FSFNet: Frequency–Spatial-Domain Fusion Network for Infrared Small Target Detection

Infrared small target detection (IRSTD) is inherently challenging due to the weak feature representation of small targets and significant interference caused by complex backgrounds. Recent mainstream IRSTD methods predominantly emphasize feature extraction in the spatial domain but often neglect crucial complementary information contained in the frequency domain, resulting in limited detection performance. To address this issue, we propose a novel Frequency-Spatial Fusion Network (FSFNet), which explicitly integrates spatial-domain features and frequency-domain information through two innovative modules: the Frequency-Spatial Block (FSBlock) and the Frequency-Spatial Feature Fusion Module (FSFFM). Specifically, the FSBlock, equipped with a Spectral Band Gate (SBG) module, selectively emphasizes informative spectral bands to capture global contextual patterns and periodic characteristics in the frequency domain, thereby complementing traditional spatial-domain local representations and enhancing the network’s ability to distinguish small targets from cluttered backgrounds. Furthermore, the FSFFM adaptively combines these multi-domain features via a frequency-guided cross-channel attention mechanism, effectively bridging the semantic gap between the spatial and frequency domains. Extensive experimental evaluations on three publicly benchmark datasets demonstrate that the proposed FSFNet consistently achieves superior detection accuracy and robustness compared with state-of-the-art methods.