Introduction to Hyperspectral Satellites
Spaceborne Spectroscopy and Imaging
While multispectral satellites, such as Landsat and SPOT satellites, have been in regular use since the 1970s, hyperspectral satellites have emerged as a new generation of remote sensing satellites since the beginning of this millennium. Imaging spectrometry, also known as hyperspectral imaging, is a combination of the traditional spectroscopy technology and the modern imaging system (Qian 2013).
Spectroscopy is a key analytical method used to investigate material composition and related processes through the study of the interaction of light with matter. Energy is absorbed by the matter, creating an excited state. The interaction creates some form of electromagnetic waves. Isaac Newton described the spectral nature of light. Spectroscopy or spectrometry deals with the measurement of a specific spectrum for identification of matters. By using a spectrometer, one can determine the level of excitement in the matter’s atoms to identify what kind of material it is. It is incredibly difficult to make the distinction just with the naked eye, as human eyes are not able to see the fine details. Joseph von Fraunhofer invented the spectroscope in 1814 and used it to characterize the optical properties of glass for the development of more powerful telescopes. He also identified the dark lines in the solar spectrum. Kirchhoff and Bunsen used spectroscopy to investigate the composition of the solar atmosphere by establishing the connection between the solar Fraunhofer lines and the spectroscopic signatures of elements observed in the laboratory. Determining composition remotely, without physical contact, is one of the most valuable capabilities of spectroscopy. Since its beginning, spectroscopy has evolved and been used to enable a broad range of scientific discoveries by Edwin Hubble to deduce the expanding nature of the universe.
Spectroscopy is widely used in laboratories in the disciplines of physics, chemistry, and biology to investigate material properties. Spectroscopic data are often represented by an emission spectrum, a plot of the response of interest as a function of wavelength or frequency. For example, colorimetry used for the investigation of cholesterol or blood sugar in chemical laboratories is a form of spectroscopy. Spectrometry has also been used for determining blood alcohol levels, checking automobile emissions, and monitoring smokestack pollution.
A modern imaging system converts the visual characteristics of an object, such as a physical scene or the interior structure of an object, into digital signals and creates digitally encoded representations that are processed by a processor or a computer and give output in the form of a digital image. Imaging systems typically consist of a camera and an imaging lens, along with an illumination source. Depending on the system setup, imaging systems magnify or enhance observed objects to ease the viewing or inspection of small or unclear details. Computers are becoming more and more powerful with increasing capacities for running programs of any kind, especially digital imaging software. Software is becoming both smarter and simpler at a fast pace.
In the late 1970s, detector, optical, and computer technologies advanced sufficiently to enable a combination of spectroscopy with an imaging system. The combination of the spectroscopy technology and the modern imaging system is referred to as imaging spectrometry, now also called hyperspectral imaging. It could measure a spectrum for every pixel in an image. This provides revolutionary ways of observing the Earth by collecting information of each pixel in a scene across the electromagnetic spectrum. Imaging spectroscopy operating in the solar-reflected spectrum senses objects on the ground in detail spectrally and spatially. Molecules and particles of the land, water and atmosphere interact with solar energy in the 400-2500 nm spectral region through absorption, reflection, and scattering processes. Imaging spectrometers in the solar-reflected spectrum measure spectra of the ground objects as images in some or the entire portion of the spectra. These spectral measurements are used to determine constituent composition through the physics and chemistry of spectroscopy for scientific research and applications over the regional scale of the image.
The primary advantage of hyperspectral imaging is that, because an entire spectrum is acquired for each pixel of the acquired imagery, an operator needs no prior knowledge of the sample, and post-processing allows all available information from the dataset to be exploited (Chang 2013). Hyperspectral imaging can also take advantage of the spatial relationships among the different spectra in a neighborhood, allowing more elaborate spectral-spatial models for a more accurate segmentation and classification of the image (Grahn and Geladi 2007). The story of hyperspectral imaging is closely tied to advances in digital electronics and computing capabilities due to its complexity for acquiring and processing large data volume.
The term hyperspectral imaging was originally defined by Goetz et al. (1985) as “the acquisition of images in hundreds of contiguous, registered spectral bands such that for each pixel a radiance spectrum can be derived.” It sets this type of spectral remote sensing apart from multispectral imaging by requiring the spectral bands to be contiguous, so that no gaps occur through which precious spectral information might slip undetected. This original definition, however, did not explicitly mention the bandwidth of the spectral bands with respect to that of multispectral imaging. It did implicitly define the bandwidth of the hyperspectral bands as being much narrower, because of hundreds of bands in hyperspectral imaging within the same wavelength range rather than a few' bands in multispectral imaging. The bandwidth of hyperspectral sensors is typically 10 nm or less, w'hich is much narrower than that of multispectral sensors w'hose bandwidth is about 100 nm or so. The narrowness of the spectral bands in hyperspectral imaging is more important than the number of spectral bands being in hundreds. An imaging sensor that produces spectral images with a bandw'idth of 10 nm or less should be classified as a hyperspectral sensor even if its total number of bands is less than a hundred. The original definition of hyperspectral imaging could be updated as “the acquisition of many images of contiguous, narrow, registered spectral bands such that for each pixel a radiance spectrum can be derived” to explicitly mention the bandwidth of the spectral bands.
Figure 1.1 shows the concept principle of a hyperspectral satellite. It acquires images of a given scene on the ground in hundreds (or tens sometimes) of continuous and narrow spectral bands over wavelengths that range from the near-ultraviolet to the shortwave infrared. Each image, often referred to as spectral image or band image, corresponds to a particular wavelength or spectral band number. The collected “datacube” contains both spatial and spectral information of the materials within the scene. Each element or pixel in the scene is sampled across hundreds (or tens) of narrow band images at a particular spatial location in the datacube, resulting in a one-dimensional (ID) spectrum. It is a plot of wavelength versus radiance or reflectance. The spectrum for a single pixel acquired by a hyperspectral satellite appears similar to a spectrum collected by a spectrometer in a spectroscopy laboratory. The spectrum can be used to identify and characterize the particular feature of the pixels w'ithin the scene based on the unique spectral signatures or “fingerprints.” Hyperspectral imagery can provide direct identification of the surface materials and has been used in a wide range of remote sensing applications, including geology, agriculture, forestry, environment, ocean, atmosphere, climate change, defence and security, and law enforcement. The fascinating detailed spectral information provided by hyperspectral imagery often provides results not possible with multispectral or other types of imagery (Qian 2016).
FIGURE 1.1 Concept principle of a hyperspectral satellite. A datacube and spectrum of each pixel is generated by a hyperspectral satellite.
Hyperspectral Imaging Approaches
Hyperspectral sensors acquire both spatial and spectral information of a scene and generate a datacube for the scene. There are at least three different approaches to acquiring the hyperspectral data in terms of the type of imaging spectrometers:
- 1. Dispersive elements based approach
- 2. Spectral filters based approach
- 3. Snapshot hyperspectral imaging
Dispersive Elements Based Approach
Figure 1.2 shows an example of the concept of a dispersive element based hyperspectral sensor with a dispersing element and a two-dimensional (2D) detector array. In this approach, a spectrometer separates the radiation light from an object into spectral components, or dispersion, by the use of an optical element possessing a known functional dependence on wavelength, specifically prisms or diffraction gratings. A dispersive element is inserted into a collimated beam of the instrument and then creates a dispersed spectrum centered on the object’s location in the instrument’s field of view (FOV). A grating or a transparent prism can be used to break up the radiation light into its constituent monochromatic components. The grating-dispersed monochromatic components are in linear distribution. The monochromatic components dispersed using a prism are in nonlinear distribution. A grism can also be used to break radiation light up into its constituent monochromatic components. A grism is a combination of a grating and a prism (also called a grating prism) so that radiance light at a chosen central wavelength passes straight through.
FIGURE 1.2 An example of the concept of a dispersive elements based hyperspectral sensor.
Dispersive element based hyperspectral sensors are the most popular ones for both airborne and spaceborne remote sensing. This type of hyperspectral sensors needs to scan the scene on the ground either by a dedicated scanner or by the entire instrument with the aircraft or satellite flight motion to obtain the spatial coverage. There are two operating modes in terms of scanning of the scene: whiskbroom and pushbroom.
For the early airborne hyperspectral sensors, the whiskbroom operating mode was often used, such as the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) developed by NASA’s Jet Propulsion Laboratory (JPL) (Green et al. 1998) (the recent AVIRIS still uses the whiskbroom mode). This was because linear detector arrays were used that could record the monochromatic components of the spectrum of only one ground pixel (or the sampling cell). The instrument needs to scan the ground sampling cells one after another in a cross-track line using the whiskbroom mode, as shown in Figure 1.3. After completion of scanning the entire current cross-track line, the instrument starts to scan the ground sampling cells of the next cross-track line when the aircraft or satellite flies forward in the flight direction (also referred to as along-track direction), and so on. The advantages of the whiskbroom operating mode are:
- 1. Simple design of the instrument.
- 2. Wide FOV, as there is no constraint imposed by the available number of pixels of a 2D detector array in the spatial direction thanks to using only ID linear detector array.
- 3. Easy calibration, since all the spectra of the ground sampling cells within the scene are generated by the same linear detector array and the same optics, which have the identical spectral characteristics. This type of hyperspectral sensor has no spatial distortion (also referred to as keystone) as does a pushbroom hyperspectral sensor.
FIGURE 1.3 Concept of a hyperspectral sensor operating in the whiskbroom mode with a linear detector array.
The disadvantages of the whiskbroom operating mode are:
- 1. Requirement of a mechanical scanner, which contains moving parts in a vacuum chamber.
- 2. Requirement of post-processing for spatial incongruence.
- 3. Constraints of high spectral and spatial resolution requirements due to low integration time.
Almost all the spaceborne hyperspectral sensors use the pushbroom operating mode. As shown in Figure 1.2, in the pushbroom mode all the ground sampling cells in an entire cross-track line are imaged into a 2D detector array simultaneously. One dimension of the detector array corresponds to the spatial direction of the cross-track line and another dimension corresponds to the spectral extension of the ground sampling cells. The advantages of the pushbroom operating mode are:
- 1. No moving parts.
- 2. Congruence spatial images.
- 3. Longer integration time for each ground sampling cell, because each of them is sensed simultaneously by a row of elements of the 2D detector array (e.g., rows А, В, C, ... G in Figure 1.2) instead of one cell after another, which omits the time sharing scanning of all the ground sampling cells in a cross-track line. Longer integration time means more photos are collected and results in higher signal-to-noise ratio (SNR).
The disadvantages of the pushbroom operating mode are:
- 1. Complex optical design and complex focal plane.
- 2. Narrow FOV, which is constrained by the available number of pixels of the 2D detector array in the spatial direction.
- 3. Complex calibration.
- 4. Spectral distortion (also referred to as smile) and spatial distortion (also referred to as keystone). The hyperspectral data collected by a pushbroom hyperspectral sensor need to be sufficiently corrected for smile and keystone distortion before being distributed to users for downstream applications (Qian 2013).
FIGURE 1.4 A block diagram of composition of a spectral filter based hyperspectral sensor.