CAJ | 학술논문

为了研究 NIR 光谱的不同波长选择算法在中药质量在线检测中的应用，应用自编的化学计量学软件 THUNIR对两大类6种波长选择算法进行建模。在模型评价上，以留一法交叉验证均方根误差（RMSECV）或测试集均方根误差（RMSEP）最小为最优。实验表明，Filter波长选择方法与Wrapper波长选择方法的RMSECV值均比全谱的RMSECV值要小的多，用于建模的波长数也比全谱也少的多。可见，NIR光谱的波长选择方法不仅可以简化模型，还可以提高模型的预测能力。
위료연구 NIR 광보적불동파장선택산법재중약질량재선검측중적응용，응용자편적화학계량학연건 THUNIR대량대류6충파장선택산법진행건모。재모형평개상，이류일법교차험증균방근오차（RMSECV）혹측시집균방근오차（RMSEP）최소위최우。실험표명，Filter파장선택방법여Wrapper파장선택방법적RMSECV치균비전보적RMSECV치요소적다，용우건모적파장수야비전보야소적다。가견，NIR광보적파장선택방법불부가이간화모형，환가이제고모형적예측능력。
In order to study the NIR spectra of different wavelengths selection algorithm used in traditional Chinese medicine quality on-line detection, application self chemometric software THUNIR two types of six kinds of wavelength selection algorithms for modeling. On the model evaluated to leave a root mean square error of cross validation(RMSECV)or test set RMSE(RMSEP)minimum for the best. Experiments show, Filter wavelength selection methods and RMSECV value Wrapper wavelength selection method RMSECV value than the full spectrum of much smaller, the number of wavelengths used for modeling are also much less than the full spectrum. Visible, NIR spectral wavelength selection method can not only simplify the model, you can also improve the predictive ability of the model.