温度限制串联相关网络-红外光谱法用于中药大黄样品的鉴定分类

将温度限制串联相关网络与红外光谱分析技术相结合,对大黄样品的真伪进行分类。采用小波变换对原始数据进行压缩,将原来的775个数据点压缩到49个数据点,既提高了网络的训练速度又保持了原来的特征谱峰。对45种样品进行了测定和鉴别,正确率可以达到84．4％。对影响分类结果的网络参数,进行了讨论。红外光谱法作为中药鉴别的一种方法与神经网络相结合,使中药鉴别更加快速、方便。

Identification of official rhubarb samples based on IR spectra and temperatureconstrained cascade-correlation networks

A temperature-constrained cascade correlation network (TCCCN) based on infrared reflectance spectrometry (IRS) was applied to develop classification models for identifying 45 official and unofficial rhubarb samples. Wavelet transformation compression was used to reduce the variables of measured spectra. The number of spectra variables was reduced to the 49 from original 775, so that the network training was improved. The effect of network parameters was investigated and the parameters were optimized. The rate of correct classification is 84. 4%. The proposed method can be a useful tool for identification of Chinese herbal medicines and quality control in the production process.