基于贝叶斯压缩感知的频率步进探地雷达成像算法

Stepped-frequency GPR imaging algorithm based on bayesian compressive sensing

针对传统压缩感知频率步进探地雷达成像算法存在计算量大和对噪声和重建正则化参数敏感的问题,提出一种基于稀疏贝叶斯学习的贝叶斯压缩感知成像算法。该成像算法的核心通过稀疏贝叶斯线性回归模型中相关向量机的学习来实现对探测场景反射系数的重构。仿真结果表明,相比其他的经典算法,所提成像算法能够更好地利用了探测场景的统计先验信息,能够更好地兼顾重构精度和计算效率。

The stepped-frequency ground penetrating radar（ GPR） based on the traditional compressive sensing is rather computationally intensive and sensitive to the regularization parameter. To solve the above problem,animaging algorithm based on Bayesian compressive sensing（ BCS） is proposed in the paper. Within the sparse Bayesian linear regression model,the proposed BCS-based imaging algorithm uses the relevance vector machine to reconstruct the reflectivity of the investigation scene. The numerical simulation results showthat the proposed imaging algorithm can take advantage of the prior statistical information of the scene reflectivies and achieve both reconstruction accuracy and computation efficiency compared with the traditional CS-based imaging methods.