建立基于激光诱导击穿光谱仪技术获取的半定量青白色软玉的微量元素含量的人工神经网络模型, 以促进人工神经网络技术在宝石产地溯源方面的应用。 以我国新疆、 广西、 江苏、 青海, 以及韩国和俄罗斯六个产地的青白色软玉为样品, 利用激光诱导击穿光谱仪在颜色均匀干净的部分获取元素含量数据。 使用数据筛选原则对数据进行了筛选和Al的归一化处理之后, 以因子分析和线性回归分析讨论了数据间的共线性, 在数据间不存在明显多重共线性的情况下建立了三层人工神经网络的判别模型。 结果表明, 所选取的每个变量的VIF值小于5, 数据间不存在明显的多重共线性, 因子分析的KMO值小于0.6, 表明变量间无明显关系。 同时利用软玉t-SNE图对数据进行降维和可视化处理, t-SNE图显示大部分数据点都重叠在一起, 表明对此数据进行简单聚类和相关分析是无法区分产地的, 因此选择人工神经网络的方法对六个产地的数据进行产地判别分析。 经人工神经网络模型迭代判别之后, 模型对我国新疆、 广西、 江苏、 青海, 以及韩国和俄罗斯六个产地的青白色软玉判别的精度达到0.933, 其中韩国软玉的数据判别结果精度最高, 达到0.995, 误差为0.028, 青海软玉的数据判别结果最低为0.803, 误差为0.090。 综上所述, 激光诱导击穿光谱结合人工神经网络的方法在宝石产地溯源方面的应用是具有很大潜力的。

To promote the application of artificial neural network technology in identifying the origin of gems, an artificial neural network model of semi-quantitative trace element content of bluish white nephrites obtained by laser-induced breakdown spectrometer was established. The element content data were obtained by laser-induced breakdown spectrometer in the uniform and clean parts of nephrites from six regions: Xinjiang, Guangxi, Jiangsu, Qinghai, Korea and Russia. After screening using data filtering principles and normalizing the data, the collinearity between data is discussed by factor analysis and linear regression, and the discriminant model of the artificial neural network is established. The results show that the VIF value of each selected variable is less than 5, so there is no obvious multicollinearity among the selected elements. The KMO value of factor analysis is less than 0.6, indicating that there is no obvious relationship between variables. Moreover, thet-SNE graph of nephrite is used to reduce and visualize the data. T-SEN graph shows that most of the data points are overlapped together, indicating that the data’s simple clustering and correlation analysis could not distinguish the origin. Therefore, the artificial neural network is selected for the identification analysis of the six origin data. After the iterative discrimination of the artificial neural network model, the accuracy of the model for the identification of the blue and white nephrite from six producing areas is up to 0.933, and the nephrite from Korea has the highest data discrimination accuracy of 0.995 with an error of 0.028,while nephrite from Qinghai has the lowest data discrimination accuracy of 0.803 with an error of 0.090. In conclusion, a laser-induced breakdown spectrometer combined with the artificial neural network has great potential in applying gem origin tracing.