A SIMPLIFIED DESCRIPTION OF FULL SPECTRAL IMAGING
or, "Full Spectral Imaging for Dummies"
or, "An Idiot’s Guide to Full Spectral Imaging"
or, "Full Spectral Imaging for Practitioners of Traditional Remote Sensing"
Top Level Summary
The goal of the Full Spectral Imaging development is to provide high quality remotely sensed information to anybody who needs it, in a format that is easy to use, in real-time.
Full Spectral Imaging (FSI) is the successor to hyperspectral imaging (HI). FSI eliminates the need to select spectral and spatial resolutions. The ultimate spectral and spatial resolutions of an FSI system are determined by the optical characteristics of the instrument, and are specified by the applications(s). The FSI system only transmits the amount of information contained in a scene. If the scene contains a lot of spatial details and many spectral signatures, the full capability of the instrument will be utilized. If the scene is spectrally and/or spatially uniform, the system will only transmit the amount of information required.
The Short Version
Full Spectral Imaging (FSI) is the successor to Hyperspectral Imaging, which is the successor to Multi-Spectral Imaging, which is the successor to Plain-old Imaging. Plain-old Imaging whether it be panchromatic (black and white) or color depends on the spatial resolution of the image. In other words, if the person looking at the picture cannot see what he is looking for because it is too fuzzy, then he will not be able to figure out exactly where and what it is. Rather than continuously try to increase the spatial resolution of Plain-old Imaging in order to get a better ‘look’, somebody figured that it might be a good idea to take advantage of the color of what you were looking at, the color of the target. The color of the target is the amount of light reflected at various wavelengths, or the reflectance spectrum.
The typical method for measuring a reflectance spectrum is to first measure the spectral reflectance of an artificial target that reflects all of the incident light equally, or a perfectly white object. This measurement is done with an instrument called a reflectometer. A reflectometer typically consists of a monochromator and a light detector. A monochromator is a device that can select a narrow range of wavelengths and which can be scanned through a wide range of wavelengths. The light detector is a device that produces an electrical signal proportional to the amount of light falling on it, and which is sensitive to all of the wavelengths covered by the monochromator.
Once the spectral reflectance of the white object, or the 100% reflectance is measured, the spectral reflectance of the target can be measured. Then, to get the absolute spectral reflectance of the target a ratio is taken between the target and the 100% standard. This simple procedure compensates for variations in the light source and in the reflectometer. The result of this procedure is a spectral reflectance curve, a plot of percent reflectance versus wavelength.
MSI, HI, and FSI
It has not been possible, until recently, to apply this simple procedure to either airborne or spaceborne remote sensing systems. Technological limitations forced the developers of early systems to select only a few spectral bands that were representative of the full spectral information that could be obtained in the laboratory. Selection of these bands has always been a difficult and sometimes controversial process. As the technology for remote sensing improved, more and more bands could be selected. Over the years an extensive science of remote sensing has grown up around this principle called Multi-Spectral Imaging (MSI).
Finally, some years ago, instrument technology became good enough so that “all” the bands could be selected. This approach to remote sensing is called Hyperspectral Imaging (HI). Even though it is now possible to select all the bands, there is still controversy as to what this means. Typically, the question boils down to issue of the spectral resolution, or the wavelength range of the individual bands. In other words, HI is just Multi-Spectral Imaging with a lot of bands.
FSI uses the same instrument technology as HI but treats the data differently. Instead of a lot of bands, FSI produces spectra. In other words, FSI takes us back to the approach to spectral reflectance measurement that is used in the laboratory. Just saying that FSI produces spectra is not the whole story. The critical difference between FSI and HI is that HI, being just MSI with a lot of bands, collects and transmits all of the data in all of the bands. This leads to one of the biggest problems with HI; too much data.
Spectra and Information
What does it mean to say that FSI produces spectra? What is so great about spectra or spectral curves anyway? The great thing about spectra is spectral features. Spectral features are the dips, humps, wiggles, and slopes of the spectral curve. The information in a spectral curve is contained in these features. What differentiates a spectral curve from a straight line, or a simple arc are those features. If a spectral curve looks like a straight line or a simple arc, or if it looks just like a spectrum that was obtained previously, then it does not contain much information.
To take advantage of spectral features or of the information in the spectral curve, one must apply Information Theory (IT). Fortunately, this is not a big deal. For many years people have been applying the principles of IT to the technology of data compression. So, we may as well take advantage of all the wonderful work that these IT people have done and employ data compression. Unfortunately, applying data compression is not as simple as it may seem. One must understand something about the way that data compression works in order to employ it most effectively. These details, which concern the performance of the FSI instrument, will be discussed later.
One advantage to employing data compression is that it can be applied to the spatial features of the FSI data as well as to the spectral features. Just as with spectral features, if the scene does not contain a lot of spatial details the amount of spatial information will be less than in a spatially complicated scene. So, in one fell swoop, by simultaneously compressing both the spectral and spatial data, one may achieve dramatic decreases in the average data rates. The data rate will vary with the spectral and spatial complexity of the scene.
As mentioned above, it is not possible to build a FSI system by simply plugging a data compression device into a HI system. The performance of the entire system, from front-end instrument optics to the data processing system, must be optimized. Typically, if one mentions ‘data compression’ to a remote sensing researcher, they tend to go ‘non-linear’. It is therefore important to demonstrate that the quality of FSI data will not be compromised by compression. The data quality can only be maintained if the complete end-to-end system is designed properly.
The Reflectance Reference and Measurement
Full spectral measurement is only one part of the analogy to laboratory measurement of reflectance. The second analogy produces the concept of Empirical Reflectance Retrieval. In the laboratory, reflectance measurements are referenced to a white reflectance standard. Though a lot of work has been done with reflectance standards on the surface of the Earth, this is only useful for instrument calibration and is not good for the task of 'reflectance retrieval' which is currently the critical task that must be performed before remotely sensed data is useful.
The two big problems for reflectance retrieval are that there are no generally available white reflectance standards, and that the atmosphere contributes a significant amount of the spectral reflectance that is measured by a spaceborne instrument. This problem is currently dealt with by measuring radiances at the instrument accurately, and by modeling the spectral reflectance contribution of the atmosphere. Accurate radiance measurement requires good instrument calibration, and atmospheric modeling required a sophisticated understanding of atmospheric reflective and radiative processes. Both tasks are difficult.
Fortunately, when looking at the surface of the Earth, we are not looking at completely unknown targets. We do know, quite accurately in fact, the spectral reflectance of many ground targets. The spectral reflectance of the known ground targets can be used to determine the spectral reflectance of unknown targets. The spectral reflectance of the known ground targets can be also used to determine the contribution of the atmosphere to the measured reflectance. Any difference in the spectral reflectance of a known target can be attributed to the atmosphere. The only assumption that we have to make is that the spectral response of the instrument remains constant. We do not have to know the response of the instrument absolutely. This significantly simplifies the instrument calibration.
The process of retrieving the spectral reflectance of unknown targets based on the spectral reflectance of known targets is called Empirical Reflectance Retrieval.
Automating the Process
The process for producing high quality remotely sensed information that is easy to use can be automated. This process is called Autonomous Remote Sensing.
A Practical Implementation of FSI
Possibly, the best example of an implementation of FSI would be a next-generation LANDSAT remote sensing system.
[To Be Continued...]
This page was last modified on Thursday, June 30, 2014