Sparse Reconstruction for Near-Field MIMO Radar Imaging Using Fast Multipole Method

Efficient physics-based learned reconstruction methods for real-time 3D near-field MIMO radar imaging

“…Lastly we note that although the developed Deep2S method is quite fast with a runtime on the order of milliseconds on a standard computer, the adjoint computation in its first stage can be further accelerated using Fourier-based methods [16,34]. Moreover, exploring the performance of the developed methods with other types of 3D network architectures (such as those mimicking transforms) may improve the reconstruction quality and is a topic for future study.…”

“…Different than direct inversion methods, regularized iterative reconstruction methods incorporate additional prior information (such as sparsity) into the reconstruction process to eliminate uniqueness and noise amplification issues arising due to limited data and measurement noise. With the advent of compressed sensing (CS) theory [23], sparsitybased reconstruction is the most commonly used analytical regularization approach and has been widely studied in various imaging problems [59,60], including radar imaging both for far-field and monostatic imaging settings [24][25][26][27][28][29][30], as well as for multistatic and near-field settings [31][32][33][34].…”

“…For example, Cheng et al [33] adapted the split augmented lagrangian shrinkage algorithm (SALSA) [60] and Oktem [32] adapted the half-quadratic regularization approach [62]. There have been also some efforts to reduce the computational cost and memory usage of such sparsity-based reconstruction algorithms by exploiting the special structure of the forward problem [8,34].…”

“…Regularized reconstruction methods on the other hand incorporate additional prior information (such as sparsity) into the reconstruction process. With the advent of compressed sensing (CS) theory [23], sparsity-based reconstruction methods are the most commonly used model-based iterative inversion methods in various imaging problems including radar imaging, both for far-field and monostatic imaging settings [8,[24][25][26][27][28][29][30], as well as for multistatic and near-field settings [31][32][33][34]. Although sparsity-based methods provide better reconstruction quality than the traditional direct inversion methods at compressive settings, they suffer from high computational cost which is undesirable in real-time applications.…”

Plug-and-Play ADMM Based Radar Range Profile Reconstruction Using Deep Priors

“…Point-based and region-based features are enhanced through regularization in synthetic aperture radar (SAR) imaging, as demonstrated by in [6]. Likewise, sparsity-based regularization has found application in near-field multiple-input multiple-output (MIMO) radar imaging [7]. Nevertheless, in any regularized inversion scheme, regularizers often fall short of capturing intricate and hidden data features, as they inherently tend to emphasize fixed and predefined characteristics based on prior assumptions about the reconstructed field, whether in its original space or a transformed one.…”