Detection Of Apple Watercore Disorder And SSC Based On Vis/NIR Transmission Spectroscopy Technology

Apple is one of the main fruits varieties planted in China and it plays an important role in the fruit industry.Watercore is a common internal physiological disorder of apples,which has a great impact on apples storage.In addition,watercore apples has a unique taste,also known as‘Honeyed apples’,which is deeply loved by the people.Apple soluble solids content(SSC)is an important index to measure the internal quality of apple,which affects consumers’purchase intention.Therefore,it is of great significance to realize the on-line rapid detection of apple watercore disorder and SSC.Taking Aksu apple as the research object,this thesis studies the on-line detection method of apple watercore disorder and SSC by visible near infrared(Vis/NIR)transmission spectroscopy.The specific research and conclusions are as follows:Based on the Vis/NIR transmission spectroscopy with the spectral range of 680-1000 nm,the effects of detection speeds and sample orientations on apple watercore disorder detection were studied,the optimal detection speed and orientation were determined,and the watercore degree was evaluated online.According to 9 groups of spectral data collected from three detection speeds(S1=0.3 m/s,S2=0.5 m/s,S3=0.8 m/s)and three detection orientations(O1:Apple stem-calyx axis vertical,stem upward;O2:Apple stem-calyx axis horizontal,stem towards light source;O3:Apple stem-calyx axis horizontal and parallel to the moving direction of the conveyor belt,calyx in front),the qualitative discrimination model of apple watercore disorder was established based on least squares support vector machine(LS-SVM)and partial least squares discriminant analysis(PLS-DA).Through comparative analysis,S2-O3 was determined as the best detection speed and orientation.Based on the spectral data of S2-O3,Monte Carlo uninformative variable elimination(MC-UVE)combined successive projections algorithm(SPA)was used to select the effective wavelengths for the qualitative discrimination of apple watercore disorder and the quantitative prediction of watercore degrees.For the qualitative discrimination of apple watercore disorder,the MC-UVE-SPA-LS-SVM and MC-UVE-SPA-PLS-DA was established based on 8 effective wavelengths have the same classification accuracy,and the recognition accuracy of prediction set was 98.48%.In order to predict the watercore degree of apples,based on the proportion of watercore area obtained by image processing of the cross-sectional picture of watercore apple and the spectral data of S2-O3 watercore apple,the prediction models of watercore degrees of partial least squares(PLS),multiple linear regression(MLR)and LS-SVM were established respectively.The results showed that MC-UVE-SPA-LS-SVM model has the best prediction effect,and its R_P and RMSEP are 0.93 and 2.12%,respectively.The discrimination model of apple watercore disorder was further simplified by analysis of variance(ANOVA).Firstly,the spectral data of S2-O3 are analyzed by ANOVA,and the local maximum value of F value was selected to determine the wavelength variable with the most significant difference between groups.Then,based on the determined two wavelength variables,the LS-SVM watercore apple discrimination model was established,and the detection accuracy of the prediction set was 98.48%.To verify the effectiveness of the ANOVA method,the spectral data of S2-O1 and S2-O2 were processed in the same way as S2-O3.The LS-SVM watercore apple discrimination model was developed based on the two corresponding wavelength variables,the detection accuracy of its prediction set was 93.94%and 89.39%,respectively.In addition,the band ratio of S2-O3 spectral data was analyzed by ANOVA,and the band ratio with the most significant difference between groups was determined from the maximum F value.The threshold discrimination model based on the optimal band ratio has a detection accuracy of 98.48%for the prediction set.At the same time,the spectral data of S2-O1 and S2-O2 were processed by using the same method.The detection accuracy of the prediction set was 95.45%and 89.39%,respectively.Based on the Vis/NIR transmission spectroscopy,the influence of detection orientation on the on-line detection of apple SSC was studied.A global orientation model integrating three detection orientation spectral data was established and optimized.Based on the spectral data of S2-O1,S2-O2 and S2-O3,the prediction models of single local orientation and global orientation apple SSC were established and compared by PLS and LS-SVM algorithms.The results show that the global orientation apple SSC prediction model has good orientation adaptability.Compared with PLS model,the LS-SVM model has higher accuracy.Based on the spectral data preprocessed by SGS combined MSC,combined with 39 effective wavelengths selected by MC-UVE and bootstrapping soft shrinkage(BOSS),the MC-UVE-BOSS-LS-SVM global orientation apple SSC prediction model was established.The R_P and RMSEP for the three detection orientations(O1,O2 and O3)were 0.879,0.838,0.851 and 0.632,0.755,0.782°Brix,respectively.