论文部分内容阅读
近红外光谱数据因为样品含有相同或高度相似的基体,往往具有相似的光谱形状,而某些光谱可能只是反映了基体信息,而与被测物含量没有相关性,称之为无信息光谱。因为光谱的高相似性,无信息光谱很容易混杂在光谱数据中。论文采用玉米近红外光谱数据,通过向其中人为添加无信息光谱数据,研究无信息光谱对近红外光谱定量分析的影响,采用偏最小二乘法,并结合去一交互检验方法建立分析模型。研究发现,当数据集中含有无信息光谱时,所建立的近红外光谱定量分析模型交互检验误差会明显增大,而且无信息光谱数目越多误差越大。所建立模型对独立检验集的预测误差也明显增高。但简单地从光谱和交互检验结果,或者主成分分析都很难发现和鉴别无信息光谱。
Near-infrared spectroscopy data often have similar spectral shapes because samples contain the same or highly similar matrix. Some spectra may only reflect the matrix information but have no correlation with the content of the analyte, and are called non-information spectra. Because of the high similarity of spectra, non-informative spectra can easily be mixed in spectral data. In this paper, corn near infrared spectroscopy data was used to study the influence of non-information spectrum on the quantitative analysis of near infrared spectroscopy by artificially adding non-information spectral data, using partial least-squares method and combining with an interaction test method to establish the analytical model. The results show that when the dataset contains no information spectrum, the error of the interactive verification of the quantitative analysis model of near-infrared spectroscopy will be significantly increased, and the more the number of non-information spectra is, the greater the error will be. The prediction error of the established model for the independent test set is also significantly increased. However, it is very difficult to find and identify non-information spectra simply from the results of spectroscopy and interaction tests, or from principal component analysis.