Target Detection Within Nonhomogeneous Clutter Via Total Bregman Divergence-Based Matrix Information Geometry Detectors

Xiaoqiang Hua, Yusuke Ono, Linyu Peng, Yongqiang Cheng, Hongqiang WangIEEE TSP 2021 · DOI · arXiv

Abstract Information divergences are commonly used to measure the dissimilarity of two elements on a statistical manifold. Differentiable manifolds endowed with different divergences may possess different geometric properties, which can result in totally different performances in many practical applications. In this paper, we propose a total Bregman divergence-based matrix information geometry (TBD-MIG) detector and apply it to detect targets emerged into nonhomogeneous clutter. In particular, each sample data is assumed to be modeled as a Hermitian positive-definite (HPD) matrix and the clutter covariance matrix is estimated by the TBD mean of a set of secondary HPD matrices. We then reformulate the problem of signal detection as discriminating two points on the HPD matrix manifold. Three TBD-MIG detectors, referred to as the total square loss, the total log-determinant and the total von Neumann MIG detectors, are proposed, and they can achieve great performances due to their power of discrimination and robustness to interferences. Simulations show the advantage of the proposed TBD-MIG detectors in comparison with the geometric detector using an affine invariant Riemannian metric as well as the adaptive matched filter in nonhomogeneous clutter.

Introduction

Radar target detection in nonhomogeneous clutter environments presents a significant challenge for traditional statistical methods. Conventional detectors, such as the Sample Covariance Matrix (SCM) or the Adaptive Matched Filter (AMF), rely on the assumption of uniform clutter across training cells. In real-world scenarios, clutter is often "spiky" or contains outliers, which can "poison" the Clutter Covariance Matrix (CCM) estimation and lead to performance degradation.

Total Bregman Divergence-Based Matrix Information Geometry

To address these limitations, this paper leverages Matrix Information Geometry (MIG). By representing radar signals as Hermitian Positive-Definite (HPD) matrices, the detection problem is transformed into measuring distances on a curved mathematical manifold.

The core innovation is the application of Total Bregman Divergence (TBD) to this manifold. Unlike standard Bregman divergences, TBD includes a normalization term that makes it more robust to outliers. The authors propose three specific detectors: Total Square Loss (TSL), Total Log-Determinant (TLD), and Total von Neumann (TVN).

Isosurfaces of different divergences on the HPD matrix manifold.

As illustrated in Figure 2, different divergences create distinct "shapes" (isosurfaces) on the matrix manifold. The TBD-based shapes provide superior separation between target and clutter compared to the standard "balls" created by the Affine Invariant Riemannian Metric (AIRM).

Detection Methodology

The TBD-MIG detection process involves modeling each data sample as an HPD matrix and estimating the background clutter using the "TBD Mean" of secondary training data. A critical theoretical finding is that TBD-based means have "bounded influence functions." This ensures that even if an outlier is infinitely powerful, its ability to bias the clutter estimate is strictly limited—a major advantage over standard geometric means.

Experimental Results and Performance

The proposed detectors were evaluated using simulated Gaussian and Compound-Gaussian clutter models. A key performance metric is data efficiency, particularly in regimes where the number of training samples is limited.

Comparison of offset error versus the number of training samples for various detectors.

Figure 3 demonstrates that the TBD-MIG detectors, especially the Total Log-Determinant (TLD) variant, achieve significantly lower offset errors than the standard SCM and AIRM-based methods as the number of samples increases.

Key results include:

  • Superior Detection: TBD-MIG detectors consistently outperformed the standard AMF and AIRM-MIG detectors in both Gaussian and non-Gaussian clutter.
  • Robustness to Outliers: Performance degraded much more slowly than competitors as the number of outliers increased.
  • TLD is King: Among the proposed versions, the TLD detector generally showed the best robustness and highest probability of detection ($P_d$).

Conclusion

The TBD-MIG framework offers a robust solution for radar detection in complex, nonhomogeneous environments. By exploiting the geometric properties of the HPD manifold and the robustness of Total Bregman Divergence, these detectors provide significant performance gains over traditional methods. Future research will focus on real-world validation and optimizing computational complexity for real-time implementation.

Summary updated

Abstract

Information divergences are commonly used to measure the dissimilarity of two elements on a statistical manifold. Differentiable manifolds endowed with different divergences may possess different geometric properties, which can result in totally different performances in many practical applications. In this paper, we propose a total Bregman divergence-based matrix information geometry (TBD-MIG) detector and apply it to detect targets emerged into nonhomogeneous clutter. In particular, each sample data is assumed to be modeled as a Hermitian positive-definite (HPD) matrix and the clutter covariance matrix is estimated by the TBD mean of a set of secondary HPD matrices. We then reformulate the problem of signal detection as discriminating two points on the HPD matrix manifold. Three TBD-MIG detectors, referred to as the total square loss, the total log-determinant and the total von Neumann MIG detectors, are proposed, and they can achieve great performances due to their power of discrimination and robustness to interferences. Simulations show the advantage of the proposed TBD-MIG detectors in comparison with the geometric detector using an affine invariant Riemannian metric as well as the adaptive matched filter in nonhomogeneous clutter.
BibTeX
@article{hua2021target,
  author = {Xiaoqiang Hua and Yusuke Ono and Linyu Peng and Yongqiang Cheng and Hongqiang Wang},
  title = {{Target Detection Within Nonhomogeneous Clutter Via Total Bregman Divergence-Based Matrix Information Geometry Detectors}},
  journal = {IEEE Transactions on Signal Processing},
  year = {2021},
  doi = {10.1109/tsp.2021.3095725},
}
Figure