红外光谱与人工神经网络相结合识别栽培、野生黄芩和粘毛黄芩

Recognition of Three Classes of Skullcaps by FTIR Spectroscopy Combined with Artificial Neural Networks

为了识别栽培黄芩、野生黄芩和粘毛黄芩 ,采用非线性 线性、线性 线性、非线性 非线性三种模式的人工神经网络 (ANN)分别分析各种黄芩的红外谱。我们采用 4 2个样本作训练集 ,34个样本作检验集 ,用各种模式的ANN进行了监督性训练。当训练目标误差平方和定为 0 0 1时 ,各类ANN对训练集中三类黄芩样本识别的正确率均为 10 0 % ,但对检验集样本识别的结果各不相同 ,其识别的正确率与隐含层节点数S1有关。我们发现当S1较大时 ,识别正确率反而下降 ,可能此时网络的非线性程度过高 ,使其不适合于该类样本集的训练。线性—线性型ANN识别的结果随S1的变化不很大 ,但识别的正确率不高 ,基本在 85 %左右。非线性—线性型ANN识别的结果最佳。当S1为 3时 ,其识别正确率超过了 97%。因此该法可用以简便、快速、准确地识别这三种黄芩药材。

In order to recognition of three classes of skullcaps (cultivated, wild Scutellaria baicalensis Georgi and Scutellaria viscidula. Bge) three kinds of models of artificial neural networks (ANN), nonlinear-linear, linear-linear and nonlinear-nonlinear model, were used combined with their infrared spectra. Skullcaps samples were collected by Fourier Transform Infrared (FTIR) spectra. 42 samples were gathered as a train set, and 34 samples as a test set, then their supervision trains were performed using three models each. When the summation of error square of train target was selected as 0. 01, the correct. rate for recognition of three classes of skullcaps using each ANN was 100% for the train set, but was different for the test set, which depended on the number of node in hidden layer, S1. It was found that with the increase of S1, the correct rate would decrease oppositely. This may be caused by the high degree of the non-linearity of the networks, so that the models of networks were not fit for the train of this kind of sample set. When using linear-linear model of ANN varied with S1 in some extent, the correct rate was generally about 85%. Recognizability obtained using nonlinear-linear model of ANN was the best. Its correct rate of recognition was > 97% when S1 = 3, and so this method can be used to recognize three of skullcaps simply, rapidly, and accurately.