人工神经网络—红外光谱法用于中药大黄样品的鉴定研究

【作者】 马书民；

【导师】 张卓勇；

【作者基本信息】 东北师范大学 ， 分析化学， 2004， 硕士

【摘要】 应用误差反向传播（BP）神经网络、径向基函数(RBF)神经网络和温度限制串联相关网络（TC-CCN），根据中药大黄样品的红外光谱数据，对正品和非正品大黄样品进行分类鉴定，并对影响分类结果的神经网络的各项参数进行研究。红外光谱数据经过小波变换压缩后输入神经网络，压缩前数据有775个点，压缩后为49个点。这样使变量数大为减少，提高网络的训练速度，又能保持特征峰。应用RBF网络，对大黄样品进行真伪分类，正确识别率为97.78%；应用BP网络，对样品的真伪的正确识别率为95.56%；应用TC-CCN网络，对样品的真伪的正确识别率为84.44%。红外光谱法作为中药的一种鉴别方法，具有快速、简便的特点。并且中药样品的红外光谱是多种官能团特征振动峰的叠加与组合，只要中药材的各成分组成相对稳定，其光谱就有一定的重现性。因此，利用红外指纹图谱有可能对中药进行鉴定。将光谱技术与神经网络方法结合起来使中药的鉴别更加简便和快速，且具有较高的正确识别率，可见该方法会成为中药鉴别的一种有效手段。更多还原

【Abstract】 Radial basis function (RBF) neural network, back-propagation (BP) neural network and temperature-constrained cascade-correlation network (TC-CCN) were applied to identify official and unofficial rhubarb samples based on infrared reflectance spectrometry (IRS). The effect of network parameters was investigated. Wavelet transformation was used to compress the original data. The original data were compressed from 775 to 49 variables. The compressed data not only preserved the feature peaks but also reduced the number of variables, and improved the speed of network training.With the RBF network, the classification correctness rates of the rhubarbs can be achieved that is 97.78%. And with the BP network, the classification correctness rates of the rhubarbs can be achieved that is 95.56%. And with the TC-CCN, the classification correctness rates of the rhubarbs can be achieved that is 84.44%.As a means of classification, the IRS makes the analyses simpler and much faster. The IRS of rhubarbs is added up. If all sorts of compositions of the rhubarbs are stable, the rhubarbs’ IRS is recurred. Using artificial neural network (ANN), we can find a smooth representation of the underlying trends in the data. The combination of IRS and ANN make the identification of Chinese herbal medicine more convenient and faster.更多还原