We will present a partnership between CERN, Intel, and the United Nations Institute for Training and Research (UNITAR) to use Deep Learning (DL) to improve the analysis of optical satellite imagery for humanitarian purposes.
Our core objective is to create spectrally valid simulated high-resolution satellite imagery depicting humanitarian situations such as refugee settlements, flood conditions, damaged infrastructure, and more, by using techniques such as Generative Adversarial Networks (GANs).
UNITAR hosts the UN Operational Satellite Applications Centre (UNOSAT), which uses satellite imagery to support disaster response, humanitarian operations, and other activities of the broader UN system. UNITAR has in recent years started studying the application of DL to satellite imagery analysis, in order to improve efficiency in the many manual tasks frequently required: for example, digitizing a refugee settlement in a satellite image can take many hours, sometimes days of effort. Mapping displaced persons across broad regions such as northwest Syria can in turn be weeks of work.
DL methods could greatly reduce the amount of time needed to complete such tasks: a key factor to an effective deployment of field operations in critical humanitarian situations.
High-resolution satellite imagery is often licensed in such a way that it can be difficult to share it across UNITAR, UN partners, and academic organizations, reducing the amount of data available to train DL models. This fact has inhibited UNITARs DL research possibilities in various ways. The creation of realistic and spectrally accurate simulated images could enable and stimulate data sharing.