ASABE Technical Library - Abstract

Abstract.

This paper proposed a multi-source information fusion identification method which based on electronic nose and physical characteristic combination, using it for kiwi fruit storage time identification. According to the change of weight loss rate, the kiwi fruit weight loss rate rose along with the storage time increase. After extracted feature value, linear discriminant analysis (LDA), euclidean distance analysis (ED), support vector machine (SVM) and loading analysis (Loadings) were used for pattern recognition. The LDA result shows that both 2 identification methods (electronic nose identification method and electronic nose and physical characteristic combination identification method) cannot classify the storage time of kiwi fruit. According to LDA and ED, the identification method of electronic nose and physical characteristic combination can gain more sample information and lager different sample group distance when comparing with single electronic nose method. SVM results show that the training set’s returned classification accuracy of 2 identification methods for kiwi fruit storage time identification are all 96%, the test set’s classification accuracy of them are 76.67% and 86.67%, respectively. According to Loadings, after combining with physical characteristic, electronic nose method can gain more sample information, keep the contribution rate of each sensor as more as possible, has better classification effect. This research proved the feasibility of using electronic nose and physical characteristic combination to improve classification effect of single electronic nose, which provide reference for future fruit quality detection.