建立支持向量机SVM识别模型对翡翠产地进行识别

Establishing Support Vector Machine SVM Recognition Model to Identify Jadeite Origin

为了实现翡翠产地的快速无损鉴别,丰富宝玉石产地鉴别方法的多样性,基于红外光谱分析得到的数据,建立支持向量机(SVM)识别模型对三个产地的翡翠进行分析。实验收集了缅甸、俄罗斯和危地马拉3种翡翠的红外光谱数据共106条,为了达到更好的模型识别效果,建模前将原始的红外光谱数据进行反射率到吸光度的转化,再对光谱进行不同的预处理。预处理的目的是降低噪声、基线漂移和散射现象等对模型识别效果的影响。本次实验预处理使用的方法有SG平滑、均值中心化、标准化、趋势校正、多元散射校正、最大最小归一化、标准正态变换以及标准正态变换后再进行趋势校正。实验结果表明,对红外光谱进行预处理后模型得到的识别准确率均高于原始光谱的73%;三个产地翡翠的红外光谱分开进行多元散射校正和最大最小归一化得到的模型识别准确率高于混合进行预处理得到的结果;一些预处理方法结合使用也会提高模型的识别准确率,如标准正态变换和趋势校正。对三个产地翡翠的红外光谱分开进行最大最小归一化处理后得到的识别准确率达到了最高的95%,说明这种采用红外光谱技术建立的支持向量机(SVM)识别模型可以实现对翡翠产地的快速识别。

To realize the rapid and non-destructive identification of jadeite origins and enrich the diversity of methods for the identification of precious jadeite origins,a support vector machine SVM recognition model was established to analyze jadeite of three origins based on the data obtained from infrared spectral analysis.The experiments collected a total of 106 infrared spectral data of three jadeite species from Myanmar,Russia and Guatemala in order to achieve better model identification,the original infrared spectral data were transformed from reflectance to absorbance before modeling,and then the spectra were pre-processed differently.The purpose of preprocessing is to reduce the effects of noise,baseline drift and scattering phenomena on the model recognition effect.The methods used for preprocessing in this experiment are SG smoothing,mean centering,normalization,trend correction,multivariate scattering correction,maximum-minimum normalization,standard normal transformation and standard normal transformation followed by trend correction.The experimental results show that the recognition accuracy of the models obtained by preprocessing the infrared spectra is higher than that of the original spectra by 73%;the recognition accuracy of the models obtained by multivariate scattering correction and maximum-minimum normalization of the infrared spectra of the three emerald origins separately is higher than that of the results obtained by mixing preprocessing;some preprocessing methods used in combination also improve the recognition accuracy of the models,such as standard normal transform and trend correction.The recognition accuracy obtained after maximum-minimum normalization of the infrared spectra of the three origins of jadeite separately reached the highest 95%,indicating that this support vector machine SVM recognition model built using infrared spectroscopy can achieve rapid recognition of jadeite origins.