Descartes Labs has launched GeoVisual Search. GeoVisual Search allows anyone to search the public satellite information of Earth for visually similar items and have them all highlighted on the map for people to check out. For example, when searching for airport runways, GeoVisual Search will highlight all of the airport runways in the United States on the map, where I can quickly go and checkout all of the airport runways that the system found. GeoVisual Search differs from Google Maps because the AI powering the system actually recognizes objects and highlights the matching ones (such as an airport runway) versus Google Maps which simply places marker on a location with the information (such as JFK Airport) which has to be previously provided.
GeoVisual Search is powered by a deep learning AI that crunches millions of images of the Earth into overlapping tiles (128-pixels per side) and then extract that information into readily accessible features (such as texture, shapes, and color). When a visual search is performed (for example on a windmill), the AI compares the features of the initial search image versus all of the extracted features stored in the database. GeoVisual Search then returns up to 1000 images that are visually similar (based on the extracted features) to the source image.
Descartes Labs offers 3 forms of resolution for GeoVisual Search based on the available data: the continental United States at a resolution of about 3.3 feet per pixel, China with a resolution of about ~16 feet per pixel and the rest of the world with a resolution of 65 feet per pixel. The program is currently just a demo but you can you can test the system yourself at the link below.