Scholarlink

<p>Routine road maintenance is essential to ensure traffic safety. This study proposes DAL-YOLO, a novel multi-object detection model specifically designed for UAV-based road maintenance. First, the feature extraction module is enhanced markedly in small object detection accuracy by integrating Deformable Attention into the C3k2 module. Second, an Adaptive Scaled Pyramid Network is introduced, dynamically fusing high-, mid-, and low-level features with adaptive fusion weights, excelling in dense targets and complex scenarios. Furthermore, a lightweight detection head, the Lightweight Shared Convolution BatchNorm Head, combines shared convolution layers with independent BatchNorm structures, improving precision and recall for dense and occluded object detection substantially. Experimental results on the road maintenance dataset show that DAL-YOLO achieves an AP of 39.8% and an AP50 of 58.8%, improving the accuracy of small, medium, and large objects notably. This model offers an efficient and accurate solution for UAV-based routine road maintenance tasks.</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

Dal-yolo: a multi-target detection model for UAV-based road maintenance integrating feature pyramid and attention mechanisms

  • 期刊论文

  • 发表日期：

    2025年5月2日

查看全文
  • Xuerui Lan,
  • Lijun Liu,
  • Xuyang Wang

摘要

Routine road maintenance is essential to ensure traffic safety. This study proposes DAL-YOLO, a novel multi-object detection model specifically designed for UAV-based road maintenance. First, the feature extraction module is enhanced markedly in small object detection accuracy by integrating Deformable Attention into the C3k2 module. Second, an Adaptive Scaled Pyramid Network is introduced, dynamically fusing high-, mid-, and low-level features with adaptive fusion weights, excelling in dense targets and complex scenarios. Furthermore, a lightweight detection head, the Lightweight Shared Convolution BatchNorm Head, combines shared convolution layers with independent BatchNorm structures, improving precision and recall for dense and occluded object detection substantially. Experimental results on the road maintenance dataset show that DAL-YOLO achieves an AP of 39.8% and an AP50 of 58.8%, improving the accuracy of small, medium, and large objects notably. This model offers an efficient and accurate solution for UAV-based routine road maintenance tasks.