自适应的EEMD残余相关基线校正算法

基线校正是光谱分析的重要环节，现有算法通常需要设定关键参数，不具备自适应性。根据总体平均经验模态分解(ensemble empirical mode decomposition，EEMD)残余量特点，提出用残余量拟合光谱基线。通过残余量与信号相关性、残余量自相关和互相关性(称为残余相关准则)判断残余量是否是基线组成部分，以此为基础提出一种自适应的EEMD残余相关基线校正算法。对叠加曲线背景和线性背景的模拟光谱数据进行实验，结果显示在已知基线数学假设情况下，EEMD残余相关法逊于多项式拟合，同非线性拟合相差不多，优于小波分解。在没有光谱背景知识情况下，对真实拉曼光谱数据进行试验。经过上述方法预处理过的玉米叶片光谱采用3层BP神经网络建立与叶绿素之间预测模型，经过残余相关基线校正的模型具有最大校正相关系数和预测相关系数，最小交叉验证标准差和相对分析误差。各种基线校正方法中，残余相关基线校正对特征峰峰位、峰强和峰宽影响最小。实验表明，该算法可用于拉曼谱图基线校正，无需分析样品成分的先验知识，无需选择合适的拟合函数、拟合数据点、拟合阶次以及基函数和分解层数，也无需基线信号分布的数学假设，自适应性很强。

Adaptive EEMD Residue Related Baseline Correction Algorithm

Baseline correction is an important part of spectral analysis; the existing algorithms usually need to set the key parameters and does not have adaptability. The spectral baseline is fitted by the residue according to the feature of ensemble empirical mode decomposition (EEMD for short). The correlation between residual and original signal, the self-correlation and the crosscorrelation of residual form the residual related rule. The residual related rule is proposed to judge whether the residual is a component of baseline, based on which adaptive EEMD residual related base line algorithm is proposed. With experiment on the simulated spectrum data of superimposing curve background and the linear background, the results showed that in the case of known baseline mathematical assumption: EEMD residual related method is not so good for polynomial fitting, it is almost no difference from linear fitting, but is better than the wavelet decomposition. In the absence of spectral background knowledge, the real Raman spectrum data are tested. The model is established between Raman spectra treated by the procedure above and chlorophyll, and the model corrected by EEMD residual related baseline method has the biggest correlation coefficient and prediction coefficient, but the smallest root mean square error of cross validation and relative prediction deviation. The effect of EEMD residual related baseline method effects on the peak position, peak intensity and peak width is the smallest in all kinds of baseline correction methods. EEMD residual method has the best baseline correction effect. Experiments show that this algorithm can be used for Raman spectra baseline correction, without prior knowledge of the sample composition analysis, and there is no need to select appropriate fitting function, fitting data points, fitting order as well as basis function and decomposition levels, also there is no need of mathematical hypothesis of baseline signal distribution, so the adaptability is very strong.