Research On Modeling Of Soluble Solids Content In Apple Based On Near Infrared Spectroscopy

China is the largest country in producing fruit and has abundant fruit resources.However,the technology of fruit post-harvest treatment and the reprocessing level are low.In addition,the fruit testing and sorting methods are relatively backward.The technical defect of fruit processing affects the quality and sales of fruits,thereby reducing the competitiveness in the international market.In recent years,near-infrared spectroscopy?NIRS?technology has been widely used in the rapid non-destructive testing of the internal quality of agricultural products,due to its advantages such as rapid,effective,and pollution-free.Therefore,this paper takes apple as the research object,studies the nondestructive detection of apple based on NIRS and quantitative modeling methods from the following four aspects:The near infrared spectroscopy characteristics of apple,the influencing factors of the non-destructive detection accuracy of soluble solids content?SSC?,the detection parameters of near-infrared spectrometer and the rapid non-destructive detection mathematical model of soluble solid content.The main work of this paper is as follows:?1?The effects of some instrument parameters such as image resolution,scanning times and sample accessories of Fourier near-infrared spectrometer on near infrared spectral response characteristics of soluble solids content in apples are researched by experimental analysis methods.The requirements of scanning speed,signal-to-noise ratio and model prediction accuracy are comprehensively analyzed.Furthermore,the optimum combination of instrument parameters should be as follows:The sample module is integrating sphere diffuse reflection module;The number of scanning times is 64;The image resolution of the instrument is 8cm^-1;The gain is 4;The number of scanning point is 1557.Relational experiments show that the matching parameters can meet the accuracy requirements of mathematical model for non-destructive testing of soluble solids content in apples.?2?The near-infrared spectral response characteristics of apples are researched,and it is occurred that there is no significant difference in the position of absorption peaks between different samples.However,different spectral acquisition locations have some impacts on the spectral response characteristics,while there is no significant difference in the near-infrared spectra of different equatorial locations,of which the correlation coefficient of the established quantitative model is high and the error of the prediction root mean square is small.Besides,the absorbency of fruit stalks and bottom side is generally higher,and the correlation coefficient of the established quantitative model is relatively low as well as the prediction error is large.The effect of the absorbency of near-infrared spectra and soluble solids content of different surface colors is not obvious,so the soluble solids content can be accurately predicted.?3?Four mathematical models,which contain principal component regression?PCR?,partial least squares?PLS?,artificial neural network?ANN?and support vector machine?SVM?,are established by combining different spectral preprocessing methods.The prediction accuracy and application scope of each model is analyzed and compared.The predictive accuracy of the predictive models obtained by different modeling algorithms has some differences.According to the experiments,PLS and BP neural network are effective methods in non-destructive testing of soluble solids content in apples.When the principal factor of the PLS is 9,the predictive correlation coefficient and root mean square error are0.924 and 0.475 respectively,which have anti-interference and non-linearity capabilities.When the principal factor of BP neural network is 16,the prediction correlation coefficient and root mean square error are 0.935 and 0.474 respectively.The application of SVM in the grading of apples is studied,and after optimizing the parameters of radial basis function by particle swarm optimization?PSO?,the classification accuracy reaches to 92.3%.The prediction accuracy of the mathematical model is generally satisfactory,indicating that there is a good correlation between NIRS of apples and soluble solids content.