LCMamNet: A Lightweight Cross-scale Mamba Network for Infrared Small Target Detection

Researchers introduced LCMamNet, a lightweight Mamba-based model for infrared small target detection.

It combines directional encoder blocks, dense cross-scale feature fusion, and a progressive decoder to preserve target structure and suppress clutter.

LCMamNet achieved strong mIoU on IRSTD-1k, NUAA-SIRST, and NUDT-SIRST while keeping parameters, FLOPs, and inference time low.

The model also ran successfully on a Jetson Orin NX 16G SUPER, making it promising for real-time edge deployment.