Mamba-Infrared Small Target Detection: Multi-Context Hierarchical Aggregation Mamba for Robust Infrared Small Target Detection

Infrared small target detection (ISTD) plays a crucial role in applications such as security surveillance, maritime rescue, and satellite remote sensing. However, the task remains highly challenging due to weak target signals, complex background interference, and significant variations in target scale. Existing methods often struggle to simultaneously achieve three essential objectives: fine-grained local feature representation for capturing subtle target edges and textures, comprehensive global context modeling for suppressing background interference, and multiscale feature fusion for handling dynamic target sizes. To address these limitations, we propose a novel ISTD model based on Mamba architecture called Mamba-ISTD. Specifically, we present the multi-context Local-Global Scanning Mamba (LGSM) module, designed to improve local feature representation using parallel fine-grained convolutions that maintain target edges and textures, alongside global scanning for modeling long-range dependencies and noise reduction. Additionally, we introduce the Cross-layer Spatial Fusion Mamba (CSFM) module, which initially extracts multiscale features using a layer-adaptive sliding window that reduces in size with decreasing sampling depth, followed by Mamba scanning on concatenated features for precise spatial fusion of deep semantics and shallow spatial details. Extensive experiments conducted on three public datasets demonstrate the superior robustness and effectiveness of the proposed method in complex real-world scenarios.