Keras Spatial is a Python library meant to make preprocessing geospatial data easier. Deep learning techniques allow researchers to investigate data extracted from remote sensing archives, but the data involved must be prepared properly. For instance, rasters are generally stored in a file larger than a single training sample.
To solve this issue, Keras Spatial gives users a data generator built to read samples directly from a raster data source. This removes the need to create smaller individualized raster files prior to model execution. WIthout the need to divvy up large remote-sensing data images into equal-sized samples, Keras Spatial allows users to increase efficiency.
Created by scientists at NCSA, the team behind Keras Spatial hopes to eliminate much of the preprocessing work necessary to conduct important geospatial research.
EXPLORE KERAS SPATIAL