Neural network based electronic nose for classification of tea aroma

Abstract

This paper describes an investigation into the performance of a Neural Net work (NN) based Electronic Nose (EN) system, which can discriminate the aroma of different tea grades. The EN system comprising of an array of four tin-o xide gas sensors was used to sniff thirteen randomly selected tea grades, which were exemp lars of eight categories in terms of aroma profiles. The mean and peak of the transient signals generated by the gas sensors, as a result of aro ma sniffing, were treated as the feature vectors for the analysis. Principal Co mponent Analysis (PCA) was used to visualise the different categories of aro ma profiles. In addition, K-means and Kohonen's Self Organising Map (SOM) cluster analysis indicated there were eight clusters in the dataset. Data classification was performed using supervised NN classifiers; namely the Multi-Layer Perceptron (MLP) network, Radial Basis Function (RBF) network, and Constructive Probabilistic Neural Network (CPNN) were used for aroma classification. The results were that the three NNs performed as follows: 90.77, 92.31, and 93.85%, respectively in terms of classificat ion accuracy. Hence the performance of the proposed method of aro ma analysis demonstrates that it is possible to use NN based EN to assist with the tea quality monitoring procedure during the tea grading process. In addition the results indicate the possibility for standardization of the tea aro ma in nu meric terms. Keywords Tea aro ma _ Electronic nose _ Neural network IntroductionThe characteristic fragrance of tea, the 'aroma' of either in fused leaf or the tea liquor, is one of the most significant parameters for evaluation of its quality [1][2][3]. The aro ma indicates that the tea comprises at least one of a certain number of odours that are desirable and are h ighly valued in 'good quality tea'. This odour is also known by the 'tea tasters' as the 'nose' or 'fragrance' and is one of the three attributes of the tea flavour relating to the smell of the tea, and is utilized for the assessment of tea quality in the tea industry. The other two flavour attributes are senses of taste and astringency, which are perceived within the taster's mouth. The source of the tea aroma is attributed to various volatile organic compounds (VOCs) present in tea. These VOCs are formed as a result of significant lip id degradations during the course of the various tea processing stages [4]. For example, the majo rity of the aro ma compounds are formed fro m amino acids, fatty acids, carotenoids and glycosides [5]. On the other hand, the variation of aroma in different teas is attributed to the variation of these VOCs, and their ratios, as pres ent in tea. There are many reasons for such variations in aroma profile of which one of the main reasons is due to tea being made fro m different varieties of 'tea clones' [6]. Moreover, the processing techniques, the pruning and the pruning time also have a major impact on the variations in VOCs; thus affecting the aroma profiles [7][8][9]. Du ring quality evaluation, exp...

Application of an electronic nose system coupled with artificial neural network for classification of banana samples during shelf-life process

“…Generally, artificial neural network (ANN) is a powerful method that can be used to classify the sensor array responses of e-nose [23,24]. The neural network model used in this study was a back-propagation multilayer perceptron (BP-MLP) algorithm (Fig.…”

“…Sensor array responses of banana samples were repeatedly fed into the network and training was accomplished by automatically adjusting the connection weights in order to reduce the error level [24]. The MLP learns by supervision, and during the training phase it is provided with training vectors together with the associated targets corresponding to the specific banana aromas [23]. Also, the MLP method is considered to be rather fast and efficient when a small data set is being used [25].…”

A novel technique for rapid evaluation of fish freshness using colorimetric sensor array

“…In the testing phase, the new data can be interpreted by the network, and the output is calculated based on the fixed weight factors. (Borah et al, 2008;Tudu et al, 2009).…”

Longjing tea quality classification by fusion of features collected from E-nose

“…In the field of sensor arrays, time domain features, such as maximum value, average value, maximum variance, maximum slope, root mean square and standard deviation [11][12][13][14][15], and frequency domain features extracted by wavelet transform, wavelet packet transform and fast Fourier transform [13,[16][17][18][19], are used for signal analysis. But for E-nose, the most frequently-used features are maximum value and average value extracted from time domain [20][21][22][23]. There are few reports about feature extraction in frequency domain.…”

“…Feature selection algorithms, such as principal component analysis (PCA), linear discriminant analysis (LDA), partial least squares (PLS) and genetic algorithm (GA), have been utilized in food analysis [7][8][9][24][25][26]. In these algorithms, PCA and LDA are the two most used algorithms in quality classification of tea [9,10,20,23,25]. However, considering the nonlinear data acquired by E-nose, the linear algorithms (PCA and LDA) may not be the most suitable ways for feature selection.…”