Answers are grounded in this paper and cite their sources.
Infrared small-target detection (IRSTD) is critical for aerospace and maritime search-and-rescue operations. However, existing methods face three technical bottlenecks: (1) background noise contaminates small-target feature maps during deep feature extraction; (2) conventional channel attention mechanisms erroneously suppress channels containing target responses; and (3) target signals in skip connections risk dilution by background information. To address these challenges, we propose FCCF-Net (Fine-grained Channel and Cross-Encoder Fusion Network). Specifically, we introduce shallow feature maps for cross-layer fusion, providing precise positional anchors that suppress noise erosion and enhance feature discriminability. Furthermore, we develop a fine-grained channel attention mechanism that strengthens target-responsive channels while attenuating invalid background channels. Finally, we propose a commonality-constraint mechanism via skip connections, leveraging shared semantic and structural features to dynamically optimize upsampling, reduce background redundancy, and reinforce target responses. Comprehensive experiments on three public datasets (SIRST, IRSTD-1K, and NUDT-SIRST) verify the effectiveness and competitiveness of our proposed method. Source code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/Bo-Hui/FCCFNet</uri>.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: