Scholarlink

<p>In recent years, the rapid expansion of the UAV industry has driven the demand for efficient detection systems. However, existing solutions typically suffer from prohibitive hardware costs and struggle to accurately perceive small targets embedded in complex backgrounds with limited visual features. To address these challenges, we present DayNight Wander YOLO, a hardware-software integrated system designed for cost-effective and precise small-target perception. On the hardware side, we design a low-cost, autonomous monitoring node based on the K230 platform, enabling wire-free deployment in off-grid environments. On the algorithmic side, we construct a multi-weather dataset and propose a tailored lightweight detection model designed to handle complex environmental interference. The network incorporates a Wave-LLG structure that decomposes feature frequencies to preserve fine-grained edge details. Furthermore, a HyperSPP module utilizes hypergraph correlations to capture global context, while a cross-dimensional attention mechanism suppresses background noise. Performance evaluation on our self-built VSUAV dataset demonstrates that DNW-YOLO achieves competitive accuracy. When deployed on the K230 board, the system delivers 77.0 mAP<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(_{50}\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mn>50</mn> <mrow /> </mmultiscripts> </math></EquationSource> </InlineEquation> at 15 FPS.</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

DNW-YOLO: a cost-effective hardware–software system for small UAVs in complex backgrounds

  • 期刊论文

  • 发表日期：

    2026年3月23日

查看全文
  • Xinyu Wu,
  • Qiang Fu,
  • Lei Pan,
  • Yuan Zheng,
  • Zhipeng Lu,
  • Chenchen Qin,
  • Chongyao Yan,
  • Yi Zeng,
  • Chenglong Li

摘要

In recent years, the rapid expansion of the UAV industry has driven the demand for efficient detection systems. However, existing solutions typically suffer from prohibitive hardware costs and struggle to accurately perceive small targets embedded in complex backgrounds with limited visual features. To address these challenges, we present DayNight Wander YOLO, a hardware-software integrated system designed for cost-effective and precise small-target perception. On the hardware side, we design a low-cost, autonomous monitoring node based on the K230 platform, enabling wire-free deployment in off-grid environments. On the algorithmic side, we construct a multi-weather dataset and propose a tailored lightweight detection model designed to handle complex environmental interference. The network incorporates a Wave-LLG structure that decomposes feature frequencies to preserve fine-grained edge details. Furthermore, a HyperSPP module utilizes hypergraph correlations to capture global context, while a cross-dimensional attention mechanism suppresses background noise. Performance evaluation on our self-built VSUAV dataset demonstrates that DNW-YOLO achieves competitive accuracy. When deployed on the K230 board, the system delivers 77.0 mAP \(_{50}\) 50 at 15 FPS.