Hyperspectral Imaging for Astronomy: 4th October 2019

Dr Hugh Mortimer (RAL Space)

The speaker began by discussing the JunoCam instrument on NASA’s Juno probe to Jupiter. This is primarily an outreach camera which is directed by citizen scientists, who also process and publicize the images. It was a late addition to the probe, and a very unusual one, which helps to satisfy astronomers’ curiosity and sense of awe about Jupiter.

The speaker’s primary interest is in developing instrumentation for space missions. Humans are very good at remote sensing our world. We can capture images with between 300 and 500 Megapixels, at 45 to 100 Hz in the visible spectrum (350 to 700 nm). We perform real-time processing and pattern recognition with only 25 Watts of power, something that cannot be done with instrumentation! We have many limitations: we cannot see radio waves, infra-red, ultra-violet or X-rays; the data cannot be stored quantitatively; the analysis is subjective, as witness the famous photo of “the dress” and many optical illusions; and around 10% of humans are colour-blind to some degree. However, we are extremely good at observing and recording (by drawing) what we observe.

Three of the most important developments in the evolution of science were (1) the invention of the telescope by Lipperhey in 1608, and its development by Galileo, (2) the development of photography as a means of permanently recording images (around 1826), and (3) the digital camera which allowed information to be stored electronically, patented in 1972. These technologies opened up new ways of exploring the universe.

There are trade-offs between different types of resolution: spectral, spatial and temporal. Spectral resolution allows us to understand processes. Temporal resolution requires us to sacrifice spatial resolution for more speed or spectral detail. Spatial resolution tells us about patterns. Finally, radiometric resolution tells us how accurate the data are in terms of calibration, accuracy and precision. In any instrument we start by collecting photons, then manipulate them by filtering, splitting and sorting, and finally detect them with either a single detector or in 1- or 2-dimensional arrays of detectors.

Imaging detectors are not colour-sensitive, but colour information can be obtained either by a Bayer filter matrix or by using multiple spectral filters with well-defined transmission bands. JunoCam uses a technique called “push-broom” imaging, with a linear detector and four strips of different filters. This method allows the generation of colour images, but with no temporal resolution.

One of the speaker’s projects is the development of a multi-hyperspectral camera, which will have narrow-band filters in a 4 x 4 matrix similar to the Bayer type. This device will trade spectral for spatial resolution, producing images of 500 x 500 pixels in 16 different wavelengths at rates up to 60 Hz. The radiometric resolution is somewhat limited because there are necessarily fewer photons per image channel. An alternative approach would be to use a spectrometer, and scan a slit across the image in a push-broom arrangement. This approach was rejected because it is hard to calibrate, has poor signal-to-noise ratio and requires moving parts, unlike the filtered camera.

The speaker finished by showing a miniaturized Fourier Transform spectrometer, based on a Michelson interferometer. This was developed for space missions as an infra-red spectrometer, and is rugged, easy to calibrate, with high temporal resolution and no moving parts. The case was made as a single piece by 3D printing, with supports for the optics and detector.

Hyperspectral imaging will become increasingly common in the future, as a holistic technique for acquiring a full set of data about any subject, in its proper context.

Notes and summary by Chris Hooker.