高斯拟合算法在光谱建模中的应用研究

采用高斯拟合算法对光谱进行特征提取，利用拟合得到的特征参量表征光谱信息，并结合多元校正方法对光谱模型进行优化和解释，建立了样品快速准确的测定方法。实验以玉米活体叶片为研究对象，建立叶片光谱与叶绿素含量之间的关系模型，采用三个高斯峰对原始光谱的1 551个数据拟合后，光谱数据转换为9个高斯特征量(约为整个波段的0.58%)，进而利用该高斯特征量来预测叶绿素含量。实验结果显示，采用高斯拟合分别与偏最小二乘法和主成分回归结合建模，其预测集相关系数分别为0.960和0.962；不采用高斯拟合算法而直接采用偏最小二乘法和主成分回归对全光谱建模，其预测集相关系数分别为0.957和0.919。可见，将高斯拟合算法运用到定量分析模型中是可行的，该方法不仅简化了模型参数，而且提高了模型的可解释性。

Study on Application of Gaussian Fitting Algorithm to Building Model of Spectral Analysis

In the present paper, Gaussian fitting algorithm is introduced. It is possible to separate some Gaussian peaks in one spectrum and get peak height value, peak position value and other parameters by this algorithm. These Gaussian parameters are used to describe original spectral information and extract spectral feature. The spectral model is optimized and explained by the combination of the Gaussian fitting algorithm and multivariate calibration methods. The relationship between the spectra and the chlorophyll of the corn’s leaves is studied in this paper, every spectrum which contains 1 551 absorbance data is fitted and separated by three Gaussian peaks, and then 1 551 data are converted to 9 Gaussian parameters (approximately 0.58% of the whole original spectral data), and the chlorophyll content is estimated by the Gaussian parameters. This modeling method is fast and accurate for the estimation of a sample. The model of chlorophyll content with the Gaussian fitting algorithm and PLS is built in the range of 400-800 nm, and the experiment results show that the correlation coefficient between the estimated values and the real values is 0.960, and the relative standard deviation is 0.0485; The model of chlorophyll content with the Gaussian fitting algorithm and PCR is built in the same wavelength range, and the experiment results show that the correlation coefficient is 0.962, and the relative standard deviation is 0.048; while the correlation coefficient is 0.957 and the relative standard deviation is 0.051 for the model of PLS without Gaussian fitting algorithm; and the correlation coefficient is 0.919 and the relative standard deviation is 0.077 for the model of PCR without Gaussian fitting algorithm. The reliability of the prediction results shows that Gaussian fitting algorithm is satisfactory for building model of spectral analysis. Compared to the conventional methods, the method of Gaussian fitting algorithm can not only simplify the parameters of models, but also improve the explanation of analysis models. The result of the study shows that it is practical and feasible to apply the Gaussian fitting algorithm to quantitative analysis models.