Classification Of China Soy Sauce Using Artificial Neural Networks And Discriminant Techniques

Soy sauce was originated in China. It has been become necessary for people in their daily life all over the world. Nearly 65% of soy sauce production in the world are comes from China. Flavor components are the key to influence the quality. Two kinds of fermentation in China are high salt liquid state fermentation and low salt solid state fermentation. The production area of China soy sauce is mostly distributed in northern China, southern and eastern. In different soy sauce production areas are used different raw material. Therefore, the volatile components in soy sauce are also different. The study in volatile components of different production regions and fermentation has great significance.Artificial neural network is a mathematical modeling method that has been used in various scientific fields. Such as pattern recognition, data fitting and forecasting purposes. This study will combine volatile components in the soy sauce and artificial neural network to classifying the different production areas of China and fermentation. In order to provides the theory basis for brewing soy sauce and improve the quality of soy sauce in China. This study mainly content includes as the following:(1) Using headspace solid phase microextraction pretreatment method to extract the volatile components of soy sauce samples. And then used meteorological chromatography- mass spectrometry coupled technique to analysis the volatile components in the samples. The results showed 28 kinds of volatile components in 78 soy sauce samples, including alcohols, esters, phenols, acids, aldehydes, ketones and heterocyclic compounds. After the sensitivity analysis for data dimension reduction, total finds 15 and 22 kinds of volatile components, respectively. These volatile components have the largest contribution for classification, and as the input of artificial neural network classification.(2) This study selected back propagation neural network to classify. BP network is error backward propagation algorithm for training which belongs to the steepest gradient descent method. It has a good performance in the field of pattern recognition. However, the problem of BP network is slowly convergence speed, easily trapped in local minimum and the number of hidden layer nodes. For overcome these defects of BP network, genetic algorithms was using by optimization algorithm. Through the way of global searching to optimized the learning rate, momentum factor and the number of hidden layer.(3) Genetic algorithm was used to optimize the BP network structure. 15 data sets were selected in 78 groups of data by Kennard- Stone selection algorithm were used to test the performance of network. The result showed 100% accuracy in the classification of fermentation and production regions. Furthermore, this study proves the feasibility of combine BP network and genetic algorithm to classified different production regions and fermentation.