基于神经网络的近红外光谱鉴别蜂蜜品种研究

Determination of floral origin of honey by near infrared spectroscopy based on artifical netural network

提出了蜂蜜品种鉴别的一个新方法。首先用傅里叶变换近红外光谱仪（FT-NIR）在4000～12000cm^-4范围扫描3种蜂蜜样本（荆条蜜、枣花蜜和槐花蜜），获得的近红外光谱图经过一阶导数和Savitzky-Golay平滑处理。其光谱数据用主成分分析（PCA）法进行分析得到主成分数据，结合BP人工神经网络（BP—ANN）建立蜂蜜品种鉴别模型。前13个主成分的累计贡献率达99．91％。其主成分得分值作为BP-ANN的输入层，品种值作为输出层建立3层BP神经网络模型，135个作为建模样本，余下15个样本用于预测。研究结果表明，预测识别准确率达100％。该方法简单、可靠，结果较满意。

A new method for discrimination of botanical origin of honey was developed. First, three kinds of honey samples were scanned by Fourier tansform near infrared spectrometer（FT-NIR） in the regions 4000 -12000 cm^-1, transformed in 1 st derivatives and Savitzky-Golay smooth. Secondly, the spectra data were analyzed by principal component analysis （PCA）, combined with BP artifical neural network （BP-ANN） for classification different floral honey samples and the model were eatablished. The eigenvalue of the first 13 principal components of which cumulative contribution reached 99.91%. were applied as BP-ANN inputs, the values of type of honey were applied as outputs, and then the three layer BP-ANN model was build. In this model, 135 honey samples were calibration sets and the rest 15 samples were used as prediction set. The analysis results showed that 100% of honey samples was correctly predicted and the technique might be simple, reliable and achieve satisfactory results for classification of honey varieties.