Decoupled Motion Representation Learning for Moving Infrared Small Target Detection

Researchers have proposed new methods to improve infrared small target detection in dynamic scenes, a challenging task due to coupled motions among targets, platforms, and backgrounds.

Existing multi-frame methods typically perform implicit temporal modeling, where background dynamics dominate motion correspondence learning^[1]. A new decoupled motion representation learning framework has been introduced, featuring an explicit motion branch to model globally coherent motion dynamics and an implicit motion branch to capture target-sensitive local motion anomalies. The explicit motion branch uses pretrained optical flow priors and a structure-preserving self-supervised adaptation strategy for infrared motion correspondence learning. Meanwhile, a coherent-motion-guided local anomaly reasoning module identifies and suppresses coherent-motion-induced false responses. Extensive experiments on two challenging benchmarks demonstrate that this method outperforms existing state-of-the-art approaches, particularly in dynamic scenes with complex motions^[1]. Another development is the REEM optimization framework, a lightweight SCR-guided difficulty-aware optimization framework that incorporates Signal-to-Clutter Ratio as a visibility prior during training. REEM applies a differentiable modulation to the soft-IoU learning signal, emphasizing low-visibility targets, and is integrated into a U-Net-based MSHNet without adding parameters or inference-time overhead^[2].