ESE-Net: Edge-Shape Enhancement Network for Infrared Small Target Detection

Infrared small target detection (ISTD) aims to segment small targets from infrared images and is widely applied in military and industrial fields. Although recent deep learning-based methods have achieved remarkable performance, they often fail when targets are indistinguishable from complex backgrounds. This is mainly due to the limited use of spatial domain features, which cannot capture subtle boundary cues, making precise segmentation challenging. To address this, we propose an Edge-Shape Enhanced Network (ESE-Net), which reinforces edge feature representations to improve target discrimination in complex infrared scenes. First, we design a Multiscale Spatial Edge Attention (MSEA) module to strengthen target edges by perceiving directional gradient changes. To suppress background noise while highlighting target boundaries, we introduce an Edge Guidance Module (EGM) that extracts edge features in the frequency domain via a wavelet transform and performs reversible down sampling, discarding low-frequency components before fusing with spatial features. Furthermore, a Multiscale Group Convolution Module (MGCM) is integrated in deep layers to preserve target details and mitigate the risk of small target loss. Experiments on the NUAA-SIRST and IRSTD-1K datasets demonstrate the effectiveness of our method.