A Magnetization Vector Inversion Method Based on Block Sparse Bayesian Learning

Tianjiao Chen

doi:10.1109/tgrs.2026.3707004

Return

A Magnetization Vector Inversion Method Based on Block Sparse Bayesian Learning

2026-06-24

PRE

DOI：10.1109/tgrs.2026.3707004

Original

Original request for help

Save

Abstract

En 中文

Under the joint effects of remanent and induced magnetization, magnetization vector inversion (MVI) is an important tool for interpreting magnetic anomaly data. However, conventional MVI faces challenges in both accuracy and computational efficiency due to fine-grid discretization and solution nonuniqueness. Nonuniqueness leads to spurious anomalies and misidentification, while fine-grid discretization causes a rapid increase in the number of unknowns, raising computational costs. To address these issues, this article proposes an MVI method based on block sparse Bayesian learning (BSBL). The three magnetic components in each grid cell are treated as a single block. A block-sparsity prior is imposed to enhance significant magnetic cells and suppress nontarget regions. As a result, the inversion accuracy is improved. To enhance computational efficiency and meet engineering requirements, the proposed method automatically compresses and removes insignificant cells during iterations to reduce the solution size and concentrates the major computations on small-scale solutions in the observation domain, avoiding large-matrix operations in the high-dimensional model domain. Meanwhile, the noise level and sparsity degree are learned adaptively to reduce manual parameter tuning. In addition, a unit-block prior is adopted to reduce the number of hyperparameters and the computational cost. The inversion results of simulated and measured data demonstrate that the proposed method improves inversion accuracy while significantly reducing computational cost, thereby providing reliable support for high-precision and high-efficiency interpretation of magnetic anomalies under complex geological conditions.

Keywords:

Block sparse Bayesian learning (BSBL)

computational efficiency

inversion accuracy

magnetization vector inversion (MVI)

remanent magnetization