建立了一种傅里叶变换红外光谱（Fourier transform infrared spectroscopy, FTIR）快速无损鉴别当归及其不同炮制品多糖的新方法。 生当归（unprocessed Angelica sinensis, UAS）经炮制为酒当归(angelica sinensis parched with wine, WAS)、 土当归(angelica sinensis parched with soil, SAS)、 油当归(angelica sinensis parched with oil, OAS)和当归炭(charred angelica sinensis, CAS), 为了有效、 合理地使用当归及其不同炮制品多糖, 对其进行鉴别分析很有必要。 首先获取当归及其不同炮制品多糖的FTIR, 运用离散小波变换方法(discrete Wavelet transform, DWT)对其进行分解, 然后选取2, 3, 4尺度下的高频信息作为特征信息, 分别提取其小波熵作为特征值, 采用BP神经网络进行训练, 再用训练所得模型对当归及其不同炮制品多糖的FTIR小波提取出的特征值进行智能判别。 通过对30个样本进行预测, 判别率高达93.3%, 说明基于FTIR离散小波特征提取及BP神经网络分类法识别当归及其不同炮制品多糖, 具有较优的可行性。

A new rapid and nondestructive method for identifying polysaccharides from Angelica sinensis and its different processed products was developed, and this method was based on Fourier transform infrared spectroscopy (FTIR). In the clinic of traditional Chinese medicine, unprocessed Angelica sinensis(UAS) is of ten used after processed, the common processed products are Angelica sinensis parched with wine(WAS), Angelica sinensis parched with soil(SAS), Angelica sinensis parched with oil(OAS) and Charred Angelica sinensis(CAS). In order to use polysaccharides from Angelica sinensis and its processed products effectively and reasonably in clinic, it is very necessary to identify them. FTIR of polysaccharides from Angelica sinensis and its different processed products was determined, and then it was decomposed by discrete wavelet transform (DWT). The high frequency information in scale 2, 3 and 4 was selected as feature information, from which the each wavelet entropy was extracted as characteristic value. BP neural network was trained with these characteristic values. The trained BP neural network was used to identify polysaccharides from Angelica sinensis and its different processed products. According to 30 prediction samples, the correct rate for recognition was 93.3%, which indicates that: it has better feasibility to identify polysaccharides from Angelica sinensis and its different processed products by this method, which is based on FTIR, discrete wavelet transform and BP neural network.