| At present,the on-line detection technology for the fruit sugar content in China is usually used for the sorting of sugar content of a single variety of fruit,a unchanged production area or a single year.In the actual production process,most fruit producers'products are not just only one species,and different types of fruit have certain differences in fruit size,peel color,texture of pulp,nutrient content,and moisture content,these differences can cause spectral differences in the spectral characteristics even if the spectrum of the fruit is collected on the same device,these in turn will cause the model to be unusable,this requires the manufacturer to continuously change the mathematical model according to the type of fruit that needs to be produced.In addition,even if it is the same variety of fruit,the internal quality of the fruit may be different due to the different origin of the fruit,fruits purchased by fruit producers vary widely in moisture,sugar content,size,and nutrient content,it often leads to a reduction in the accuracy of sorting on fruit produced in another area based on a fruit sugar sorting model established in one place of production.The same species of fruit from the same origin will also cause differences in precipitation,temperature,and light due to different years,which will also affect the quality of the fruit,and will reflect on the collected spectrum,causing changes in spectral characteristics,thus affecting the versatility of the model.Stability and accuracy.In order to study the universal model of on-line detection of different kinds of fruit sugar content and the model upgrade and maintenance methods of the same kind of fruit,the visible/near-infrared dynamic online detection device developed by the laboratory was used to collect the near-infrared spectrum of fruit samples.The detection speed of the device was set to 300 per minute.(1)This paper analyzes the spectral differences of different kinds of fruits,and uses multiple scattering correction,second derivative,coefficient of variation,and other pretreatment methods for their spectral characteristics,on the basis of screening out the similar bands with smaller coefficient of variation,the spectral information that is valid for modeling is preserved on maximum degree.Establish a universal mathematical model for on-line detection of different types of thin-skinned fruit sugars at the same time,which enables the simultaneous detection of different types of fruit sugars with similar physical and chemical properties.Then by comparing and analyzing the prediction effect of the model established by the various methods used to determine the coefficient and stability,etc.Finally,considering the advantages and disadvantages of the model,the best general mathematical model is selected.For the best similarity band,the PLS and LS-SVM algorithms are used for modeling.The accuracy and stability of the two are compared.Finally,the mathematical general model established by the PLS algorithm is determined as the best model.The rc and rp of the model reached 0.93,RMSEC and RMSEP were 0.52°Brix and 0.51°Brix respectively.Models were tested using data other than modelling to verify the model's versatility and compatibility.The performance of the universal mathematical model was tested on four fruits,and the results showed that Fuji apple's verification result was the best.(2)In this paper,the apples of the candy heart in three years(2015,2016,and 2017)were used as the research object.Visible/near-infrared spectroscopy online detection technology was used to establish a mathematical model of soluble solids in fruits using PLS algorithm.To study the upgrade and maintenance methods of infrared dynamic on-line detection model for near fruit soluble solids.The"basic mathematical model"was established using the fruit of the earliest year(2015).After the mathematical model was established,the accuracy and stability of the fruits from different years but from the same origin were tested,and the reasons were analyzed.Then,the basic model is upgraded and maintained,and the method for updating and maintaining the mathematical model of dynamic on-line detection of fruit is discussed.Adopted three mathematical models to upgrade maintenance methods,a series of pre-processing methods have been adopted to reduce the spectral variability and then select the most similar bands,after comprehensively considering the most effective information of the retained spectrum,the insensitive bands with the least influence on the environment are selected and modeled to improve the scalability of the model.Based on the basic model,adding fruit data of different years,increasing the data amount of the model to increase its stability and accuracy;plotting the true and predicted values of the model,and analyzing the model's true value and prediction Function of the value to adjust the intercept of the model to improve the prediction performance of the model.The advantages and disadvantages of the three model maintenance methods are discussed.The model maintenance methods using the updated data in the three upgrade methods have the largest improvement in the prediction correlation coefficient in two years,increasing by 0.28 and 0.20 respectively,indicating that the model maintenance method using updated data is correct.The stability of the model is most effective.The use of direct correction method has the largest decrease in root mean square error in the two years,which is reduced by 0.69°Brix and 0.39°Brix,respectively,indicating that the direct correction method has the best effect on the accuracy of the model.However,using the direct correction method,the prediction correlation coefficient of the model is not much higher,only 0.1.After a comprehensive comparison,in the case of a modest maintenance effect,the priority is to adopt the method of selecting insensitive bands for model upgrade and maintenance,in the case that the effect of using the similar band method is not greatly improved,the method of adding samples and new data can be used to update and maintain the model. |
PDF Full Text Request