Electronic Nose-Based Technique for Rapid Detection and Recognition of Moldy Apples

Abstract

In this study, the PEN3 electronic nose was used to detect and recognize fresh and moldy apples inoculated with Penicillium expansum and Aspergillus niger, taking Golden Delicious apples as the model subject. Firstly, the apples were divided into two groups: individual apple inoculated only with/without different molds (Group A) and mixed apples of inoculated apples with fresh apples (Group B). Then, the characteristic gas sensors of the PEN3 electronic nose that were most closely correlated with the flavor information of the moldy apples were optimized and determined to simplify the analysis process and improve the accuracy of the results. Four pattern recognition methods, including linear discriminant analysis (LDA), backpropagation neural network (BPNN), support vector machines (SVM), and radial basis function neural network (RBFNN), were applied to analyze the data obtained from the characteristic sensors, aiming at establishing the prediction model of the flavor information and fresh/moldy apples. The results showed that only the gas sensors of W1S, W2S, W5S, W1W, and W2W in the PEN3 electronic nose exhibited a strong signal response to the flavor information, indicating most were closely correlated with the characteristic flavor of apples and thus the data obtained from these characteristic sensors were used for modeling. The results of the four pattern recognition methods showed that BPNN had the best prediction performance for the training and testing sets for both Groups A and B, with prediction accuracies of 96.3% and 90.0% (Group A), 77.7% and 72.0% (Group B), respectively. Therefore, we demonstrate that the PEN3 electronic nose not only effectively detects and recognizes fresh and moldy apples, but also can distinguish apples inoculated with different molds.