| As a non-destructive detection tool,near infrared spectroscopy(NIRS)technology has been intensively researched and widely used in real-life applications.However,the post-production processing technology of China’s fruit is relatively backward,and the export volume of apples is disproportionate to the planted area,which lacks competitiveness in the international market.NIR spectroscopy can achieve rapid grading of fruit quality,improve the quality and scale of China’s fruit exports and enhance international competitiveness.However,NIR spectroscopy is easily affected by the temperature of the sample,and the model established at a specific temperature may not be ideal for the detection of samples with temperature variations,which limits the application of NIR technology and is an urgent problem to be solved.In this thesis,the effect of temperature difference on the detection model of soluble solids content of apples and the calibration method were investigated to achieve rapid and high accuracy detection of soluble solids content of apples,using the fruit quality dynamic online detection device developed by the laboratory.The main results and conclusions of the study are as follows.1、The effect of temperature differences on the NIR spectra and SSC detection models of apples was investigated.The spectra of the same batch of samples at six different temperatures were collected by a fruit quality dynamic online detection device developed in the laboratory and the SSC was measured.The effect of temperature on the spectral characteristics of the samples was investigated and a partial least squares discriminant analysis(PLS-DA)model was developed to better analyse the spectral differences of the samples at different temperatures.In addition,the partial least squares regression(PLSR)method was used to develop an apple SSC model.A global temperature model was developed based on the temperature compensation method to reduce the effect of temperature on the results.The results showed that the spectra of the same apple at different temperatures had significant differences,and the overall classification of the spectra at different temperatures was as high as 89.9%correct,which could better distinguish the spectra at different temperatures,and the sample temperature had a significant effect on the spectra.The model at different temperatures has better predictive performance when the samples are at the same temperature.The model built by collecting spectra at 20°C has better prediction results.The root mean square error of the local temperature model is larger than that of the single temperature model.This indicates that the local model is sensitive to temperature fluctuations and lacks the ability to accurately predict the results.As the temperature deviation increases,the root mean square error increases.The global temperature model has good predictive capability with a reduced root mean square error model and higher prediction accuracy.The global temperature model contains a larger sample of temperatures,reducing the effect of temperature fluctuations,and also has redundant information and uses a variable selection algorithm to filter variables to simplify the model.One of the competing adaptive reweighting algorithms(CARS)was able to effectively remove redundant information and the model performance was improved with an increase in R_P from 0.904 to 0.917 and a reduction in RMSEP from 0.645°Brix to 0.600°Brix,resulting in an improvement in prediction accuracy.2、A study on the temperature correction of apple SSC models based on NIR spectroscopy.To further investigate the effect of temperature,spectra collected at 20°C were used as the modelling set and spectra of multiple batches of samples at different temperatures were collected as the prediction set to build the apple SSC model.The use of commonly used pre-treatment methods could not eliminate the effect of temperature,and the global temperature compensation model was also adapted to remove the effect of temperature in multiple batches of samples.The model accuracy was improved but still could not meet the actual sorting requirements,and other methods to eliminate the effect of temperature on the apple SSC model needed to be further investigated.Three spectral calibration methods,orthogonal signal correction(OSC),generalised least squares weighting(GLSW)and external parameter orthogonalisation(EPO),were used to calibrate the apple spectra.The results showed that the prediction accuracy of the models was significantly improved after correction,with the best model accuracy after EPO correction,with R_P and RMSEP reaching 0.947 and0.489°Brix,respectively,which indicated that the external parameter orthogonalisation method played a good role in temperature correction. |
PDF Full Text Request