FDMNet: frequency decoupled modulation network for infrared small target detection
作者
Zhipeng Wang,Feng Zhou,Jiayang Yin,Jinxin Xia
出处
期刊:Sixth Symposium on Novel Optoelectronic Detection Technology and Applications [SPIE] 日期:2026-05-11卷期号:: 143-143
标识
DOI:10.1117/12.3107581
摘要
To address the issue where small target signals in infrared images are severely affected by high-intensity background clutter, we propose a Frequency Decoupling Modulation Network (FDMNet) for infrared small target detection. FDMNet models infrared images as high-frequency and low-frequency components, guiding the network to learn robust discriminative clues from a differentiated frequency domain feature space. First, a frequency decoupling modulation module is used to construct a dual-branch encoder-decoder structure, mapping the input image into high and low-frequency feature components. By designing efficient intra-frequency and inter-frequency communication structures, the contrast between target signals and interfering backgrounds can be adaptively enhanced as computation deepens, achieving robust detection. Finally, a rectification enhancement module is designed in the model’s prediction head, using a multiscale receptive field fusion attention mechanism to fuse high and low-frequency features, aiming to further enhance small target signals and suppress background noise. Results on two public datasets show that the proposed method achieves better performance compared to various state-of-the-art (SOTA) methods in recent years.