基于GAMP的近场毫米波成像快速算法

Fast Near-field Millimeter-Wave Imaging Algorithm via Generalized Approximate Message Passing

  • 摘要: 传统近场毫米波均匀采样成像由于扫描时间长和计算代价大等问题无法实现实时成像。为此,该文构建了近场毫米波压缩采样成像模型及相应的观测矩阵,提出了一种基于广义近似消息传递的近场毫米波压缩采样成像快速算法。该算法将广义近似消息传递有效嵌入到期望最大化框架,加快了收敛速度;并利用快速傅里叶变换、小波滤波等方式构造了观测矩阵的快速算子,避免了大型观测矩阵的构造、存储与计算,进一步提升了算法运算速度。实验结果表明该算法可以快速、有效地从压缩采样数据中重建近场毫米波二维图像,并在重建效果与运算时间上都优于主流的快速迭代阈值收缩算法。

     

    Abstract: The conventional near-field millimeter-wave imaging from uniform sampling measurements does not work well in realizing real-time imaging due to high computational cost and long scan time. To this end, we present a near-field millimeter-wave imaging model from under-sampled measurements and the corresponding sensing matrix. Meanwhile, a fast millimeter-wave imaging algorithm based on generalized approximate message passing is proposed. The algorithm accelerates the convergence rate by embedding the generalized approximate message passing into the expectation maximization framework. In order to circumvent the construction and storage of large-scale observation matrix, we construct a fast implementation of the observation matrix using fast Fourier transform and wavelet filtering, which leads to an improved computing speed at the same time. Our experiments results confirm that the algorithm can reconstruct 2-D millimeter-wave images rapidly from under-sampled measurements, and it exhibits better performance in terms of both reconstruction quality and execution time compared with the fast iterative shrinkage-thresholding algorithm.

     

/

返回文章
返回