Scholarlink

<p>UAV-based infrared small target detection is challenging due to extremely small target size, weak texture information, complex background clutter, and strict real-time requirements. To address these issues, this paper proposes EHR-Net, a real-time Edge-Enhanced Hierarchical Residual Network built upon the YOLOv11 framework. Specifically, an Edge-Information Enhanced Stem (EIEStem) is introduced to preserve shallow details during early downsampling with limited computational overhead. A Residual MetaFormer-CGLU (RMF-CGLU) module is designed to enhance backbone feature representation through efficient context modeling and gated channel interaction. In addition, a Hierarchical Residual Aggregation Module (HRAM) is incorporated to improve multi-scale feature fusion and semantic aggregation. By integrating these components, EHR-Net can effectively enhance weak target responses and alleviate background interference while maintaining efficient inference. Experimental results on the HIT-UAV and NUDT-SIRST datasets show that EHR-Net improves mAP@0.5 by 4.9% and 4.1%, respectively, over the YOLOv11n baseline. Meanwhile, EHR-Net achieves end-to-end GPU inference speeds of 302.19 FPS and 204.70 FPS on an NVIDIA RTX 3090, demonstrating a favorable accuracy–efficiency trade-off for GPU-based real-time UAV infrared small target detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

EHR-Net: a real-time Edge-Enhanced Hierarchical Residual Network for UAV infrared small target detection

  • 期刊论文

  • 发表日期：

    2026年8月5日

查看全文
  • Guanxun Cui,
  • Xiaolong Li

摘要

UAV-based infrared small target detection is challenging due to extremely small target size, weak texture information, complex background clutter, and strict real-time requirements. To address these issues, this paper proposes EHR-Net, a real-time Edge-Enhanced Hierarchical Residual Network built upon the YOLOv11 framework. Specifically, an Edge-Information Enhanced Stem (EIEStem) is introduced to preserve shallow details during early downsampling with limited computational overhead. A Residual MetaFormer-CGLU (RMF-CGLU) module is designed to enhance backbone feature representation through efficient context modeling and gated channel interaction. In addition, a Hierarchical Residual Aggregation Module (HRAM) is incorporated to improve multi-scale feature fusion and semantic aggregation. By integrating these components, EHR-Net can effectively enhance weak target responses and alleviate background interference while maintaining efficient inference. Experimental results on the HIT-UAV and NUDT-SIRST datasets show that EHR-Net improves mAP@0.5 by 4.9% and 4.1%, respectively, over the YOLOv11n baseline. Meanwhile, EHR-Net achieves end-to-end GPU inference speeds of 302.19 FPS and 204.70 FPS on an NVIDIA RTX 3090, demonstrating a favorable accuracy–efficiency trade-off for GPU-based real-time UAV infrared small target detection.