Scholarlink

<p>The rapid proliferation of unmanned aerial vehicles (UAVs) has brought many valuable applications, but has also introduced significant security concerns, which require robust detection frameworks. Infrared imaging provides a viable solution for UAV detection under low-visibility conditions, yet existing approaches suffered from high computational costs and inefficiencies in detecting small targets. This work presents YOLO-LIRTU, which is a lightweight framework designed to address hardware constraints, small-target feature degradation, and parameter redundancy in infrared UAV detection. By integrating an optimized ShuffleNetV2 backbone with depthwise separable convolutions and channel shuffle mechanisms, a dynamic multi-scale feature enhancement module, and a decoupled feature generation technique using GhostConv operations, the framework achieves notable computational efficiency while maintaining detection accuracy. The redesigned backbone reduces parameter count through spatial feature reorganization, while the dynamic downsampling strategy preserves critical details for distant targets. Evaluated on the Anti-UAV dataset containing 211,527 infrared images, YOLO-LIRTU requires only 1.73M parameters and 7 GFLOPs, delivering 76.04% <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(AP_{75}\)</EquationSource> </InlineEquation> and 64.72% <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(AP_{50:95}\)</EquationSource> </InlineEquation>. These results suggest its potential as a computationally efficient solution for edge-based UAV surveillance in challenging environments. Our code can be found at: <a href="https://github.com/jinxinhuo/YOLO-LIRTU">https://github.com/jinxinhuo/YOLO-LIRTU</a></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

YOLO-LIRTU: a lightweight infrared real-time UAV detection framework

  • 期刊论文

  • 发表日期：

    2026年1月17日

查看全文
  • Hao Yu,
  • Huan Jiang,
  • Wangming Lan,
  • Qian Jiang,
  • Hongyue Huang,
  • Xin Jin

摘要

The rapid proliferation of unmanned aerial vehicles (UAVs) has brought many valuable applications, but has also introduced significant security concerns, which require robust detection frameworks. Infrared imaging provides a viable solution for UAV detection under low-visibility conditions, yet existing approaches suffered from high computational costs and inefficiencies in detecting small targets. This work presents YOLO-LIRTU, which is a lightweight framework designed to address hardware constraints, small-target feature degradation, and parameter redundancy in infrared UAV detection. By integrating an optimized ShuffleNetV2 backbone with depthwise separable convolutions and channel shuffle mechanisms, a dynamic multi-scale feature enhancement module, and a decoupled feature generation technique using GhostConv operations, the framework achieves notable computational efficiency while maintaining detection accuracy. The redesigned backbone reduces parameter count through spatial feature reorganization, while the dynamic downsampling strategy preserves critical details for distant targets. Evaluated on the Anti-UAV dataset containing 211,527 infrared images, YOLO-LIRTU requires only 1.73M parameters and 7 GFLOPs, delivering 76.04% \(AP_{75}\) and 64.72% \(AP_{50:95}\) . These results suggest its potential as a computationally efficient solution for edge-based UAV surveillance in challenging environments. Our code can be found at: https://github.com/jinxinhuo/YOLO-LIRTU