TAMamba: Texture-Aware Mamba With Physical-Oriented Scanning for Robust Infrared Small Target Detection

Infrared small target detection is severely challenged by directional background clutter, which often conceal weak target signals. Existing CNN, Transformer, and Mamba-based detectors struggle to adaptively suppress such clutter due to fixed receptive fields, global attention bias, or rigid scanning paths. To overcome these limitations, we present TAMamba, an orientation-guided Mamba framework equipped with a Texture- Aware State-Space Module (TSSM) and a Deep Focus Algorithm (DFA). TSSM introduces explicit physical orientation priors to generate dynamic texture masks, guiding selective scanning toward clutter-sensitive regions. DFA further enhances target representations by performing selective normalization that amplifies sparse small-target signals while suppressing background interference. Extensive experiments on NUAA-SIRST, IRSTD- 1k, and NUDT-SIRST benchmarks demonstrate that TAMamba achieves state-of-the-art performance with only 2.3M parameters, reducing parameters by 51% compared to the lastest models, while maintaining superior accuracy and robustness in complex scenarios. These results highlight TAMamba as a compact yet highly effective solution for robust ISTD. We will release Ours codes onhttps://github.com/SparrowTought/TAMamba.