To Boldly Go Where No Data Has Gone Before

As a child I would often gaze up at the stars and wonder what lied behind the darkness of space. Today we know a lot more about the solar system thanks to satellite technology. And while space exploration has traditionally been controlled and owned by government agencies, the advent of cheaper and more advanced technology has spurned an invest boom in space exploration by the private sector. Satellites that cost millions of dollars to build and launch are on par with the cost of most luxury sports cars. With the increased investment in space as a business, there’s suddenly an enormous amount of data – and opportunity – for others to learn and profit from what we know from space.

Viewing Earth observation (EO) data in combination with geo-spatial data from mobile or industry-specific data offers a potential to provide unique augmentation data that can be very useful in predictive and prescriptive analytics. Google Maps, for example has successfully used satellite imagery to provide navigation, traffic density alerts and much more intelligence from up high. Satellite imagery from Earth observation has some very interesting potential use cases when combined with artificial intelligence and machine learning.

·       Agriculture – Satellite imagery is being used for crop management, irrigation management, pest control and better resource utilization.

·       Vegetation index– Governments utilize satellite imagery for vegetation management.

·       Vehicle density – Satellite imagery is used for vehicle density measurement and change detection to manage traffic and enable smoother vehicular movement management.

·       Disaster management – Natural disaster risk management is closely connected with satellite imagery. Post-disaster relief management is supported by effective utilization of satellite imagery.

·       Autonomous vehicles – Satellite imagery is used for autonomous vehicle path planning and then combined with mobile geo-spatial data to enable personalized experiences for the rider. (e.g. – servicing a request that a rider left a laptop in the car and bringing it back to the rider.)

For the effective application of EO + ML, a crucial need is to create learning data models, and this requires satellite imagery annotation.

Annotating Images from Space

Annotation, also referred to as data labeling, is the process of tagging everything from text to images to classify what’s in a document or asset to create more accurate training data for machine learning models. There are two distinct segments for application of ML to satellite imagery.

One-level application where ML is applied straight to satellite imagery. In this case, “object detection” is one of the most used data points for identifying buildings, airplanes, streets, intersections, rivers, parking areas, etc. The other use case is “change detection”, such as coastal line shifts, addition of new buildings over time, roads, homes, etc. These provide useful datasets that can be used to understand changes in population and thus used in multiple ways to predict demand changes or traffic management, and so on.

Next, there’s Multi-level application, where ML is applied by extracting information from satellite imagery and augmented with non-satellite imagery data to build learning models. In multi-level application satellite imagery is one of the inputs for the ML for prediction. For example, counting the number of cars in the parking lots of a large retail chain is a good indicator of the number of customers they attract.  If this can be done across large demographics, this data can be used to predict seasonal changes in footfalls in retail and predict sales changes. Likewise, crop satellite imagery for high value crops can be augmented with other local weather data to predictive supply patterns that can be used to predict the price changes for commodities.

All these offer great opportunities by using satellite imagery data with M\machine learning for predictions. It’s hard to count the number of boats in the harbor from land, but it is quite the opposite when doing so from space with the best camera on (and out of?) the planet.

A recent example of this was the excellent visuals provided by Refinitiv during Saudi Arabia drone oil attack that caused the biggest global oil disruption. The ship tracking data plotted on satellite imagery made it possible for economists and traders to understand and visualize the impact it will have on supply and demand. The day is not far off when satellite imagery and AI can help thwart such attacks and become a key strategic differentiator in how countries handle global risks.  The key will be understanding and accurately labeling all the info that’s coming from them.

Check out how we’re helping companies annotate data from satellite images and other image sources to meet their business needs.