Shulou(Shulou.com)11/24 Report--

CTOnews.com, September 3 (Xinhua)-- the Allen Institute of artificial Intelligence (Allen Institute for AI), founded by Microsoft co-founder Paul Allen, has released a new tool called Satlas, which includes the world's first map to use generative artificial intelligence technology to improve satellite image resolution, showing renewable energy projects and forest cover around the world.

CTOnews.com noted that the map used satellite images from the European Space Agency (European Space Agency) Sentinel-2 (Sentinel-2) satellite. However, these images still do not clearly show the details of the ground, so they used a solution called "Super-Resolution". Basically, a deep learning model is used to fill in details, such as what a building might look like, to produce a high-resolution image.

The above image is a high-resolution image of Nakuru, Kenya, generated by artificial intelligence, and the following is a low-resolution image taken from the same location by satellite. Currently, Satlas focuses on global renewable energy projects and forest cover. The data are updated monthly, including parts of the earth monitored by Sentinel-2, which includes most areas except Antarctica and the high seas far from land.

The map shows solar farms and onshore and offshore wind turbines, and can be used to see how canopy coverage changes over time, which is important for policy makers trying to achieve climate and other environmental goals.

According to the Allen Institute, this is the first tool with such wide coverage and free access to the public, and its developers say it may be the first time that super-resolution technology has been demonstrated on a global map.

Of course, there are still some problems that need to be solved. Like other generative artificial intelligence models, Satlas is prone to "hallucinations", sometimes drawing buildings in a strange way, such as a rectangular building, and the model may think it is a trapezoid or something, which may be due to differences in architectural styles in different regions, making the model difficult to predict. Another common "illusion" is to place cars and ships where the model believes there should be cars and ships, based on the images used by the training model.

To develop the Satlas, the Allen team had to manually browse satellite images to mark 36000 wind turbines, 7,000 offshore platforms, 4,000 solar farms and 3,000 tree canopies. For super-resolution, they input many low-resolution images taken in the same place at different times into the model. The model uses these images to predict sub-pixel details in high-resolution images.

The Allen Institute also plans to expand Satlas to provide other types of maps, including one that identifies the types of crops grown around the world.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.