Scholarlink

<p>Unmanned aerial vehicle (UAV) imagery poses significant challenges for small object detection due to complex backgrounds, scale variation, and dense object distribution. To address these issues, this paper proposes AeroFineFusion-Det, an enhanced YOLOv8-based detection framework tailored for UAV aerial scenes. The proposed method introduces a fine-scale bidirectional multi-scale fusion strategy to improve information flow across feature levels, together with a dual-attention weighted fusion mechanism that adaptively enhances discriminative representations. In addition, a hybrid unit is embedded into the C2f module to strengthen shallow-detail perception, while a cross-level local-aware fusion head further improves the coordination between classification and localization. Extensive experiments on the VisDrone2019 and UAVDT datasets demonstrate that AeroFineFusion-Det consistently improves detection performance over representative detectors. Compared with YOLOv8s, the proposed method improves AP<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(_{50:95}\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mrow> <mn>50</mn> <mo>:</mo> <mn>95</mn> </mrow> <mrow /> </mmultiscripts> </math></EquationSource> </InlineEquation> by 4.4% and AP<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(_s\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mi>s</mi> <mrow /> </mmultiscripts> </math></EquationSource> </InlineEquation> by 5.3% on VisDrone2019, while reducing model parameters by 33.9%. Although additional feature-fusion computation is introduced, the experimental results validate the effectiveness of the proposed design for UAV aerial small-object 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

Enhancing UAV aerial small object detection through fine-scale bidirectional multi-scale fusion

  • 期刊论文

  • 发表日期：

    2026年7月19日

查看全文
  • Xuanyu Cai,
  • Guanxun Cui

摘要

Unmanned aerial vehicle (UAV) imagery poses significant challenges for small object detection due to complex backgrounds, scale variation, and dense object distribution. To address these issues, this paper proposes AeroFineFusion-Det, an enhanced YOLOv8-based detection framework tailored for UAV aerial scenes. The proposed method introduces a fine-scale bidirectional multi-scale fusion strategy to improve information flow across feature levels, together with a dual-attention weighted fusion mechanism that adaptively enhances discriminative representations. In addition, a hybrid unit is embedded into the C2f module to strengthen shallow-detail perception, while a cross-level local-aware fusion head further improves the coordination between classification and localization. Extensive experiments on the VisDrone2019 and UAVDT datasets demonstrate that AeroFineFusion-Det consistently improves detection performance over representative detectors. Compared with YOLOv8s, the proposed method improves AP \(_{50:95}\) 50 : 95 by 4.4% and AP \(_s\) s by 5.3% on VisDrone2019, while reducing model parameters by 33.9%. Although additional feature-fusion computation is introduced, the experimental results validate the effectiveness of the proposed design for UAV aerial small-object detection.