Temporal-Emerged Prompting for Segment Anything in Multiframe Infrared Small Target Detection

Accurately localizing and segmenting small targets in low signal-to-noise ratio (SNR) infrared sequences remains a challenging task. Since targets are often indistinguishable from the background in individual frames, existing methods, even when equipped with advanced foundation model and powerful inter-frame association mechanisms, still fail to detect them. Motivated by the observation that targets tend to emerge gradually from the background over time and become distinguishable, we propose Temporal-Emerged Prompting for Segment Anything Model (TEP-SAM), a principled framework designed to explicitly exploit such temporal-emerged cues to modulate and prompt SAM. TEP-SAM operates by jointly modeling global motion patterns and local motion deviations to locate potential targets. It further enhances target region features by leveraging motion discrepancy, thereby generating temporal-emerged cues for SAM and enabling non-interactive segmentation. By bridging large-scale semantic pretraining with task-specific temporal modeling, TEP-SAM effectively adapts SAM to the challenging multiframe infrared small target detection task. Extensive experiments demonstrate the effectiveness of our approach, particularly under severely low-SNR conditions and in complex dynamic background.

Detecting small objects in infrared videos is important for applications such as remote sensing and safety monitoring, but it is difficult when the object is only a few pixels and almost blends into a noisy background. Looking at one frame at a time is often not enough: a real object may be invisible in a single image, while background clutter can look similar. Our key idea is to use time as a clue. As the video progresses, a true object gradually separates from its surroundings through consistent motion, while most background noise does not behave in the same way. We build TEP-SAM, a method that learns these temporal clues and uses them to guide a powerful segmentation model called the Segment Anything Model, without requiring a human to click on the target. By combining SAM’s general image understanding with motion information from infrared videos, our method can locate and segment very small and weak objects more reliably. Experiments show that TEP-SAM works especially well in noisy scenes and complex moving backgrounds.