The quantity of information available from typical imaging satellites is compromised by the volume of data that it must transmit.  The FSI system will extract the information from the data before transmission.

Full Spectral Imaging (extended abstract)

Full Spectral Imaging (FSI) is the successor to hyperspectral imaging, much as hyperspectral imaging was the successor to multispectral imaging.  FSI s imaging spectroscopy as it would have been done originally if the technology had been available.  Imaging spectrometry is basically, two-dimensional spectral reflectance measurement, or the measurement of the spectral measurement of scenes or targets.  In the laboratory, one-dimensional spectral or spot  reflectance measurements produce a reflectance spectrum.  The information in that spectrum are contained in the features of that spectrum.  The reflectance spectrum is obtained as a continuous measurement, with some spectral resolution as determined by the instrumentation.

The concept of spectral bands was developed as a solution to the technological limitation of early remote sensing systems.  Bands were carefully selected (and always a matter of debate) that would best represent the reflectance spectrum.  When hyperspectral systems were introduced, it became possible to have many more bands.  The primary advantage of hyperspectral systems is that the bands are continuous, with at worst only small spectral intervals between them.  Unfortunately, most researchers treat hyperspectral data as simply multispectral with a lot of bands.  In many cases, they just select the bands they want.

FSI eliminates the concept of bands.  As in the laboratory, the spectral resolution (corresponding to the number of bands in a hyperspectral system) is determined by the instrument characteristics.  The FSI instrument collects all of the light and divides it up into spectral intervals depending on the characteristics of the instrument.  There is no need for accurate characterization of bandwidth, center band wavelength, and spectral distribution with the bands.

One of the most common objections to hyperspectral systems is that they produce “too much data”.  A typical hyperspectral system collects every bit for every band for every image pixel.  If we consider what we are actually looking for, spectral and spatial features, then we must consider what we really want from the raw data.  What we want is the information in the spectral and spatial features.  We want to extract the information from the raw data.

Fortunately, there is a very well developed technology for extracting information from raw data; it is called “data compression”.  For all practical purposes, a hyperspectral or a FSI instrument is a video camera.  Instead of producing a sequence of scenes, it produces o sequence of two-dimensional frames, one dimension being the spatial data and the other being the spectral data.  A video camera compresses the raw data using algorithms that preserve the best visual quality.  A hyperspectral of FSI imager would compress the raw data using algorithms that preserve the spectral and spatial information.  The same, well-developed technologies can be applied.

An instrument that produces information in the form of spectral features has many advantages over one that produces information in the form of band ratios.  First, there is a lot more information in the features of a complete reflectance spectrum.  Secondly, the need for highly accurate radiometric correction and instrument characterization is not as great then full spectral information is available.

The technology to build FSI systems is all currently available.  Most of the hardware has been developed for hyperspectral systems.  The information extraction (data compression) algorithms have been developed by people who have nothing to do with remote sensing.  The same general principles that have been developed to compress data for video cameras can be adapted to compress spectral and spatial data.  It should be noted that we are not talking about compression factors of 2 or three, but 10 to 100 or more.  The compression factor is entirely dependent on the scene, on the amount of spectral and spatial information that the instrument collects.  For example, the data rate will be very low when viewing the wheat fields of Saskatchewan, and very high when viewing the northeast coast of the United States.

FSI is just one example where advantage can be taken of technological developments outside traditional remote sensing and photogrammetry.  The only possible disadvantage is that it might introduce some new ways of doing remote sensing with which people are unfamiliar.  On the other hand, if the remote sensing community does not keep up with new methods, it risks losing out to those who are familiar with the new methods and who might simply take over their business.

Full Spectral Imaging (Draft of GIM International article)

  Benefits of the technology compared to multi- and hyperspectral remote sensing

Full Spectral Imaging (FSI) is the successor to hyperspectral imaging, much as hyperspectral imaging was the successor to multispectral imaging.  FSI provides many of the improvements to multispectral imaging that hyperspectral imaging does not.  The primary improvements are:

·         More information

·         Smaller data volume

·         Simpler instrumentation

·         Reduced calibration requirements

·         Simpler interpretation of the information

The key principle of FSI is that it provides information rather than data.  Rather than transmit every byte for every pixel for every spectral interval as does hyperspectral imaging, FSI only transmits the information contained in the data.  In remote sensing for example, a large wheat field with homogeneous spatial and spectral features would have low information content and consequently a low data rate.  An urban area, on the other hand would have large amount of information and a high data rate.

FSI information is contained in spectral features rather than band ratios.  Historically, spectral reflectance measurements have always been full spectral measurements.  Band ratioing was developed in response to the technological limitations of early remote sensing systems.  Over the years is has become a successful and widely accepted technique, but it has its limitations.  The various expansions of band ratioing such as, end-members, principle components, clusters, subspace projections, convex hull projections, manifold maps, and abundance maps have similar limitations.  These limitations are addressed by FSI.

The instrumentation requirements are simpler for a FSI system than for a multispectral system.  This is due to the fact that an FSI instrument collects all of the light across the entire spectrum and simply divides it up into spectral intervals depending on the capability of the instrument.  There is no need for accurate bandwidth and band center characterization as there is in a multispectral system.

Absolute radiometric calibration is nowhere near as critical for a FSI system as for a multispectral system.  One simplifying factor is mentioned above, no need for accurate bandwidth and band center characterization.  Another is the fact that FSI information is contained in spectral features rather than absolute values of band ratios.

Finally, FSI information is easier to process that multispectral information.  The full spectral reflectance curve is available rather than just bands.  It is possible to look at differences in the spectral features all across the spectrum to differentiate targets.

  Present state of the art of the technology

High quality instrumentation is required to extract the information from the data.  The primary requirements are response stability and high signal-to-noise ratios.  Fortunately, the technological developments that have gone into hyperspectral imaging over the past several years can all be used for FSI.  In addition, technologies that have been developed for applications unrelated to remote sensing can also be used.

A FSI or hyperspectral system is for all practical purposes, a digital video camera.  An enormous amount of work has gone into perfecting digital video camera technology.  Hyperspectral imagers have taken some advantage of this work, most notably in the use of the detectors that have been developed for video cameras.  It is also possible to take advantage of the data processing technology that has been developed for video cameras.  For a few hundred Euros, one may buy a video camera that produces high quality products with excellent color correction and stability.  Though the mode of operation of an FSI system is quite different than that of a video camera, these technologies both in the hardware and the software, can be adapted to FSI systems.

  Which applications may benefit from the technology?

Virtually all applications of passive optical remote sensing would benefit from FSI.  One set of applications that may benefit most is real-time disaster and environmental monitoring.  The high information content and relatively low data rates of an FSI system would allow direct broadcast technology to be used which would make the information available to a greater number of users in real time.  This is an application of particular interest to the author.

A specific advantage of FSI is that the information is easier to process.  One is dealing with spectral features rather than bands.  This reduces absolute calibration requirements and makes it easier to compensate for atmospheric effects.  The principles of Pattern Recognition can be applied to FSI information, both spectrally and spatially.  Basically, this is figuring out what you do not know based on what you do know.  This would contribute significantly to any application that requires the automation of information processing.

  How to handle the abundance of data?

FSI answers one of the primary complaints about hyperspectral imaging which is, “too much data”.  As mentioned above, this problem has been addressed by commercially available digital video cameras.  The applications and the exact algorithms that are needed for remote sensing systems are somewhat different than those needed for video cameras, but the basic principles have been established.  By extracting the information from the raw data, reductions in data rate by factors of 10 to 100, and even more in special cases, can be achieved.  Also, as mentioned above FSI information is particularly suitable for automated processing. 

  Which are the main limitations at present?

The primary problem faced by FSI at present is its acceptance by the traditional remote sensing community.  Hyperspectral imaging is really only multispectral imaging with a lot of bands, so it is not too difficult to accept.  FSI on the other hand uses a different approach to the information.  Fortunately, some sectors of the remote sensing community are already using these techniques for specialized applications.  Again, it is a good idea to look outside the traditional areas of expertise to develop new capabilities.

It is not possible to discuss and explain in detail all the features and potential spin-off applications of FSI systems in this short article.  During the past couple of years, a significant amount of work has been done by the author and a diverse set of collaborators in the design of complete FSI systems, most notably for real-time disaster and environmental monitoring .  Most of this work on the development of FSI systems has been done in collaboration with people and organizations outside the traditional remote sensing community.  It is very important to take advantage of these “outside the box” capabilities to advance passive optical remote sensing capabilities.  The author is currently working on forming a consortium of academics and businesses to develop the next generation of passive optical Earth observing systems.  Any questions and/or comments are welcome.