基于太赫兹时域光谱和模式识别技术软玉和仿品鉴别

Identification of Nephrite and Imitations Based on Terahertz Time-Domain Spectroscopy and Pattern Recognition

玉石是一种稀有的矿物质,自古以来备受国人喜爱,其真伪鉴别一直是珠宝鉴别行业的棘手难题,传统的鉴别方法已经难以实现对真假玉石的准确鉴别。太赫兹检测技术可以实现快速无损检测,在混合物的分类鉴别方面,有广泛的应用。基于太赫兹时域光谱技术和模式识别技术,对来自我国新疆、青海,以及巴基斯坦、阿富汗四个地区的软玉样品及玻璃、大理石、石包玉三种仿品,使用透射模式测得样品在0.1~1.5 THz频率范围内的太赫兹谱,通过参数提取得到其折射率谱线。由于其化学成分的复杂和多样性,仅靠其特征谱线图,并不能正确的区分软玉和仿品,为了更好的对其进行鉴别,需要建立分类模型。采用主成分分析(PCA)对实验得到的原始折射率数据进行降维和特征提取,作出样品在第一、二主成分上的二维得分图,在图中可以看出软玉和仿品能够很明显的区分开来。在经过降维处理之后的数据中,随机选取其中的四分之三作为训练集,剩下的作为测试集,输入到支持向量机(SVM)建立的分类模型中,并引入网格搜索(GridSearch)、遗传算法(GA)和粒子群算法(PSO)对支持向量机参数进行优化。结果显示,基于网格搜索的支持向量机最优参数c=2.828 4,g=2,识别率为97.7%,运行时间为1.39 s,用时最短;基于遗传算法的支持向量机最优参数c=1.740 1,g=4.544 6,识别率为98.3%,运行时间为3.6 s;基于粒子群算法的支持向量机最优参数c=11.287 2,g=1.833 1,识别率为98.6%,运行时间为6.13 s,用时最长。虽然三种优化算法得到的最优参数不同,但均可实现正确的分类。研究结果表明,使用太赫兹时域光谱技术结合模式识别方法可以快速、准确的鉴别软玉和仿品,这为玉石的鉴别提供了一种新手段。

Jade is a rare mineral that people have favored. The identification of jade authenticity has always been a thorny problem in the jewelry identification industry. Traditional identification methods are difficult to identify the nephrite and their imitations.Terahertz standoff detection technology can realize quick non-destructive testing and has a variety of applications in the classification and identification of mixtures. In this paper, Terahertz Time-domain Spectroscopy(TDS) and pattern recognition are applied to identify nephrite and imitations. The terahertz spectrum of several nephrite jade samples from Afghanistan, China’s Qinghai, Pakistan and China’s Xinjiang and imitations, like glass, marble, and raw gemstone is measured with TDS in the frequency range 0.1~1.5 THz. Due to the complexity and diversity of the sample’s chemical composition, the nephrite jade and the imitation cannot be distinguished correctly withtheir characteristic spectrum. In order to distinguish Jade with their imitations, a classification model is established.Principal Component Analysis(PCA) performs dimension reduction and feature extraction on the refractive index. The scores of the first and second principal components of the sample were obtained. It can be found that nephrite and imitations can be clearly distinguished from each other. Based on the extracted data, third quarters of them are randomly selected as the training set, the rest as the test set, a Support Vector Machine(SVM) model is established, and the parameters of the Support Vector Machine is optimized by GridSearch, genetic algorithm(GA) and particle swarm algorithm(PSO). The optimal parameters of SVM based on grid search are c=2.828 4 and g=2 while that based on GA are c=1.740 1, g=4.544 6 and based on PSO c=11.287 2, g=1.833 1. The recognition rates of the three optimization algorithms are 97.7%, 98.3% and 98.6%, and the running time is 1.39, 3.6, 6.13 s respectively. Although the optimal parameters obtained by the three optimization algorithms are different from each other, all of them can achieve a correct classification. The results show that the Terahertz spectrum combined with the pattern recognition method is a promising technique for identifying nephrite with their imitations.