Automatic Dataset Builder for Machine Learning Applications to Satellite Imagery

TL;DR

This paper introduces an innovative tool architecture that automatically generates suitable datasets for AI applications in Earth Observation, reducing manual effort and making dataset creation accessible to non-experts.

Contribution

The paper presents a novel automatic dataset builder architecture for satellite imagery AI applications, including two implemented versions with a user-friendly GUI.

Findings

Two versions of the dataset builder are available on GitHub.

The tool simplifies dataset creation for satellite imagery AI applications.

It enables non-experts to generate datasets efficiently.

Abstract

Nowadays the use of Machine Learning (ML) algorithms is spreading in the field of Remote Sensing, with applications ranging from detection and classification of land use and monitoring to the prediction of many natural or anthropic phenomena of interest. One main limit of their employment is related to the need for a huge amount of data for training the neural network, chosen for the specific application, and the resulting computational weight and time required to collect the necessary data. In this letter the architecture of an innovative tool, enabling researchers to create in an automatic way suitable datasets for AI (Artificial Intelligence) applications in the EO (Earth Observation) context, is presented. Two versions of the architecture have been implemented and made available on Git-Hub, with a specific Graphical User Interface (GUI) for non-expert users.