SWAN: Synergistic Wavelet-Attention Network for Infrared Small Target Detection

Abstract

Infrared small target detection (IRSTD) is critical in both civilian and military applications. However, the existing IRSTD methods still do not fully exploit the frequency-domain characteristics of the targets. To tackle this issue, we propose the synergistic wavelet-attention network (SWAN), achieving collaborative optimization through the deep coupling of frequency-spatial features, efficient long-range dependency modeling, and adaptive multiscale feature enhancement. Specifically, we introduce Haar wavelet convolution (HWConv) for a deep, cross-domain fusion of the frequency energy and spatial details of small targets. The shifted spatial attention (SSA) mechanism efficiently models long-range spatial dependencies with linear computational complexity, enhancing contextual awareness. Additionally, the residual dual-channel attention (RDCA) module adaptively calibrates channel-wise feature responses to suppress background interference while amplifying target-pertinent signals. Extensive experiments demonstrate that SWAN achieves significant improvements in detection accuracy and robustness, particularly in challenging scenarios. The source codes are available at https://github.com/jing2024star/SWAN