Optical Properties Of Pomelo And Online Nondestructive Detection Method Of Soluble Solids Content

As an important and famous fruit in China,pomelo has excellent quality,unique flavor,and rich nutrition.However,the postharvest grading and processing technology of pomelo in China is relatively backward.With the continuous increasing of consumers’demands for fruit quality,especially the internal quality such as flavor and taste,there is an urgent demand for nondestructive detection technology for the internal quality of pomelo.Therefore,this paper adopted Visible and Near Infrared Spectroscopy(Vis-NIRS)technique to explore the photon transportation characteristics in pomelo.Aimng at the characteristics of thick peel,large size and complex multi-layer structure,an online nondestructive detection method for soluble solids content(SSC)was proposed,and a set of online detection equipment was designed and developed.The main contents and results of this paper are as follows:(1)According to the characteristics of multi-layer structural of pomelo,the optical properties of different layers in pomelo were explored.Firstly,the optical properties of exocarp,endocarp and pulp of pomelo were measured by a single integrating sphere system,and the differences in optical properties were significant.The correlation between the optical properties(absorption coefficientμ_a and reduced scattering coefficientμ_s’)and the quality attributes of tissues were studied.Combined with competitive adaptive reweighted sampling(CARS)and partial least squares regression(PLSR)methods,the quality prediction models were established with optical properties parameters.For pulp tissue,the prediction model established by the measured total transmittance(MT)performed best,in which the optimal determination coefficient of cross-validation R~2 was 0.97 for SSC,with the minimum root mean square error of cross-validation(RMSECV)of 0.11%,and the R~2 was 0.93 for moisture content with RMSECV of 0.20%.(2)The thick peel of pomelo would affect the information extraction from the pulp.In this paper,a Monte Carlo(MC)light transmission simulation method was used to investigate the light transmission properties of pomelo in multiple tissue layers at 808 nm.Firstly,the spherical multilayer tissue transport model was established,and the internal light distribution and spatial resolution spectra of pomelo were measured by a self-designed multifunctional bedstand.The simulation results were consistent with the changes of measured light intensity.Then,The simulation results of different light-detector angles showed that with the increase of the angle,the information weight of pulp in spectral gradually increased,and the partial pathlength in pulp tended to be stable when the angle was greater than 120°.Considering the information weight of the pulp and the detected spectral intensity,the effect was best when the light source-detector angles were arranged at 90-120°,and the absorbance weight of the pulp was between 77.19%and 78.49%.(3)There might be local distribution differences in the internal structure and components of pomelo,which would affect the performance of the model.This paper explored the impact of the internal light distribution and SSC distribution of pomelo on the accuracy of the static detection model.First,it was found that the difference in the internal tissue structure would cause the local light intensity changes through the measurement of the puncture fiber.Then,the spectra of six illumination points and the SSCs of the‘photon accumulation area’were collected for 103 samples respectively,and the spectra and SSCs of each point were all cross combined as independent data sets to establish the PLSR prediction model of SSC.The results showed that the SSC prediction model could be established by cross-matching the spectral data and the SSC data of different points,and the prediction performance of the models established by the average spectra were improved,which may be related to the difference in the light distribution and the multiple scattering of light in the pulp.Therefore,multi-point illumination using multiple light sources helps to improve the model accuracy.(4)The on-line detection system of SSC in pomelo was designed,developed and tested.Firstly,a double-layer parallel light source system was designed with symmetrical arrangement.The number of light sources and light source patterns were analyzed in a comparative test to determine the lighting scheme of 16 lights on both sides.After that,a rotary table automatic reference system was designed,and the neutral density filter with a diameter of 20 mm and transmittance of 0.1%was optimized through the comparative test.Finally,220 samples of two cultivars(‘Guanximiyou’and‘Hongroumiyou’)were used for the on-line detection testing.The stability of on-line spectra acquisition was evaluated through outlier detection,and the spectral anomaly rate was 1.8%.The PLSR prediction model has achieved good performance,the RMSEP of‘Guanximiyou’and‘Hongroumiyou’was 0.41%and 0.57%,respectively.The mixture model of two cultivars got the performance with RMSEP of 0.55%.The results showed that the on-line detection system could realize the rapid and non-destructive detection of SSC with 3 pomelos per second.(5)Aiming at the issue of model accuracy and universality in industrial applications,the effects of different types of data and wavelength optimization methods on the prediction accuracy were explored,and a latent variable updating(LVU)method was proposed to correct the prediction deviation in external validation.Firstly,it was found that different types of data would have a significant impact on the accuracy of the model,which was mainly reflected in the amplification of dark noise at both ends of the absorbance spectrum.The results showed that the accuracy of the transmittance model was better than that of the absorbance.Secondly,the wavelength optimization method significantly improved the performance of models in internal prediction of the same batch with RMSEP of 0.240%-0.570%.However,for the external validation data in different years,the model showed over-fitting which led to large prediction deviation.Finally,a model updating method of LVU was proposed,and verified.A model update method with LVU was proposed and validated to address the problem of poor model applicability.Compared with recalibration and slope and bias correction(SBC),the CSMW-PLSR-LVU achieved the best performance in the external validation with an RMSEV of 0.599%when combined with the changeable size moving window(CSMW)method.The residual distribution of 100 external validation samples showed that this method could meet the actual production needs of SSC detection,with 91%of the residual less than±1.0%and 60%of the residual less than±0.5%.The on-line detection equipment and modeling method have been applied to the actual production of SSC detection of‘Majiayou’in Shangrao,Jiangxi Province.