将移动窗口偏最小二乘(MWPLS)应用于脐橙中重金属Cd含量的激光诱导击穿光谱(LIBS)定量分析模型中, 通过改变MWPLS窗口宽度并结合标准归一化处理、一阶导数、二阶导数、中心化处理和多元散射校正等5种数据前处理方法, 优选与脐橙中Cd元素相关性高的光谱区间, 并与传统偏最小二乘法进行对比分析。模型评价及验证结果显示, 当优选移动窗口为61个波长宽度、优选区域为218.61~222.55 nm时, 结合一阶导数数据前处理方法所构建的模型效果最佳, 验证集决定系数、预测均方根误差、主因子数、平均预测相对误差分别为0.9953, 15.10×10-6, 12, 7.43%。MWPLS结合合适的数据前处理方法可以筛选出脐橙中Cd元素的LIBS光谱区域, 提高定量分析模型的预测能力。

The quantitative method named moving window partial least square (MWPLS) is applied to predicting the content of Cd in navel orange by analyzing the spectral information obtained by laser induced breakdown spectroscopy (LIBS). Five methods like standard normal variate (SNV), first derivative (FD), second derivative (SD), centralization and multivariate scatter correction (MSC) are utilized to process LIBS data and acquire better spectral quality, and the window width of MWPLS is optimized to select the area of wavelength correlated with Cd element. Meanwhile, MWPLS is compared with conventional partial least square method. The results display that the model built by FD is better when moving window is 61 wavelength wide and the range of wavelength is 218.61-222.55 nm. At this condition, the determinant coefficient of validation set, root mean square error of prediction (RMSEP), principal component and average relative error are 0.9953, 15.10×10-6, 12 and 7.43%, respectively. This study shows that the LIBS area for predicting Cd in navel orange can be screened by MWPLS coupled with suitable data preprocessing, and the prediction ability can be improved effectively.