Developments in numerical treatments for large data sets of XPS images

Abstract

Empirical data include signal with superimposed variations in intensity of an unwanted nature. These signal fluctuations are of particular interest in XPS measurements when the data are partitioned into both spatial and energy collection bins. In separating signals into a data cube with axes of energy and displacement in x and y, this division creates many more data binning locations than is typical of spectroscopy or classical imaging-type measurements resulting in only small numbers of counts per data bin. Under these circumstances, errors can easily dominate the signal of interest. It is usually assumed that pulse counted data have Poissonian statistics, so that the magnitude of the noise is proportional to the square root of the number of counts. When data are distributed throughout a 3D cube, the counts per voxel are typically low and any random errors can cause deviation from the expected Poisson distributed intensities, which has consequences for data treatment. This paper shows how raw intensities from 3D data sets can be preprocessed to recover a pseudo-Poisson behaviour, which allows principal component analysis with prior data scaling to be used to process the data. Data acquired from crystal formations on photosensitive films based on titanium clusters are used to demonstrate these techniques.

Application of quantitative X‐ray photoelectron spectroscopy (XPS) imaging: investigation of Ni‐Cr‐Mo alloys exposed to crevice corrosion solution

“…An outlier filter was first applied to the image data sets with a 70% threshold value with five iterations. The outlier filter was applied before PCA analysis to help discriminate between potentially weak XPS signals and noise by reducing local anomalous pixel intensities . The data were then sorted using an Optimum Scaling (OpS) routine to determine the number of significant factors.…”

“…The application of multivariate analysis on these large XPS data sets provides a method for the extraction of spectral and image information not apparently obvious . Over the years, principal component analysis (PCA) methods have become more efficient at removing artifacts and noise from XPS image data sets by sorting and reducing the number of factors to compute using singular value decomposition and nonlinear iterative partial least squares routines . XPS imaging has been successfully applied to a wide variety of applications from adhesion issues of polymethyl methacrylate, 3D imaging of nanocomposites, evaluation of wear scars and the differentiation of similar carbon chemical states …”

Photoemission and Characterization of Neutrophils and Nanoparticles : Energy Mapping and Elemental Composition with sub-µm Resolution

“…This generates images with good pixel intensities for further processing and analysis. 5,124,125 Every pixel in the image contains a spectrum and these spectra are used for data processing. Multivariate techniques such as principal component analysis (PCA) have been used in this thesis to create enhanced images with a lower content of noise.…”

“…Those abstract factor images that contains no features can be attributed to noise and by using the abstract factor images with significant features, which have the greatest variance, to reconstruct the entire image-data set, a new image-data set will be generated with smoothing in the spatial domain and with reduced noise. 124,125 After data reduction procedures, pixel-by-pixel analysis can be made by measuring the photoelectron peak areas and normalizing using Scofield relative sensitivity factors. 126 Figure 10.…”

Imaging XPS for industrial applications

“…Since such developments, over the last decade or so, development of traceable and quantitative analysis methods for XPS image analysis have been reported [9,[17][18][19]. However, XPS spectrum image data sets acquired using standard laboratory spectrometers tend to have inherently poor signal-to-noise, and therefore require the use of multivariate analytical techniques to achieve data scaling and avoid prohibitively long acquisition times [19][20][21][22][23][24].…”