Research On Nondestructive Testing Of Citrus Quality Based On Visible-near Infrared Spectroscopy Technology

Citrus,as the first major category of fruit,have always been loved by people.However,due to the large number of citrus varieties on the market and the uneven quality,it can’t meet consumers’ increasing demand for citrus quality.Timely understanding of the moisture,hardness and nutrient content of citrus is beneficial to the regulation of the growth environment of citrus during the growth period,and improves the quality of planted citrus.After the citrus is picked,it can be stored and transported separately according to needs,which not only ensures the economic benefits of producers and broker,but also ensure the needs of consumers and improve consumer satisfaction.Therefore,rapid and non-destructive testing of the internal quality of citrus is of great significance.In this study,the late-maturing citrus variety of Owari in Yichang City,Hubei Province was used as the research object.The five indicators of citrus moisture,hardness,sugar content,acidity and vitamin C were studied through visible-near infrared spectroscopy technology and quantitative analysis model,so as to achieve the the purpose of rapid non-destructive testing of citrus quality.The specific research content is as follows:(1)Changes in the quality of citrus.The fresh and intact citrus was stored indoors at room temperature,and physical and chemical indicators were collected at different intervals.The test duration was about 20 days.Among them,the moisture,hardness,acidity and vitamin C content show a gradual decrease trend with time,and the sugar content will increase slightly and then decrease.(2)Determination of the way of spectrum collection.For the five physical and chemical indicators,100 citrus samples were used for spectrum collection and physical and chemical indicators measurement experiments.Spectral collection is divided into reflectance spectra of citrus horizontal,vertical,and oblique placement,and transmission spectra of citrus horizontal,vertical,and oblique placement.After preprocessing the spectrum with MSC,SNV and SG,PLS model was established.The modeling results of the five indicators are consistent,and the modeling results show that the PLS model was established by using the transmission spectrum of citrus vertical placement and the transmission spectrum of citrus oblique placement is better than the other four.Compared with reflection,the modeling effect of transmission spectrum is better;compared with horizontal placement,vertical placement has the best effect,followed by oblique placement.Let’s make a further comparison between the vertical and oblique placement of citrus.With moisture index as representative,MSC and SNV are selected as pretreatment methods,and representative feature points are screened out through CARS to establish the PLS model.The results show that the optimal PLS model is obtained when using SNV to preprocess the transmission spectra collected under the vertical placement of citrus and using the CARS algorithm to obtain 67 characteristic wavelength points.The correlation coefficient Rc of the training set of the model is 0.9313,and the root mean square error The RMSEC is 0.0079,the test set correlation coefficient Rp is0.9216,and the root mean square error RMSEP is 0.0070.Therefore,it is determined that the later test spectrum collection mode is the transmission spectrum of the citrus vertical placement mode.(3)The establishment of prediction models for citrus moisture,hardness,sugar content,acidity and vitamin C.The Owari late-maturing citrus before and after the picking period were selected as test samples,and the samples used for moisture test after the picking period were dried to increase the water content gradient.The transmission spectra of the citrus vertically placed are collected and the physical and chemical values are measured and then modeled.The Mahalanobis distance method and the standardized scoring method are used to detect and eliminate the abnormal data,and then the KS(Kennard Stone,KS)algorithm is used to divide the sample set,and the PLS,SVM and ELM models are established to determine that the subsequent modeling uses the PLS model,and then The spectral data extracted from different pre-processing and different characteristic wavelength points were established to establish a PLS model for comparison.The modeling results show that the best pretreatment method corresponding to moisture and acidity is SNV,and the best pretreatment method corresponding to dimension C is MSC.Hardness and sugar content do not need to be pretreated;the best feature wavelength extraction method is CARS,the best model are PLS models.The corresponding RPD values of the PLS models established for moisture,hardness,sugar content,acidity and vitamin C are 2.9061,2.4112,3.9615,2.5035,and 2.8056,respectively.The prediction performance is good,indicating that the established PLS model has good prediction performance.(4)Development of citrus quality testing software.On the basis of the previous modeling analysis,Py Qt is used to develop the software interface and realize the prediction function,so as to achieve the purpose of non-destructive testing of citrus moisture,hardness,sugar content,acidity and vitamin C.