Machine Learning @ JEODPP: Convolutional Neural Networks

Background

Motivated by the increasing availability of open and free Earth observation data through the Copernicus Sentinel missions, the JRC Big Data Analytics team investigates the capacity of advanced computational models, convolutional neural networks (CNN) for solving the multi-class 10m Sentinel-2 image segmentation problem.

A great range of CNN models have been tested, from standard CNNs to U-net, SegNet and FCN. The annotated learning set was derived from TOP10NL data.

The data processing and analysis has been performed on the JRC Earth Observation Data and Processing Platform (JEODPP) by I.3 Big Data Analytics project, with conceptual contribution of B.6 Digital Transformation and Artificial Intelligence project.

See JRC publication
V. Syrris et.al. (2019): Evaluation of the potential of convolutional neural networks and random forests for multi-class segmentation of Sentinel-2 imagery