基于深度展开网络的SFGPR压缩感知成像方法

SFGPR Compressive Sensing Imaging Method Based on Deep Unfolding Network

针对频率步进探地雷达(SFGPR)传统压缩感知成像方法中参数选取敏感、成像精度较低的问题,提出一种基于深度展开网络的SFGPR压缩感知成像方法。该方法首先将快速迭代收缩阈值算法的迭代过程映射到深度网络架构中,然后加入卷积神经网络模块作为成像区域的稀疏表示及其逆过程,需要手动调整的参数设置为可学习的网络参数,最后使用经过杂波抑制的降采样回波数据对网络进行训练和测试。仿真和实测数据处理结果表明该方法能够在无需人工调优参数的情况下,提高地下目标的成像精度。

Aiming at the problems of sensitive parameter selection and low imaging accuracy in the traditional compressive sensing imaging method of stepped frequency ground penetrating radar(SFGPR),a SFGPR compressive sensing imaging method based on deep unfolding network is proposed.This method first maps the iterative process of the fast iterative shrinkage threshold algorithm to the deep network structure,and then adds the convolutional neural net⁃work module as the sparse representation of the imaging area and its inverse process.The parameters that need to be manually adjusted are set to learnable network parameters.Finally,the network is trained and tested using the down⁃sampling echo data after clutter suppression.The simulation and measured data processing results show that this method can improve the imaging accuracy of underground targets without manual adjusting parameters.