Scholarlink

<p>Subsurface defect detection in Ground Penetrating Radar (GPR) imagery is challenged by strong background clutter and multi-scale hyperbolic signatures. We propose FLS-YOLO, a lightweight detection framework with task-specific refinements to existing neural operators. Rather than directly stacking modules, we modify the forward computations of PConv, LWGA, and SCConv to inject GPR-oriented inductive biases. Specifically, channel partitioning is redefined using directional energy statistics, attention weights are augmented with a vertical gradient bias to model electromagnetic anisotropy, and saliency estimation incorporates second-order curvature priors to improve hyperbolic boundary reconstruction. Experiments on RTSTdataset show that FLS-YOLO achieves 0.6958 mAP@0.5 at 164 FPS with 2.41M parameters, yielding a better accuracy–efficiency trade-off for large-scale subsurface inspection.</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

FLS-YOLO: a multi-scale subsurface defect detection network via heterogeneous grouping and spatial reconstruction

  • 期刊论文

  • 发表日期：

    2026年4月15日

查看全文
  • Wenbo Zhang,
  • Zican Liu,
  • Qing Yang,
  • Xin Qiu

摘要

Subsurface defect detection in Ground Penetrating Radar (GPR) imagery is challenged by strong background clutter and multi-scale hyperbolic signatures. We propose FLS-YOLO, a lightweight detection framework with task-specific refinements to existing neural operators. Rather than directly stacking modules, we modify the forward computations of PConv, LWGA, and SCConv to inject GPR-oriented inductive biases. Specifically, channel partitioning is redefined using directional energy statistics, attention weights are augmented with a vertical gradient bias to model electromagnetic anisotropy, and saliency estimation incorporates second-order curvature priors to improve hyperbolic boundary reconstruction. Experiments on RTSTdataset show that FLS-YOLO achieves 0.6958 mAP@0.5 at 164 FPS with 2.41M parameters, yielding a better accuracy–efficiency trade-off for large-scale subsurface inspection.