AI-driven Acquisition of spaceborne hyperspectral data
Abstract : AI learning techniques have been widely applied in hyperspectral imaging to enhance the extraction of valuable information and improve the performance of various tasks. With the advent of commercial companies offering hyperspectral imaging and analysis from space, AI became a game changer in automating the acquisition and streamlining the process of capturing hyperspectral data.
This talk provides insights into how AI and deep learning models are used to analyse real-time data and environmental conditions for adjusting parameters such as exposure time, integration time, and sensor settings to optimise the acquisition process, reduce the need for manual intervention and ensure optimal data quality. Furthermore, it illustrates how explainable AI can optimise the sampling strategy by identifying the most informative locations or spectral bands for capturing hyperspectral data in a way that maximises information gain and minimises redundancy. The presentation concludes with a demonstration of how lightweight deep learning models can be used to process hyperspectral data in orbit, reducing the need for large-volume data transmissions. This onboard processing capability enables rapid decision-making and can be particularly useful in applications where real-time insights are critical, such as hazard monitoring or surveillance.