Vancouver-based Metaspectral has announced that a team created by the company has reached the semi-finals of the XPRIZE Wildfire competition. Eight other Canadian entries failed to make the cut.
The XPRIZE Wildfire competition is a four year, multi-track, $11 million USD competition focused on developing technologies that can mitigate wildfires. According to XPRIZE, it is intended to “transform current wildfire management approaches through the development of new technologies that can rapidly and accurately detect, characterize, and respond to wildfires before they become destructive.”
This is especially salient in Canada, where many parts of the country were covered with record-breaking wildfires, leading not only to widespread destruction but to choking smoke in many of the country’s most densely-populated areas. Cities like Calgary and Toronto wound up with some of the worst air quality in the world at various points during the summer of 2023.
Perhaps with that in mind, Metaspectral has turned their technological eye to identifying these wildfires. Metaspectral is a company focused on deploying machine learning and artificial intelligence technologies for use in advanced computer vision and hyperspectral imaging, and in particular on managing the immense data flows created by hyperspectral cameras.
In a press release, Francis Doumet, CEO and co-founder of Metaspectral, said that “our technology uses hyperspectral imagery captured by satellites. Hyperspectral images include up to 300 unique spectral bands instead of the usual three that conventional cameras capture,” which provides “an extremely high level of detail.”
According to a SpaceQ email exchange with Doumet, the company has “a planned deployment of a payload on the International Space Station to analyze hyperspectral data in real-time,” and has “also partnered with SkyFi to integrate satellite imagery into their hyperspectral data analysis platform for enhanced data analysis capabilities.”
Team SpectraCan
Metaspectral formed a team for the XPRIZE Wildfire competition, “SpectraCan.” The team is competing in the Space-based Detection and Intelligence Track (Track A) of the competition, which tasks teams to develop solutions that can accurately detect fires across large landscapes in under one minute, and precisely characterize and report data without false positives in under ten minutes.
SpectraCan “proposes real-time fire detection using existing satellites for continuous coverage in high-risk areas with geostationary and orbital satellites,” by employing an algorithm to track satellite coverage and to “pull data for specific regions of interest as it becomes available.” The team is aiming to detect wildfires in less than sixty seconds, by applying “deep learning to spectral data in real time.” Doumet said that their technology will “[make] it possible to detect wildfires earlier and with much higher accuracy.”
At the moment, SpectraCan consists solely of Metaspectral employees, but Doumet said to SpaceQ that “Metaspectral is in discussion with several parties on partnering and/or joining the Spectracan team.”
Doumet told SpaceQ that this will be a mix of existing and new technologies. Metaspectral’s SpectraCan team will “use its existing AI technology to spectrally unmix signals to look for hidden targets,” but will also “integrate new data sources and potentially train new models to achieve the highest possible accuracy.” Doumet also told SpaceQ that Metaspectral has been focusing on “using Generative AI to automate complex tasks to make the data format more accessible;” this may become part of SpectraCan’s approach to wildfire detection.
Metaspectral CTO and co-founder Migel Tissera emphasized in the news release that the volume of data from hyperspectral imagery, and Metaspectral’s role in managing it, as a key part of SpectraCan’s strategy. “Hyperspectral image capture produces gigabits per second of data, which requires significant computing resources for storage, analysis, and transmission,” he said, and Metaspectral manages that by using their proprietary, advanced data compression algorithms and machine learning to make it possible to process, analyze, and transmit this data in real time without compromising the data quality.”
Tissera pointed to how the technology “can also be applied to identify high-risk zones by distinguishing between healthy and deteriorating vegetation, and even moisture levels on the ground.”
XPRIZE Semi-Finalists
Along with SpectraCan, 20 teams made it to the semi-finals. The teams will share a milestone prize purse of $750,000 USD, of which SpectraCan received the equivalent of $50,000 CAD, which they’ll be using to continue to develop the technology.
Reaching the semi-finals, according to the XPRIZE Foundation, involved being selected by “XPRIZE Wildfire’s independent judging panel,” which is “composed of external experts who possess the subject matter experience to evaluate, verify, and validate the teams’ solutions,” including experts in fire science, geophysics, and engineering.
Each team, including SpectraCan, provided a Qualifying Technical Submission—a written proposal detailing their team’s solution to the competition challenge—and the judges assessed and rated the submissions before a collective review. Doumet said that SpectraCan’s goal is employing their technology for “detecting fires of sizes 10m x 10m within 1 minute,” though they’re ideally looking to be able to detect fires of as little as 1 meter squared within that time.
Now that they’ve created the technical plan, Doumet said the next step for the semi-finalists is to apply their solutions to a shared dataset that they will all be receiving. “Those with the highest detection accuracies” on the dataset, he said, “will proceed to the final.”
The First Place team in the track will receive $3.5 million USD, and a total of $1.5 million USD will go towards the milestone prizes. There is also a $1 million USD bonus prize for teams whose entries “successfully demonstrate Accurate, Precise, and Rapid Detection.” Other semi-finalists in the Space-based detection track include teams from Loft Orbital, Muon Space, MyRadar, Orbital Sidekick, OroraTech, Guardian Space, and Akula Tech.