采用激光诱导击穿光谱（LIBS）技术结合偏最小二乘判别分析（PLS-DA）对新疆、青海和俄罗斯的白色软玉进行产地研究。选取产自新疆（和田、于田、且末）、青海（格尔木）、俄罗斯（贝加尔湖）的146个白色软玉样品作为样品集，从样品集中随机抽取111个样品作为校正集，用于建立PLS-DA识别模型，剩余35个样品作为验证集，用于检验PLS-DA识别模型的预测效果。采用LIBS对三个产地的软玉样品进行成分分析，选择Na、K、Al、Li、Be、Mn、Sr、Zr、Ba、Y、Ce作为目标元素，并选取589.995，766.490，396.152，670.793，313.042，257.610，407.771，389.138，455.403，437.493，401.239 nm处的谱线作为目标元素的分析谱线，选取Si元素作为内标元素，以其在 288.158 nm处的谱线作为内标元素分析谱线，分别计算各目标元素与内标元素的谱线强度的比值Rx，由Rx组成自变量矩阵，用于模型的建立与预测。实验结果表明，采用LIBS结合PLS-DA建立的产地识别模型，其校正自变量和验证自变量与实际分类变量的相关系数都大于0.9，PLS-DA识别模型的交叉验证均方根误差和预测均方根误差均小于0.29，PLS-DA产地识别模型对验证集中新疆（和田、于田、且末）、青海（格尔木）、俄罗斯（贝加尔湖）产地的35个白色软玉样品的识别正确率为92%。研究表明，采用LIBS结合PLS-DA能够快速有效识别三大产地的白色软玉。

Identifying the origins of nephrite from three different places has been studied, using laser-induced breakdown spectroscopy (LIBS) coupled with partial least squares discriminant analysis (PLS-DA) approach. 146 nephrite specimens from Xinjiang (Hotan, Yutian, Qiemo), Qinghai (Golmud) and Russia (Baikal) were selected as research subjects. 111 of the specimens were chosen as calibration samples to build the PLS-DA model, and the rest 35 specimens were used as prediction samples to test the PLS-DA model. In this study, LIBS was used to test and analyze the element composition of nephrite specimens. Na, K, Al, Li, Be, Mn, Sr, Zr, Ba, Y and Ce were chosen as analysis object elements, and their spectral lines at 589.995, 766.490, 396.152, 670.793, 313.042, 257.610, 407.771, 389.138, 455.403, 437.493, 401.239 nm were selected as analysis spectral lines. The LIBS spectrum of Si at 288.158 nm was chosen as internal spectral line. Ratio between the intensity of analysis spectral lines Rx and the intensity of analysis internal spectral line was calculated. The matrix of independent variables composed by Rx was applied to calibrate and test the PLS-DA mode. This study shows that for discrimination mode built by PLS-DA approach coupled with LIBS, the correlations between category variables of calibration or prediction and the measured category variables are all remarkable with a correlation coefficient over 0.9, and low root mean squared error of cross calibration and root mean square error of prediction (less than 0.29). The discrimination accuracy for the nephrite from three different origins is 92% by PLS-DA model based on the validation set of samples. The results indicate that using LIBS coupled with PLS-DA approach can achieve a well recognition of the origins of nephrite specimens.