Abstract: To address the issues of missed and false detections in infrared small-target detection caused by small target sizes, insufficient texture features, and complex background interference, this paper proposes a dual-domain feature extraction and coordinate attention detection network. Firstly, to enhance detection probability, a dual-domain feature extraction module is designed. This module extracts local features in the spatial domain using central difference convolution and aggregates global image information in the frequency domain through a combination of signal processing and adaptive deep learning, effectively avoiding missed detections. Secondly, a coordinate-activated attention module is proposed. By leveraging the characteristics of attention mechanisms, this module emphasizes the saliency of the target in feature maps while suppressing background interference, significantly reducing false alarms in complex scenarios. Lastly, to address the issue of feature loss caused by repeated downsampling, a global context extraction module is designed. This module combines multi-scale standard convolution with multi-scale dilated convolution to effectively extract multi-level feature representations of the target, further reducing missed detections. Experimental results on four public datasets demonstrate that the proposed method outperforms existing advanced methods. Specifically, on the IRSTD1k dataset, the proposed method achieves mIoU, nIoU, and F1 score of 72.23%, 68.85%, and 83.89%.