A collection of background subtraction algorithms for spectroscopic data

To install the package, simply type "]" followed up by "add BackgroundSubtraction" in the Julia REPL.

The main function is based on the multi-component background learning model (MCBL), with the corresponding function mcbl:

mcbl(A::AbstractMatrix, k::Int, x::AbstractVector, l::Real)

• A is the data matrix, each column of which is assumed to be a spectrogram.
• k is the number of components in the multi-component background model.
• x is the index vector corresponding to rows of A. For example, if a column of A is an X-ray diffraction spectrogram, x should be the angle of diffraction of each row.
• l is the length scale of the background component. It controls how quickly the background model is allowed to vary with x. This functions as an important regularization for medium-sized data (100s-1000s spectrograms).

There are 3 parameters controlling the algorithm, which can optionally be passed as keyword arguments:

• minres is the minimum residual standard deviation after which the algorithms terminates.
• nsigma is the number of standard deviations above the noise level after which a data point is classified as a peak. A smaller number will be more agressive in classifying points as peaks.
• maxiter is the maximum number of iterations between updating the noise and background model.

If you use the MCBL for work or a publication, please cite the original article:

Ament, S.E., Stein, H.S., Guevarra, D. et al. Multi-component background learning automates signal detection for spectroscopic data. npj Comput Mater 5, 77 (2019). https://doi.org/10.1038/s41524-019-0213-0