← 研踪 PaperTrailarXiv:2611.19820 · v1
Hermitian Positive Definite Neural Networks for Matrix-CFAR Geometric Signal Detection in Nonhomogeneous Clutter
首次提交 2026-11-15 14:00
版本更新 2026-11-15 14:00
雷达观测 2026-11-17 12:06
摘要
Matrix information geometry (MIG) detectors exploit the Riemannian geometry of Hermitian positive definite (HPD) matrices to achieve robust object detection in nonhomogeneous clutter. The matrix constant false alarm rate (matrix-CFAR) method is a representative MIG detector that relies on the geometric structures with respect to specific metrics. In most cases, computing representative HPD covariance matrices requires numerical methods, typically involving computationally intensive iterative optimization algorithms. Although explicit formulas are available for certain cases, such as the log-Euclidean (LE) mean, they are generally unavailable for many other matrix means and medians. To address this challenge, we propose integrating the Hermitian positive definite neural network (HPDNet) into the matrix-CFAR framework, replacing conventional numerical optimization procedures with a learned, geometry-preserving mapping that directly estimates representative HPD covariance matrices from data. In the simulations, the framework is evaluated for MIG detectors using both means and medians associated with the affine-invariant Riemannian metric, the LE metric, and the Bures--Wasserstein metric, on both simulated compound-Gaussian clutter and real IPIX sea clutter. In both cases, the HPDNet is trained solely on simulated data to assess its applicability to the study of real data. The results show a significant reduction of computational cost in testing, while maintaining detection performance and robustness.
作者备注:11 pages, 9 figures
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-15 14:00 · 最新观测版本
Hermitian Positive Definite Neural Networks for Matrix-CFAR Geometric Signal Detection in Nonhomogeneous Clutter
Matrix information geometry (MIG) detectors exploit the Riemannian geometry of Hermitian positive definite (HPD) matrices to achieve robust object detection in nonhomogeneous clutter. The matrix constant false alarm rate (matrix-CFAR) method is a representative MIG detector that relies on the geometric structures with respect to specific metrics. In most cases, computing representative HPD covariance matrices requires numerical methods, typically involving computationally intensive iterative optimization algorithms. Although explicit formulas are available for certain cases, such as the log-Euclidean (LE) mean, they are generally unavailable for many other matrix means and medians. To address this challenge, we propose integrating the Hermitian positive definite neural network (HPDNet) into the matrix-CFAR framework, replacing conventional numerical optimization procedures with a learned, geometry-preserving mapping that directly estimates representative HPD covariance matrices from data. In the simulations, the framework is evaluated for MIG detectors using both means and medians associated with the affine-invariant Riemannian metric, the LE metric, and the Bures--Wasserstein metric, on both simulated compound-Gaussian clutter and real IPIX sea clutter. In both cases, the HPDNet is trained solely on simulated data to assess its applicability to the study of real data. The results show a significant reduction of computational cost in testing, while maintaining detection performance and robustness.
查看此版本解读 →查看此版本原文 ↗