基于LM-BP神经网络模式的酸性染料分类方法

Classification Method of Acidic Dyes with the Pattern of Levenberg-Marquardt Backpropagation Neural Network

提出用Levenberg MarquardtBackpropagationNeuralNetwork(LM BP)网络对酸性偶氮染料进行分类,网络结构为4-6-5。优化了隐含层神经元数和网络训练次数,表明隐含层神经元数应比输出层神经元数多一个。考察了训练集样本的选择对结果的影响,测试集的样本参数大小要处于训练集样本之间。本网络把其中22种染料作为训练集,把另外18种染料作为测试集,与采用GCEDM逐次分类法比较,测试集识别率为83%。

The acidic dyes were classified by using Levenberg-Marquardt backpropagation neural network. The best structure of network is 4-6-5.The learning times of training group and number of neurons in layer were optimized. It is showed that the number of neurons in layer is more one than the elements of output vector. Chosen training group to the influence of the results has been studied, the size of parameters of testing group must be among the parameters of training group.Among all the dyes,22 kinds of dyes are taken as training group,and other 18 kinds of dyes are taken as testing group.Compared with GCEDM,the recognition ratio of testing groups is 83%.