IRSTD-SAM: Segment Any Infrared Small Target with Parallel Attention

Infrared small target detection (IRSTD) is challenging due to targets’ small size, low contrast, and complex backgrounds. We propose IRSTD-SAM, which adapts the SAM2 vision model for IRSTD by inserting lightweight adapters into Transformer MLP layers for efficient task-specific tuning. A dual-branch decoder with multi-scale fusion, spatial–channel attention, and Canny edge-guided refinement further enhances small-target features. Experiments show improved localization accuracy and background suppression in complex scenes.