Project

In PINEAPPLE, we aim to address the lack of “trust” in (deep) machine learning algorithms for earth observation. We tackle two real-life earth observation downstream tasks: (1) estimating soil parameters from hyperspectral imaging (HSI), and (2) detecting methane in HSI, using new deep and classic machine learning algorithms empowered by explainable AI (XAI) methods.

We believe that PINEAPPLE will be an important step toward not only “uncovering the magic” behind learning algorithms (hence building trust in them in earth observation downstream tasks), but also in showing that XAI methods can be effectively used to improve such data-driven algorithms, ultimately leading to better predictive models.

We put special effort into:

  1. unbiasing the validation of existing and emerging algorithms through ensuring their full reproducibility, both at the algorithm and at the data level, and
  2. understanding & improving the generalization of such algorithms when fundamentally different data is used for testing, e.g. noisy, with simulated other atmospheric conditions, captured in different area/time.

Research project done in collaboration between KP Labs and MI2.AI (Warsaw University of Technology); funded by the European Space Agency grant no. ESA AO/1-11524/22/I-DT.

Logo: European Space Agency Logo: Warsaw University of Technology

Team

Jakub Nalepa Paweł Skorupa Lukasz Tulczyjew Agata Wijata
Jakub Nalepa
Head of HSI		 Paweł Skorupa
Project Manager		 Łukasz Tulczyjew
Research Scientist		 Agata Wijata
Research Scientist
Przemysław Biecek Hubert Baniecki Tymoteusz Kwiecińsk Vladimir Zaigrajew
Przemysław Biecek
Head of XAI		 Hubert Baniecki
Research Scientist		 Tymoteusz Kwieciński
Software Engineer		 Vladimir Zaigrajew
Research Scientist

Results

Paper https://arxiv.org/abs/2403.08017

Red Teaming Models for Hyperspectral Image Analysis Using Explainable AI

ICLR 2024 Workshop on Machine Learning for Remote Sensing (ML4RS)

Remote sensing applications require machine learning models that are reliable and robust, highlighting the importance of red teaming for uncovering flaws and biases. We introduce a novel red teaming approach for hyperspectral image analysis, specifically for soil parameter estimation in the Hyperview challenge. Utilizing SHAP for red teaming, we enhanced the top-performing model based on our findings. Additionally, we introduced a new visualization technique to improve model understanding in the hyperspectral domain.

Software https://github.com/xai4space/meteors

Python package for explanations of remote sensing imaginery.

Contact

Email us at pineapple (at) kplabs.pl for any queries.