Research Of Nondestructive Detection For Red Non-soft Peach Based On Hyperspectral Imaging

Fresh red non-soft peach is rich in vitamin C,protein,organic acid and other nutrients.It also has medicinal value and is a fruit that people must buy in summer.However,with the gradual improvement of living standards,people’s demand for taste and internal quality has gradually increased.In this paper,the red non-soft peach in Shanxi Province was used as the object,and the rapid non-destructive detection of red non-soft peach was studied by hyperspectral imaging technology.This experiment proposes a red-pure peach detection model based on hyperspectral technology,which provides a theoretical basis for the development of por Tab.equipment for red non-soft peach.Main research contents and the conclusion are as follows:(1)The spectral information of 300 red non-soft peachs was preprocessed by preprocessing methods such as SNV,MSC,SG(3,1),SG(3,2),baseline,first-order derivative and second-order derivative.The results show that the first derivative preproccssing results was the best,Rc~2 and Rp~2 were 0.9383 and 0.9321,respectively.The difference between Rc~2 and Rp~2 was the smallest,and the Correct root mean square error(RMSEC)and prediction root mean square error(PMSEP)were 0.2148 and 0.2262,respectively(2)In the discrimination of the origin of the red not-soft peach,full-band-PLS,full-band-LS-SVM,full-band-ELM,the PCA-PLS model,PCA-LS-SVM,PCA-ELM MC-UVE-PLS,MC-UVE-LS-SVM,MC-UVE-ELM modeling were established and compared.The results show that the accuracy of the ELM model based on principal component analysis(PCA)was highest 98.7%,which can realize the correct discrimination of red non-soft peach in different producing areas.(3)In the SSC quantitative detection study of red non-soft peach,8 abnormal samples deviating from the overall mean and variance were excluded by Monte Carlo sampling,and the remaining 104 spectral data were used for quantitative detection of SSC.Then,the data of104 samples were corrected by different pre-processing methods.It is found that the model based on MSC combined with second derivative(2Der)predicts the best performance.Finally,the characteristic wavelength were extracted by the SPA,and the principal component were extracts by PCA algorithm.And full-band-PLS,full-band-LS-SVM,SPA-PLS,SPA-LS-SVM,PCA-PLS,PCA-LS-SVM quantitative detection model were established.The results show that the PLS model based on SPA extracted feature wavelength has the highest prediction accuracy.The prediction coefficient of the SSC model is 0.8921,the root mean square error of the prediction set is 0.6451,and the model input is only 8 variables,which greatly improves the model operation speed.