基于深度学习的探地雷达地下目标类别检测方法
Deep-Learning Method for Underground Target Detection in Ground-Penetrating Radar
摘要
针对目前城市地下空间探测中探地雷达(Ground Penetrating Radar,GPR)图像类别检测时存在的精度低、结构异常特征双曲线识别困难、模型计算复杂度高等问题,文章提出一种基于改进YOLOv11的轻量化地下目标检测模型RBD-YOLOv11.利用GprMax软件进行金属管、PVC管及钢筋网三个类别的GPR图像仿真并通过CycleGAN对仿真图像进行域迁移,构建数量充足且高质量的数据集.首先,采用重参数化视觉Transformer轻量主干网络,减少模型参数量与计算量.其次,在Neck部分采用多尺度特征融合网络双向特征金字塔网络和空间-通道解耦下采样模块,以提升多尺度特征融合能力,降低计算成本.最后,采用动态上采样实现复杂背景下结构异常特征双曲线的有效识别.该方法能在更轻量化条件下实现较高的识别准确率,在GPR图像类别检测领域具有优势.
To address low accuracy,complex hyperbolic feature recognition of structural anomalies,and high computational demands in urban underground image detection using ground-penetrating radar(GPR),this study proposes a lightweight detection model named RBD-YOLOv11,based on an improved YOLOv11n.A high-quality dataset comprising metal pipes,PVC pipes,and rebar meshes was constructed using GprMax for simulation and CycleGAN for domain transfer.Firstly,a re-parameterization vision transformer is adopted as a lightweight backbone network to significantly reduce model parameters and computational load.Secondly,a bidirectional feature pyramid network and spatial-channel decoupled downsampling modules are integrated into the Neck architecture to enhance multiscale feature fusion capabilities while reducing computational costs.Finally,a lightweight dynamic upsampler is employed to effectively recognize structural anomaly hyperbolas under complex backgrounds.Experimental results show that the proposed method achieves high recognition accuracy with a lightweight architecture,thereby benefiting GPR image category detection.
年,卷(期):
2026, 47(5)
分类号:
TP391.41
U418
P631.3
机标分类号:
TP391
TP183
P631.3
在线出版日期:
2026年11月4日