Detection Of Soluble Solids Content Of Fruit Based On Vis-NIR Spectroscopy And Imaging Technology

Soluble solids content(SSC)of fruit is an important parameter which could evaluate the taste and flavor and determine the fruit picking time.It directly influences the storage life and quality of fruit.The traditional quality detection methods for fruit would destruct the samples,which exist some disadvantages including the complex preprocessing,long detection time and strong subjectivity.Therefore,research on rapid and non-destructive detection technology has become an urgent demand in the fruit industry.In recent years,visible and near infrared(Vis-NIR)spectroscopy and hyperspectral imaging technique have been widely used for non-destructive detection of fruit quality.However,Vis-NIR spectroscopy technique is mainly used for detecting the internal quality of small and medium-sized fruits with thin peer(such as apples,pears,peaches,etc.),while the research on large and thick peer fruits(such as 'Hami' melon)is relatively less.More researches using hyperspectral imaging technology focused on the detection of fruit refractory defects,such as subcutaneous damage,slight rot,etc.For the detection of internal quality for fruit,the applicability of hyperspectral imaging technology needs further study.In this paper,'Hami'melon(with large size and thick peer)and 'Dangshan' pear(with small size and thin peer)were both used as research objects.The Vis-NIR spectroscopy combined with stoichiometry methods was applied to establish and optimize the SSC prediction model for 'Hami' melon,and the hyperspectral imaging technique was used to measure and visualize the SSC of 'Dangshan' pear.The main research contents and methods of this paper are as follows:(1)The effective penetration depth of Vis-NIR light into 'Hami' melon tissue was studied.According to the research requirement,the experiment platform of penetration depth of light was designed and constructed.And the diffuse reflectance spectroscopy was collected.The attenuation of light into melon tissue was explored by slicing.The light penetration depth into melon tissue in range of 720-880nm was calculated by least square fitting.The results indicated that when d(the distance between the light source and the detector)increased,the light penetration depth would increase and the more spectral information would be obtained,while the spectral intensity would weaken.Based on the above research,a ring-shaped hand-held fiber optic probe(radius was 20mm)which was suitable for detecting internal spectrum information for 'Hami' melon was designed.The effective spectral information in melon tissue could be obtained using this probe,so that the internal quality of'Hami'melon could be better detected.(2)The effect of different measurement positions of 'Hami' melon on the accuracy of prediction model based on Vis-NIR spectroscopy for SSC was studied.The calibration sets of three positions(calyx,equator and stem)were used to establish local prediction models,respectively.The calibration set from all local positions was applied to build a global prediction model.Comparing the local and global models,it could be found that the local model based on equator position obtained the best predictive result.Furtherly,different selection methods of characteristic wavelengths including UVE(uninformative variable elimination),CARS(competitive adaptive reweighted sampling)and combined with SPA(successive projections algorithm)were employed to choose the effective wavelengths from full spectra for improving the accuracy and stability of models.Then,the local prediction model based on equator position and global prediction model were established based on PLS(partial least square)and LSSVM(least square support vector machine),respectively.The overall results showed that the local model based on equator obtained the better performance than global model,and the optimal prediction model were UVE-SPA-PLS and CARS-SPA-LSSVM.It indicated that it was feasible to build the models using only the equator position for predicting the SSC of whole 'Hami' melon.(3)The feasibility of non-destructive detection and visualization of SSC in'Dangshan' pear were studied based on hyperspectral imaging technology.Mean normalization method was employed to decrease the influence of sample curvature on spectral profiles.Spectral information from whole surface of fruit was extracted to construct the data set.Effective wavelength selection methods including MC-UVE(Monte Carlo-uninformative variable elimination),SPA,CARS,GA(generation algorithm),CARS and combination algorithm(CARS-SPA and GA-SPA)were employed to improving the performance of model,respectively.Linear PLS model and nonlinear LSSVM,BPANN(back propagation artificial neural network)models were established and then compared.The results showed that the best predictive models were linear CARS-PLS and GA-SPA-PLS,which could be effectively used for predicting SSC of pear.Finally,the corresponding SSC value of each pixel in hyperspectral image was predicted by developed models and the distribution of SSC was showed in pseudo-color image.The overall results indicated that it was feasible to develop PLS model combined with intensity calibration of hyperspectral image and wavelength selection methods(CARS and CARS-SPA)for detecting and visualizing SSC in'Dangshan'pear non-destructively and rapidly.In summary,Vis-NIR spectroscopy and hyperspectral imaging technologies can be used for SSC detection with rapidness and accuracy of fruit,which provide some theoretical references for developing non-destructive online detection system of internal quality of fruit.