Lockheed Martin has developed a satellite imagery recognition system using open-source deep learning libraries for the identification and classification of objects or targets in large areas.

Manual categorisation and labelling of items within a satellite image is a time-consuming process. However, the new deep learning model speeds up and automates satellite image analysis, saving substantial time.

Running in the Cloud, the Global Automated Target Recognition (GATR) artificial intelligence model uses Maxar’s Geospatial Big Data platform (GBDX) to access a 100 petabyte (PB) satellite imagery library and curated data labels that speed up the training of deep learning algorithms.

GATR scans a large area quickly with the assistance of fast graphics processing units (GPUs). Deep learning methods reduce the need for extensive algorithm training by automating object recognition.

The AI-based GATR teaches itself the identifying characteristics of an object area or target. For instance, it will be able to distinguish between a cargo plane and a military transport jet.

The satellite image analysis system can also scan large areas such as countries. It uses deep learning techniques found in the commercial sector to identify aircraft, ships, buildings, seaports and several other categories.

Lockheed Martin space mission solutions vice-president and general manager Maria Demaree said: “There’s more commercial satellite data than ever available today, and up until now, identifying objects has been a largely manual process.

How much of an impact will the COVID-19 outbreak have on the revenue of aerospace companies?

View Results

Loading ... Loading ...

“Artificial intelligence models like GATR keep analysts in control while letting them focus on higher-level tasks.”

The company claimed that GATR has an accuracy rate of more than 90% on the models that have been tested. It said that the model searched the 120,000km² state of Pennsylvania for fracking sites in two hours.

“The AI-based GATR teaches itself the identifying characteristics of an object area or target.”

Lockheed Martin senior fellow and GATR principle investigator Mark Pritt said: “I’m not an expert on what oil production sites are, and I don’t have to be.

“This system teaches itself the defining characteristics of an object, saving valuable time training an algorithm and ultimately letting an image analyst focus more on their mission.”

GATR is a result of the research conducted by Pritt’s team during the ‘Functional Map of the World’, an Intelligence Advanced Research Projects Activity (IARPA) challenge.