小波变换和偏最小二乘法在烟草常规成分预测中的应用

Application of Wavelet Transform and Partial Least Square in Prediction of Common Chemical Compositions in Tobacco Samples

摘要: 为了实现烟草样品的快速近红外光谱(NIR)分析,将小波变换(WT)用于烟草样品NIR的数据压缩,并结合偏最小二乘法(PLS)对压缩后的数据进行建模。与直接采用PLS相比,WT-PLS可有效地压缩原始谱图的数据,消除谱图中噪声和背景的干扰,降低所建模型的随机性,从而大大提高了运算速度,并获得了更高的预测精度。

Abstract: In order to analyze tobacco samples with near infrared spectrum quickly, the data was compressed by wavelet transform (WT) and the model on the compressed data was developed by partial least square (PLS).In comparison with PLS algorithm, the WT-PLS effectively compressed the original spectra data, removed the interference of noise and background, and reduced the randomness of the developed model.Therefore, the computation speed was significantly improved and the precision of prediction was increased.