Label Maker is basically a python library that helps to draw out the insight from satellite imagery. Satellite imagery is pictures of earth or other celestial bodies that are captured by imaging satellites. Label Maker plays a major role in data preparation for satellite machine learning. It provides us with processed machine-learning training data that can be used with different Machine Learning frameworks, including TensorFlow, MXNet, and Keras. It extracts data from OpenStreetMap, a platform that helps in data extraction which basically deals with the expansion and development of freely-reusable geospatial data. The data extracted is collaborated with the imagery sources like Digital Globe or Mapbox that leads to the formation of a file that can be used for training in machine learning algorithms.

The basic requirement of Label Maker includes- Python 3.6 and Tippecanoe. Python is a programming language that allows work to be done efficiently and compile your systems more effectively whereas Tippecanoe enables the formation of a scale-independent view of data that allows accessing the density and texture of the data.

The Supervised learning methods require two things satellite imagery and ground-truth labels. While training a model there might be a good dataset already available or you may need a custom class label to form a new dataset. Through Label Maker you can define your required classes as OSM features, specify imagery tileset and rest will be handled.

The tool downloads satellite imagery tiles and OpenStreetMap QA Tile information which is further stored as a .npz file for use in Machine Learning training. The command provides us with npz file which consists of all the imagery and labels as numpy arrays, distributed between training and testing sets. The labels can be created and formulated that corresponds to most types of machine learning problems. Label Maker enables us to quickly plug satellite imagery into your favourite machine learning framework and expands the growth prospect.

More Information: GitHub