home Geo Events Aerial Imagery Machine Learning Challenge

Aerial Imagery Machine Learning Challenge

World Bank, WeRobotics, and OpenAerialMap have joined hands to launch open Machine Learning (ML) challenge for classification of very high-resolution aerial imagery. It can also be termed as Artificial Intelligence (AI) Challenge because the baseline is to harness the AI computational abilities for identification (classification) of different features on aerial imagery.

It is a step towards automated recognition and classification of aerial imagery into different features to assist disaster response activities. On contrary, manual image classification is time taking as well as resource and skills intensive task. Thus it has become very important to have tested classification algorithms in place that at least produce good results on the local/regional scale.

Natural Disasters are a global reality. Every decade an estimated one million people are killed by natural disasters. Knowing the extent of damage is vital for the appropriate response. Previously, the use of remote sensing was considered as an efficient source for disaster impact assessment. But, in recent years, the advancement of Unmanned Aerial Vehicle (UAV) technology has caught more attention and focus is gradually shifting to the use of high-resolution aerial imagery. Now UAVs can easily be used for estimating damages and losses by acquiring post-disaster imagery.

In this World Bank AI challenge, approx. 80km2 of high spatial resolution (under 10 cm) imagery has been provided. The imagery is captured on Kingdom of Tonga (south pacific) region during October 2017. So it is quite recent aerial imagery. Optical Remote Sensing imagery is also available for the same area. Besides that, training data sets of roads, buildings, and trees are also provided.

The Actual Challenge:

World Bank now invites individuals or teams to develop machine learning classifiers and automate the analysis of this imagery.

The following classifiers are required in order of priority:

  • Trees (counts and location of individual coconut trees)
    • Coconut trees
    • Banana trees
    • Papaya trees
    • Mango trees
  • Road type (size and surface type)
    • 2-way road vs. 1-lane road
    • Paved vs. dirt road

The accuracy of the classifiers should be  greater than 80%.

Deadlines for submitting your work:

Classifiers for coconut trees should ideally be available by March 1, 2018 with remaining classifiers delivered by June 1, 2018.

For universities operating on a semester program, challenge invite them to submit their classifiers by June 2018 so that evaluation committee can compare accuracies.

Benefits, Awards and Prizes:

The winning team(s) will receive public praise and a Certificate of Achievement. The organizers may also announce monetary prize in future. Nevertheless, students are highly encouraged to participate, because participation in such activities is always a plus point on students’ profiles.

More importantly, the results from this activities will enable the World Bank and partners to significantly accelerate the analysis of aerial imagery before and after major humanitarian disasters. This will help accelerate and improve humanitarian and development efforts across the South Pacific.

Winning teams will also have the opportunity to engage in other related projects around the world.

Please follow this document for further details.

Training data sets are available through following links:

world bank aerial imagery challenge training dataset
Snap Shot of Training Data

Roads and Buildings:

https://export.hotosm.org/en/v3/exports/8a5ba924-1f34-4ed8-a4f6-7b0e2921c06e

Coconut Trees:

https://drive.google.com/file/d/1rumWHzO3_CO40uXhaP69roUyfFzYCe20

 

Please sharing this opportunity far and wide so that challenge can see maximum participation.

Sign up for all updates in your inbox!

Hi, I am a professional Geoinformatics Engineer who has passion for knowing more and more about Geospatial technologies. My particular areas of interest are Spatial Databases, Remote Sensing, and Geospatial Analytics. In addition, I am a strong proponent of free and open-source GIS, Volunteer Geographic Information (VGI), and open geo-data such as OSM.

TAGS:

Muhammad Kamran

Hi, I am a professional Geoinformatics Engineer who has passion for knowing more and more about Geospatial technologies. My particular areas of interest are Spatial Databases, Remote Sensing, and Geospatial Analytics. In addition, I am a strong proponent of free and open-source GIS, Volunteer Geographic Information (VGI), and open geo-data such as OSM.

One thought on “Aerial Imagery Machine Learning Challenge

Leave a Reply

NEVER MISS AN INTERESTING UPDATE!