Hyperspectral imaging
For broader coverage of this topic, see Spectral imaging.
Hyperspectral imaging collects and studies information from across the electromagnetic spectrum. Its goal is to get the spectrum for each pixel in a picture of a scene. This helps scientists find objects, identify materials, or notice changes. There are three general types of spectral imagers. Push broom scanners and the related whisk broom scanners build images over time. Band sequential scanners take pictures of an area at different wavelengths. Snapshot hyperspectral imagers use a staring array to make an image in an instant.
The human eye sees the colors of visible light in mostly three bands. Long wavelengths look red, medium wavelengths look green, and short wavelengths look blue. Spectral imaging splits the spectrum into many more bands. This method can also reach beyond what we can see. Hyperspectral imaging records spectra with fine detail across a wide range of wavelengths. It measures continuous bands, while multiband imaging measures spaced-out bands.
Hyperspectral imaging is often used to study surfaces, but it can also look beneath the surface. How deep it can see depends on the range of wavelengths used and the properties of the sample.
Engineers build hyperspectral sensors and processing systems for many uses. These include astronomy, agriculture, molecular biology, biomedical imaging, geosciences, physics, and surveillance. These sensors look at objects across a large part of the electromagnetic spectrum. Some objects leave unique "fingerprints" in this spectrum, called spectral signatures. These help identify the materials in a scanned object. For example, a spectral signature for oil helps geologists find new oil fields.
Sensors
Hyperspectral sensors collect information as a set of "images." Each image shows a narrow range of light wavelengths, called a spectral band. These images are combined into a three-dimensional data cube. Two dimensions (x and y) show the scene, and one dimension (λ) shows the range of wavelengths. Sensors can gather this cube in four different ways: spatial scanning, spectral scanning, snapshot imaging, and spatio-spectral scanning.
Hyperspectral cubes come from airplanes, such as NASA's Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). They can also come from satellites, such as NASA's EO-1, which carries a hyperspectral instrument called Hyperion. For many development and testing studies, handheld sensors are used. A sensor's quality is measured by its spectral resolution, which is how narrow each captured band is. Sensors that detect many narrow bands can identify objects, even when those objects fill only a few pixels. Spatial resolution also matters. If pixels are too large, several objects can share one pixel and become hard to tell apart. If pixels are too small, each sensor cell captures less light, and the signal-to-noise ratio makes the measurements less reliable. Collecting and processing hyperspectral images is also called imaging spectroscopy, or 3D spectroscopy when talking about the data cube.
Scanning techniques
There are four basic ways to collect the data for a hyperspectral image. The best choice depends on the job, since each method has its own strengths and weaknesses. In spatial scanning, each sensor output shows a full slit spectrum, and the image is built up one line at a time with a push broom scanner. These line-scan cameras are common in remote sensing and are also used to check materials moving on a conveyor belt. A special case called point scanning uses a whisk broom scanner, which looks at a single point instead of a slit. In spectral scanning, each sensor output shows one wavelength of the scene. The scene is scanned by changing filters while the platform stays still.
Main article: Snapshot hyperspectral imaging
In non-scanning, one sensor output holds all the spatial and spectral data at once, so no scanning is needed. Many computed tomographic imaging spectrometry and integral field spectroscopy with lenslet arrays systems have been designed, along with others such as integral field spectroscopy with image slicing mirrors. These snapshot systems can collect light quickly and efficiently, but they need a lot of computing power and are expensive to build. Spatiospectral scanning combines some strengths of spatial and spectral scanning, and it is described in more detail in its own article.
Distinguishing hyperspectral from multispectral imaging
Hyperspectral imaging is part of a group of methods called spectral imaging or spectral analysis. The name came from the development of NASA's Airborne Imaging Spectrometer (AIS) and AVIRIS in the mid-1980s. NASA prefers the older term "imaging spectroscopy," but "hyperspectral imaging" is now used more often in science and everyday talk. Experts suggest using "imaging spectroscopy" or "spectral imaging" and avoiding big prefixes like "hyper-," "super-," and "ultra-," so that misnomers do not cause confusion.
Hyperspectral imaging is related to multispectral imaging. Hyperspectral imaging uses many narrow, joined ranges of wavelengths, so it can show the spectrum of every pixel in a scene. Multispectral imaging uses a smaller set of chosen wavelengths and does not show the full spectrum of an object. Its separate bands are what set it apart from color photography in the visible range. Landsat is a well-known example. The term "ultraspectral" could be used for very precise interferometer sensors. These often have low spatial resolution and cover only several pixels because they produce a large amount of data.
Applications
See also: Multispectral imaging § Applications
Hyperspectral remote sensing is used in many ways. It was first developed for mining and geology, because it can identify minerals and help look for ore and oil. Later it was used in ecology, surveillance, and historical manuscript research, such as imaging the Archimedes Palimpsest. Organizations such as NASA and the USGS have posted lists of minerals and their spectral signature online for researchers. In Australia, people use imaging spectrometers to detect grape varieties and to build an early warning system for disease. Hyperspectral data can also show the nutrient and water level of wheat. It can help check animal proteins in animal feed, which helps prevent bovine spongiform encephalopathy (BSE), also known as mad-cow disease, using near infrared microscopy. Scientists also use it in zoology to study coloration. Some tropical frogs and leaf-sitting insects reflect a lot of near-infrared light. It can help with waste sorting and recycling by telling different materials apart. Cameras can work with machine vision systems. One system that uses machine learning can tell 12 types of plastic apart, which helps sort unstandardized plastic products and packaging. Researchers at the Université de Montréal, working with Photon etc. and Optina Diagnostics, are testing whether it can find retinopathy and macular edema early. This could help an ophthalmologist treat the eye before damage happens.
In the food processing industry, hyperspectral imaging helps optical sorters find and remove defects and foreign material that ordinary cameras cannot see. Hyperspectral sorters are most common in the nut industry, and people are working on using them in potato processing. Geologists can quickly map drill cores for many useful minerals, including those in the feldspar, silica, calcite, garnet, and olivine groups. Hyperspectral imaging is useful for surveillance. Military groups now use countermeasures to avoid being watched from the air. It has also shown promise for facial recognition. Thermal systems once needed liquid nitrogen or helium cooling. In 2010, Specim introduced a camera that can be used outdoors on a UAV without an outside light source. In astronomy, the same idea is called integral field spectroscopy. It is used on the Very Large Telescope and the Advanced CCD Imaging Spectrometer on the Chandra X-ray Observatory.
Main article: Chemical imaging
Soldiers can face chemical hazards that are mostly invisible. Hyperspectral imaging can detect them from up to 5 km away. Cracks in pavements that are hard to see with regular cameras may also be found with this technology.
Advantages and disadvantages
A main advantage of hyperspectral imaging is that it collects a full spectrum at each point. The operator does not need to know about the sample ahead of time, and the useful information can be studied later. The method can also look at how nearby spectra relate to each other. This allows more detailed models for a more accurate segmentation and classification of the image.
The equipment is expensive and complicated, so it is not used very often. Analyzing the data needs a lot of computing power, especially for real-time or mobile uses. The data also needs a lot of storage and transfer space, because hyperspectral cubes are large, multidimensional datasets. For example, NASA's AVIRIS sensor, used since 1987, produces about 500 megabytes per datacube. Researchers have worked on processing data on satellites so that only the detection results need to be sent. Deep learning is also being studied as a way to lower the computing cost of processing hyperspectral data.

Quiz time!
4 questions about Hyperspectral imaging
What does hyperspectral imaging try to get for each pixel in a picture?
Pictures
Keep exploring
This article is a child-friendly adaptation of the Wikipedia article on Hyperspectral imaging, available under CC BY-SA 4.0.
Images from Wikimedia Commons. Tap any image to view credits and license.