摘要
利用激光诱导击穿光谱(LIBS)技术结合机器学习算法,对9个同基国家标准样品合金钢进行牌号识别.采用统计敏感的非线性迭代剥峰(SNIP)算法对原始合金钢LIBS光谱进行了连续背景扣除.主成分分析(PCA)算法用于对光谱数据降维处理,消除光谱冗余信息.前10个主成分累计贡献率为94.3%,9个同基合金钢LIBS光谱数据按7∶3划分训练集与测试集,以主成分分析前 10个主成分为输入量建立了PCA-支持向量机(SVM)、PCA-决策树、PCA-K最近邻(KNN)和PCA-线性判别分析(LDA)同基合金钢识别模型.4种模型训练集 9种牌号的平均准确度分别为 99.06%、97.47%、90.47%和100%,测试集平均准确率分别为96.29%、79.63%、67.04%和100%.其中PCA-LDA算法的同基合金钢牌号识别率达到了100%.本研究为激光诱导击穿光谱技术在同基合金钢牌号的快速识别提供了方法和参考.
Abstract
In this study,laser-induced breakdown spectroscopy(LIBS)combined with machine learning algorithms was employed to identify the grades of nine homogeneous,national,standard alloy-steel samples.The original LIBS spectra of the alloy steels were processed using a statistically sensitive nonlinear iterative peak-clipping(SNIP)algorithm for continuous background subtraction.Principal component analysis(PCA)was used to reduce the dimensionality of spectral data and eliminate redundant information.The first 10 principal components constitute 94.3%of the total variance.The LIBS spectral data of the nine homogeneous alloy steels were partitioned into a 7∶3 ratio to create training and testing datasets.Based on the first 10 principal components obtained from PCA,PCA-support vector machine(SVM),PCA-decision tree,PCA-K nearest neighbor(KNN),and PCA-linear discriminant analysis(LDA)models were established for alloy-steel identification.The average accuracies of the four models for the training set are 99.06%,97.47%,90.47%,and 100%for the SVM,decision tree,KNN,and LDA,respectively,whereas those for the testing set are 96.29%,79.63%,67.04%,and 100%,respectively.The PCA-LDA model achieves a 100%identification rate for homogeneous alloy-steel grades.This study provides method and reference for the rapid identification of homogeneous alloy-steel grades using laser-induced breakdown spectroscopy.
NETL
NSTL
万方数据