Skip to contents

Changelog

Source: NEWS.md

hyperspectaculR 0.1.0

First release. Publication-grade artistic visualisation of hyperspectral imagery, built on the hsi_cube class from hyperspectR.

Compositions

Each requires the spectral axis: none could be produced from a single wavelength, which is the rule the package holds itself to.

  • hsa_spectral_density() draws the joint distribution of reflectance against wavelength across every pixel, with the mean spectrum and a 5–95% envelope read back off the same binned matrix. The only non-image composition, and the one the others should be read against — every image applies a contrast stretch, and this shows the distribution that stretch is applied to. show_limits = TRUE draws those limits onto it.
  • hsa_mandala() renders concentric annuli, each drawn from a different wavelength, sweeping the spectrum outward from a chosen centre.
  • hsa_spectral_flux() sums the absolute change between consecutive bands, giving a field of spectral variability that responds to shape rather than spread — unlike a variance map, it is sensitive to band order.
  • hsa_fusion() builds a colour composite from three averaged band groups, which suppresses per-band sensor noise relative to picking three wavelengths.

Design decisions

  • Every function returns a ggplot, so figures compose with patchwork, re-theme, and export reproducibly. Nothing writes a file as a side effect.
  • Every rendering captions the contrast stretch and limits it applied. Stretching is what makes this data legible and also what makes it easy to mislead with, so it is never silent.
  • hsa_palette() offers only ramps monotonic in lightness, so ordering survives greyscale printing and stays legible to colourblind readers. Asking for turbo warns that it does not.
  • Cubes are duck-typed rather than imported, so hyperspectR stays in Suggests — it is not on CRAN — while the art functions work anywhere a cube can be obtained.

Not included, deliberately

Two visualisations from an earlier draft were dropped rather than ported, both failing the package’s own rule. A 3-D height surface is one scalar per pixel by definition, so its third dimension is a single band or a derived index. A band animation is a sequence of single-band images, so it fails frame by frame. Neither becomes defensible by being rendered more attractively.

Dropping them also removes any need for rayshader, rgl or gifski.