Hierarchical Deep Fusion Detail Feature Enhancement Network for Infrared Small Target Detection

Infrared small target detection aims to identify tiny targets from complex infrared backgrounds, a task that is both critical and highly challenging. Existing methods still face difficulties in effectively extracting discriminative features of small targets, suppressing noise, and preserving fine details, which often limits detection performance. To address these issues, this paper proposes a Hierarchical Deep Fusion Detail Feature Enhancement Network (HFE-Net) based on recent U-shaped neural architectures for infrared small target detection. Specifically, a Channel Feature Enhancement Attention Module (CFEAM) is designed for infrared small target detection, which enhances both local and global information in a unified manner. This module strengthens channel-wise representations while preserving fine-grained details of small target features. Moreover, a Multi-branch Strip Perception Attention Module(MSPA) is introduced, which integrates convolution-based strategies to enhance global contextual information and effectively suppress interference in deeper layers. Furthermore, a Shallow Information Enhancement Processing Module (SIEP) is developed to efficiently exploit shallow features by combining spatial-frequency attention mechanisms, thereby preserving local details, suppressing shallow noise, and expanding the receptive field to improve detection performance. Extensive experiments conducted on multiple public datasets demonstrate that the proposed method significantly improves mean Intersection over Union (mIoU) compared with baseline models, achieving gains of 4.52% on NUDT-SIRST and 2.25% on NUAA-SIRST. The results confirm that the proposed algorithm not only achieves superior detection accuracy over state-of-the-art methods but also exhibits strong noise robustness and reliable detection effectiveness.