A Context-Adaptive Hyperspectral Sensor and Perception Management Architecture for Airborne Anomaly Detection.

The deployment of airborne hyperspectral sensors has expanded rapidly, driven by their ability to capture spectral information beyond the visual range and to reveal objects that remain obscured in conventional imaging. In scenarios where prior target signatures are unavailable, anomaly detection provides an effective alternative by identifying deviations from the spectral background. However, real-world reconnaissance and monitoring missions frequently take place in complex and dynamic environments, requiring anomaly detectors to demonstrate robustness and adaptability. These requirements have rarely been met in current research, as evaluations are still predominantly based on small, context-restricted datasets, offering only limited insights into detector performance under varying conditions. To address this gap, we propose a context-adaptive hyperspectral sensor and perception management (hSPM) architecture that integrates sensor context extraction, band selection, and detector management into a single adaptive processing pipeline. The architecture is systematically evaluated on a new, large-scale airborne hyperspectral dataset comprising more than 1100 annotated samples from two diverse test environments, which we publicly release to support future research. Comparative experiments against state-of-the-art anomaly detectors demonstrate that conventional methods often lack robustness and efficiency, while hSPM consistently achieves superior detection accuracy and faster processing. Depending on evaluation conditions, hSPM improves anomaly detection performance by 28-204% while reducing computation time by 70-99%. These results highlight the advantages of adaptive sensor processing architectures and underscore the importance of large, openly available datasets for advancing robust airborne hyperspectral anomaly detection.