Evaluation of Three Peak Detection Algorithms for XPS Spectra

Abstract

In order to evaluate three peak detection algorithms, the influences of parameter values used were examined using digitized synthetic XPS spectra with different levels of noise. The three peak detection algorithms are the Threshold Curve of the Second Derivative (2nd DER method), the Directly Calculating Peak and Background Relations at a Candidate Peak (PB method), and the Rough Estimation of Spectrum Background (BGD method). The peak detection results clearly showed that particular combination of parameter values produce the best performance for each algorithm. The validity of those parameter values was assessed by comparing the results from the peak detection algorithms with those of visual detection. On the whole, the BGD method was found to be the most practical. The importance of choosing effective parameter values and the advantages and disadvantages of these three peak detection methods were summarized.

Quantitative analysis of trace levels of surface contamination by X‐ray photoelectron spectroscopy. Part I: Statistical uncertainty near the detection limit

“…In this context, we define the achievable detection limit as the smallest atomic concentration that can be considered statistically different from zero when a specific background‐subtraction method is applied to a measured spectrum (e.g., automated measurements of atomic concentration). This quantity may differ from the ultimate detection limit of an acquired spectrum corresponding to the smallest atomic concentration capable of producing a peak that is detectable above the background by eye or by a peak‐detection algorithm . Derivation of the RBSV for such peak‐detection methods is beyond the scope of this work.…”

synapse: interactive support on photoemission spectroscopy measurement and analysis for non-expert users

“…As a preliminary experiment, we tried two implementable methods for peak identification: the threshold curve of the second derivative method (2nd DER method) and the directly calculating peak and background relations at a candidate peak method (PB method) (Furukawa et al, 2016). As a consequence, we found that it is difficult to distinguish between broad peaks and large noise in the 2nd DER method, and the PB method hardly identifies peaks with narrow widths and low intensities from noise.…”

“…We applied the methods described in Section 4.2 to actual data and plotted intensity and SNR according to kinetic energy and sweeps, respectively. We used a window width of 11 and a polynomial degree of 4 for the parameters of the filter, referring to the work of Furukawa et al (2016). The results are shown in Fig.…”

“…, N À 1) be the discrete data of I i to which the Savitzky-Golay filter is applied. Based on the results of Furukawa's experiment (Furukawa et al, 2016), we let the window length be 11, while, based on our preliminary experiment, we temporally let the polynomial order be 4. Also, we define the noise as follows,…”