Scholarlink

<p>Detecting small targets in remote sensing images is challenging due to their extremely small size, complex backgrounds, and dense distributions. Existing general object detectors often suffer from severe feature loss and background interference, leading to high missed detection rates. To address this, we propose HB-YOLOv11, a novel model built upon the YOLOv11 framework for small-target detection in remote sensing. First, a Background Attenuated Pooling (BAPOOL) module is introduced. It adaptively fuses max and min pooling via a probabilistic weighting mechanism, effectively suppressing complex background noise while preserving critical target features. Second, a Hypergraph Parallelized Patch-Aware Attention (HyperPPA) module is proposed. It extracts global, local, and sequential features through a multi-branch structure and employs hypergraph convolution to model higher-order semantic relationships among these features, enhancing discriminative capability in complex contexts. Experiments on two public remote sensing datasets, USOD and AI-TOD, demonstrate that HB-YOLOv11 outperforms mainstream models such as YOLOv8 and YOLOv11 in terms of recall and mean average precision (mAP), validating its effectiveness in reducing missed detections and improving detection accuracy.</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

HB-YOLOv11: A model focused on enhancing the detection of remote sensing small targets in complex backgrounds

  • 期刊论文

  • 发表日期：

    2026年4月9日

查看全文
  • Jie Liu,
  • Qi Liu,
  • Ruijie Wang,
  • Mengyan Qu

摘要

Detecting small targets in remote sensing images is challenging due to their extremely small size, complex backgrounds, and dense distributions. Existing general object detectors often suffer from severe feature loss and background interference, leading to high missed detection rates. To address this, we propose HB-YOLOv11, a novel model built upon the YOLOv11 framework for small-target detection in remote sensing. First, a Background Attenuated Pooling (BAPOOL) module is introduced. It adaptively fuses max and min pooling via a probabilistic weighting mechanism, effectively suppressing complex background noise while preserving critical target features. Second, a Hypergraph Parallelized Patch-Aware Attention (HyperPPA) module is proposed. It extracts global, local, and sequential features through a multi-branch structure and employs hypergraph convolution to model higher-order semantic relationships among these features, enhancing discriminative capability in complex contexts. Experiments on two public remote sensing datasets, USOD and AI-TOD, demonstrate that HB-YOLOv11 outperforms mainstream models such as YOLOv8 and YOLOv11 in terms of recall and mean average precision (mAP), validating its effectiveness in reducing missed detections and improving detection accuracy.