| Apple(Malus pumila)is consistently and deeply getting the favor of people due to its rich nutritional values.However,big differences among the internal quality indices of apple which can easily result in the interest loss of producer,processor and consumer and many other problems.For example,the brand benefit of producer can be damaged,the high price for good quality cannot be achieved.The processing quality of processor can hardly be guaranteed and the marketing standard of product can be affected.The purchasing desire of consumer can be reduced.With the rising living standard of people,more and more people have started pursuing high quality,safe and reliable product of apple,therefore,there is a need to do the detection research of apple quality.As a relatively mature detection method,near infrared spectroscopy has been widely applied in the detection study of agricultural products.This technique can meet the needs of detection on the condition of nondestructive or even non-contact occasions.But,concerning the problems,such as optic transmission mechanism of internal part of postharvest apple,correction or compensation method for online detection model under different influencing factors and optimization of online discriminant model for internal hidden defects,are still short of systematical analysis and research.In this study,apple was selected as research object,its spectral online detection method and mechanism were mainly studied.The specific objectives are to:(1)design a SRS acquisition system to study the spectral information of different parts of apple tissues,and collect the spatially resolved signals of different distances between detector and light source,use this signal to study the prediction of SSC and firmness in apple;(2)establish a diffuse reflectance online detection system,combine SNR and ACR to analyze,and study the effect of different spectral detection orientations on SSC prediction models for apple;(3)use a full transmission online detection system,combine with speed compensation methods,and study the effect of different convey speeds on SSC prediction models for apple;(4)use a transmission online detection system,propose and compare fruit diameter correction methods,and study the effect of different fruit diameter sizes on SSC prediction models for apple;(5)utilize a full transmission online inspection system,combine with the spectral information of sample middle part and characteristic wavelength range,study the effect of spectral information of moldy apple under different detection orientations on discriminant correct rate.Main results are as follows:(1)A multi-fiber based SRS system was employed to study the spectra distribution law of different parts in sunlit and shaded sides of apple,and its SSC and firmness were effectively determined.Results showed that the obtained CARS-PLS model using reference-free processing method could achieve ideal prediction result for both SSC and firmness in apple.The established local prediction model of 4 different parts could obtain better results of SSC and firmness than its corresponding global model.And it was helpful for model accuracy to some extent when using the absorption and scattering properties of spectral information under the individual or combined utilization of different distances(1,2,3,4 mm)between detector and light source.(2)A Vis/NIR diffuse reflectance mode was applied to study the effect of apple detection orientation on online detection of its SSC,the relative compensation models were established for optimization.Results showed that the fruit orientation T1 achieved the best performance in the self-designed system.PLS model based on spectra after SGS pretreatment and LS-SVM model based on spectra after 2-Der pretreatment could achieve good results of SSC prediction towards 6 orientations(T1,T2,T3,T4,T5 and T6).The detection orientation had obvious effects on prediction results,while T4 of local model showed poor prediction performance which might due to its high ACR.Global model based on spectra of all 6 orientations had higher prediction accuracy.When using CARS-SPA-PLS model based on diffuse reflectance spectra to predict samples in prediction set of all orientations,their rp and RMSEP were 0.815,0.818,0.837,0.731,0.807,0.842 and 0.487,0.484,0.460,0.573,0.497,0.453 °Brix,respectively.This illustrated that the proposed global model could eliminate the interferences of spectral detection orientation variation on SSC prediction performance to some extent,which was helpful to the study of all-orientation online detection for fruit SSC.(3)A full-transmittance online detection system was used to study the effect of fruit moving speed on sample SSC prediction and its corresponding compensation method.Results showed that model based on global speed had better robustness and higher prediction accuracy than local speed model.The global-P13 model based on 36 EWs after SGS pretreatment of all speeds could all obtain good results(rp = 0.8419,0.8895,0.8948 and RMSEP = 0.6281,0.5318,0.5196 °Brix).The combined spectra pf P13 could also improve the prediction accuracy of model to some extent.This was beneficial to the practical application of transmittance online detection of fruit SSC under all speeds,which could effectively reduce the effect of speed variation on model accuracy.(4)A full-transmittance online detection system was utilized to study the effect of fruit diameter size on sample SSC prediction and its different diameter correction method.Results showed that fruit diameter correction method based on diameter attenuation coefficient(M3),whose model results were obviously better than method only using spectra information(M1)and method fused spectra and fruit diameter information method(M2),was proposed.The established CARS-PLS model after M3 correction under T1 in710-980 nm could achieve the best result(rp = 0.92、RMSEP = 0.50 °Brix),which was helpful to the transmittance online detection of fruit SSC with multi-diameter information,and effectively corrected the effect of fruit diameter differences on model accuracy.(5)A full-transmittance online detection system was employed to study the effect of different inspection orientations on online discriminant model for moldy apple.The corresponding discriminant model were established combined with LDA,SVM and KNN,Using the mean spectra of middle part in 700-720 nm,SVM model combined with PCA for dimension reduction after BOC pretreatment could achieve the best results.Furthermore,model based on mean spectra of middle part was more beneficial to promote the discriminant rates for moldy apple sample,which illustrated that the middle part of sample spectra could provide more effective information than whole sample spectra for moldy sample determination to some extent.While model based on spectra in 700-720 nm made the model variable more concise,and at the same time,the discriminant rates could be also promoted to some extent,which also meant higher efficiency for model processing,it was helpful to the application of full-transmittance online discrimination for moldy apple sample. |
PDF Full Text Request