Scholarlink

<p>Object detection, as a core technology in computer vision, has achieved large-scale application in multiple fields by leveraging deep learning. However, the detection of small and dense targets in complex scenarios remains a challenging problem that urgently needs to be overcome. This issue becomes even more prominent in the task of unmanned aerial vehicle (UAV) remote sensing images due to the constraints of data, environment, and the inherent characteristics of the targets. Therefore, this paper proposes the UAV remote sensing image detection model NCAF-YOLO based on the improvement of YOLOv8s. This model improves the YOLOv8s network structure, designs the C2f-DCN module, integrates the CARAFE module, and proposes the FL-ShapeIoU loss function. The improved NCAF-YOLO model outperforms the YOLOv8s model on the VisDrone2019 dataset, with an increase of 5.9% in precision, 7.0% in recall, and 8.3% in <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(mAP_{50}\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>m</mi> <mi>A</mi> <msub> <mi>P</mi> <mn>50</mn> </msub> </mrow> </math></EquationSource> </InlineEquation>. Through structural optimization, the size of the model has been reduced, which provides favorable conditions for its potential deployment on mobile or edge devices. To verify the performance of the method, an extension experiment was conducted on the AI-TOD dataset. The comparison results show that the proposed method has significantly improved the accuracy of small-target detection on other benchmark datasets.</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

Ncaf-yolo: a remote sensing image detection model for unmanned aerial vehicles’ vision

  • 期刊论文

  • 发表日期：

    2026年5月14日

查看全文
  • Lei Tian,
  • Furun Guo,
  • Yanhao Guo,
  • Zhanhao Yang,
  • Chengshen Lao

摘要

Object detection, as a core technology in computer vision, has achieved large-scale application in multiple fields by leveraging deep learning. However, the detection of small and dense targets in complex scenarios remains a challenging problem that urgently needs to be overcome. This issue becomes even more prominent in the task of unmanned aerial vehicle (UAV) remote sensing images due to the constraints of data, environment, and the inherent characteristics of the targets. Therefore, this paper proposes the UAV remote sensing image detection model NCAF-YOLO based on the improvement of YOLOv8s. This model improves the YOLOv8s network structure, designs the C2f-DCN module, integrates the CARAFE module, and proposes the FL-ShapeIoU loss function. The improved NCAF-YOLO model outperforms the YOLOv8s model on the VisDrone2019 dataset, with an increase of 5.9% in precision, 7.0% in recall, and 8.3% in \(mAP_{50}\) m A P 50 . Through structural optimization, the size of the model has been reduced, which provides favorable conditions for its potential deployment on mobile or edge devices. To verify the performance of the method, an extension experiment was conducted on the AI-TOD dataset. The comparison results show that the proposed method has significantly improved the accuracy of small-target detection on other benchmark datasets.