Renormalise spectral data with a custom reference

2020-04-01

Some use cases require more flexibility than the high-level user-friendly functions provides by lightr. For this use case, lightr also exports the low-level individual parsers, which allow the user to code its own custom workflow.

We don’t recommend the use of those functions unless you absolutely have to. Most users should use lr_get_spec() and lr_get_metadata() instead.

Here, we take the example of the method presented in Gruson et al (2018) where reflectance spectra need to be normalised in an unusual way.

Raw, un-normalised spectral data depends on both the spectrometer and the lamp as well as the conditions during the recording (including ambient light, temperature, etc.). To allow for comparison between studies, it is thus normalised by a white and a dark reference with the following formula:

$\text{Processed}=\frac{\text{Raw}-\text{Dark}}{\text{White}-\text{Dark}}$

For this example here, we need to normalise the raw data by a white reference contained in another file. This can’t be done with with lr_get_spec() because lr_get_spec() returns reflectance spectra that have already been normalised by the white reference contained in the same file.

library(lightr)

Step 1: import un-normalised data

We manually import the data using the appropriate low-level parser:

reflect_data <- lr_parse_procspec(
system.file("testdata", "procspec_files", "OceanOptics_Linux.ProcSpec",
package = "lightr")
)
length(reflect_data)
## [1] 2

The result contains 2 elements:

• the spectral data itself
• the metadata captured during the recording
head(reflect_data[[1]])
##         wl      dark     white     scope processed
## 1 176.3604 32822.795 32822.795 32822.795   0.00000
## 2 176.5816 32822.795 32822.795 32822.795   0.00000
## 3 176.8027 32822.795 32822.795 32822.795   0.00000
## 4 177.0238  1483.549  1517.545  1496.656  38.55422
## 5 177.2449  1492.150  1506.486  1510.991 131.42857
## 6 177.4660  1965.640  1934.102  1976.290 -33.76623

Step 2: find the matching white reference

We import that white reference in the same way:

white_data <- lr_parse_procspec(
system.file("testdata", "procspec_files", "whiteref.ProcSpec",
package = "lightr")
)

Step 3: normalise the reflectance data

We can now normalise the reflectance spectrum with the equation stated at the beginning of this vignette:

$\text{Processed}=\frac{\text{Raw}-\text{Dark}}{\text{White}-\text{Dark}}$

But first, we verify that the integration times:

We can now get rid of the metadata part and focus on the data only:

reflect_data = data.frame(reflect_data[[1]])
white_data = data.frame(white_data[[1]])

As a last step before being able to normalise the data, we also need to check if the reflectance spectrum and the white reference are sampled with the same wavelengths:

all.equal(reflect_data$wl, white_data$wl)
## [1] TRUE
res <- (reflect_data$scope - reflect_data$dark) /
(white_data$white - white_data$dark)
head(res)
## [1]        NaN        NaN        NaN -5.3333333 46.0000000  0.6190476