A spatio-temporal cooperative expanded U-Net for multi-frame infrared small target detection

Infrared small target detection (IRSTD) technology plays a vital role in various security applications. However, existing methods still struggle to detect low signal-to-clutter ratio (SCR) targets against complex backgrounds. General single-frame detection algorithms exhibit poor performance when missing spatial information. While multi-frame detection methods incorporating temporal information have improved performance, their spatio-temporal fusion remains insufficiently refined, failing to fully leverage target motion patterns. To address these challenges, we propose a novel Spatio-temporal cooperative expanded U-shape network (STCEUNet) for IRSTD. Specifically, we first adopt a progressive spatio-temporal hybrid enhancement architecture in the feature extraction backbone. Building upon single-frame feature extraction, this architecture establishes cross-frame information transmission channels at different levels and scales, enabling deep interactive fusion of complementary spatio-temporal information. Second, we further introduce the central difference attention group (CDAG) and multi-frame feature fusion module (MFFM) for intra-frame feature aggregation and inter-frame feature fusion, respectively. CDAG performs adaptive gradient enhancement based on local saliency to provide effective spatial features, facilitating the preservation of latent target information. MFFM explicitly encodes target motion patterns by calculating the correlation and difference in spatio-temporal context between adjacent frames. It distinguishes targets from background based on motion consistency, thereby improving response to moving infrared small targets. Finally, experiments conducted on two public datasets validate the effectiveness of our approach. The proposed approach achieves detection rates of 0.983 and 0.959, with false alarm rates of 3.28E-05 and 5.14E-06, exhibiting superior performance compared to other methods.