激光诱导击穿光谱(LIBS)结合字典学习对气溶胶光谱数据筛选方法的研究

Study on the Screening Method of Aerosol Spectral Data by LIBS Combined with Dictionary Learning

气溶胶是大气中的重要组分,对气候、生态环境等均有重要的影响。激光诱导击穿光谱(LIBS)在用于气溶胶检测时,由于气溶胶的离散分布,导致采集到大量无效光谱。通过制备7种不同浓度的NaCl气溶胶样品,选取NaCl溶液(10%)的5000条光谱数据进行分类,其中70%作为训练集,30%作为测试集,建立一种结合字典学习对有效光谱数据进行筛选的方法——K-SVD-SVM。当字典基向量数设置为3时,模型分类性能最优,准确率(Accuracy)、精确率(Precision)、召回率(Recall)、精确率和召回率的调和平均(F1)分别达到96%、95%、95%、0.95。此外,采用K-SVD-SVM方法对7种不同浓度的气溶胶样品进行筛选后,输入结合遗传算法的极限学习机(GA-ELM)模型开展定量分析,同时将未筛选的原始光谱数据输入定量模型进行对比。未筛选的原始数据测试集RMSE和R^(2)分别是0.0303和0.8726,筛选光谱后,分别提升至0.0187和0.9809。结果表明,K-SVD-SVM方法有着较好的分类性能,且采用此方法筛选出的有效数据可以为气溶胶中元素定量分析提供数据支撑。

Aerosol is an important component of the atmosphere,which has an important impact on climate and ecological environment.When laser-induced breakdown spectroscopy(LIBS)is used for aerosol detection,a large number of invalid spectra are collected due to the discrete distribution of aerosols.In this study,a method to filter effective spectral data by combining dictionary learning,a K-SVD-SVM method,was proposed.By preparing seven different concentrations of NaCl aerosol samples,5000 spectral data of 10%NaCl solution were selected for classification,of which 70%were used as training sets and 30%as test sets.When the number of dictionary basis vectors was set to 3,the model classification performance was optimal,and the harmonic mean(F 1)of accuracy,precision,recall,precision,and recall went to 96%,95%,95%and 0.95,respectively.In addition,the K-SVD-SVM method was used to screen seven aerosol samples with different concentrations,and the GA-ELM model was input for quantitative analysis.At the same time,the unscreened original spectral data was input into the quantitative model for comparison.The RMSE and R 2 of the unscreened original data test set were 0.0303 and 0.8726,and were increased to 0.0187 and 0.9809 after screening the spectrum.The results showed that the K-SVD-SVM method had good classification performance,and the effective data selected by this method could provide data support for quantitative analysis of elements in aerosols.