Linguistic Description Generation for Non-cooperative Spacecraft Time-Series Behaviors Using Transformer Models

Automatically generating accurate and dynamically consistent descriptions of non-cooperative spacecraft maneuvers is a critical challenge in advancing Space Situational Awareness. To address this challenge, we propose a cross-modal Transformer framework that translates raw image sequences into precise linguistic narratives. First, a dataset of spacecraft action descriptions is constructed by incorporating constraints from actual spacecraft behaviors. A progressive semantic description framework is then designed, which begins by using a semantic segmentation model to extract the target region and remove background interference. Subsequently, normalized semantic features are extracted from the temporal sequence of the target region to generate the final linguistic description of the spacecraft’s behavior. Experimental results demonstrate that this method generates high-quality linguistic narratives for complex on-orbit spacecraft behaviors, with its efficacy further validated in hardware-in-the-loop simulations.