Scholarlink

<p>To address the shortcomings of existing vision-based anti-drone detection methods—such as insufficient accuracy in small-target recognition, excessively high model complexity, and difficulty in meeting real-time, high-throughput processing demands on resource-constrained edge computing platforms—this paper proposes a lightweight, high-precision detection method named LGCE-YOLOv11 for high-performance edge computing. Firstly, the C3K2_GOMF module is designed, which employs orthogonal regularization to reduce feature redundancy and enhance fine-grained features of small targets, while incorporating an improved multi-scale attention mechanism (STMEA) to extract drone features across different scales. Secondly, a lightweight global context enhancement module (LGCE) is constructed, leveraging parallel multi-branch multi-scale pooling, semantic memory mechanisms, and adaptive gated fusion to strengthen global semantic correlations while reducing the parameter count. Finally, to optimize computational load, the model introduces a high-resolution P2 detection branch and removes the redundant P5 detection head, achieving a balance between detection accuracy and computational efficiency. Experiments conducted on the DUT Anti-UAV dataset demonstrate that, compared to the baseline model YOLOv11n, the proposed method improves precision (P), recall (R), mAP50, and mAP50-95 by 2.8%, 6.0%, 4.5%, and 6.7%, respectively, while reducing the parameter count by 40% and maintaining a frame rate (FPS) above 120 frames per second, thereby meeting real-time processing requirements.</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

LGCE-YOLOv11: a lightweight anti-drone detection method

  • 期刊论文

  • 发表日期：

    2026年5月8日

查看全文
  • Jiang Wei,
  • Li Ke,
  • Yang Junjie,
  • Tao Kunyu,
  • Chen Zeyu,
  • Fu Chao

摘要

To address the shortcomings of existing vision-based anti-drone detection methods—such as insufficient accuracy in small-target recognition, excessively high model complexity, and difficulty in meeting real-time, high-throughput processing demands on resource-constrained edge computing platforms—this paper proposes a lightweight, high-precision detection method named LGCE-YOLOv11 for high-performance edge computing. Firstly, the C3K2_GOMF module is designed, which employs orthogonal regularization to reduce feature redundancy and enhance fine-grained features of small targets, while incorporating an improved multi-scale attention mechanism (STMEA) to extract drone features across different scales. Secondly, a lightweight global context enhancement module (LGCE) is constructed, leveraging parallel multi-branch multi-scale pooling, semantic memory mechanisms, and adaptive gated fusion to strengthen global semantic correlations while reducing the parameter count. Finally, to optimize computational load, the model introduces a high-resolution P2 detection branch and removes the redundant P5 detection head, achieving a balance between detection accuracy and computational efficiency. Experiments conducted on the DUT Anti-UAV dataset demonstrate that, compared to the baseline model YOLOv11n, the proposed method improves precision (P), recall (R), mAP50, and mAP50-95 by 2.8%, 6.0%, 4.5%, and 6.7%, respectively, while reducing the parameter count by 40% and maintaining a frame rate (FPS) above 120 frames per second, thereby meeting real-time processing requirements.