【摘要】 以三级大豆油脂酸价检测为研究对象,采用db4小波对36个油脂样品的近红外光谱进行预处理,选择出最佳的小波分解层数及阈值选取方式,并对原始光谱和经小波去噪后的光谱利用偏最小二乘回归建模,以决定系数、交互验证误差均方根和相对误差均值3个统计量来评价回归模型性能。结果表明原始光谱决定系数0.8668、交互验证误差均方根0.0641、相对误差均值3.28%;小波去噪后光谱决定系数0.9772、交互验证误差均方根0.0536、相对误差均值2.26%。小波变换能在消除噪声的基础上最大限度地保留原始信号的能量和特征信息,为建立稳健可靠的预测模型打下基础。更多还原

【Abstract】 The detection of third degree soybean oil acid value was used as research object,near infrared spectral datas of 36 oil samples were preprocessed by db4 wavelet,best wavelet decomposition layer and selection way of threshold were chose,then models of initial spectrum and wavelet denoised spectrum were constructed using partial least square regression,and the performances of models were evaluated by three statistics named decision coefficient r2,RMSECV and the mean of relative error,the r2,RMSECV and the mean of relative error were 0.8668,0.0641 and 3.28%,respectively,for initial spectrum,and were 0.9772,0.0536 and 2.26%,respectively,for wavelet denoised spectrum.Obviously,the wavelet transform can remain as much as possible energy and feature information of initial signals,at the same time eliminate the noise,which laid the foundation for constructing stable and reliable forecasting model.更多还原