Study Of Cluster Analysis Used With Laser-induced Breakdown Spectroscopy In Explosive Recognition

Breakdown Spectroscopy Laser-induced(LIBS) is a new technology of material analysis based on atomic emission spectrum in recent decades. The basic principle is to focus the laser on the sample surface to induce the ablation of the plasma, and the spectrum of the plasma emitted is collected for qualitative and quantitative analyses. LIBS has been used in the detection of explosives for 12 years, while the current research focuses on the use of supervised learning methods for spectral data processing which may cause the problem of low recognition rate of training sample. In comparison, the spectral recognition algorithm based on unsupervised learning does not need to establish the model, but directly to the obtained sample spectral data for classification and identification, thus avoiding the above problems.This thesis mainly studies the application of clustering analysis in the unsupervised learning methods in the identification of explosive LIBS spectrum. With plastic as interferents, K-means clustering analysis, hierarchical clustering, and ISODATA method of explosives spectrum’s recognition effect is studied. The results show that the methods can be used to identify the explosive, if the parameters are chosen properly. Classification Rate Correct(CCR) can reach more than 0.95 when the ratio of the spectral intensity is used as the input variable. This result is better than the test results obtained from the PLS-DA model when the test set is not fully included in the training set.