A LLS operator based S‐I WT de‐noising algorithm applied in EDXRF

Abstract

An improved shift‐invariant wavelet (S‐I WT) de‐noising algorithm based on LLS operator is proposed for high‐resolution energy dispersive X‐ray fluorescence. Sym8 is chosen as the wavelet basis function and performed noise reduction on the analog signal. Comparison of the de‐noising effect of S‐I WT, improved WT and LLS S‐I WT (where LLS is the log square root operator) method are quantitatively evaluated by using evaluation criteria signal‐to‐noise‐ratio (SNR), root mean square error and Pearson correlation coefficient. Meanwhile, a new evaluation criterion of de‐noising effect, called peak area relative difference, is also proposed to evaluate the counting deviation. The results show that the LLS‐SI WT is simple and reliable, can effectively reduce pseudo‐Gibbs artificial signals and statistical fluctuation. Besides, this method simplifies the calculation, reduces the running time and improves the running efficiency. The LLS‐SI WT is also applied to reduce the noise after adding strong noise to the signal, the SNR has been improved from 14.0040 to 14.7552, and most of the characteristic peak information retains to the greatest extent.

XRFast a new software package for processing of MA-XRF datasets using machine learning

“…Prior to undertaking the data reduction through the dictionary learning algorithm, the MA-XRF data is compressed through a log-log-square root operator. 52 This operation is typically performed in conjunction with baseline removal (as with the SNIP algorithm). 53 Here the log-log-square root operator is used to emphasize minor spectral bands or features that would otherwise be suppressed or eliminated by the SVD decomposition.…”