对应分析在近红外有效波长选择的应用

针对分立波长型近红外判别仪器开发过程中遇到的组合波长优选问题,提出一种基于对应分析的有效波长选择方法.对11种蘑菇样品及其在701-2500nm的光谱信息作对应分析,在对应图上计算各样品点与波长点的距离后,首先由各样品点的最近距离波长点确定出最稳健波长组合,在此基础上,对各样品点的前10、50、200个最近距离波长点进行优选,得出用921nm、1376nm、1424nm、1720nm、2233nm、2454nm这6个波长建立的SVM模型,能在输入变量最少的情况下达到用最稳健波长组合所建模型的预测能力.

Sensitive Wavelength Selection for Near Infrared Spectral Based on Correspondence Analysis

Sensitive wavelength selection is important in development of near infrared spectrum（NIR） discriminative instrument with discrete wavelengths,a effective wavelength selection method was proposed based on correspondence analysis. 11 kinds of mushroom samples and the infbrmation of reflectance spectrum in 701-2500nm was studied based on Correspondence analysis, calculating the distance between each sample point and wavelength points on correspondence figure, the most robust wavelength group was selected by the nearest distance wavelength points to each sample point. And one of wavelength groups including 921nm, 1376nm, 1424nm 1720nm, 2233nm, 2454nm, were selected within the 10,50,200 nearest distance wavelength points for each sample point, a SVM developed based on them which can get the highest correct discrimination rate with the least input variables.