The Inter-Agency Implementation and Advanced Concepts (IMPACT) Machine learning (ML) team has launched a beta version of an open-source tool called ImageLabeler. The tool provides a catalog of Earth science events for scientific research and centralizes the creation of training data used to train machine learning algorithms to automatically detect specific events. The tool was developed to facilitate the creation and management of labeled Earth science images for use by supervised machine learning algorithms. With this tool, users can directly access NASA’s archive of satellite imageries and collaboratively label them to indicate the presence or absence of a target Earth science phenomena. Additional features include extraction of images from satellite imagery, drawing bounding boxes to isolate the event in an image and assigning internal teams for groups of users to work simultaneously on labeling images for Earth science events. Several large-scale training datasets have already been generated using this tool, including the flood extent dataset for the Institute of Electrical and Electronics Engineers (IEEE) data science challenge.
The tool can be accessed here: https://impact.earthdata.nasa.gov/labeler/ .
