× ⚠️ You are reading the development version documentation.
Read here the documentation of the last stable release.

NDDataset baseline correction

In this example, we perform a baseline correction of a 2D NDDataset interactively, using the multivariate method and a pchip/polynomial interpolation.

For comparison, we also use the asls`and `snip models.

As usual we start by importing the useful library, and at least the spectrochempy library.

import spectrochempy as scp

Load data

datadir = scp.preferences.datadir
nd = scp.read_omnic(datadir / "irdata" / "nh4y-activation.spg")

Do some slicing to keep only the interesting region

ndp = nd[:, 1291.0:5999.0]
# Important:  notice that we use floating point number
# integer would mean points, not wavenumbers!

Plot the dataset

ndp.plot()

Remove a basic linear baseline using basc:

ndp = ndp.basc()

Make it positive

offset = ndp.min()
ndp -= offset
ndp.plot()

Define the Baseline object for a multivariate baseline correction model. The n_components parameter is the number of components to use for the multivariate baseline correction. The model parameter is the baseline correction model to use, here a pchip interpolation (piecewise cubic Hermite interpolation).

blc = scp.Baseline(
    log_level="WARNING",
    multivariate=True,  # use a multivariate baseline correction approach
    model="polynomial",  # use a polynomial model
    order="pchip",  # with a pchip interpolation method
    n_components=5,
)

Now we select the regions ( ranges ) to use for the baseline correction.

blc.ranges = [
    [1556.30, 1568.26],
    [1795.00, 1956.75],
    [3766.03, 3915.81],
    [4574.26, 4616.04],
    [4980.10, 4998.01],
    [5437.52, 5994.70],
]

We can now fit the baseline correction model to the data:

blc.fit(ndp)

<spectrochempy.processing.baselineprocessing.baselineprocessing.Baseline object at 0x7fcff19bc410>

The baseline is now stored in the baseline attribute of the processor: (note that the baseline is a NDDataset too). The corrected dataset (the dataset after the baseline subtraction) is stored in the corrected attribute of the processor:

baseline = blc.baseline
corrected = blc.corrected

Plot the result of the correction

corrected.plot()

We can have a more detailed representation using plot

ax = blc.plot(nb_traces=2, offset=50)
blc.show_regions(ax)

We can also plot the baseline and the corrected dataset together: for some individual spectra to, for example, check the quality of the correction:

corrected[0].plot()
baseline[0].plot(clear=False, color="red", ls="-")
ndp[0].plot(clear=False, color="green", ls="--")

corrected[10].plot()
baseline[10].plot(clear=False, color="red", ls="-")
ndp[10].plot(clear=False, color="green", ls="--")

The baseline correction looks ok in some part of the spectra but not in others where the variation seems a little to rigid. This is may be due to the fact that the pchip interpolation is perhaps not the best choice for this dataset. We can try to use a n-th degree polynomial model instead:

We don’t need to redefine a new Baseline object, we can just change the model and the order of the polynomial:

blc.model = "polynomial"
blc.order = 5  # use a 5th degree polynomial

and fit again the baseline correction model to the data:

blc.fit(ndp)

baseline = blc.baseline
corrected = blc.corrected

corrected[0].plot()
baseline[0].plot(clear=False, color="red", ls="-")
ndp[0].plot(clear=False, color="green", ls="--")

corrected[10].plot()
baseline[10].plot(clear=False, color="red", ls="-")
ndp[10].plot(clear=False, color="green", ls="--")

corrected.plot()

This looks better and smoother. But not perfect.

We can also try to use a asls (Asymmetric Least Squares) model instead. This model is based on the work of Eilers and Boelens (2005) and performs a baseline correction by iteratively fitting asymmetrically weighted least squares regression curves to the data. The asls model has two parameters: mu and assymetry. The mu parameter is a regularisation parameters which control the smoothness of the baseline. The larger mu is, the smoother the baseline will be. The assymetry parameter is a parameter which control the assymetry if the AsLS algorithm.

blc.multivariate = False  # use a sequential approach
blc.model = "asls"
blc.mu = 10**9
blc.asymmetry = 0.002
blc.fit(ndp)

<spectrochempy.processing.baselineprocessing.baselineprocessing.Baseline object at 0x7fcff19bc410>

baseline = blc.baseline
corrected = blc.corrected

corrected[0].plot()
baseline[0].plot(clear=False, color="red", ls="-")
ndp[0].plot(clear=False, color="green", ls="--")

corrected[-1].plot()
baseline[-1].plot(clear=False, color="red", ls="-")
ndp[-1].plot(clear=False, color="green", ls="--")

corrected.plot()

Finally, we will use the snip model

blc.multivariate = False  # use a sequential approach
blc.model = "snip"
blc.snip_width = 200
blc.fit(ndp)

<spectrochempy.processing.baselineprocessing.baselineprocessing.Baseline object at 0x7fcff19bc410>

baseline = blc.baseline
corrected = blc.corrected

corrected[0].plot()
baseline[0].plot(clear=False, color="red", ls="-")
ndp[0].plot(clear=False, color="green", ls="--")

corrected[-1].plot()
baseline[-1].plot(clear=False, color="red", ls="-")
ndp[-1].plot(clear=False, color="green", ls="--")

corrected.plot()

This ends the example ! The following line can be uncommented if no plot shows when running the .py script with python

# scp.show()