Research On Internal Quality Inspection Of Juicy Peach Based On Near Infrared Spectroscopy And Data Fusion

In this study,near infrared spectroscopy combined with chemometrics methods to quantitatively analyze content of soluble solids of peach produced in Wenzhou.However,the predictive results was poor when only a single instrument are used.To overcome this problem and improve the model performance,we explored the fusion of spectral data from different sources.The main contents and research conclusions are as follows:First,taking the peach in Wenzhou area as the experimental object,the flame-NIR spectrometer and the USB2000~+near-infrared spectroscopy detection system were used to collect the NIR data of 270 peach samples.Second,in order to improve the detection accuracy,this study eliminated the outliers in flame-NIR and USB2000~+spectral data during the calculation process,and then a variety of pre-processing and variable selection methods were used to establish a quantitative detection model of peach soluble solids.Among them,the best model of flame-NIR spectral data GA-PLS,the correlation coefficient of the prediction set reaches 0.855,and the root mean square error of prediction(RMSEP)is 0.909°Brix;as for USB2000~+,the best model is the smoothed full-range PLS.The correlation coefficient of the prediction set is 0.772,and the root-mean-square error is 1.253°Brix.Based on the above results,it can be found that both NIR spectrometers can achieve quantitative detection of peach soluble solids,but there is a significant difference in their predictive capabilities.Third,in order to obtain a better detection model,this study used data fusion strategy to establish two fusion models for the spectral data collected by flame-NIR and USB2000~+spectrometers.One is simple data fusion.The best result is the GA-PLS fusion model,whose the correlation coefficient of the prediction set is 0.853 and the root mean square error of the prediction set is 0.908°Brix.Compared with single detection model of flame-NIR and USB2000~+,the value of the root mean square error of the prediction set were reduced by 0.1%and 34.5%respectively.The other is complex data fusion.The best result is the CARS-PLS model.The correlation coefficient of the prediction set is 0.934,and the root mean square error is 0.87°Brix.Compared with single detection model of flame-NIR and USB2000~+,the value of RMSEP reduced by 3.9%and 38.3%respectively.The results show that data fusion improves the reliability of the model and obtains better prediction results.Therefore,two data fusion methods can be used to optimize the internal quality detection model of peach by using the richer spectral information of different instruments.