Spatial-Frequency Feature Learning for Infrared Small Target Detection

The precision of small target detection plays a pivotal role in infrared image analysis. Currently, various methods are proposed to solve the issue of insufficient small target features. However, many infrared target detection methods often fail to fully consider how to enhance detection accuracy of small targets through effective information guidance. Frequency domain analysis indicates that small targets typically exhibit significant and prominent differences in high-frequency regions. On the other hand, as a key manifestation of high-frequency content, edge information naturally connects the spatial and frequency domains. Inspired by these, this paper proposes an infrared small target detection method called Spatial-Frequency Feature Learning Network (SFLNet). It aims to solve the challenges in small target detection through collaborative modeling of the spatial and frequency domains. SFLNet adopts an edge-guided decoder structure that helps the network preserve the structural integrity and shape information of small targets during reconstruction. Moreover, this module introduces a spatial-frequency joint learning mechanism that enables complementary and enhanced information exchange between the spatial and frequency domains, which improves the perception and recognition of small target features. Experimental results show that SFLNet outperforms comparison methods on multiple datasets, particularly in precise localization of targets and preservation of target shapes.