Authenticating cherry tomato juices-Discussion of different data standardization and fusion approaches based on electronic nose and tongue

The study presented six approaches (two e-nose measurements, an e-tongue measurement and three fusion approaches using both of the instruments) for recognition and quantitative analysis of four tomato juice groups: unadulterated and three adulterated tomato juices with different adulteration levels. Recognition of the juices was performed by principle component analysis (PCA) and cluster analysis (CA). Quantitative calibration with respect to pH and soluble solids content (SSC) was performed using four regression methods (principle components regression (PCR) based on stepwise selection, multiple linear regression (MLR) based on raw feature vector, forward selection and stepwise selection features). CA based on different data standardization and distance calculation methods were compared, and precision-recall measure was applied to quantify clustering outcomes. The result implies that it is important to explore the optimum standardization and distance calculation methods for every dataset studied prior to CA. Humidity effect was also explored and the result showed that employing desiccant for e-nose measurement presented no improvement. The fusion dataset that consists of variables selected by analysis of variance (ANOVA) presented the best authentication ability, and the quality indices highly correlated to this dataset.