.png){kind=icon}
.png){kind=icon}
Return
Spatial-frequency collaborative implicit representation for infrared small target detection
.png){kind=icon}
PRE
AI
Z
G
DOI：10.1007/s00371-026-04685-7.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Infrared small target detection (IRSTD) is a critical task in visual surveillance and search systems; yet, it remains challenged by weak target signals and complex background clutter. Existing methods mostly focus on spatial feature learning and neglect frequency-domain information, while discrete upsampling causes detail loss for tiny targets. Here, we propose a spatial-frequency collaborative implicit neural representation framework (SFC-INRNet) for IRSTD. Specifically, the method features an adaptive frequency enhancement and reconstruction (AFER) module within a dual-domain encoder to dynamically mine high-frequency target priors and suppress clutter. Furthermore, a multi-scale heterogeneous feature fusion (MSHN) module aligns and deeply integrates cross-domain representations using scale-aware strategies. Finally, a frequency-aware implicit decoder (FAID) reformulates discrete upsampling into a continuous function regression, enabling arbitrary-scale feature sampling for the high-fidelity reconstruction of sub-pixel structures. We show that our method achieves 94.84% IoU on NUDT-SIRST, 76.94% IoU on NUAA-SIRST, and 65.53% IoU on IRSTD-1 K, outperforming most state-of-the-art methods. This work provides a new dual-domain and continuous representation paradigm for infrared small target analysis, with potential value for infrared vision applications. Our code will be made public at https://github.com/xzc6666/SFC-INRNet .
Keywords:
Infrared small target detection
Spatial-frequency collaboration
Implicit neural representation
Heterogeneous feature fusion
Journal
IF:
2.9
Papers:
4.5K
Citations:
6.5K
Organization
Cited Papers