Hyperspectral imaging gets a makeover

AusCover’s Perth node is leading a national collaborative project to coordinate Australian scientific work in hyperspectral imaging that will improve the country’s capability and increase our international standing in the field.

The project will encompass the development of standardised software to process and display hyperspectral data recorded by satellite and airborne sensors.

Hyperspectral imaging makes use of the fact that every natural material has a unique colour or spectral ‘signature’. A hyperspectral sensor may record data in hundreds of narrow bands or wavelengths of the visible, near infrared and thermal infrared light that is reflected or scattered by every natural material. The data from each measurement needs to be processed to form an image. The images provide very clear, detailed pictures of vegetation, soil types, water quality, and shallow-water marine habitats, making it useful in environmental monitoring and understanding land use.

Professor Merv Lynch, in the Department of Imaging and Applied Physics at Curtin University, works with AusCover. He reported that scientists from several Australian research institutions agreed at a recent workshop in Perth to develop generic processing software that would achieve much greater efficiencies and result in a higher quality and more versatile product. These advances would position Australia to be well prepared to take advantage of the sequence of the next generation of satellite and airborne sensors.

AusCover itself will acquire and use hyperspectral images taken from the supersites and national transects. Supersites are relatively large areas that are representative of the range of vegetation types, such as tropical forests and rangelands, found on the Australian continent, and are used to collect data about those environments. Terrestrial transects are long corridors across a region that are designed to capture gradients, for example annual rainfall along a transect that runs from the coast to the interior and may change by 500 mm or more a year. By monitoring variability along transects over long periods scientists expect to be able to more readily detect the impact of climate change.

Existing sensors on Earth-observing satellites typically have just a few spectral bands. For example, Landsat and SPOT (Satellite Pour l’Observation de la Terre, or Satellite for Earth Observation) may have four or five sensing bands or spectral channels, such as blue, red, green, near infrared and thermal infrared. However, the spectral properties of the range of cover on the Earth’s surface or on the seafloor environment is huge, and several hundred spectral bands may be required to identify and discriminate particular species and even sub-species. So, hyperspectral data will provide a far more detailed inventory of the Earth system by being able to classify in detail its biogeochemical components and how these change over time.

‘Historically, several Australian centres have processed hyperspectral datasets, and they have been operating more or less independently,’ Merv said.

‘We agreed that it’s important to build on what we have learned from experiences using Hyperion, which was the first hyperspectral Earth-observation satellite, launched by the USA in 2000. The US Geological Survey has provided the Hyperion dataset recorded for Australia free to the research community.

‘Much of the code to process its data is currently in the form of ENVY scripts, which you could run on a desktop computer. Although it’s not decided yet, the scientists at the workshop favour reworking the existing code so that it can also be executed on supercomputers,’ Merv said.

‘High-resolution hyperspectral datasets are rather large – you measure them in gigapixels – and they will be larger in the future, so desktop computers can’t really handle remote-sensing data any more, other than for testing code or doing test runs while developing code. As we move into the next generation of sensors we see the need to use supercomputers to process the data.

‘Open-source generic software will also make it much easier for end-users to optimise processing for their specific requirements and in no way will it limit the specific application they wish to develop.’

Scientists from CSIRO, the Defence Science and Technology Organisation, the University of Adelaide, Curtin University, and the United States Department of Defense took part in the workshop.

At the morning session researchers summarised their approach to imaging and their achievements. In the afternoon, they discussed how these might marry with international activities, particularly with new sensors and the need for the associated hyperspectral remote-sensing software.

For more information, contact Professor Merv Lynch.

Share Article