Chinese Peril

AI CAN HELP HUNT DOWN MISSILE SITES IN CHINA
By Jeremy Hsu

A surface-to-air missile is seen through a doorway in Zhuhai, China.

INTELLIGENCE AGENCIES HAVE a limited number of trained human analysts looking for undeclared nuclear facilities, or secret military sites, hidden among terabytes of satellite images. 

But the same sort of deep learning artificial intelligence that enables Google and Facebook to automatically filter images of human faces and cats could also prove invaluable in the world of spy versus spy. 

An early example: US researchers have trained deep learning algorithms to identify Chinese surface-to-air missile sites—hundreds of times faster than their human counterparts.
The deep learning algorithms proved capable of helping people with no prior imagery analysis experience find surface-to-air missile sites scattered across nearly 90,000 square kilometers of southeastern China. 

Such AI based on neural networks—layers of artificial neuron capable of filtering and learning from huge amounts of data—matched the overall 90 percent accuracy of expert human imagery analysts in locating the missile sites. 

Perhaps even more impressively, the deep learning software helped humans reduce the time needed to eyeball potential missile sites from 60 hours to just 42 minutes.
"The algorithms were used to find the locations where they said there is a high confidence of a missile site, and then humans reviewed the results for accuracy and figured out how much time the algorithms saved," says Curt Davis, a professor of electrical engineering and computer science, and director of the Center for Geospatial Intelligence, at the University of Missouri. 

"To my knowledge that’s never been studied before: How much time did you save, and how does that ultimately impact the human performance?"
The University of Missouri study, published on October 6 in the Journal of Applied Remote Sensing, comes at a time when satellite imagery analysts are figuratively drowning in a deluge of big data. DigitalGlobe, a leading commercial satellite imagery company, generates about 70 terabytes of raw satellite imagery each day, never mind all the imagery data coming from other commercial satellites and government spy satellites.
Davis and his colleagues showed how off-the-shelf deep learning models—heavily trained and modified for satellite imagery analysis—could identify objects of potentially great interest to intelligence agencies and national security experts. 

The deep learning models, including GoogleNet and Microsoft Research's ResNet, were initially created to detect and classify objects in traditional photo and video imagery. 

Davis and his colleagues adapted such models to the challenges and limitations of interpreting satellite imagery, such as training some deep learning models to interpret both color and black-and-white imagery, in case only black-and-white images of SAM sites were available.
They did so with satellite imagery representing a huge swath of Chinese territory, not that much smaller than the entire country of North Korea.
And in fact, analysts rely extensively on satellite imagery to keep track of how North Korea's weapons programs evolve. 

Human analysts have already likely identified most, if not all, existing SAM sites within the relatively small country. 

But similar deep learning tools could help automatically flag new SAM sites that appear in North Korea or other countries. 

Knowing the location of existing and new SAM sites can sometimes lead analysts to other locations of interest, because countries often place SAM sites in specific areas to defend valuable nearby assets from air attack.
The latest study also illustrates the challenges of applying deep learning AI to satellite imagery analysis. 

One major problem is the relative lack of large training datasets that include the hand-labeled examples needed to train deep learning algorithms to accurately identify features in satellite imagery. The University of Missouri team combined public data on the worldwide locations of about 2,200 SAM sites with DigitalGlobe satellite imagery to create their training data, and then tested four deep learning models to find the best-performing one.
The researchers ended up with only about 90 positively identified Chinese SAM site examples to train their AI. 

Such a puny training dataset might normally fail to yield accurate deep learning results. 

To get around that problem, Davis and his colleagues transformed the 90-odd training samples into about 893,000 training samples by shifting the original images slightly in different directions.
The impressive deep learning performance in the study likely benefited from SAM sites being fairly large, and having distinctive patterns when viewed from above in satellite images. 

Davis cautioned that deep learning algorithms face a much greater challenge when trying to analyze smaller objects such as mobile missile launchers, radar antennas, mobile radar systems, and military vehicles, because the available satellite imagery data will have fewer pixels to work with in extracting identifying features.
"It is an open question in our mind how well convolutional neural networks will work on smaller scale objects like this, especially when tested against large area datasets like we did with the China study," Davis says.
Even imperfect AI tools could prove incredibly helpful for intelligence gathering. 

For example, the International Atomic Energy Agency has the unenviable task of monitoring all declared nuclear facilities and also searching for undeclared facilities among nearly 200 countries. 

Deep learning tools could help the IAEA and other independent organizations use satellite imagery to monitor development of nuclear power and related weapons of mass destruction, says Melissa Hanham, a senior research associate in the East Asia Nonproliferation Program at the Middlebury Institute of International Studies at Monterey, Calif.
"We're in a world where there is just so much data that the best way to approach it is to do a good job on a lot of it rather than a perfect job on a small bit of it," Hanham says. 

"I'm looking forward to automating all the tedious and redundant parts of my job."

You Can Strike Oil In China: Four Reasons The Satellite Market Is Taking Off

By Michael Kanellos

China has more oil than people thought.
Orbital Insight, a Palo Alto-based startup hoping to capitalize on the growing interest in mining satellite imagery, undertook an interesting assignment: it captured, and then analyzed, satellite images regarding the world’s 20,000+ oil storage depots to try to determine the real-time supply of oil.
How? 

It focused on the shadows being cast on the inside of the world’s 20,000+ oil storage tanks. 

Oil storage tanks have floating ceilings: a short shadow indicated a well-stocked silo while a long one indicated potential shortages. 

And while doing their shadow analysis, Orbital discovered a funny thing. 

China had 2,000 more coastal tanks than people thought, said Orbital’s Shwetank Kumar at the Center for Effective Global Action in Berkeley, California this week.
The hunt may not be over either: South American nations regularly do not report their oil inventories either, he added.
So what do you need to know?

It’s a Huge Opportunity
The global satellite market is growing at 19.54%, according to some estimates. 

Venture capitalists invested $1.8 billion in space startups in 2015.
But more importantly, a growing raft of companies want to couple their data with imagery. 

I attended the conference to speak about the epidemic of water leakage—over 30% of the water in many areas of the world drips away before it gets to your tap. 

Satellite imagery can be matched with pressure data to pinpoint leaks. 

Farming and health care companies are mining data. 

Even stock traders are getting into the act.
The world takes an estimated 1 trillion images a year: it’s only going to expand.
Hello from Beijing: an image from Orbital insight

It’s Multifaceted
Like other digital imaging markets, there are a number of entry points and companies are carving out areas of specialization. 

Orbital’s stock-in-trade, for instance, revolves around blending satellite data with other streams like commodity prices or poverty index. 

The satellite images aren’t the end goal: they become a source of data for macroeconomic analysis.
Planet, meanwhile, is putting up an armada of Dove CubeSats – or very small, but capable —satellites for capturing images of life on earth. (I wrote about CubeSats back in 2005 when it was a research project so it’s gratifying to see it come to fruition.) 

Planet’s goal is to use “space to help life on earth,” said Planet’s Tara O’Shea. 

The satellites capture images at three to five meters of earth space per pixel. 

That’s fine enough to capture the progress of road building in India or the Amazon and far greater than the 30 meters per pixel resolution of Landsat. 

At the same time, it’s not deep enough to pinpoint faces, license plates or personal information. 

In Palawan, an island in the Philippines, the company is helping map the 50% of roads that wouldn’t ordinarily show up on maps.
To avoid the cost creep that can impact companies developing hardware, Planet tries to develop and launch quickly. 

It has developed 13 different versions of its satellite in 3 years and a new vehicle gets launched every three to four months. 

By contrast, it typically takes three years to launch a conventional satellite.
A Dove Cubesat from Planet.com

It has 70 satellites in orbit today and should have 100 by the end of the year. 

That will give the company the capability to take a new, comprehensive image of the earth every day.
By contrast, you have DigitalGlobe, a publicly-held company that recently bought the Radiant Group for $140 million. 

It specializes in super-high resolution photos. 

A traditional Landsat image might take up half a gigabyte, said Shay Har-Noy. 

An image from Planet might come to 16GB. 

One of DigitalGlobe’s images can weigh in at a hefty 941GB: more data, more insight. 

The company takes on projects like mapping Australia to help map out water and agricultural issues.

It’s Not Just About Hardware
Software is where you will see most of the startup activity, says Ruchit G Garg, a Microsoft alum who has founded Harvesting, which uses satellite imagery to optimize harvests. 

The data will help farmers, but mostly it’s for banks and insurance companies. 

Ideally, better data will lower the barriers to credit, he says. (SlantRange is taking a similar tack with drones.).
Likewise, Facebook is using satellite data as part of its plan to bring Internet access to emerging nations and rural communities. 

1.6 billion people worldwide live outside the reach of mobile networks, said Andreas Gros of the Facebook Connectivity Lab. 

99% of the world, however, lives within 80 kilometers of a city with a population of 10,000 or more. 

The idea is to leverage that as much as possible.
Combining imagery with census data is giving the company greater insight into population density and settlement patterns. 

So far, Facebook has scanned 27 million square kilometers and amassed 500TB of data. 

It can analyze some countries for density in population density eight hours.

It’s Not Just About Space
If there’s one meme that kept getting past around at the event, it was “ground truth,” i.e., information from people on the ground that confirms or expands satellite data.
Prabal Dutta at the University of Michigan is overseeing a project to pinpoint grid outages by using cell phones as sensors. 

73.8% of Kenyans, after all, have cell phones.
Traffic? 

70% of the traffic in Kenya comes from Matatus, or drivers for hire, said UC Berkeley’s David Schonholzer. 

Three million people use this precursor of Uber on a daily basis. 

His department is integrating sensors to monitor drivers—how fast they accelerate, whether the go off-road, how many sudden stops or sharp turns they take—to improve road safety. 

If you tried to analyze grid or road traffic with imagery alone, you’d only get a partial answer.