Hyperspectral imaging data is, by definition, spatial. However, typically there is no easy and immediate anchor to real world units. Most of the time ENVI or GeoTIFF will load in GIS software in WGS84 or with NULL CRS.
Getting the units
To express HSI data in real world units we need, at the very minimum, to know real world resolution. Internally, we use micrometers per pixel throughout the HSItools package and offer options to convert to mm and cm. We also offer three complementary ways to get and store real world resolution. We assume throughout the workflow, that pixels are square.
Known resolution
In an instance where real world pixel size is known, for example in a lab where field of view is used to determine scanning resolution, or once one scan is done, and others follow exactly the same settings, it is possible to simply provide a number.
um <- HSItools::hsi_calibration_direct(60, units = "um")
um
#> um_per_px
#> 60Digitized scale
If there is a vertical or horizontal measure scanned along the specimen it is easy to digitize it and use as a reference. Simply read your digitized data as a terra SpatVector pointing to a line or two-point feature and provide length that was read out. Importantly, be sure that the CRS is NULL.
# Construct SpatVector by hand
line_vect <- terra::vect(
matrix(c(1000, 2000, 1000, 2100), ncol = 2, byrow = TRUE),
type = "lines"
) |>
HSItools::hsi_drop_crs()
# Distance of 10000 micrometers is default
um <- HSItools::hsi_calibration_from_scale(
line_vect,
distance = 10000,
units = "um"
)
#> Warning: [perim] unknown CRS. Results can be wrong
um
#> um_per_px
#> 100Known dimensions
If you know the exact scan length that was used in the controlling software, you can use it as a robust reference. We use known scan length and number of rows in a SpatRaster to calculate resolution. This requires that kind of metadata is written down, but ensures very good resolution estimate.
r <- terra::rast(
system.file(
package = "HSItools",
"testdata/products/SAVGOL_testdata.tif"
)
)
um <- HSItools::hsi_calibration_from_dims(
terra::nrow(r),
distance = 10000,
units = "um"
)Building a coordinate raster
An important step, and also data which is “free of charge” is building a SpatRaster that holds in its cells simple information about real world coordinates. This is an opitnionated framework, where we construct a SpatRaster that has two layers (bands). The first one is "row_um" and stores Y-axis position, the second one is "col_um" and stores X-axis. This provides both an interesting dataset, and a lookup table for other spatial data.
Anchoring to a reference point
Often, the extent of a captured specimen does not reflect the ture origin, or a reference point anchored at some position along the Y-axis. For example, sediment core scanning often calls for specific point to represent real zero depth. For this, you can provide a point within a SpatRaster extent and use it as spatial anchor. By default we treat this as a point equal to zero, but any other value in real world units can be provided.
Shifting extent
Once coordinate SpatRasters are available, either plain coordinates or anchored, they can be used to change the SpatExtent of a SpatRaster of choice. This operation, and further, downstream operations need to be done in concert. Maning, for example, cropping has to be applied to both SpatRasters so they keep their reference frame.