Non-destructive Quality Evaluation of Litchi Fruit Using e-Nose System

Abstract

In this study, we have developed a conveyor e-Nose system for the non-destructive quality evaluation of litchi fruit. A pair of electronic noses with an optimized sensor array of six sensors was adopted to test two litchi fruits at a time when passing through two channels of the running conveyor belt and halt below the e-Nose for a few seconds. The study started with sensor array optimization using a sensitivity test. Three methods were employed for sensor optimization. And the results of all the three methods are quite similar to each other. After acquiring data about the aroma volatiles of the sample and from the electronic nose, Principal Component Analysis (PCA) was employed to see the pattern differences between good litchi and rejected litchi. The experimental results showed that the patterns of the two categories are different from each other. As the clustering of two patterns is prominent, data analysis for classification is employed. 123 data were taken for two categories of litchi. After acquiring aroma volatiles from a sensory array of good and rotten litchi, the data was analyzed and compared using SVM, Logistic Regression, KNN, decision tree and random forest classifier models. SVM and Logistic Regression showed the lowest accuracy rate. As the decision tree model shows the lowest error rate, it is applied and integrated into the system to allow the classification of good and rejected litchi.