Discrimination And Total Polyphenol Content Prediction Of Wild Edible Boletes By Near Infrared Spectroscopy

Edible fungus,commonly known as"mushrooms",have existed in human civilization for thousands of years.China was the first country to recognize and use edible mushrooms,and they are known as"God’s food"in western countries.The World Food and Agriculture Organization and the World Health Organization have also suggested that the best diet structure for human beings is"one meat,one vegetable,and one mushroom".The wild edible boletes mushroom grows in the forest area with little or no pollution,it is a natural organic food.Their fruitbodies have the characteristics of high protein and low fat and contain many kinds of trace elements and essential amino acids which are beneficial to human health.Yunnan Province is the richest region in China in terms of natural production and distribution of wild boletes mushrooms,boletes are an important source of income for local farmers and herders.The wild edible boletes mushrooms are diverse,widely distributed and of great edible,medicinal and economic value.At present,there is commercial fraud in the supply chain of boletes mushrooms.Total polyphenols are the main antioxidant substances of boletes,but the conventional determination methods are time-consuming,and cannot be widely applied in the edible mushroom market.In this study,we used Fourier transform near infrared(FT-NIR)spectroscopy combined with chemometrics to(1)collect FT-NIR spectra of the main edible boletes in Yunnan and use multivariate analysis to discriminate the species,storage duration and edibility of boletes;(2)fit FT-NIR spectral data and total polyphenol content measurements to establish prediction models for total polyphenol content of boletes to provide a new method for quality control and evaluation of wild edible boletes.The main study results are as follows:(1)To develop a rapid and effective method for discriminating edible boletes species using FT-NIR spectroscopy.FT-NIR spectra of two parts(stipe and cap)of 418fruit bodies of five boletes species from Yunnan Province were collected and combined with the data fusion strategy to establish the partial least squares discrimination analysis(PLS-DA)model and residual convolutional neural network(Res Net)model using two-dimensional correlation spectroscopy(2DCOS)to discriminate boletes species.Differences in FT-NIR spectral features exist between the stipe and cap of fruitbodies.The data fusion strategy can provide comprehensive and complementary information.Data-level fusion can provide rich information and help optimize the PLS-DA model;feature-level fusion can reduce the input variables and improve the classification ability of the PLS-DA model by screening spectral feature variables.In the Res Net model based on 2DCOS images of a single part of the stipe or cap,the correct classification rate of the training set,test set,and external validation samples are 100%,and the loss value is close to 0.The model has good convergence and generalization ability.This method does not require pre-processing and other operations on FT-NIR spectra and only needs to convert the raw FT-NIR spectra into 2DCOS images to build the Res Net model,which is a fast and effective method for the discrimination of boletes species.(2)In response to the substandard phenomenon in the boletes market,this study proposes a strategy for the discrimination of boletes slices based on NIR spectroscopy,i.e.,firstly,to discriminate the storage duration of boletes slices to ensure that they are within the shelf life;and then to discriminate the species of boletes slices within the shelf life to assess their economic value.FT-NIR spectra of 1376 boletes samples were collected,including four species collected from 2017 to 2020(different storage periods).Three supervised methods,PLS-DA,extreme learning machine(ELM)and Res Net model,were used to validate the feasibility of this strategy.The results showed that the Res Net model built using 2DCOS images achieved 100%accuracy for both storage period and species discrimination of boletes.Compared with the PLS-DA and ELM models,the Res Net model does not require complex data pre-processing and has better discrimination ability.(3)To verify the feasibility of using NIR spectroscopy to directly discriminate the edibility of wild boletes.FT-NIR spectra of 420 samples of seven species of boletes were collected,the 2DCOS images were combined with Red-Green-Blue(RGB)image analysis and multivariate analysis methods to establish the model for the discrimination of boletes.The data driven version of soft independent modeling of class analogy(DD-SIMCA)model was used to discriminate the edibility of boletes with good sensitivity(0.98)but specificity of 0.59.The model is not suitable for accurate discrimination of edibility of boletes but can be used for preliminary screening of conditionally edible Caloboletus calopus.For the discrimination of boletes species,the random forest(RF)model had good classification and generalization ability,with 97.22%and 100%accuracy in the training and test sets,respectively.Therefore,FT-NIR spectroscopy combined with DD-SIMCA and RF models can be used for rapid discrimination of edibility and species of wild boletes.(4)To develop a new method for rapid prediction of total polyphenol content in boletes using NIR spectroscopy combined with chemometric methods.The FT-NIR spectra of the 316 boletes samples were collected,and the total polyphenol content of the samples was determined by the Folin-phenol method,and the FT-NIR spectra and the measured values were fitted to establish the prediction model.Three regression algorithms,namely partial least squares regression(PLSR),support vector regression(SVR),and RF,were used to build the prediction models.The effects of sequential orthogonal preprocessing techniques(SPORT)on the prediction ability of the models were investigated.The results showed that the SVR model developed by the second-order derivative(SD)pre-processing method had the best prediction performance with a coefficient of determination(R~2)of 0.94,root mean square error of prediction(RMSEP)of 0.07,and relative prediction deviation(RPD)of 3.18 for the prediction set.FT-NIR spectroscopy combined with the SVR model can be used as a fast and accurate method to predict the total polyphenol content in boletes mushrooms,and the method can also be used as a reference for the rapid prediction of other chemical components(such as proteins,polysaccharides)in edible mushrooms.