A New Motion Feature-Enhanced Multiframe Spatial–Temporal Infrared Target Detection Network

Performing moving target detection in space-based infrared detection systems has long been a critical challenge and research focus. The primary difficulties stem from the small and textureless nature of targets, complex backgrounds, and substantial noise. Additionally, satellite platform mobility causes the backgrounds to move, further complicating the task of distinguishing moving targets from dynamic clutter. Moreover, the lack of publicly available multiframe moving target datasets derived from space-based platforms has hindered the development of space-based infrared moving target detection technologies. Therefore, a moving target dataset (MIRSat-QL) with a complex dynamic background was constructed from a space-based low-Earth orbit satellite platform. A novel motion feature enhancement-based dynamic infrared space target detection network (MFE-Net) is subsequently proposed. Specifically, MIRSat-QL is a semisimulated dataset that combines space-based low-Earth-orbit satellite imagery with ground-truth target data, providing reliable data support for detecting dynamic targets in complex space-based scenarios and facilitating algorithmic development and validation. To address detection challenges and reduce false alarms caused by dynamic clutter interference in complex scenes, a motion-guided feature refinement (MGFR) module is introduced. This module effectively extracts motion features, suppresses dynamic backgrounds, and enhances target signals via dynamic background correspondence mapping (DBCM) and multiframe differential enhancement (MDE). Additionally, a multiattention-guided feature fusion (AGFF) module, which efficiently performs dynamically weighted fusion and multiscale feature refinement, is designed, significantly enhancing the ability of the model to detect small targets. Extensive experiments demonstrate that the proposed algorithm achieves state-of-the-art (SOTA) performance. The dataset and code are available at https://github.com/lifenghong/MFE-Net.