近红外光谱的数据预处理研究

进行了小麦样品近红外光谱数据的预处理研究，一般仪器记录的样品近红外光谱数据中包含有一系列噪声和干扰信号，因此适当的预处理是进行后续光谱定标、建模及模型传递的基础，对可靠地获得准确分析结果具有很重要的作用。结合小麦样品蛋白质含量近红外光谱分析工作，对由近红外光栅光谱仪和傅里叶变换近红外光谱分别记录的66种小麦样品光谱数据，采用高斯一阶、二阶导数小波变换方法进行了预处理。对比常用的一阶差分预处理，证明高斯函数导数小波变换方法是十分有效、实用的，预处理后光谱曲线非常光滑、噪声消除效果明显，富含有用光谱分析信息的区域更加清晰显示，因而非常有助于后续的光谱定标、建模和模型传递工作。

Preprocessing of Near Infrared Spectroscopic Data

Spectroscopic data of samples recorded by spectroscopic instruments are confused by a series of noises, and interferences, therefore the proper data preprocessing is the basis of the following spectroscopic calibration, model establishment and transference, which is very important for the achievement of accurate analytical results. This paper reports our research work, combined with NIR spectroscopic analyses of the protein contents of wheat, that is the preprocessing of NIR spectral data recorded for 66 different wheat samples by a NIR grating spectrophotometer and a NIR Fourier transform spectrometer, respectively. The preprocessing algorithm is wavelet transform with the Gaussian first and second order derivatives. Compared with the result of preprocessing by normal first order difference algorithm, the wavelet transform algorithm by Gaussian derivatives was proved to be very effective and applicable, the spectra were smoothed perfectly, noises were eliminated obviously, and the spectral sections, which include all useful information for spectral analyses, were displayed clearly. So, it is very beneficial to the following spectral calibration, model establishment and transference.