The aerospace market was closed off to private companies for many years: only sovereign countries were able to bear the huge expenditure of taking off and rising above the earth's surface. During the Cold War years, the main advances in the 'space race' were achieved by the Russian and US space agencies and this state of affairs lasted until well into the 2000s. Since then however, things have changed a lot, as demonstrated by the spectacular progress made by Elon Musk's space technology company, SpaceX, which often steals the limelight from NASA nowadays. And while SpaceX is perhaps the most talked-about private company in this sector, it is far from being the only one to make a business out of reaching for the stars. An increasingly well-equipped constellation of startups is hot on its heels and attracting investor interest.
will be invested in the satellite imagery business by 2026
In 2015, US venture capital funds invested twice as much money in the space sector as during the preceding 15 years put together. And last year, two young startups – Orbital Insights and Descartes Labs – raised $50 million and $30 million respectively. There are in fact several distinct space markets. While companies such as Virgin Galactic and Blue Origin specialize in futuristic tourism, others, notably Planetary Resources and Deep Space Industries, are focusing on mining valuable resources on asteroids, Facebook is looking to expand public access to the Internet and SpaceX is on a mission to colonize other planets.
Without doubt however, one of the most promising areas is satellite imagery. This market is currently seeing a spectacular boom, driven by a number of startups, including Planet and DigitalGlobe, which are launching into orbit small-sized devices able to capture more precise and more frequent images of the entire globe. In 2017 the global satellite imagery market attained a total value of $3 billion, and is expected to continue growing at 12.5% per year over the next few years, reaching an annual $8.6 billion by 2026, according to Statistics MRC, a US consultancy and research firm.
satellite imagery: a promising market
Using AI to separate the wheat from the chaff
However, these images still need analyzing in order to extract the essence of the information and identify the most relevant data so as to develop practical applications. Satellites generate huge quantities of images which are simply impossible to process by hand. This is where Artificial Intelligence (AI) comes into play. This invention from the world of computing which goes back all the way to the 1950s, has made a spectacular comeback in recent years with highly-publicized performances such as IBM Watson's exploits in winning the television quiz show Jeopardy! and the more recent victories by AlphaGo in the board game Go over top players Lee Sedol and Ke Jie. The most recent advances in Artificial Intelligence are mainly due to machine learning, a branch of the discipline which enables computers to learn for themselves by training up on huge volumes of data.
AI TEAMING UP WITH SATELLITES
This technique has enabled spectacular progress in image recognition. When you show a computer millions of images of different cats, it ends up being able to identify a cat. The technique is also behind the advances made in self-driving cars and also Facebook's algorithms that can recognize an individual person in a photo. Such image recognition algorithms have arrived at just the right moment in the aerospace field as they provide an excellent means of processing satellite photos. "Five years ago we couldn't have done what we're able to do today", Pavel Machalek, co-founder and CEO SpaceKnow, a company specializing in satellite data analysis for commercial applications, told Fortune magazine. He believes that the combination of huge amounts of computing power, machine learning and satellite imagery is now opening up new horizons.
Like SpaceKnow, there are a number of startups that are building their business models around extracting data from satellite images. This is precisely what Descartes Labs, Cape Analytics and Orbital Insights are doing. OrbitalInsights has recently signed a partnership with US commercial vendor of space imagery and geospatial content DigitalGlobe, a company which has spent the last 17 years taking high-definition photos of the earth's surface. Orbital Insights now intends to extract data from the images for use in practical applications.
Mapping the digital divide...
kilometers of roads to map
So what sort of applications are we talking about? The first major set is all about listing static objects. Researchers at MIT have recently harnessed AI to map roads from satellite images. At the moment, this work is partly done by hand, which takes a lot of time. Even Google, with its impressive financial resources, is still far from completing all the 30 million kilometers of roads which criss-cross planet Earth. In the era of self-driving vehicles, having a highly accurate mapping system has become a key asset. This is what prompted researchers at MIT to develop RoadTracer, a software package that uses neural networks – algorithms loosely based on the way the human brain works – in order to fully automate the road identification process, which is usually based on a combination of machine learning and human correction. Not content with being more efficient, the software is also 45% more accurate than the currently-available techniques.
Meanwhile Facebook is combining AI and satellite imaging in order to achieve its goal of bringing Internet connection to 4.2 billion people who today lack one. We know that these people exist; but we do not know exactly where to find them. Using image recognition algorithms similar to those deployed on social networks, Facebook researchers have analyzed 14 billion images supplied by DigitalGlobe in an effort to find out where the individual victims of the digital divide are. Two billion people in twenty countries have been identified.
... and poverty
Some other initiatives have set out to investigate poverty-stricken areas, so as to help charitable organizations focus their spending better, and to enable governments to get a better view of the situation in order to adjust their policies. There are already several methods of mapping deprived areas, but they all have some gaps. One option is to poll the inhabitants directly. However, this technique is very costly in terms of both time and resources, with the result that surveys are often only carried out once every ten years. Another option is to base conclusions on night-time lighting: the more a given area is lit at night, the richer and more developed it tends to be. This method is not however very precise and researchers are now starting to use new tools which take more variables into account.
satellite photo taken by digitalglobe
Penny is one such tool. Developed by researchers at the Carnegie MellonUniversity in collaboration with DigitalGlobe. Penny uses neural networks, trained on huge quantities of images, aggregated with data on how corresponding revenues are distributed. Researchers have supplied Penny with images from the cities of New York and St Louis. Having trained itself, the software has been able to demonstrate that it is capable of automatically recognizing the better-off and less well-off geographical areas, basing its judgements solely on the architecture in the area. Accordingly, the poorest urban spaces have more car parks, more basketball courts and more buildings of similar height and shape, while the better-off areas boast more green spaces, larger buildings and houses with gardens.
A similar initiative has been undertaken by researchers at Stanford University. They combined satellite images and machine learning algorithms fed by three different data streams – night-time lighting, photos taken during the day, and data on poverty drawn from surveys – in order to precisely identify the most disadvantaged zones in five Africa countries. Cheap and easy to roll out on a large scale, this technique enables efficient mapping of poverty worldwide. In another project, Orbital Insights, in partnership with the World Bank, trained an AI program to recognize poverty and opulence indicators in the region, based on such factors as the shape, size and number of buildings, the number of cars circulating, and the existence of farms. Following a promising initial test period in Sri Lanka, the duo is now testing its system in Mexico.
Algorithms underpinning sustainable development
Last but not least, Google is deploying its AI expertise to analyse images so as to boost the renewable energy sector. The Sunroof project, launched in 2015, enables the number of houses with solar panels in a given geographical area to be worked out. By combining this data with weather information, Google's algorithm is also able to identify areas that enjoy lots of sunshine and list how much the households in these zones could save in terms of both energy and hard cash by using solar panels. The Californian city of San Jose has been using the tool to decide the best locations for solar panels.
So much for analyzing static phenomena. There are however other applications which focus on studying change on the earth’s surface. This typeof analysis is very useful for precision agriculture. Meanwhile, Brazilian firm Agronow has developed a platform for farmers which combines image analysis and weather data so as to provide them with a means of tracking precisely how their crops are growing, deciding the ideal moment to harvest and assessing the quality of the crops harvested. By incorporating market data, the platform can even give a price which the farmer could expect to obtain for his/her produce. Descartes Labs, launched in 2014, has made its reputation in satellite imaging analysis by monitoring corn (maize) fields in the UnitedStates. The company has been able to accurately assess crop content and quality and calculate potential yields. At the same time, Microsoft is using satellite imaging to help farmers in India. Its algorithms, which analyse weather conditions and assess crop status, are able to predict the risk of an insect or parasite attack so that farmers can be sent an automated warning call. The system also figures out the ideal moment to harvest and will work out a reasonable sales price depending on current market rates.