基于FCOS算法的地下目标重建方法

Underground target reconstruction method based on FCOS algorithm

摘要:

针对从复杂多样的探地雷达(GPR)成像中检测和定位被埋藏的物体会耗费大量人力时间成本的问题，提出一种基于深度学习的方法。采用一阶全卷积目标检测算法(FCOS)对任意目标进行定量分析，然后对目标区域进行跟踪与聚类标记，曲线拟合获取地下目标的精确位置，重构被埋藏的地下目标信息。仿真结果表明，该方法避免了传统处理算法所需的对数据进行复杂的计算，能够快速检测到目标，并且能对目标的位置与介电属性进行高精确度估计，在深度上的定位误差不大于3 cm。该方法有效实现了地下场景重构目标的位置、深度和大小。

Abstract:

Detecting and locating buried objects from complex and diverse Ground Penetrating Radar(GPR) imaging is labor-intensive and time intensive. A method based on deep learning is proposed. The quantitative analysis on arbitrary targets is performed by using the Fully Convolutional One-Stage(FCOS) object detection algorithm. The target area is tracked and labeled with clustering tags. The precise location of underground target is obtained by the curve fitting. The information is reconstructed for the buried underground target. The simulation results show that this method avoids the complex calculation required by the traditional processing algorithm, and can quickly detect the target. The position and dielectric properties of the target are estimated with high precision, and the positioning error in depth is below 3 cm. Therefore, this method effectively realizes the reconstruction of the position, depth and size of the target in the underground scene.