HYPERVIEW 2

Overview

KP Labs and the ESA’s φ-Lab, together with Warsaw University of Technology and Poznan University of Technology rejoin forces for a second time: With HYPERVIEW2, the AI4EO Challenges return to AI-enhanced in-orbit processing to advance more efficient and sustainable farming.

Remember HYPERVIEW? We are back!

Back in 2022, we challenged you to the first HYPERVIEW Challenge, where we asked you to make agriculture more sustainable and efficient by leveraging the power of hyperspectral Earth observation (EO) data and on-board processing tools.

With HYPERVIEW2 we are returning to this challenge, with upgraded tools and a new approach to Explainable Artificial Intelligence (XAI) systems.

Increasing efficiency and sustainability by leveraging EO-powered tools is one thing – making the system understandable and comprehensible, another.

HYPERVIEW2 Challenge is part of the EASi workshop at ECAI 2025 – see the webpage at https://ai4eo.eu/portfolio/easi-workshop-hyperview2/

Making sense of it: Explainable Artificial Intelligence

As Earth observation increasingly relies on AI for mission-critical tasks such as satellite operations, data analysis, and decision-making, ensuring the reliability, interpretability, and accountability of AI systems becomes paramount.

Regarding farming applications, AI systems play an increasing role in analysing vast amounts of satellite imagery and data to monitor climate change and ecosystem health. Explainability and interpretability of those systems are essential to ensure that scientists, policymakers, and stakeholders in the agricultural sector can understand and trust AI recommendations, thus enabling more informed and effective decision-making in tackling environmental issues.

HYPERVIEW2 Challenge addresses this need by fostering effective and explainable AI models applicable to airborne and satellite imaging.

The objective of this challenge is to advance the state of the art of AI models for soil parameter retrieval using airborne hyperspectral images, Sentinel-2 multispectral satellite images and PRISMA hyperspectral satellite images.

The task for participants to HYPERVIEW2 Challenge is to develop an explainable AI model to estimate the concentration of six, equally-important soil contaminants/trace elements: Boron (B), Copper (Cu), Zinc (Zn), Iron (Fe), Sulphur (S), and Manganese (Mn) in top soil using Earth observation imagery.

Challenge structure

The HYPERVIEW2 Challenge focuses on two aspects of machine learning models: their prediction capability (i.e., the quality of soil parameters’ estimations), as well as on their robustness investigated through applying appropriate XAI techniques.

Accordingly, the HYPERVIEW2 Challenge will comprise two stages:

1. Assessing the prediction performance of the models (the quality of soil parameters’ estimation should be maximized)
2. Assessing the robustness of the models using XAI methods

The final ranking will be impacted by both the HYPERVIEW Score and the quality of the XAI analysis developed by the participants.

Challenge timeline and deadlines

• Challenge launch: 21st May 2025
• Online workshop (introduction to HYPERVIEW2): Date announced soon
• XAI checkpoint: 30th June 2025
• Reproducibility check by organizers (top 10): 30th June - 11th July 2025
• Release of top-performing models to other teams for XAI: 14th July 2025
• Challenge closing and submission of XAI reports: 31st August 2025
• Award ceremony (during EASi workshop!): 25-26th October 2025

One EASi workshop registration fee for the winning teams (up to 5) will be covered by the organizers, with a maximum amount of ca. 500 EUR per team.