The learnings of combining AI with aerial and satellite images

This is a translation of the original article published in Dutch on May 19, 2020 by the Data Science Initiative and the NL AI Coalition.

One of the most important requirements for AI to excel is the correct data. Image based datasets are very suitable, because AI systems are good at recognizing objects in images. Thousands of images from the air and space are collected on a daily basis in the Netherlands, with enormous potential to be used for AI purposes.

“Most people have no idea what is possible with all these aerial and satellite photos when combined with AI” says Camiel R. Verschoor of Birds.ai. In an effort to raise awareness about the application of these kinds of images, he wrote a guide together with Luc van den Ende, Joost Dorscheidt (Geronimo.AI) and Nout van Deijck (NSO). "We hope that this will teach people what is available and to encourage ideas which make use of these applications in their (governmental) organization", Verschoor says.

Below are the most important learnings about the combination of AI and aerial images.

Lesson 1: This is also interesting if you don’t have a business case (yet)

With most AI applications it is important to determine a good business case to start with. However, with aerial image projects it often works the other way around. First, see what pictures and applications are available, then, determine how detailed the images are and which (potential) AI applications already exist in combination with aerial photos. Compare all these insights with your own organisation and soon you’ll find new possibilities. From monitoring infrastructure to the inspection of riverbanks, as seen with Rijkswaterstaat, the amount of possibilities is enormous. AI will play an important role in this, as it can execute a great part of the searching work. 

Lesson 2: Use satellite images for a time-lapse analysis

One of the biggest advantages of space images is that they are made on a regular basis. The satellite orbits the earth and creates images on a periodic basis. Through this, minor changes can be determined and mapped on satellite images by AI. Consider the growth of refugee camps for example, or any other issues where change over time is involved can be used. The projects do not necessarily have to be on an enormous scale, it can also be interesting for local authorities. For example, municipalities can assess changes in the infrastructure very efficiently, such as the use of solar panels.

Lesson 3: Aerial photography for more detail

Despite aerial images being taken with a smaller intervals (six times a year) than satellite images, they have the advantage of showing details very well. Where satellite images determine that solar panels are on a roof, aerial images (from a plane / done) can give us more details. Aerial images can tell us more about the status of the panel, such as glass damage. The data of these detailed images can be converted into actions such as repairs or adjustments. Adding AI to this can lead to predictions about maintenance, for example. 

Lesson 4: Expensive is not always your best option 

Both satellite and aerial pictures are often freely available. This means that the costs can remain quite low. Unfortunately, this is not always the case as governments still make use of images from commercial organisations. Only if the images concern very specific objects, higher costs cannot be avoided. One way to limit costs is by using drones. In the guide you will find an overview of where you can obtain all sorts of images. 

“We hope these lessons and the included guide will inspire people,” says Verschoor. “I am sure that with the right combination of AI and aerial / satellite images, better policies can be made in the public sector. A great opportunity to let people and technology really work together. ”