Classificação de fases em imagens hiperespectrais de raios X característicos pelo método de agrupamento por deslocamento para a média

Abstract

In the present work we introduce the Mean Shift Clustering (MSC) algorithm as a valuable alternative to perform materials phase classification from hyperspectral images. As opposed to other multivariate statistical techniques, such as principal components analysis (PCA), clustering techniques directly assign a class (phase) label to each pixel, so that their outputs are phase segmented images, i.e. , there is no need for an additional segmentation algorithm. On the other hand, as compared to other clustering procedures and classification methods based on cluster analysis, MSC has the advantages of not requiring previous knowledge of the number of data clusters and not assuming any shape of these clusters, i.e., neither the number nor the composition of the phases must be previously known. This makes MSC a particularly useful tool for exploratory research, allowing automatic phase identification of unknown samples. Other advantages of this approach are the possibility of multimodal image analysis, composed of different types of signals, and estimate the uncertainties of the analysis. Finally, the visualization and interpretation of results are also simplified, since the information content of the output image does not depend on any arbitrary choice of the contents of the color channels. In this paper we apply the PCA and MSC algorithms for the analysis of characteristic X-ray maps acquired in Scanning Electron Microscopes (SEM) which is equipped with Energy Dispersive Detection Systems (EDS). Our results indicate that MSC is capable of detecting minor phases, not clearly identified when only three components obtained by PCA are used.