In transmission electron microscopy (TEM), 3D tomographic reconstruction can be achieved by acquiring a series of images at different tilt angles. A different approach is obtaining 3D chemical reconstructions from energy filtered images in the TEM (EFTEM)[1-3], and more recently, by acquiring EELS spectrum images (EELS-SI), each pixel containing a complete EELS spectrum [4,5]. However, in both techniques only a limited amount of information is effectively reconstructed. In this paper we aim to derive a full EELS dataset in 4D, where every voxel of a whole volume contains a complete spectrum of energy losses, as schematized in Fig. 1. By analogy to the spectrum image notation, we will name this 4D dataset as EELS spectrum volume (EELS-SV).
Our approach to EELS-SV reconstruction is based upon SI, thus taking a single SI for every tilt angle. It takes advantage of Multivariate Analysis (MVA), and more precisely of blind source separation (BSS)[6], to find a new spectral basis (Fig. 2a) which can describe all the spectra in the dataset as a weighted sum of its components. Therefore only the 3D reconstructions of the weighting components (Fig. 2b) will be necessary to recover the spectra in each voxel (Fig. 2c-e). We will apply this approach to analyze a BFO/CFO nanocomposites, enabling the characterization of a CFO nanocolumn embedded in BFO matrix.
References
[1] G. Möbus, et al, Ultramic, 96 (2003) 433.
[2] M. Weyland, P.A. Midgley, Microscopy and Microanalysis, 9 (2003) 542.
[3] R.D. Leapman, et al, Ultramic, 100 (2004) 115.
[4] K. Jarausch, et al, Ultramic,. 109 (2009) 326.
[5] L. Yedra et al., Ultramic., 122 (2012) 12
[6] N. Dobigeon et al., Ieee Transactions on Signal Processing, 57 (2009), 4355