Development Of Spectroscopic Non-Destructive Detection Method And Device For Protein Content And Other Qualities Of Eggs During Storage

The evaluation and prediction of egg quality during storage and its relationship with storage time and conditions have been one of the hot issues in the field of food processing and preservation.At present,nondestructive detection methods seldom consider the edible quality and safety of eggs from the perspective of changes in specific protein content.If the variation of egg quality can be interpreted by the biochemical changes in the composition of eggs,that is,to find the most essential characteristics that cause the change of egg quality and to establish the external performance method of the that characteristics.this is of great theoretical significance to further reveal the change mechanism of egg quality and achieve the effective non-destructive monitoring of egg quality.From the perspective of the biochemical changes in the composition of eggs,this project took the S-ovalbumin content,ovalbumin content,Huff unit,yolk index and other quality indicators as the research objects,and used biochemical methods,hyperspectral imaging technology and visible near-infrared spectroscopy technology to study the multiple quality indicators of eggs.The main research contents and conclusions are as follows:1)Variation and correlation of egg quality and main specific protein content during storageThe changes of egg quality index,S-ovalbumin content and ovalbumin content during storage were statistically analyzed by the traditional biochemical testing method.The correlation between the index values of Huff unit,yolk index and p H value and the content of S-ovalbumin and ovalbumin was investigated respectively.It was found that the correlation between each quality parameter and S-ovalbumin content was higher than that with ovalbumin content..The gray correlation degree of each egg quality parameter with S-ovalbumin content and ovalbumin content was analyzed respectively.The comprehensive gray correlation degree of each quality index with S-ovalbumin content and ovalbumin content was all greater than 0.5.An equivalent egg age prediction model was established with S-ovalbumin content or ovalbumin content as independent variables.The determination coefficients of the model were both greater than 0.9（p≤0.01）.Taking the results of correlation analysis into consideration,the content of S-ovalbumin was selected as the research object to further study the correlation between S-ovalbumin content and HU,yolk index of two different varieties of eggs（eggs with blue brown shell eggs and roman pink shell）.It was found that the HU and yolk index of different varieties of eggs were significantly negatively correlated with S-ovalbumin content,the correlation coefficient between S-ovalbumin content of pink-shelled eggs and S-ovalbumin content of brown-shelled eggs was 0.950,there was a significant correlation between them（p≤0.01）.That is,under the same storage condition,the variation of S-ovalbumin content in different egg varieties was less affected by the egg varieties.Therefore,S-ovalbumin content was determined to be the most essential characteristic factor that caused the change of egg quality.2)Non-destructive detection of egg freshness,p H and viscosity based on hyperspectral imaging technologyHyperspectral imaging system was used to collect transmission spectrum of eggs during storage,and p H meter and viscometer were used to measure the p H and viscosity of eggs.It was found that there was a strong correlation between egg freshness,p H and viscosity.The prepossessing results showed that the PLS model based on the full band after the first order differential treatment has the best prediction performance for each quality index.The competitive adaptive reweighted sampling algorithm（CARS）and the successive projections algorithm（SPA）were used to select the characteristic wavelength.The partial least squares regression model and the multiple linear regression model established based on the feature bands screened by CARS and SPA were compared and analyzed.The MLR models established had better prediction performance for the values of HU,p H and viscosity.Based on the characteristic wavelength collection extracted by CARS,SPA was used to further screen the characteristic wavelength.The optimal number of screened band for HU,p H and viscosity based on CARS-SPA were 13,10 and 6,respectively.The determination coefficient（R_p²）of the prediction set of HU,p H and viscosity of the MLR models was 0.884,0.903 and 0.903 respectively,and their RPDs were all greater than 2.0,suggesting that it was reliable for the in assessment of HU,p H and viscosity.3)Nondestructive detection and visualization of S-ovalbumin content based on hyperspectral imaging technologyVisible-near-infrared（300-1100nm）hyperspectral imaging technology was used to collect the image of eggs during storage.Seven image parameters were extracted from the corrected hyperspectral image,and three image feature variables(mean value of red component(R_avg),fractal dimension（D）and minor axis（n）)were screened out by using the correlation coefficient method.The PLS models were established and it was found that the modeling performance remains unchanged after excluding the other four image parameters.The image of whole egg was taken as the region of interest,and the average spectrum of each egg was extracted.Bands in the range of 450nm-1000nm were selected.After the first-order differential and smoothing pretreatment,the characteristic wavelengths closely related to the S-ovalbumin content were extracted by CARS,SPA and the two-dimensional correlation synchronous spectroscopy（2D）.The PLS model and the BP neural network optimized by genetic algorithm model（GA-BP）were established respectively by using the selected characteristic wavelengths.The results showed that GA-BP model based on 20 characteristic wavelengths screened by SPA had the best prediction performance of S-ovalbumin content,for the model,its R_c²was 0.857,RMSEC was 0.084,R_p²was 0.806,RMSEP was 0.120,RPD was 2.012.The three image feature parameters were fused with the feature wavelength screened by the above three wavelength selecting methods respectively,after that,principal component analysis（PCA）was used to reduce dimensionality,and prediction models were established based on the features after fusion.It was found that the GA-BP model established by 5 principal components obtained by fusion of 3 image features and 14 bands screened by CARS had better prediction performance,and its R_c²was 0.856,RMSEC was 0.084,R_p²was 0.845,RMSEP was 0.143,RPD was 1.918.The optimized model was used to compute each pixel of the image,visualization of S-ovalbumin content distribution in the egg was obtained using pseudo-color image.4)Nondestructive detection of egg freshness indexes and S-ovalbumin content based on visible-near infrared spectroscopyBy using the self-built visible near-infrared spectroscopy system,the transmission spectral of eggs with blue brown shell and roman pink shell in the storage period were collected respectively,and HU,yolk index and S-ovalbumin content were determined.451 bands in the range of 500nm-950nm were selected to analyze.After spectral preprocessing,the feature wavelength was screened by uninformative variable elimination algorithm（UVE）,genetic algorithm（GA）and stepwise regression algorithm（STP）,and the prediction models of partial least squares regression（PLS）,gaussian process regression（GPR）,multiple linear regression（MLR）and support vector machine regression（SVM）were established respectively.For eggs with blue-brown shell,the results showed that the best prediction model for HU was the GPR model based on the 10characteristic wavelengths screened by UVE,its R_c²was 0.981,RMSEC was 0.031,R_p²was 0.708,RMSEP was 10.825,RPD was 1.603.The best prediction model for yolk index is the PLS model based on 18 feature wavelengths selected by GA,its R²were0.792,0.770 for calibration set and validation set,respectively.RMSE was 0.021 for calibration model and 0.030 for validation mode,and RPD was 1.765.The best prediction model for S-ovalbumin content was the PLS regression model based on 19 wavelengths screened by GA,its R_c²was 0.919,RMSEC was 0.058,R_p²was 0.917,RMSEP was 0.079,RPD was 3.236.For eggs with roman pink shell,the optimal prediction model for HU is the MLR model based on 15 feature wavelengths screened by STP,its R_c²was 0.926,RMSEC was 5.380,R_p²was 0.765,RMSEP was 10.416,RPD was 2.322.None of the established models could predict the yolk index of eggs with roman pink shell（all RPDs were less than 1.5）.The optimal prediction model of S-ovalbumin content is the GPR model based on the 9 characteristic wavelengths selected by STP,its R_c²was 0.954,RMSEC was 0.047,R_p²was 0.846,RMSEP was 0.109,RPD was 2.223.The prediction model of HU,yolk index and S-ovalbumin content of single egg varieties has laid a foundation for the development of a general quantitative and rapid detection model of egg freshness indexes and S-ovalbumin content.5)Prediction model Maintenance of freshness indexes and S-ovalbumin content of different varieties of eggThe prediction model of HU,yolk index and S-ovalbumin content of a single variety of egg was used to predict the corresponding quality index of another variety.It was found that the prediction of the same index of another variety by the model established by a single variety was poor.The average spectrum and the spatial distribution of principal components of the two varieties of eggs were compared and analyzed.It was found that the average spectra of the two varieties of eggs were not significantly different in the near infrared region while there were obvious differences in the visible region.The spatial distribution of principal components of two spectral matrix showed that the spatial distribution of principal components of two egg varieties cannot completely cover each other,that is,the applicability of the prediction model of single egg variety was poor.Three algorithms,global update,direct correction and slope/intercept correction,were used to maintain the detection model of HU,yolk index and S-ovalbumin content of two varieties of eggs respectively.By comparing the results of different model maintenance methods,it was found that the three model maintenance methods can improve the prediction performance of the best model in different degrees.Among them,the general model maintained by global update has the best effect on the prediction performance.The performance of the general prediction model of HU based on the global update method for the blue brown shell eggs was 0.770 for R_p²,9.063 for RMSEP and 2.017 for RPD,and for roman pink eggs,R_p²was 0.834,RMSEP was 8.753,RPD was 2.231.For the yolk index,the performance of the general prediction model of yolk index for blue brown shell eggs was 0.780 for R_p²,0.029 for RMSEP and 1.837 for RPD,and for roman pink eggs,R_p²was 0.684,RMSEP was 0.038,RPD was 1.555.For S-ovalbumin content,the performance of the general prediction model of S-ovalbumin content for blue brown shell eggs was 0.936 for R_p²,0.069 for RMSEP,3.649 for RPD,for roman pink eggs,R_p²was0.839,RMSEP was 0.112,RPD was 2.035.The optimized models were more stable,reliable and accurate,and they can be applied to detect HU,yolk index and S-ovoalbumin content of two varieties at the same time,it laid a foundation for developing a non-destructively detection device for egg freshness indexed and S-ovoalbumin content.6)Development of nondestructive detection device for egg freshness indexes and S-ovalbumin contentThe overall design of device for rapid detection of egg freshness indexes and S-ovoalbumin content was carried out by Using USB2000+spectrometer.Appropriate key components,including adjustable light source,heat dissipation device,spectrum acquisition unit,etc.,were selected and assembled and debugged.According to the established nondestructive testing model of HU,yolk index and S-ovalbumin content,Qt software was used to develop a software of rapid detection of egg freshness indexed and S-ovalbumin content.After verification,HU,yolk index and S-ovalbumin content can be non-destructively detected by using the hardware and the software developed.