DAS-Tool: Multispectral Data Analysis Toolbox for SNAP

Computer-based image analysis is essential to assist human understanding and semantic annotation of satellite images of the Earth’s surface. The technical objectives of this project were the elaboration and implementation of dedicated algorithms for Sentinel-2 data analysis. The project proposes an unitary data mining concept that uses advanced data visualisation and explainable features, together with specific graphical instruments to enhance relevant aspects of Sentinel-2 imagery and enable semantic analysis based on a two-stage process:

1. Exploratory visual analysis, aiming to highlight predominant data features for the scene content and help the user perceive certain aspects that are not always reflected in the visible part of the spectrum, maximising the data impact on the human visual system to help image understanding and interpretation. The added value becomes important as the data content representation (the second functionality) will focus more on extracting numerical patterns rather than visual characteristics and the image analysis will provide similarity by data processing. Therefore, the results will not always correspond to the intuitive user perception on the scene. In order to correlate the classification map with its understanding and application, the user can exploit data visualisation to understand the computed correlations and modify the parameters for appropriate content representation.
2. Data content representation, focussing on the identification of relevant spectral, texture, and physical parameters, scene-related features that are further included in a learning process modelling the data content according to statistical similarities. This step will result in a classification map emphasizing the existing categories of objects inside the scene. The analysis entails a compact workflow interconnecting feature extraction and feature classification to describe the Sentinel-2 data content characteristics.

Designed to enhance the exploitation of Sentinel-2 data through fast image understanding and analysis, the concept was implemented as Sentinel-2 dedicated data analysis (DAS-Tool) plugin for the Sentinel Application Platform (SNAP) and deployed as an open-source tool empowering the Earth Observation (EO) community with fast and reliable results.

Driven by the characteristics of Sentinel-2 data, the project aims at increasing the accuracy of traditional algorithms by combining processes that are fit to the image content. The plugin accommodates multiple solutions for each processing phase and enables flexibility in data exploration and multilevel analysis (locally – at pixel level, contextually – at patch level). The methodology reduces the semantic gap by revealing to the user the kind of patterns that are statistically similar through exploratory visual analysis. This will increase the relevance of the training samples and the accuracy given a specific application.

You can find DAS-Tool on the SNAP Community Plugins page.

See also the following publication:

A.C. Grivei, I. Neagoe, F.A. Georgescu, A. Griparis, C. Vaduva, Z. Bartalis, M. Datcu, “Multispectral data analysis for semantic assessment – A SNAP framework for Sentinel 2 use case scenarios”, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Vol. 13, pp. 4429 – 4442, 2020, DOI: 10.1109/JSTARS.2020.3013091. https://ieeexplore.ieee.org/document/9153144

DAS-Tool Demo Videos:

• SNAP – Getting started with DAS-Tool. Installation and pixel based Sentinel-2 data analysis
• SNAP – Interactive Sentinel 2 data analysis with DAS-Tool (patch-based)

The project was part of the ESA Romanian Industrial Incentive Scheme.