基于粒子群-支持向量机算法的激光诱导击穿光谱钢铁快速检测与分类

doi:10.3964/j.issn.1000-0593(2024)06-1559-07

摘要
参考文献
相关文章 (15)

全文: PDF (6438 KB)
输出: BibTeX | EndNote (RIS)

摘要：钢铁是国民经济中的支柱性产业，由于受生产技术的限制，我国钢铁产品主要集中为质量参差不齐的中低端产品，废品率较高，易造成资源浪费和环境污染。因此，钢铁产品的快速检测与鉴别分类，对保护环境以及提高钢铁资源的回收利用率有着重要意义。利用激光诱导击穿光谱技术(LIBS)进行10种钢铁样品光谱数据的快速采集，并采用支持向量机(SVM)算法对其数据进行学习建模，得到钢铁快速分类模型。然而，由于不同钢铁样品的光谱数据特征是复杂且相似的，导致设置的模型参数也会对SVM模型的分类结果有着较大的影响。为了实现对不同牌号钢铁合金的快速检测分类，实验中采用粒子群算法(PSO)与网格寻优法两种不同方法来优化模型参数，并分别选取样品中6种微量元素(Mn、Cr、Cu、V、Mo、Ti)的17条特征谱线，和经主成分分析法(PCA)对全谱数据降维提取得到的前17个主成分作为模型的输入，建立PSO-SVM、PSO-PCA-SVM、PCA-SVM和SVM四种分类模型。实验结果表明，相比于精度最高的PCA-SVM模型的优化时间（257.84 s），PSO-SVM模型优化时间最短（11.5 s），且识别精度可达96.67%，与PCA-SVM模型的精度（97.5%）几乎相当。该结果表明LIBS结合PSO-SVM算法可实现快速的钢铁检测与分类，该方法为钢铁产品的快速检测与分类提供了一种新的解决途径。

关键词：激光诱导击穿光谱；支持向量机；粒子群算法；钢铁分类

Abstract：The steel industry has become a mainstay of the Chinese national economy. Due to the limitation of production technology, Chinese steel products are mainly concentrated in the middle and low-end products of uneven quality. It could result in the severe waste of steel resources and the pollutionofmetal garbage wastes. Therefore, the rapid identification and classification method of steel products is significant for environmental protection and for improving steel resources' recycling rate. This work utilised laser-induced breakdown spectroscopy (LIBS) to quickly collect the spectral data of 10 kinds of special steels. Then, a support vector machine (SVM) learned and modelled the spectral data to obtain the rapid steel classification model. However, due to the element composition of different special steels being complex and similar, the performance of classification results may be directly and significantly affected by SVM model parameters. To realise the rapid classification and detection of different grades of steel alloys, the two different methods of particle swarm optimisation (PSO) and grid search optimization were used to optimize the model parameters and speed up the training efficiency. Then, the spectral intensity of 17 characteristic lines of 6 major trace elements (Mn, Cr, Cu, V, Mo and Ti) in samples and 17 feature information variables extracted from the LIBS spectrum data with full variables by principal component analysis (PCA) were chosen as the input to establish the PSO-SVM, PSO-PCA-SVM, PCA-SVM and SVM models for steel classification respectively. The experimental results show that compared with the SVM model's optimization time of 115.64 s, the shortest optimization time of PSO-SVM is 11.5 s, and its classification accuracy (96.67%) is not significantly inferior to the accuracy of the PCA-SVM model (97.5%). The results show that LIBS combined with the PSO-SVM algorithm can achieve rapid and high-precision steel classification, which provides a new solution to detect and classifythedifferent steel products rapidly and precisely.

Key words：Laser-induced Breakdown spectroscopy; Support vector machine; Particle swarm optimization; Steel Classification

收稿日期: 2022-07-23 修订日期: 2023-12-13

中图分类号:

O433.4

基金资助: 国家自然科学基金项目（61705064），湖北省自然科学基金项目（2021CFB607），湖北省教育厅团队研究项目（T201617），孝感市自然科学计划项目（XGKJ2021010003）资助

通讯作者: 肖永军 E-mail: 346917126@qq.com

作者简介: 曾庆栋，1982年生，湖北工程学院物理与电子信息工程学院教授 e-mail: jerry-z@hbeu.edu.cn

引用本文:

曾庆栋，陈光辉，李文鑫，孟久灵，李耿，童巨红，田志辉，张晓林，李国辉，郭连波，肖永军. 基于粒子群-支持向量机算法的激光诱导击穿光谱钢铁快速检测与分类[J]. 光谱学与光谱分析, 2024, 44(06): 1559-1565.
ZENG Qing-dong, CHEN Guang-hui, LI Wen-xin, MENG Jiu-ling, LI Geng, TONG Ju-hong, TIAN Zhi-hui, ZHANG Xiao-lin, LI Guo-hui, GUO Lian-bo, XIAO Yong-jun. Classification of Special Steel Based on LIBS Combined With Particle Swarm Optimization and Support Vector Machine. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2024, 44(06): 1559-1565.

链接本文:

https://www.gpxygpfx.com/CN/10.3964/j.issn.1000-0593(2024)06-1559-07 或 https://www.gpxygpfx.com/CN/Y2024/V44/I06/1559

[1] Afgan M S, Hou Z, Wang Z. Journal of Analytical Atomic Spectrometry, 2017, 32(10): 1905.
[2] Almessiere M A, Altuwiriqi R, Gondal M A, et al. Talanta, 2018, 185: 61.
[3] Wang Z, Deguchi Y, Shiou F, et al. ISIJ International, 2016, 56(5): 723.
[4] Li X, Zhang L, Tian Z, et al. Journal of Analytical Atomic Spectrometry, 2020, 35: 2928.
[5] Zeng Q D, Deng F, Zhu Z H, et al. Plasma Sci. Technol, 2019, 21(3): 034006.
[6] Zhang D C, Ding J, Feng Z Q, et al. Spectrochimica Acta Part B: Atomic Spectroscopy, 2021, 180, 106192.
[7] Yue Z Q, Sun C, Chen F Y, et al. Biomedical Optics Express, 2021, 12(5): 2559.
[8] Zeng Q D, Guo L B, Li X Y, et al. Journal of Analytical Atomic Spectrometry, 2016, 31(3): 767.
[9] Gazeli O, Bellou E, Stefas D, et al. Food Chemistry, 2020, 302: 125329.
[10] Yelameli M, Thornton B, Takahashi T, et al. Journal of Chemometrics, 2019, 33(2): e3092.
[11] Yao M, Fu G, Chen T, et al. Journal of Analytical Atomic Spectrometry，2021, 36: 361.
[12] Shin S, Moon Y, Lee J, et al. Plasma Science and Technology, 2019, 21(3): 034011.
[13] Kim E, Kim Y, Srivastava E, et al. Chemometrics and Intelligent Laboratory Systems, 2020, 203(1): 104072.
[14] Chen G H, Zeng Q D, Li W X, et al. Optics Express, 2022, 30(6): 9428.
[15] Wang J M, Liao X Y, Zheng P C, et al. Analytical Letters, 2018, 51(4): 575.
[16] Wang J M, Zheng P C, Liu H D, et al. Analytical Methods, 2016, 8(15): 3204.
[17] Liu Y Y, Zhao S Y, Gao X, et al. RSC Advances, 2022, 12(53): 34520.
[18] LI Ming-liang, DAI Yu-jia, QIN Shuang, et al(李明亮，戴宇佳，秦爽，等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2022, 42(2): 587.
[19] ZHAO Shang-yong, ZHOU Zhi-ming, SONG Chao, et al(赵上勇，周志明，宋超，等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2020, 40(8): 2629.
[20] Cortes C, Vapnik V, et al. Mach. Learn.，1995, 20(3): 273.
[21] Marini F, Walczak B. Chemometrics and Intelligent Laboratory Systems, 2015, 149: 153.
[22] Porizka P, Klus J, Kepes, E, et al. Spectrochimica Acta Part B: Atomic Spectroscopy, 2018, 148: 65.