Red Teaming Models for Hyperspectral Image Analysis Using Explainable AI

TL;DR

The paper tackles reliability and biases in hyperspectral soil-parameter estimation by applying a SHAP-based red-teaming approach to the winning HyperView model, eagleeyes, and its Intuition-1 onboard deployment. It introduces novel SHAP-driven visualizations that incorporate band/wavelength context and preprocessing transformations, and demonstrates a SHAP-guided feature-pruning method that reduces inputs to less than $<1\%$ with at most $\le 5\%$ performance loss. The analysis reveals that the model relies on a tiny fraction of features, producing predictions concentrated in a narrow range and underperforming on outliers, which motivates a lighter, edge-friendly alternative with comparable accuracy. Overall, the work advances explainable-AI–driven red-teaming in hyperspectral imaging, improving model robustness and enabling efficient onboard inference for remote-sensing missions.

Abstract

Remote sensing (RS) applications in the space domain demand machine learning (ML) models that are reliable, robust, and quality-assured, making red teaming a vital approach for identifying and exposing potential flaws and biases. Since both fields advance independently, there is a notable gap in integrating red teaming strategies into RS. This paper introduces a methodology for examining ML models operating on hyperspectral images within the HYPERVIEW challenge, focusing on soil parameters' estimation. We use post-hoc explanation methods from the Explainable AI (XAI) domain to critically assess the best performing model that won the HYPERVIEW challenge and served as an inspiration for the model deployed on board the INTUITION-1 hyperspectral mission. Our approach effectively red teams the model by pinpointing and validating key shortcomings, constructing a model that achieves comparable performance using just 1% of the input features and a mere up to 5% performance loss. Additionally, we propose a novel way of visualizing explanations that integrate domain-specific information about hyperspectral bands (wavelengths) and data transformations to better suit interpreting models for hyperspectral image analysis.

Figures (9)

• Figure 1: Variations in Shapley values in response to changing feature values in the eagleeyes model for phosphorus predictions, highlighting the most influential feature (left) and a feature with lesser impact (right).
• Figure 2: Detailed visualization of Shapley values, depicting the significance of various aggregated feature transformation groups with their corresponding wavelengths aggregations on the $x$-axis.
• Figure 3: A flowchart presenting the eagleeyes soil parameters' estimation approach. Four regression models (random forests) are trained to retrieve one parameter each, and they operate on manually-designed feature extractors obtained for an input HSI of size $w\times h\times b$, where $w$ and $h$ is its width and height, respectively, and $b$ denotes the number of spectral bands (in this study, $b=150$ or $b=192$).
• Figure 4: Residuals visualization for each soil parameter across different model versions. Each row represents a soil parameter, in the order of phosphorus, potassium, magnesium, and pH. Within each row: the left panel shows a scatter plot (ground truth on the $x$-axis, predicted values on the $y$-axis), the middle panel displays a boxplot of residuals, and the right panel presents a histogram of the residuals.
• Figure 5: Visualization of the top $10$ features identified by Shapley values, ordered by their contribution to model predictions in the eagleeyes model trained for the hyperview challenge. The panels show, clockwise from top left, phosphorus, potassium, magnesium, and pH.