【摘要】 目的利用小波包多尺度分析去噪感康近红外光谱,建立有效测定感康中对乙酰氨基酚、盐酸金刚烷胺的近红外光谱分析模型。方法使用最优分解方法和合适的阈值处理得到重构光谱的小波包分解系数,达到降噪的目的。结合偏最小二乘法(PLS)建立不同尺度分解重构光谱的感康定量分析模型。结果模型的校正和预测精度在小波包尺度为3的时候最好,模型的相关系数对对乙酰氨基酚和盐酸金刚烷胺分别为0.995 95和0.913 07。交互验证均方与误差(RM-SECV)分别为0.005 62和0.006 86,预测均方与误差(RMSEP)分别为从0.005 79和0.007 88减小为0.004 78和0.00603,与原始光谱相比模型的稳定性和预测性能都得到了优化。结论使用小波包变换的多尺度分析能够消除原始光谱的噪声,使建立定量分析的PLS模型具有更好的稳健性和预测精度。更多还原

【Abstract】 Objective To develop a novel algorithm of wavelet packet transform multi-scale analysis for denoising near infrared diffuse reflectance spectroscopy(NIRS) to determine paracetamol and amantadine hydrochloride in Gankang.Methods The partial least squares(PLS) model with restructure Gankang NIRS was advanced and it was used to determine prediction sets.Results The optimum scale for the model was 3.The regression coefficients(R) of model for paracetamol and amantadine hydrochloride were 0.995 95 and 0.913 07,respectively.The root mean square error of cross-validation(RMSECV) of them reduced from 0.00709 and 0.008 81 to 0.005 62 and 0.006 86,and the root mean square errors of prediction(RMSEP) of them were 0.004 78 and 0.006 03,respectively.Conclusion Wavelet packet transformation multi-scale analysis is an effective method for reducing noises in NIRS,which makes PLS model more typically and moderately.The model’s efficiency and precision are improved.更多还原