紫外拉曼光谱具有拉曼散射强度高、 易于荧光光谱分离、 受环境干扰影响小以及人眼安全性高等特性, 所用的紫外拉曼光谱仪采用波长266 nm激光器, 拉曼和荧光光谱会有部分重叠, 增加了准确获取拉曼光谱特征信息的难度, 进一步影响样品的辨识。 因此, 需要在分析拉曼光谱之前进行基线校正来消除荧光干扰。 根据紫外拉曼+荧光混合光谱中, 荧光光谱具有逐渐增加且接近分段线性递增的特点, 利用分段线性函数拟合荧光光谱基线是一种较简捷的方法, 于是针对传统分段线性拟合基线校正方法基线点定义过度依赖操作人员、 自动化水平较低等问题, 研究了一种改进的紫外拉曼光谱分段线性拟合基线校正方法: (1)首先求原始信号经不同次平滑迭代后的光谱数据。 由于波峰相对于基线是高频信号, 在多次平滑过程中, 波峰附近的光谱强度逐渐下降且变化较大, 基线部分逐渐上升且相对变化很小, 经不同次迭代平滑的光谱波峰和基线点处的光谱强度标准差SD差异较大。 (2)然后通过对光谱强度偏差的比较确定准有效基线点位置。 通过适当设定的阈值SD0提取出准有效基线点位置; (3)再利用线性迭代拟合法提取并修正过校正基线点。 准有效基线点将整个拉曼光谱分割成N个特征峰区间, 分别连接特征峰区间两端点得到一条直线, 若特征峰全部在直线以上表明不存在过校正, 否则区间端点向其峰方向移动并再次直线连接, 重复以上过程, 直到特征峰全部在直线以上, 得到有效基线点; (4)最后逐段直线连接所有相邻有效基线点得到整个光谱的基线。 原始光谱减去基线就是基线校正后的拉曼光谱。 通过对模拟和实际测量的紫外混合光谱的基线校正处理实验表明: 该方法能自动确定基线点位置, 且较传统方法能获得更好的基线校正效果, 为下一步的光谱分析提供更准确的光谱信息。

UV Raman spectroscopy has the characteristics of high Raman scattering intensity, easy fluorescence spectrum separation, little influence by environmental interference and safety to the human eye. In this paper, the ultraviolet Raman spectrometer uses a laser with a wavelength of 266 nm. The Raman spectrum and the fluorescence spectrum will partially overlap, which increases the difficulty of accurately obtaining the characteristic information of the laser Raman spectrum, and further affects the identification and analysis of the sample. Therefore, baseline corrections need to be performed prior to analyzing Raman spectroscopy to eliminate fluorescence interference. According to the distribution characteristics of the mixed spectrum of ultraviolet Raman and fluorescence, the fluorescence spectrum has a gradual increase and is close to a piecewise linear increase. Therefore, fitting a fluorescence spectral baseline using a piecewise linear function is a relatively simple method, so that the troughs of the characteristic peaks just fall on the baseline. Aiming at the problem that the traditional piecewise linear fitting baseline correction method is over-reliant on the operator and the low level of automation, improved UV-Raman spectroscopy piecewise linear fitting baseline correction method is studied. (1) First the spectral data of the original signal after different smoothing iterations is obtained. Since the peak is a high-frequency signal with respect to the baseline, the spectral intensity at the peak position gradually decreases and changes greatly, while that at the baseline portion gradually rises and the relative change is small during the multiple smoothing process. So the standard deviation (SD) of the spectral intensity at the spectral peaks and the baseline points is different after different smoothing iterations. (2) Then the position of the quasi-valid baseline points is determined by comparing the spectral intensity deviations. The quasi-valid baseline points can be extracted by appropriately setting the threshold; (3) Next the quasi-valid baseline points divide the entire Raman spectrum into N characteristic peak intervals. Comparing the lines obtained by connecting the two ends of the characteristic peak interval with the spectral intensity of the characteristic peak interval, if the characteristic peaks are all above the straight line, there is no over-fitting, otherwise the endpoints of the characteristic peaks move toward the peak direction and are connected again by straight lines. The above process is repeated until the characteristic peaks are all above the line connecting the two ends of the interval, and the valid baseline points are obtained. (4) Finally, all adjacent valid baseline points are connected in a straight line by segment to get the baseline of the entire spectrum. The corrected Raman spectrum is obtained by subtracting the baseline from the original spectrum. Baseline calibration experiments of simulated and actual measured UV and fluorescence hybrid spectra show that the method of this paper can automatically determine the position of the baseline point and obtain better baseline correction effect than the traditional method, which will provide more accurate spectral information for the next spectral analysis.