SatImNet: Structured and Harmonised Training Data for Enhanced Satellite Imagery Classification

TL;DR

SatImNet provides a structured, harmonized open-source training dataset for satellite imagery classification, demonstrating improved deep learning model performance through data integration and management.

Contribution

This work introduces SatImNet, a novel open-source, structured training data collection for satellite imagery, and showcases its practical application in CNN-based classification and segmentation.

Findings

SatImNet improves data interoperability for satellite imagery classification.

CNN models trained on SatImNet achieve higher accuracy.

The dataset facilitates effective remote sensing image analysis.

Abstract

Automatic supervised classification with complex modelling such as deep neural networks requires the availability of representative training data sets. While there exists a plethora of data sets that can be used for this purpose, they are usually very heterogeneous and not interoperable. In this context, the present work has a twofold objective: i) to describe procedures of open-source training data management, integration, and data retrieval, and ii) to demonstrate the practical use of varying source training data for remote sensing image classification. For the former, we propose SatImNet, a collection of open training data, structured and harmonized according to specific rules. For the latter, two modelling approaches based on convolutional neural networks have been designed and configured to deal with satellite image classification and segmentation.