Four Cool Artificial Intelligence Technologies
Lockheed Martin is Developing to Find Wildfires, Ice Cream Shops, Planes, Ships, Blue Cars and More
Artificial intelligence isn’t about robots taking control – it’s about the efficient and effective use of vast quantities of available data to solve problems in a timely manner.
The truth is, in today’s society, the amount of geospatial intelligence (GEO INT) data that is collected every day around the world – from cameras, sensors and satellites – is staggering. It’s literally more than what humans can physically sort through, process and try to understand – let alone do that when quick decisions are required.
To solve this massive data crunching problem, Lockheed Martin is utilizing digital transformation tools like artificial intelligence and machine learning (AI/ML) every day. By using AI, we can ensure our customers always have the greatest situational awareness and actionable intelligence to perform critical missions to protect our nation and our allies.
Here’s a look at four cool AI/ML technologies Lockheed Martin is developing to take advantage of the world’s growing amount of GEO INT data.
Finding and Fighting Fires Faster
Dry summer months and global warming aren’t doing any favors for our brave firefighters. According to the National Interagency Fire Center, wildfires have burned 2,990,255 acres this year (June 17).
Adding to firefighters’ challenges, using current resources, it often takes hours to map a growing wildfire’s perimeter and heat spots. It sometimes takes days using fuel property data – often 3-5 years old – to help predict fire behavior. Time is not on their side and the situation on the ground is always changing.
Lockheed Martin is using AI/ML to help get critical data to firefighters faster. The company is developing the Cognitive Mission Manager, a command and control suite of services designed to use AI/ML to deliver timely, accurate, and actionable intelligence to the ground commanders and operators to enable rapid response and fire suppression. The system uses local sensor data and current commercially-available space-based imagery to generate simulations that provide a more accurate and current picture of a wildfire situation and to predict wildfire growth.
“AI allows us to reduce data gather cycle times and to deliver more accurate and timely data in an actionable way for firefighters. Using AI, we can provide precise fire mapping in minutes vs. hours and predictions of the rate of a spread and the direction of a fire within hours vs. days,” said Dan Lordan, Senior Manager at Lockheed Martin’s AI Center.
Help Finding Your Tank, Car ... and Ice Cream?
It doesn’t matter if you’re looking for tanks on a battlefield, blue cars or all the ice cream shops in Philadelphia, finding things in today’s growing data volume and diversely-sourced GEO INT data environment is a challenge. Especially if you need that data to make critical decisions faster.
Lockheed Martin’s Cognitive Tip & Cue product addresses the shortcomings of manually intensive data collection management and orchestration by using AI/ML algorithms to solve this problem. It starts with feeding the system a simple needs statement in English.
“We can take a sentence like, ‘I need to know where all the ice cream shops are in Philadelphia,’ and the system’s cognitive components break that statement down into target and strategy information for searching through the data from all our data sources,” said Jacob Carroll, Senior AI/ML Research Engineer at Lockheed Martin. “Using AI/ML techniques, Cog Tip and Cue can run autonomous at machine speed and produce actionable intelligence in seconds.
And because it is a machine learning model, the system can learn and evolve.
"We can modify the parameters or modularities of every search to produce new information and when the system encounters new situations, it can adapt to the changes much faster than a human could,” Carroll added.
Detecting Changing Data Faster
GEO INT intelligence data users are often interested in detecting changes over areas they are monitoring. They might want to know if a ship left or arrived into harbor traffic. Are buildings expanding in an urban sprawl? Have roads or infrastructure been damaged after a natural disaster?
Lockheed Martin has developed AI/ML tools that provide Object Level and Pixel Level Change Detection using time series satellite imagery to deliver automated, rapid change detection. Trained on real data, the algorithms running inside the tool allow for evaluation and inference for end users.
“In the case of urban development, deep learning techniques can be used to show how satellite imagery over time can be monitored for rapidly urbanizing areas,” explained Dom LeDuc, Image Processing Engineer at Lockheed Martin. “Post-processing algorithms can provide object level change detection specifying the location and time new buildings were constructed.”
With enough pixel-level data inputs for labeling, the AI can actually be taught to identify and recognize objects – like types of aircraft and ships or damage to infrastructure after storms – in areas you’re monitoring.
“Pixel level data can be ingested by our AI algorithms to provide aircraft dimensions or propulsion type, as well as discerning types of aircraft, ships and buildings,” added Dagen Braun, AI/ML Engineer at Lockheed Martin. “This allows a field analyst to quickly gain a better understanding of the situation and confidently suggest a course of action."
Teaching AI to do ... AI
The truth is AI doesn’t know what it’s looking for until someone – a human – teaches it what to search for. But teaching AI can be a tedious job. And teaching AI takes people.
These important people spend countless hours “labeling” real data that is fed to an AI system until it “recognizes” a search item on its own. For instance, if you wanted AI to identify a “brown car” every time it is appears in satellite imagery, a labeler might need to introduce the AI to thousands of images of “brown cars.” As the hours go by, the risk is the labeler could become less engaged and more prone to mistakes or burnout.
To solve this problem, Lockheed Martin created TruthTrail, a prototype gamification application built off a gaming engine which presents a simulated environment that “rewards” labelers for their work. Similar to various step count health apps used by millions every day, labelers reach different “achievements” as they introduce more images to AI systems.
“We’ve seen other industries use gamification to continue to motivate people to ‘get to the next level.’ With TruthTrail we’re applying similar tools to help keep our labeling employees more engaged,” explained Ainoghena “AI” Igetei, Software Engineer at Lockheed Martin.
On the backend, TruthTrail intends to capture and leverage data on how the labelers react and solve problems. This allows for designing, discovery, and testing new ideas and architectures in support of meeting future customer needs, in particular for artificial intelligence.