Study Of Fingerprint Identification And Traceability For Fresh Apple Juices Based On Electronic Nose And Electronic Tongue

Fresh apple juice is one of the most frequently consumed juices all over the world,due to its various nutritional constituents and pleasant organoleptic qualities.In this study,electronic nose(EN)and electronic tongue(ET)were utilized to detect the fresh apple juice samples and the experiment parameters were optimized to ensure the response curve fully characterizing sample information.Under the optimized detection conditions,discrimination models were built based on EN and ET response signals of different fresh juice samples according to variety.The partial least square(PLS)regression technique was utilized to establish the prediction models that described the relationship between sensors signals and the main quality indices of juice samples.And we explored the feasibility of using EN and ET electronic fingerprints to quantify the influence of meteorological factors on apple juice quality.Finally,the combination of EN and ET was used to trace apples according to apple variety and geographical origin by detecting the squeezed fresh juices.Different pattern discrimination models were established based on chemometrics and the discriminant ability of different traceability models were compared based on the recognition and prediction accuracy.The main findings are as follows:(1)The optimal detection conditions of EN were 10 m L of sample volume,90 min of headspace equilibrating time,20±1℃and 50-60% relative humidity(RH).ET detection should be carried out in quiet environment at 20±1℃and 50-60% RH.Under this condition,EN and ET can detect samples with high repeatability.(2)The apple variety had a significant effect on the EN and ET response signals of its fresh juice.The PCA model based on the EN response signals could not distinguish the fresh apple juice samples of different varieties.But the LDA model revealed satisfactory discrimination of juices,and provided both 100% success rates in terms of recognition and prediction ability.The PCA model based on the ET response signals could not identify the different fresh apple juice samples,while the LDA model can identify some samples with the overall recognition rate of 63.6%.The PCA model based on the fusion dataset of EN and ET could not completely distinguish juice samples,but the LDA model could accurately identify them with 100%success rate for both recognition and prediction ability.(3)Good correlations were found between EN response signals and SSC,p H,TAC and TSC,but the correlations between ET response signals and SSC,p H,TAC and TSC were poor.There were good correlations between fusion dataset and SSC and TAC,but the quantitative prediction ability for p H was generally poor,and the correlation with TSC was poor and could not predict its content well.The fusion signals can improve the quantitative prediction ability of sensor signals to SSC and TAC,but could not improve that of sensor signals to p H and TSC.The EN sensor response signals had good correlations with the content of most volatile components,and had good prediction precision for the content of the main volatile components.The ET response signals could not be used to quantitatively predict the content of volatile components in apple juices because of poor correlations between them.The fusion dataset can improve the correlation between the response signals and the content of volatile components.(4)The meteorological factors and their interactions had significant effects on the main quality indexes of different fresh apple juices,and the response signals of EN and ET.The regression expressions of the juice quality indices and meteorological factors,the juice quality indices and sensor response signals,and sensor response signals and meteorological factors were established.The influence of meteorological factors on the quality of fresh apple juices was characterized by the change of EN and ET response signals.(5)The EN and ET analysis combined with LDA models can accurately distinguish juice samples with 95.5% success rate according to apple variety or origin.The PCA models based EN dataset,ET dataset and fusion dataset of all samples could not separate the samples completely.The results of load analysis showed that the EN sensors S6,S8,S1,S3,S5 and ET sensors ZZ,BB,HA,GA and CA have strong discriminant ability for sample discrimination.The LDA,SVM and PLS-DA discriminant models were established,and all the three models achieved satisfactory discrimination and prediction performances.The discriminant success rates of LDA models based on EN dataset,ET dataset and fusion dataset were 96.8%,75.1% and99.2%,respectively.For SVM model,the correctness rate of training set was98.25-100% and the correctness rate of prediction set was 96.67-100%.The prediction accuracy of PLS-DA discriminant model was 100%.The discrimination ability of classification model based on EN dataset was better than that based on ET dataset.The data fusion method made it possible to build a more robust classification model,and the discrimination ability was better than models based on solely EN dataset or ET dataset.