【摘要】 提出了一种扣除背景和噪声干扰的新方法一小波包正交校正法。首先将原始光谱进行离散小波包变换,消除噪声及部分背景信息,然后采用正交信号校正法滤除与分析物浓度无关的全部信息。与单纯的离散小波包变换及正交信号校正方法相比,小波包正交校正法能有效地扣除背景和噪声干扰,使模型具有更强的抗干扰能力,提高了模型的预测精度。用该法对牛奶样品的近红外光谱进行处理,并将扣除干扰后的数据采用偏最小二乘法建立校正模型,其脂肪、蛋白质和乳糖的预测均方根误差分别为0.086 5%、0.093 6%和0.1115%,实现了牛奶样品常规组分的定量分析,进一步证明这种算法是切实可行的。更多还原

【Abstract】 A new hybrid algorithm(DWPTOSC) is proposed for eliminating the interferences of background and noise in near-infrared(NIR) data,which combines discrete wavelet packet transform and orthogonal signal correction. First,discrete wavelet packet transform is employed to remove noise and partial background information;then orthogonal signal correction is applied to remove the information uncorrelated to the concentrations of the analyte. The prediction ability and robustness of the proposed DWPTOSC algorithm are superior to those obtained using either discrete wavelet packet transform or orthogonal signal correction independently.The DWPTOSC algorithm was validated in a milk sample processing experiment,in which the contents of the main constituents in the milk sample were estimated based on the NIR spectra of the milk.The predicted root mean square errors of the calibration models for fat,protein and lactose are 0.086 5%,0.093 6%and 0.111 5%,respectively.The results show that DWPTOSC is an effective method for eliminating the interferences in NIR spectra.更多还原