小波变换在提高油菜籽脂肪酸近红外分析精度中的应用

利用小波变换预处理结合偏最小二乘回归建立光谱定量分析模型,比较其与传统建模结果的差异。采用db4小波9层分解光谱,保留部分中间层细节系数,其他细节系数软阈值处理,低频逼近系数舍去处理,最后重构光谱用于建模。结果表明,该方法与传统误差校正方法相比,具有预处理简单,预测精度高的优点。

Application of Wavelet Transform in Enhancing the Accuracy of Near-infrared Analysis of Rapeseed Fatty Acids

Wavelet transform was employed to preprocess NIRS and the model was acquired by partial least squares(PLS)regression which was evaluated by comparing the results with conventional preprocessing methods.The coefficients of original NIRS decomposed at 9 layers by the wavelet function of Db4 were treated as follows:the details of some middle layers were remained and the others were filtered by soft-thresholding,then approximation of low frequency was removed.The NIRS reconstructed based on the treated coefficients of WT were used for model-building and predicting.The results showed that the method had the advantages of simplifying pre-treatment and improving prediction accuracy comparing with conventional method.