| China is a large fruit producing country with a wide variety of fruit cultivars.However,even for the same cultivar,the quality may vary in producing area;and even for the same producing area,the quality varies in cultivar differences.Therefore,identifying fruit producing area and cultivar are major challenge in the post-harvest fruit processing industry.Existing studies showed that NIR spectroscopy had the ability to identify fruit producing area and cultivar,but the spectra were a macroscopic representation of photons being absorbed and scattered in the transmission of light in fruit tissues.However,currently,it is unclear how exactly the differences in fruit producing area and cultivar affect the optical properties,and it is also unclear the reason that NIR spectroscopy can nondestructively detect producing area and cultivar.Hence,in this research,‘Fuji’(Changfu No.2)apples from three different producing area(Yangling,Luochuan and Jingning)and three different cultivars(‘Qinguan’,‘Fuji’,and‘Ruiyang’)apples from the same producing area were used as samples to measure their light absorption coefficients μa and scattering coefficients μs’ in the wavelength of 950-1650 nm using the single integrating sphere technology.The major internal qualities of the fruit,including soluble solids content(SSC),moisture content(MC),firmness(FI)and peel color were measured using traditional methods,the microstructure of the flesh tissue was also observed,the effects of producing area and cultivar on the optical parameters,internal qualities and microstructure of the fruit were analyzed.The reasons for the differences in optical parameters from the perspectives of quality and microstructure were analyzed,and the relationship between optical characteristics and internal qualities was also analyzed.Based on this,the model was developed to predict the internal qualities of apples and identify different producing areas and cultivars based on the spectra of optical parameter.Furthermore,five different types of spectra were used as research objects,and different pre-processing methods were used to process the original spectra,and partial least squares regression(PLSR)and convolutional neural network(CNN)regression models were then established to predict apple SSC,respectively,to explore the differences in prediction performance between deep learning models and traditional models under different conditions,and to analyze the spectra suitable for predicting apple SSC.The main research contents and conclusions are as follows:(1)A single integrating sphere system was constructed for measuring optical parameters of fruits and the reliability of the system was verified in the wavelength range of 950-1650 nm.The results showed that in the range of 950-1650 nm,the average relative error between the measured value of μa and the reference value was 8.23 %,and the average relative error of the measured value of μs’ was 3.71 %,which indicated that the single integrating sphere system built in this study could accurately measure the optical parameters.(2)The effects of producing area differences on the internal qualities,optical parameters and the relationship between optical parameters and internal qualities of apples during the storage period of 70 d after harvest were investigated,and a linear discriminant analysis(LDA)model of apple producing area was developed.The results showed that the SSC of apples of all three producing area decreased with the increase of storage time.The SSC of apples of all three producing area were significantly different at each storage time except for 50 d of storage.At the beginning and end of storage,apple flesh cells underwent wrinkling and deformation,and the degree of cell structure breakage was higher than that at the beginning of storage.The flesh tissues of Luochuan apples were more compact than those of Yangling and Jingning apples,but the cell sizes were relatively uneven.All apple flesh μa had three distinct absorption peaks in the 950-1650 nm band,located at 985 nm,1200 nm and 1430 nm,respectively.At the 0 d and 70 d of storage,the pulp μa values of Jingning apples were the highest and those of Luochuan apples were the lowest.In the wavelength range of 950-1650 nm,the pulp μa showed negative correlation with SSC and positive correlation with MC.The LDA model based on μa spectra showed the highest accuracy in classifying apple producing area,with98.87 % accuracy in the calibration set and 86.88 % accuracy in the prediction set.The PLSR model based on μa spectra had the best prediction performance for SSC and MC,and the PLSR model based on μs’ spectra had the best prediction performance for FI.(3)The influence of apple cultivar differences on their optical characteristics and their relationship with internal quality was investigated.The results showed that the internal quality of the three apple varieties tended to change in the same way,i.e.,SSC,MC and FI decreased gradually with increasing storage time,but their values and rates of change varied from cultivar to cultivar.There were some inter-varietal differences in the color parameters of the pulp of the three apple cultivars,and there was a greater potential for identifying apple cultivar based on color parameters.The absorption peaks at 980 nm,1192 nm and 1425 nm were observed for all varieties of apple flesh μa,and the values of pulp μa of ‘Qinguan’ apple at 1and 36 d after harvest were larger than those of the other two cultivars.The pulp μa values of‘Qinguan’ apple were significantly higher than those of ‘Fuji’ and ‘Ruiyang’ apples at 1 and 6d of storage.The apple pulp of the three cultivars showed a large varietal difference in optical parameters in general.The pulp μa values of ‘Qinguan’ apple correlated best with SSC and MC before 1425 nm,and the ‘Fuji’ apple had the best Rμa-SSC and Rμa-MC in the wavelength range of 1425 nm-1650 nm.The pulp μa values of three apple cultivars showed significant differences in correlation with internal qualities.The LDA model based on μa spectra had the best classification accuracy for apple cultivars with 94.22 % accuracy in the calibration set and 78.70 % accuracy in the prediction set.(4)The differences between traditional linear models and deep learning models in predicting the internal quality of apples for different spectra under different preprocessing methods were investigated.The results showed that for the diffuse reflectance spectra of intact apple,the PLSR model established after S-G+MSC processing had the best predictive performance(Rp = 0.96,RMSEP = 0.54 %).The performance of CNN models based on μs’ spectra or original total diffuse transmission spectra was better than that of PLSR models.Except for a few preprocessing methods,spectral preprocessing reduced the prediction performance of the CNN models for SSC to different degrees,indicating that CNN had powerful competence to process original spectra than the traditional PLSR.The results showed that the producing area and cultivar of apples indeed affected their physicochemical properties and optical parameters,which in turn affected the correlation between optical parameters and internal qualities,and that producing area identification and quality prediction could be performed based on optical parameters.In addition,the deep learning model could better predict the internal quality of apples based on the original spectra compared to the traditional PLSR.This study provides important theoretical basis for the use of NIR spectroscopy in fruit quality detection and producing area and cultivar identification. |
