Hi.
How should I normalize noisy spectra? Normalization by 0 to 1 will change the spectral shape while normalization by dividing by maximum will not change the spectral shape, however, the spectra will remain noisy and difficult to combine with other spectra (e.g., spectra for NIR and VIS regions). Thank you.
I'm proposing the application first an other preprocessing such as spectra smoothing or else to reduce the noisy part and then normalize as you want
There are currently four options:
- "PQN".This is probably the best option for many data sets.
- "TotInt" normalizes by total intensity.
- "Range" allows one to do something similar to "TotInt" but rather than using the sum of the entire spectrum as the denominator, only the sum of the given range is used.
- "zero2one" scales each spectrum separately to a [0 ... 1] scale. This is sometimes useful for visual comparison of chromatograms but is inappropriate for spectral data sets.
- https://bryanhanson.github.io/ChemoSpec/reference/normSpectra.html
Spectra normalization is commonly used for reducing the variance of devices (such as laser fluctuation) and environmental factors. It is not designed for reducing noises. You need to perform noise reduction prior to that. You can try smoothing by using digital filters. You can also try multivariant analysis to decompose the noise spectrum from the signal, such as MCR-ALS.
Greetings Pistach Nyo,
As said by Issam Barra and Xu Feng, normalization is not the best option to reduce noise.
There are smoothing algorithms that help. The one I used commonly is Savitzky-Golay's smoothing algorithm. It can be used in Matlab and R.
Normally, an order 2 polynomial fit is made and a 15 smoothing points window is selected. The smoothing points depend on the spectra itself, if the window is too large, the smoothing will reduce information.
Normalization-wise, enclosed is a equation that can be used in Excel.
- Badges
- Science method
- Similar topics
- Methodology
- Absorption Spectroscopy