Deep learning for automated feature detection in climate, weather, and space

In this presentation, we discuss ongoing projects in deep learning computer vision for automated feature detection in large spatiotemporal datasets. Examples include the detection of severe weather events on Earth and magnetic storms on the surface of the sun. In each case we need to account for spherical image distortion, large pixel-wise imbalances in the training data, and temporal continuity. We will describe in detail the methods used and how they may be employed as a template for related feature detection tasks.