was 9.7258;the best prediction processing method for the moisture prediction model was the standard normal transformation,and the best feature variable screening method was CARS,with the PLSR model test set correlation coefficient Rp was 0.9348,the root mean square error RMSEP was 0.4631 and RPD is 2.8155.The model for predicting the main physical and chemical indexes showed good performance.（4）Development of software for whole egg powder quality testing.Based on the above research,the adulteration identification model,quantitative adulteration content prediction model and main physicochemical index content prediction model of whole egg powder were implanted into the software,and the whole egg powder quality testing software was developed using MATLAB 2017a.The software contains the functions of spectral data reading and display,authenticity detection and main physicochemical index prediction,and realizes the visual operation of the whole egg powder quality testing function.The study shows the feasibility of near-infrared spectroscopy for testing the quality of whole egg powder.This study provides technical support and theoretical basis for developing a low-cost and high-efficiency whole egg powder quality testing system and provides a reference for regulatory agencies to test the quality of whole egg powder.