Home

Welcome to the aeronetlib wiki!

This library is created in AeroNet Lab at Skolkovo Institute of Science and Technology (Skoltech) and aims to provide an easy to use connection between the remote sensing data and the deep learning methods.

The essential parts are datasets.vector, datasets.raster and datasets.io.

Raster data

The main idea behind the raster data handling is that the remote sensing data often comes as very large images (100 GB per file is not a limit), so reading it in memory is of little use. In the meantime, they are spatially contiguous, so we would like still handle them as a whole, rather than cut into pieces in advance. That is why we open the image handles and keep them at hand, being able to read any part of the raster when needed.

We support single-channel GeoTiff data (and multi-channel images should be split previously into separate files).

The raster data handling is located in the dataset.raster module; the image handle is represented by the Band class; Bands representing different channels of the same image (with equal size and georeference) can compose a BandCollection (see the Raster data page for more information)

Vector data

We use geojson as a primary vector data format.

Data coordinate systems

Data transformations

Data sampling (io)