Crowdsourcing applied to satellite imagery enables governments and commercial organizations to leverage the power of massive human networks to reveal insights rapidly at a global scale. Crowdsourcing campaigns engage hundreds to thousands of people with specific assignments to discover or validate important visible features and locations within targeted areas of satellite imagery. Crowdsourcing makes it possible for humans to analyze hundreds of thousands of square kilometers of imagery in a single day, a task that might take a single imagery analyst weeks or months to complete.
GeoHIVE (Geospatial Human Imagery Verification Effort) is Radiant Solutions’ satellite imagery crowdsourcing initiative, built on DigitalGlobe’s Tomnod crowdsourcing platform. We combine the power of the crowd with DigitalGlobe’s high-resolution imagery to create accurate and insightful geospatial data. Radiant Solutions and DigitalGlobe are part of a family of industry-leading space and technologies companies under Maxar Technologies, which also includes SSL and MDA.
Recently, Radiant Solutions completed our 500th GeoHIVE campaign since we launched in 2015. The majority of our campaigns have supported work for the U.S. government, commercial clients, development of machine learning training sets, and production of our country-level human-geography datasets we call Human Landscape. Our crowds have enriched more than 1.5 million square kilometers of DigitalGlobe imagery by tagging 2.2 million geographic features.
The GeoHIVE team is committed to three major innovation goals in 2018:
- Extend the Tomnod platform’s capabilities to include more complex feature discovery and validation campaigns
- Enable emerging technologies like machine learning to automate these extended capabilities
- Collaborate and synchronize with our partners to accelerate innovation in the Tomnod platform
Here are a few examples of how we are achieving these goals, and why our next 500 campaigns will drive even greater efficiency and new applications.
Extend the capabilities of the GeoHIVE platform
To fulfill our first innovation goal, we extended the Tomnod platform’s capabilities and enhanced the GeoHIVE user experience to be able to issue directives to identify complex features. For example, we can now ask the crowd to “find all tire tracks that cross an administrative border.” We used this capability recently for a customer campaign to identify cross-border smuggling and human trafficking activity in Algeria. We asked the crowd to tag all roads and tire tracks that crossed the Algerian border into Tunisia, Libya and Mali. In less than three hours, the crowd was able to locate 2,315 tire tracks and five paved roads across 2,236 square kilometers of search space.
We can also ask the crowd to “find all of the new buildings” by marking existing building footprints into the imagery and asking the crowd to tag all buildings that aren’t masked.
The GeoHIVE team is excited to explore new ways of extending our capabilities by asking the crowd more detailed questions. The more we ask, the more we can answer, and the more we learn.
Enable emerging technologies like machine learning
Crowdsourcing began as a way to distribute work over large networks of people to scale manual tasks. It’s evolved to an even more powerful tool through its ability to automate feature detection with the addition of machine learning. We are leveraging the GeoHIVE crowd to produce training data to fine-tune machine learning algorithms for our customers. For example, we have developed a new crowdsourcing capability called feature bounding box annotation that enables us to task the crowd to draw precise polygons around features of interest within imagery. The resulting campaigns produce feature classification training datasets that can be used to train machine learning algorithms to automatically detect features within satellite imagery. We used this capability recently for a customer. We asked the crowd to look for objects such as ships, aircraft, construction equipment, cars, trucks and buildings. From this, we produce an authoritative training dataset inclusive of more than 1 million labeled features across 60 different feature classes of relevance to humanitarian aid and disaster response. Through crowdsourcing, we are able to produce the human-in-the-loop-accurate training data required to improve feature detection algorithms at a speed and scale that was previously impossible. GeoHIVE is committed to driving innovations in machine learning to help our customers reveal insights within imagery at greater scale and efficiency.
Collaborate and synchronize with partners
The GeoHIVE team regularly collaborates across technical teams within Radiant Solutions to drive innovation. For example, we recently partnered with Radiant Solutions’ Data to Insight line of business to develop image chipping workflows that significantly reduce our setup time for crowdsourcing campaigns. Because of this collaborative effort, image processing in preparation for campaigns is now completed in minutes, not hours. We also worked internally with the Radiant Solutions’ SkyNet machine learning platform team on a customer project to validate open source global datasets for airfield-specific features. We asked the crowd to validate 6,456 airfield locations produced from the platform team’s automated feature detection algorithm. In under two hours, the crowd validated 5,154 of them (79.8 percent). The project was a huge success because both our machine learning platform team and GeoHIVE proved complementary in rapidly validating points of interest. This example illustrates the benefit of using crowdsourcing to quickly validate machine learning results on a global scale.
By partnering and collaborating with other technical teams across Radiant Solutions, GeoHIVE is driving time-savings and efficiency into our campaign workflow and innovation into geospatial data creation and validation. GeoHIVE is open to partner collaborations that will help us better support our customers, accomplish our innovation goals, and complete the next 500 GeoHIVE campaigns that will reveal insights when and where it matters.
To see more and learn how to join the GeoHIVE crowd, please visit https://geohive.digitalglobe.com and click on “Join the HIVE.”