Development Of Detection Of Physicochemical Indexes And Portable Detection Equipment For Snow Pears Based On The Near Infrared Spectroscopy

The rapid detection of the quality grade of snow pears is the order to distinguish the quality of the fruit and promote the high quality and competitive price of snow pears.On the one hand,high quality and high price of snow pears can promote the fruit growers,fruit supply chain companies and fruit brand operators to actively improve the planting technology,logistics technology and marketing strategy of snow pears,improve the fruit quality,brand value and market competitiveness of snow pears;on the other hand,the improvement of the competitiveness of the snow pears brand can increase the export volume of pears in China,changing the current situation that the total output of snow pear is high but the export rate is low in China.In addition,the general improvement of the quality of snow pears can also meet consumers’growing demand for fruit quality upgrading.According to the national standard and agricultural industry standard,the detection of physicochemical indexes of snow pears is currently conducted by chemical experimental analysis method,which has many problems such as high cost,long time,physical damage of samples and chemical pollution to the environment.The"absence"of online rapid nondestructive detection of the quality of snow pears,which has largely restricted the production and sales of high-quality snow pears and the growth of brand value.Therefore,it is of great research value to rapidly detect the quality grade of snow pears.In this paper,the rapid nondestructive detection method and portable multi-parameter detection device for physicochemical indexes of snow pear were studied based on the near infrared spectroscopy.Based on the spectrum collected by the Fourier transform near infrared（FT-NIR）spectrometer,studying the influence of detection of lignin content in snow pear by the spectrum collected when the incident light is perpendicular to the fruit core（vertical incident fruit core spectrum）,the spectrum collected when the incident light departure from the incident fruit core（deviation from incident fruit core spectrum）,and the location and number of spectral collection points.The prediction performance for detecting lignin in snow pear was studied by establishing a PLS model based on the spectrum collected by a micro-fiber spectrometer.Near infrared diffuse reflectance spectra of 10 different locations around the equator line of samples were collected to study the influence of the detection of physicochemical indexes in snow pear by the location and number of spectral collection points and the optimal spectral collection points were determined.Four updating methods of calibration models were designed to compare and analyze the prediction performance of the physicochemical indexes of new batches of samples.The robustness of the updating model based on SS-FPME was the best.A portable multi-parameter detection device for physicochemical indexes in snow pear was developed,and a quality grade determination model was designed to realize the rapid non-destructive detection of five physicochemical indexes including soluble solid content（SSC）,lignin content（LC）,moisture content（MC）,titratable acidity（TA）and solid acid ratio（SAR）.The main research work of this paper is as follows:（1）The research on the rapid nondestructive detection of lignin content in snow pears.The Klason chemical method was used to determine the standard reference value of lignin content in 195 snow pear samples.FT-NIR spectrometer was used to collect the spectra of perpendicular and deviated fruit core at 10 different locations around the equatorial line of samples.Firstly,the mean spectra of the perpendicular incident fruit core and the deviation from the incident fruit core collected at the 1^st,4^thand 7^th point were preprocessed.Then synergy interval partial least squares（Si PLS）,competitive adaptive weighted sampling（CARS）,continuous projection algorithm（SPA）,bootstrap soft shrinkage algorithm（BOSS）and improved bootstrap soft shrinkage algorithm based on frequency and regression coefficient（FRCBOSS）were used to select characteristic wavelengths from the preprocessed spectra.Partial least squares（PLS）model was established for nondestructive detection of lignin content in snow pears Based on characteristic variables.The influence of the detection model of lignin content in snow pears by the light source incident mode was analyzed and compared.For vertical incidence core spectrum,the number of characteristic variables was 29,the number of maximum principal components factor was 13,Si PLS+PLS+FRCBOSS model for detecting lignin content in snow pear and the correlation coefficient of calibration set（Rc）was 0.894,root mean square error of cross validation（RMSECV）was 0.909,correlation coefficient of prediction set（Rp）was 0.863,root mean square error of prediction（RMSEP）was 0.903,the ratio of performance and deviation（RPD）was 2.278.For deviated incidence core spectrum,the number of characteristic variables was 55,the number of maximum principal component factor was 11,and the values of Rp,RMSEP and RPD of the detection model were 0.822,0.90 and 2.244,respectively.At the same time,FT-NIR spectra collected from 10 different locations of the samples were combined into 1023groups of average spectra to establish the full-spectral PLS model.The influence of the model for detecting lignin content in snow pear by the location and number of spectral collection points was analyzed and compared.Based on the average spectra collected from 2,4,6 and 10 different locations on the equator,the SIPLS+CARS+PLS model established for measuring lignin content was better.The values of Rc,RMSECV,Rp and RMSEP for predicting the lignin content in snow pear were 0.902,0.412,0.888 and 0.361,respectively.Based on the research results of FT-NIR spectrometer,the vertical incidence fruit core spectra of the 7th and 8th point of 195 Pear samples were collected by a micro fiber spectrometer.SG+Normalize+D1 was used to preprocess the spectra,and Si PLS,CARS,BOSS,FRCBOSS,Si PLS+CARS,Si PLS+BOSS and Si PLS+FRCBOSS variable selection algorithms were used to select characteristic variables for modeling.FRCBOSS+PLS model had the best performance in predicting the lignin content in snow pears,with Rp,RMSEP and RPD values of 0.834,0.711 and 2.12,respectively.The experimental results show that the model established by using FT-NIR spectrometer to collect the vertical incident fruit core spectra at the optimal spectral location can realize the non-destructive detection of lignin content in Snow pear,and determine the incident mode of light source,the optimal location and number of spectral collection point.Based on the results of FT-NIR spectroscopy,the model established by using a micro spectrometer to collect the vertical incident fruit core spectra at the 7th and 8th point of the equator of snow pear samples can realize the non-destructive detection of lignin content,and provide a certain technical support for the development of portable detection devices（2）Research on the influence of the nondestructive detection model of physicochemical indexes of snow pears by the location and number of spectral collection points.160 samples of Snow pear were purchased,and the SSC reference value was measured by refraction method,MC reference value was measured by azeotropic distillation method,TA reference value was measured by titration method,and SAR was calculated at the same time.A set of spectral acquisition system for snow pear samples was designed and built,horizontal,vertical and horizontal rotation angles could be set.The near-infrared diffuse reflectance spectra of samples at different positions around the equator could be collected by using a micro-fiber optic spectrometer.The micro spectrometer collected the spectra at 10 different locations on the equator of each sample,which could be combined into 1023 groups of mean spectra data sets.Based on these combined spectral pairs,a full-spectrum PLS detection model was established,and the optimal spectral collection points corresponding to the four physicochemical indicators of SSC,MC,TA and SAR were selected respectively.And the optimal spectrum corresponding to different indicators to collect combination average spectral preprocessing,and then indicators using SIPLS,CARS,SPA,GA,BOSS,FRCBOSS,GA+CARS,GA+BOSS,GA+FRCBOSS,GA+SPA and improved the variable selection algorithm based on the variable stability and frequency（VSFAA）,and GA+VSFAA,a total of 12 methods selected the effective feature wavelength to build the PLS model.The experimental results showed that VSFAA+PLS model based on the spectra collected at three different locations at the equator had the best performance for SSC detection in Snow pear,with Rp,RMSEP and RPD values of 0.967,0.186 and 4.09 respectively.For the detection of MC in snow pear,the CARS+PLS model based on the spectra collected from six different locations at the equator had the best performance,with Rp,RMSEP and RPD values of 0.959,0.08 and 3.24,respectively.For detection of TA in Snow pear,the VSFAA+PLS model based on the spectra collected at three different locations of the sample at the equator had the best performance,with Rp,RMSEP and RPD values of0.953,0.00067 and 3.39,respectively.For detection of SAR in snow pear,the VSFAA+PLS model based on the spectra collected at three different locations at the equator has the best performance,with Rp,RMSEP and RPD values of 0.961,0.451 and3.6,respectively.The results show that the nondestructive detection of SSC,TA and SAR in Snow pear can be realized by using PLS model to collect the spectra of three different locations around the equator of the sample.PLS model was established for nondestructive detection of MC based on spectra collected from 6 different locations around the equator.The detection performance is good and meets the test requirements.（3）Research on the robustness of model for detecting the physicochemical indexes in snow pear based on NIR spectroscopy.Four batches（512 samples）of A,B,C and D were prepared at different time periods,and the standard reference values of five physicochemical indexes（SSC,LC,MC,TA and SAR）were determined according to the national standard.Based on the research results in Chapters 3 and 4,the NIR diffuse reflectance spectra of four batches of samples were collected using a micro fiber spectrometer,respectively.The correction models for detecting the corresponding to physicochemical indicators in snow pears were established based on A batch（master batch）samples,respectively,and then using independent join new batch sample update model,sequential join the new batch sample update model,based on A semi-supervised learning model parameters of enhanced architecture（SS-FPME）supervision learning new batch sample and independent update model,based on the SS-FPME supervision learning new batch sample and sequential update model,A total of 4 methods update master batch calibration model to predict five physicochemical indicators of B,C,D batches of snow pear samples,respectively.The physicochemical indexes of the 4 batches of pear samples were tested with the same instruments,procedures and methods,but with different testing time.The results were as follows:For the detection of physicochemical indexes of batch B,The RMSEP of SSC,LC,MC,TA and SAR detection models updated based on SS-FPME supervised learning new batch independently decreased from 0.662,0.747,1.074,0.00348 and 3.03 to 0.434,0.583,0.370,0.00340 and 2.67,respectively.For batch C,RMSEP of SSC,LC,MC,TA and SAR detection models updated based on SS-FPME supervised learning new batch sequentially decreased from 0.750,1.6,1.054,0.00519 and 3.86 to 0.561,0.503,0.384,0.00370 and 2.69,respectively.For batch D,the RMSEP of SSC,LC,MC,TA and SAR detection models updated based on SS-FPME supervised learning new batch sequentially decreased from 1.047,0.865,1.085,0.01158,9.42 to 0.489,0.280,0.419,0.00418,5.36,respectively.The results show that the correction model based on SS-FPME supervised learning of new batch of samples can predict the corresponding physicochemical indexes of new batch of snow pear samples,and the prediction performance is good,which improves the prediction performance and robustness of the model.The model based on SS-FPME supervised learning with sequential updating of new batch samples is superior to the model with independent updating,due to the enhancement specifically obtained from the previous batch supervised learning.（4）Research and develop a portable multi-parameter detection device for physicochemical indexes of snow pears to achieve rapid non-destructive detection of SSC,LC,MC,TA and SAR in Snow pear.Firstly,the measurement values of five physicochemical indexes（SSC,LC,MC,TA and SAR）were standardized,and each value was added with different weights to obtain the quality parameters of snow pears.Then,according to the quality parameters,the quality of snow pear was divided into five grades:super grade,first grade,second grade,qualified and unqualified.The structure of portable multi-parameter detection device for physicochemical indexes of snow pear is mainly composed of a portable detection terminal equipment,intelligent terminals and a cloud platform.The portable detection terminal consists of a spectrum acquisition system,a micro optical fiber spectrometer and a main board and a display screen.In the spectral acquisition system,a set of sample spectral collection point position control system was designed,which has manual and automatic control functions,and can control horizontal,vertical and horizontal rotation at the same time.Light source test,micro fiber spectrometer parameter setting test,sample spectral collection point position control test and functional test were carried out.The equipment was assembled and the cloud platform server was configured to test 5 physicochemical indexes of 160 snow pear samples.Firstly,based on the spectral pretreatment of 60 samples in the calibration set,the characteristic spectra were selected and the PLS models corresponding to different physicochemical indexes were established.Then 40 snow pear samples from the prediction set were detected,and SS-FPME was used to supervise the learning of the 40 snow pear samples to update the calibration model.The 5 physicochemical indexes in 60 snow pear samples were tested.The results were as follows:For SSC,the Rp,RMSEP and RPD values were0.903,0.204 and 3.22,respectively.For LC,Rp,RMSEP and RPD values were 0.832,0.473 and 2.44 respectively.For MC,Rp,RMSEP and RPD values were 0.874,0.162 and2.82,respectively.For TA,the Rp,RMSEP and RPD values were 0.882,0.0025 and 2.95respectively.For SAR,the Rp,RMSEP and RPD values were 0.890,1.991 and 3.01,respectively.The results showed that the portable multi-parameter physicochemical indexes detection system based on NIR spectroscopy technology met the requirements of physicochemical indexes detection,and realized the instrumentation of SSC,LC,MC,TA and SAR detection of snow pear samples.The mean variance of the predicted values of SSC,LC,MC,TA and SAR were low,which met the test requirements and had high detection accuracy.The rapid non-destructive quantitative detection task of the 5physicochemical indexes of the snow pear samples was well completed.