The Allen Institute's Satlas interactive map uses AI to create high resolution images of the world, even when only low resolution satellite images are available. The Allen Institute has also trained the AI to identify the location of wind turbines, solar farms and tree canopy coverage around the globe.
Satlas uses satellite imagery from the European Space Agency’s Sentinel-2 satellites. The Allen Institute manually scanned this ESA imagery to identify 36,000 wind turbines, 4,000 solar farms, and 3,000 tree cover canopy percentages. It then trained an AI on this data in order for it to be able to globally identify the locations of other wind turbines, solar farms and tree cover around the world.
If you select the 'super resolution' option on Satlas you can explore the world with satellite imagery which has been increased four times in resolution by AI from the original ESA Sentinel-2 captured imagery. Like many other products 'enhanced' by AI Satlas does have a tendency to 'hallucinate'. For example Satlas has replaced the 5-aside football pitches in my local park with some neat rows of trees (presumably it mistook the green AstroTurf for canopy cover).
Despite these 'hallucinations' Satlas claims its data has a 'high accuracy'. Satlas has actually sampled and validated the wind turbine, solar farm and tree cover data for each continent to estimate this data's accuracy. You can view the estimated precision and recall of the data for each product on each continent on Satlas's Data Validation Report.
The Allen Institute is just one of a growing number of geospatial companies that are using AI to identify and classify objects in satellite imagery. Some of the other companies working in this area are Descartes Labs, Radiant Earth and Orbital Insight.
No comments:
Post a Comment