PCA-K-Means Based Clustering Algorithm for High Dimensional and Overlapping Spectra Signals

This paper applied a PCA-K-means method to exploit photo-thermal infrared imaging spectroscopy based trace explosives with overlapping spectral absorption bands. We intend to explore the underlying patterns that affect the clustering performance using top principal components. We also strive to investigate the effectiveness of the clustering algorithm on different analytes and substrates. We reduced the dimensions by applying the principal component analysis (PCA) on the data to transform the original data to the top principal components' feature space. The data were revealed in the feature space and formed into clusters. Then we used the K-means based clustering algorithm to classify them into six classes including RDX, PC, Copper/Steel, TNT, DNT, and PE. After that, we conducted the performance evaluation. We found that the FI score of the classification of RDX, PC, Copper/Steel, TNT, DNT, and PE is 85%, 39%, 71% 99% 92% and 18% respectively. The results demonstrated that the proposed algorithm can effectively reduce dimension and accurately determined the classes of those analytes and substrates.