摘要
塑料是一种生活中常用的高分子聚合物,随着废旧塑料数量的不断增加,造成的环境污染越来越严重,废旧塑料的分类和回收已经成为一个紧迫的问题。不同种类塑料需要不同的回收方式,因此研究塑料的分类方法具有重要意义。激光诱导击穿光谱技术(LIBS)是一种以原子发射光谱为基础的元素分析技术,具有分析快、无需样品预处理、原位分析等优势。拉曼光谱技术(RS)是一种以拉曼散射理论为基础的分子结构表征分析技术,具有多元素同时分析、样品量要求低、对样品损害小等优势。将利用LIBS技术和RS技术采集塑料的原子和分子两个角度的光谱信息,并将两个角度的光谱信息进行拼接得到融合光谱。利用LIBS光谱、RS光谱、融合光谱分别结合随机森林机器学习算法(RF)构建模型,对塑料进行分类识别,通过对三个模型分类准确率的对比,得出融合光谱可以提高分类准确率的结论。在构建模型的过程中,在相同测试集数量的情况下,训练集数量对模型构建时间以及分类准确率都有影响,针对不同的测试集与训练集比例进行准确率和模型构建时间的实验,得到测试集与训练集比例1∶3最合适的结论,并达到了96%的准确率。除了训练集的影响,光谱数据的预处理方法对塑料融合光谱的分类准确率也有影响,实验利用稀疏度基线估计去噪法处理融合光谱数据,并重新构建模型,将塑料的分类准确率提升到100%。实验结果表明,在测试集与训练集比例为1∶3时,融合光谱数据对比单一光谱数据在分类准确率上有明显的优势,且经过预处理的融合光谱数据分类准确率可以提高到100%。
Plastic is a commonly used polymer in daily life.With the increasing amount of waste plastic,the resulting environmental pollution has become more severe,making the classification and recycling of waste plastic an urgent issue.Different types of plastics require different recycling methods,so researching plastic classification methods is of great significance.Laser-Induced Breakdown Spectroscopy(LIBS)is an elemental analysis technique based on atomic emission spectroscopy,offering advantages such as rapid analysis,no sample pretreatment required,and in-situ analysis,which provides convenience for plastic classification.Raman Spectroscopy(RS)is a molecular structure characterization technique based on Raman scattering theory,which offers advantages such as simultaneous multi-element analysis,low sample quantity requirements,and minimal sample damage,also facilitating plastic classification.This paper will utilize LIBS and RS technologies to collect spectral information from both atomic and molecular perspectives of plastics,and then merge the two types of spectral information to obtain a fused spectrum.By using LIBS spectra,RS spectra,and fused spectra in conjunction with the Random Forest machine learning algorithm(RF)to build models for plastic classification and identification,a comparison of the classification accuracy of the three models reveals that the fused spectrum can improve classification accuracy.During the model-building process,with the same number of test sets,the number of training sets affects both the model construction time and classification accuracy.Experiments were conducted on the accuracy and model construction time for different ratios of test sets to training sets,concluding that a ratio of 1∶3 is the most suitable,achieving an accuracy of 96%.In addition to the impact of the training set,the preprocessing methods of spectral data also affect the classification accuracy of the plastic fusion spectrum.The experiment-employed a sparsity-based baseline estimation denoising method to process the fusion spectral data and rebuild the model,thereby increasing the classification accuracy of plastics to 100%.The experimental results indicate that when the ratio of the test set to the training set is 1∶3,the fused spectral data has a significant advantage in classif ication accuracy compared to single spectral data.The classification accuracy of the preprocessed fused spectral data can be improved to 100%.
作者
房家萱
董茜文
徐梓睿
曲东明
杨光
孙慧慧
FANG Jia-xuan;DONG Xi-wen;XU Zi-rui;QU Dong-ming;YANG Guang;SUN Hui-hui(College of Instrumentation and Electrical Engineering,Jilin University,Changchun 130026,China)
出处
《光谱学与光谱分析》
北大核心
2025年第9期2484-2490,共7页
Spectroscopy and Spectral Analysis
基金
吉林省教育厅科学研究项目(JJKH2025009ZKJ)资助。
关键词
激光诱导击穿光谱
拉曼光谱
融合光谱
随机森林
光谱预处理
Laser-induced breakdown spectroscopy
Raman spectroscopy
Fusion spectroscopy
Random forest
Spectral pretreatment
作者简介
房家萱,2002年生,吉林大学仪器科学与电气工程学院硕士研究生,e-mail:fangjiaxuan_jlu@163.com;通讯作者:杨光,e-mail:yangguang_jlu@163.com;通讯作者:孙慧慧,e-mail:sunhuihui@jlu.edu.cn。
