Hyperspectral imaging from sensors mounted on satellites has profoundly changed the way objects are viewed on Earth, other planets, and the space in between using different wavelengths of light reflected by objects that create a detailed spectral fingerprint. Extreme high-resolution details of images on the ground and in space can be seen, monitored, and tracked with greater precision than with multispectral imaging.
While the technology has been primarily used by government missions, a number of startups are now targeting this area, and have launched the first commercial hyperspectral satellites with plans to deploy constellations. These startups are targeting use cases like crop analysis, or monitoring mining operations from space. But the field is wide open, more applications are being discovered, and costs for launch are coming down.
“Hyperspectral imaging is still in the stages of demonstration and validation,” Aravind Ravichandran, a satellite data strategist for TerraWatch Space, said. “From a market perspective, I think the focus will be for them to just provide this to the users who have the ability to process [the data], because hyperspectral data is extremely complex to understand and process.”
Hyperspectral imaging is a type of sensor that allows us to see closer, not in terms of its spatial resolution or the image quality, but in terms of the electromagnetic spectrum, Ravichandran said. “Instead of just seeing a plant, hyperspectral imaging can go deeper and see the chlorophyll content of the plant, or the composition of the soil. I call it kind of zooming to the properties and characteristics of whatever we're seeing, as opposed to just zooming into it as a picture.”
Hyperspectral imaging takes a spectrum of light and divides it into hundreds of much narrower spectral bands in the 10-20 nm range. Multispectral imaging breaks light into just four to 36 bands, from visible to near infrared in relatively large bandwidths (70-400 nm).
In the late 1950s, advanced radar-based methods were used to observe selected areas of Earth from high altitude planes, which later evolved into the multispectral imaging process. The first U.S. government satellite-based multispectral imaging sensor was Landsat 1, launched in 1972.
The first hyperspectral imaging satellite, NASA's EO-1, was launched in 2000, and there are now at least 25 hyperspectral imagers deployed in space. Three government-delivered hyperspectral image missions include PROBA-1, PRISMA, and EnMap, with the EnMap as the first hyperspectral satellite launched by the German Space Agency in April, 2022 for Earth observation, climate change, environmental protection, and nature conservation.
What Hyperspectral Imaging Does
Hyperspectral imaging captures images of an object or scene from visible to infrared, providing detailed information about the object's physical properties and chemical composition based on getting a spectrum for each pixel in the image of a scene, instead of just assigning primary colors of red, blue and green to each pixel.
A hyperspectral image is made up of a stack of images (one for each spectral band) often portrayed as a 3D image cube, with two spatial dimensions and one spectral dimension represented by the stacked layers.
Hyperspectral analysis is a complicated process requiring specialized understanding of how to read and manage the huge load of image data that includes the classification of pixels based on their spectral signature. New information extraction methods for hyperspectral image processing are under development. A new hyperspectral image classification (HSIC) framework-depth multiscale spatial spectral feature extraction algorithm has been proposed by a group of Chinese engineers.
Russell Hannigan, chief innovation officer of Xplore, founded in 2017, says hyperspectral imaging has benefits over synthetic aperture radar (SAR), which sends radar signals to the ground, which is then used to reconstruct an image. SAR can be used despite darkness or through cloud cover.
However, wind can change the appearance of the surface of water, for example, making it appear slightly red instead of blue. “If you're trying to look down from above and understand what something is made of, or composed of, or what its status is, the more channels of color that you have, the better you can differentiate it from something else,” Hannigan said. “Up to this point, most satellite systems have had multispectral sensors which are typically six, eight, 10 colors. Hyperspectral just means essentially dozens of colors. It just means you have many more to choose from.”
Xplore is working to develop novel hyperspectral capabilities based on its Xcraft spacecraft platform, but Hannigan did not share the specifics of its constellation plans, only that it’s on the scale of “seeing anywhere in the world in a couple of hours.”
Xplore also plans to use its hyperspectral sensors for space domain awareness applications by turning the spacecraft on an axis (slewing) to look at other satellites and orbital debris in Earth orbit and beyond. “There's nothing especially magical about hyperspectral imagery. It's just that it is in its infancy in the commercial space domain,” Hannigan said.
Dan Katz, CEO of Orbital Sidekick, said his company is building a constellation of hyperspectral satellites to be launched by the end of this year, with a plan to build a total of 14 that will give the company the ability to map “every square inch of the globe multiple times a week with high res hyperspectral imagery.” The company announced the launch of the first two satellites of the constellation on April 17.
Orbital Sidekick, founded in 2016, is one of the first companies to use hyperspectral imaging to detect hydrocarbon or gas leaks. People are starting to wake up to the power of hyperspectral, Katz said.
“It's a different way of thinking about how to analyze and extract intelligence from these datasets,” he said. “It's more of a computer problem or computer vision problem or machine learning problem as opposed to an analyst staring at a hyperspectral data cube. It’s not really going to help you. You need the really powerful algorithms and intelligence behind it. I think now that we have that coming, that catching up on cloud-based computing capability.”
A Myriad of Use Cases
Hyperspectral imaging can be used for crop analysis, where someone interpreting the data can look at disease, pest infestation, nitrogen levels, and water levels in a crop, according to Derek Woods, CEO, Hypersat.
Hyperspectral imaging can also be used as part of an Earth remote sensing satellite package offering, Woods said. “I think it should be,” he said. “Our satellite will fly a panchromatic context imaging camera. So we'll be able to do both panchromatic and hyperspectral. I would certainly think we're going to be combining with SAR, but also with ground-based data, weather data, and historic crop data.”
Hannigan said that it’s important to understand whether a crop is stressed in some way, and hyperspectral imaging is good for seeing that kind of thing. “Beyond that, it’s good for looking at plankton in the ocean for example, or detecting when ships dump oil, because you can see the oil on the surface of the water.”
India’s Pixxel, founded in 2019, showed its first-ever set of images from its hyperspectral pathfinder satellites in February. The company, now with three satellites in space and six more being built for a future SpaceX launch, also works with mining companies to help reduce the carbon footprint of their operations by helping a mining company keep a track on the tailings ponds, which is the waste and poisonous material that comes out of the mining.
“They create a dam for it, and we are there to ensure that it is not seeping into the soil, and it's not feeding into the trees and vegetation,” Awais Ahmed, CEO of Pixxel, said. “These companies want us to help monitor the impact of the mining on the biodiversity around their operations of the mining industry, and that we're helping make it as less harmful to the environment as possible.”
Bill Schuster, co-founder of HySpecIQ, explains another use for hyperspectral imaging in mining. HySpecIQ, founded in 2015, has been using an investment from billionaire Peter Thiel to accelerate the development of a 12-satellite constellation to deliver hyperspectral imagery.
“If you want to look for gold, it can be very deep in the ground,” Schuster said. “We do not have X-ray eyes. We cannot see underground. However, there are indicator minerals known to be associated with the presence of sought-after ore. In the case of gold, elements such as copper, lead, and cobalt can be indicators of its presence. So you search for these minerals whose location would be indicative that there is gold nearby. That's the tradecraft: Don't look for gold, because you're not going to see it. Instead, look for these other indicator minerals."
With hyperspectral imaging, Schuster said, flooding potential can be detected by looking at the saturation of soil. Working with light detection and ranging (LIDAR) to get a topographical map enables a combination of data with the water saturation data to report where the water's going to go, and what the risk is, he said.
The NRO Takes a Closer Look
The U.S. National Reconnaissance Office (NRO) has taken a greater interest in what these new commercial companies are doing with hyperspectral imaging. In March, the NRO awarded six companies — BlackSky Technology, HyperSat, Orbital Sidekick, Pixxel, Planet, and Xplore — study contracts for commercial hyperspectral imagery to allow the office to explore the potential of commercial hyperspectral imagery.
“They wanted to actually start getting used to our data,” Ahmed said. “It's just an evaluation study, where we are working with them, and where they're buying data from our satellite, with whatever capacity that exists. We are looking at different terrains, we're looking at deserts, we're looking at the mountain areas, we're looking at the vegetation area,” he said. “The NRO wants to verify how the satellite performs in different terrain.”
HySpecIQ successfully completed a contract for the NRO that was awarded a year and a half ago, according to Schuster, but the recent awards indicate that the NRO is expanding their search for companies to work with. “What I think the NRO is doing, as they're starting to depend more on commercial companies, is building up a rich industrial base,” he said. “What they're trying to do is to see who else is out there, what capabilities exist or are emerging so that they can avail the country to the best and most competitive solutions with a strong industrial base behind them.”
Woods said that Hypersat is using their NRO invite as a boost of confidence for their capability. “We're able now to work closely with the government and sort of weave their needs into our offering, and determine exactly what they want to see, what's beneficial to them, what grand sample that they need, what the revisit rates are that would be best for them,” he said. “So it's a very cooperative venture.”
Once the NRO verifies image quality, there'll be like a “stepping stone” of a year or two years when the NRO, if it thinks that all the providers or some of them make sense for acquisition, would have an RFP for a longer, larger scale contract, according to Ravichandran.
There is increasing interest for using hyperspectral imaging from space, coming from customers who are watching the steady startup company growth.
Most of their early customers actually reached out to Pixxel when they heard about the company doing hyperspectral imaging from space, Ahmed said. “We understand that there will be a challenge in making sure of widespread market adoption, because it will require education, it will require providing tools and infrastructure to actually make use of hyperspectral imaging. Because not everyone might have the expertise to do it.”
These new startup companies need to demonstrate what they can do, Ravichandran said, there have been only a few hyperspectral missions launched and commercial customers may not know how to work with the data.
“From a market perspective, I think the focus will be for them to just provide this to the users who have the ability to process the data, because hyperspectral data is extremely complex to understand and process. You cannot go and try and sell this to an agriculture company that has never used hyperspectral before, because they need to understand what it is and what they can do with it, before they will pay money for it. So that's a very long sales cycle.” VS
Lead photo credit: Pixxel
Correction: A previous version of this story misattributed a quote from Bill Schuster to Derek Woods.