基于主成分分析和人工神经网络的激光诱导击穿光谱塑料分类识别方法研究  

Classification of Plastics with Laser-Induced Breakdown Spectroscopy Based on Principal Component Analysis and Artificial Neural Network Model

作　　者：王茜蒨[1] 黄志文[1] 刘凯[1] 李文江[1] 阎吉祥[1] 

机构地区：[1]北京理工大学光电学院,北京100081

出　　处：《光谱学与光谱分析》2012年第12期3179-3182,共4页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(60978035)资助

摘　　要：研究了人工神经网络在激光诱导击穿光谱(LIBS)塑料分类识别方面的应用。选用七种常见的塑料作为实验样品,获得每种样品的170组LIBS光谱数据,利用主成分分析获得前五个主成分的得分矩阵。用每种塑料样品的130组光谱数据的主成分得分矩阵作为训练集,建立反向传播(BP)人工神经网络模型。将其余40组主成分得分作为测试数据输入训练好的模型进行分类识别,其识别准确度达到97.5%。实验结果表明,通过采用主成分分析与BP人工神经网络相结合的方法,可以很好地进行塑料激光诱导击穿光谱的分类识别,对塑料的回收利用有重要价值。The classification of seven kinds of plastic（ABS, PET, PP, PS, PVC, HDPE and PMMA） with the laser-induced breakdown spectroscopy based on artificial neural network model was investigated in the present paper. One hundred seventy LIBS spectra for each type of plastic were collected. Firstly, all 1 190 plastics LIBS spectra were studied with principal compo- nent analysis. The first five principal components （PC） totally explain 78. 4% of the original spectrum information. Therefore, the scores of five PCs of 130 LIBS spectra for each kind of plastic were chosen as the training set to build a back-propagation arti- ficial network model. And the other 40 LIBS spectra of each sample were used as the testing set for the trained model. The clas- sification accuracy was 97.5%. Experimental results demonstrate that plastics can be classified by using principal component analysis and artificial neural network （BP） method.

关 键 词：激光诱导击穿光谱 塑料 主成分分析 BP人工神经网络 分类识别 

分 类 号：O657.3[理学—分析化学]