Multiscale Channel Attention and Cross-Layer Fusion Network for Infrared Small Target Detection

Infrared Small Target Detection faces challenges due to limited global perception and feature ambiguity in complex scenarios. To address these issues, we propose a novel Multi-Scale Channel Attention and Cross-Layer Fusion Network (MACFNet). The framework integrates three key innovations: (1) The Feature Convolution Attention Transformer (FCAT) addresses limited global perception by combining local features and global contexts to enhance target representation; (2) The Efficient Channel and Spatial Attention (ECSA) module resolves feature ambiguity by optimizing discriminative feature weighting; (3) An enhanced M-UNet architecture incorporates Channel-wise Cross Fusion Transformer (CCT) modules to enable effective cross-scale semantic alignment. Extensive experiments on the SIRST and NUDT-SIRST datasets demonstrate state-of-the-art performance, achieving significantly higher IoU of 0.8396 and 0.9346, respectively, surpassing existing model-driven and data-driven methods while maintaining a real-time capable inference speed of 29.7167 FPS.