An Active Learning Tool for the Generation of Earth Observation Image Benchmarks

Abstract

This paper describes an active learning tool for the genera-tion of Earth Observation (EO) benchmark datasets. This toolis able to generate training datasets, based on its active learn-ing strategy with a classification accuracy of around 90%.Afterwards, a data cleaning tool is needed, in order to cor-rect noisy data and provide a clean dataset to be stored in thebenchmark database, and for subsequent benchmark verifica-tion. The data cleaning procedure is supported by unsuper-vised learning, using clustering algorithms to group similarpatterns, and dimension reduction algorithms to embed themin lower dimension with annotated labels. Moreover, interac-tive visualizations are implemented in most modules to helpbetter manipulate datasets and get better understandings.