Discrimination of geographical indication of Chinese green teas using an electronic nose combined with quantum neural networks: A portable strategy

Abstract

Sensory attributes of teas are related to their geographical origin and directly affect consumer preference and thus the value of products. In this paper, we propose a portable strategy of identifying five famous green teas with official geographical indications of China using a lab-made electronic nose combined with a quantum neural network (QNN). QNN can be regarded as a modified version of back propagation neural network (BPNN), in which transfer functions are superposed by several sigmoids forming multilevel intervals and can be adjusted during training. Firstly, principal component analysis and linear discriminant analysis are used for low-dimensional visualization. Then different QNNs with 2–6 quantum intervals are trained and tested compared with BPNN, support vector machine (SVM) and k-nearest neighbor (k-NN). The results show that averaged classification accuracies of QNNs are higher than that of BPNN, SVM and k-NN, and they raise with increasing of initial quantum intervals but vary a little when quantum intervals exceed 3, and reach highest to 99.38 %, indicating 3 quantum intervals is adequate for this application considering the trade-off between calculating cost and classification accuracy. The results demonstrate that QNN is a promising pattern recognition method for portable electronic noses due to its better generalization ability but simple implementation as a conventional BPNN.