Application of BP-ANN and LS-SVM to discrimination of rice origin based on trace metals

In this work, a total of twenty nine rice samples, including nineteen Xiang-shui rice samples and ten non-Xiang-shui rice samples, were employed to establish discrimination models used back propagation artificial neural networks (BP-ANN) and least squares support vector machine (LS-SVM) based on the content value of B, Zn, Fe, Cu, Mn, Na, K, Mg and Ca determined by Inductively coupled plasma atomic emission spectrometry (ICP-AES). Although the 100% accuracy of training set achieved by LS-SVM, for test set the accuracy was only 90.91%. So the optimal result, the accuracy of 100% both for Xiangshui rice samples and for non-Xiangshui rice samples in test set, was obtained by BP-ANN models with five nodes in hidden layer. A new discrimination method to discriminate rice from different geography origin was provided, which is significant to detect and prevent fraud and adulteration. And in order to improve expansibility and accuracy of models, a further study through adding the number of samples and comparison of more discrimination methods, was needed.