Hyperspectral imaging is a spectroscopic technique. The technique collects hundreds of images at different wavelengths over a linear spatial area with the aim of collecting spectra for each pixel in a sample. This analysis of wide spectrum looks at more than the primary colors (red, green, blue) at each pixel, breaking down the light hitting each pixel in a hyperspectral imaging system into many different spectral bands to provide more information on what is imaged. This includes measuring the visible light and the near-infrared, which are collected as data to form what is called a hyperspectral cube, in which the two dimensions represent the spatial extent of a scene and the third dimension represents the image's spectral content.
Example of a hyperspectral imaging system or camera.
Each material scanned by a hyperspectral imaging system possesses a specific spectral signature, also known as a "fingerprint," which allows for its unique identification. This offers hyperspectral imaging a range of applications in remote sensing, especially in its non-destructive capability while recognizing the components of matter. The technique has been employed in fields such as astronomy, agriculture, molecular biology, biomedical imaging, minerology, physics, cultural heritage, food processing, environment, and surveillance.
Hyperspectral imaging sensors work to image across electromagnetic spectrum bands to identify features of objects or areas that are otherwise not visible to common cameras or the human eye. These features that can be detected are related to the optical properties of the analyzed materials, and the different spectral signatures enable users to separate different materials from others and allow users to make qualitative statements on the analyzed objects and materials. Materials and objects scanned by hyperspectral sensors leave spectral signatures, also known as "fingerprints," which enable the identification of materials an object is composed of.
Example of a hyperspectral data cube compiled to form an image.
Hyperspectral sensors generate data, which is known as a hyperspectral cube. The cube comprises a series of images that create the information the imaging system present. Each image represents a range of the electromagnetic spectrum, which is known as a spectral band, and the images are then layered to form a three-dimensional hyperspectral data cube that can then be processed and analyzed.
A hyperspectral camera is built with an integration of an imaging spectrograph and a monochrome matrix array sensor (or a camera). A lens images the sample onto a slit of the transmission spectrograph, which then produces a spectrum imaged on a focal plane array detector and preserves the location of the respective points, which is important when putting those images back together as a hyperspectral cube. The results of the camera are then capable of being processed, scaled, smoothed, and compressed to produce information such as measurements, composition, color, coordinates, and thickness.
The difference between multispectral imaging and hyperspectral imaging partially explains hyperspectral imaging. The standard definition is that hyperspectral imaging contains more than one hundred bands, whereas multispectral has fewer. This means that an imaging camera with fifty bands can be considered a multispectral imaging system rather than a hyperspectral imaging system. However, it is often considered more correct to note that hyperspectral imaging systems are better defined and differentiated by the spectral resolution, which is called full width at half maximum (FWHM) and highlights the hyperspectral camera's capability to separate two consecutive spectral peaks from each other.
Two broadly defined technologies have enabled the development and advancement of hyperspectral imaging. The first is the development of inexpensive and high-quality diffraction gratings. The second is the advance of multiple dimension data processing.
Examples of diffraction gratings.
Hyperspectral imaging systems consist of imaging optics, a narrow slit, a diffraction grating, and a two-dimensional focal plane array detector. An image is projected through the slit onto the diffraction grating, which splits the light into discrete wavelengths before being projected on the focal array. Having high-quality, lightweight, and affordable diffraction grating makes these systems inexpensive, as they have been made cheaper and more durable from high-quality polymer gratings rather than the traditional precision blazed on high-quality glass, which is a fragile and expensive process.
A volume Bragg grating (VBG) is one such diffraction grating, which offers a periodic modulation of the refractive index through the volume of a photosensitive material. The modulation can be oriented either to transmit or reflect the incident beam. This filter includes a grating with variable thicknesses, a refractive index of the photo-thermal-refractive glass, the period of the grating, the angle between the incident beam and the normal of the entrance surface, and the inclination of the Bragg planes defined as the angle between the normal and grating vector.
Example of how a tunable volume Bragg filter can adapt and change the spectral findings of a hyperspectral camera.
Tuning such a filter allows a user to exploit the gratings to extract a small bandwidth of wavelengths out of a polychromatic input, and tuning the angle of grating makes it capable to scan the output wavelength over hundreds of nanometers. This allows the detection of a whole image at a specific wavelength band.
At the same time, the second piece of the hyperspectral imaging puzzle has been the improvement in high-speed computer data analysis. The software has been developed to tie in GPS coordinates to each data point, interleaves the pushbroom data (while accounting for overlap), can automatically check against known features, and can operate at cycles in excess of 500 frames per second. This is a crucial capability for making a jump from processing a single line of data to processing an entire composite image, especially as these imaging systems can produce over 800 million pixels of data.
The techniques of hyperspectral imaging emerged as a product of military research, where it was used primarily to identify targets among objects and background clutter. The technology has also been used in civil applications, such as earth imaging satellites, when NASA launched the EO-1 satellite with a hyperspectral imaging sensor. And the technology has been used as a quick tool for the diagnosis of tissue conditions at diagnosis and during surgery.
