SLRD-IMC: Structured Low-Rank Matrix Decomposition for Magnetic Anomaly Detection With Iteratively Merging Clusters

Magnetic anomaly detection (MAD) is pivotal for geophysical prospecting, particularly in defense and resource exploration. However, weak signals are often corrupted by electromagnetic interference and industrial noise, while conventional methods excessively focus on signal-to-noise ratio (SNR) improvement at the expense of structural fidelity. This study introduces a novel MAD framework integrating structured low-rank matrix decomposition with iteratively merging clusters (SLRD-IMC). First, implementing a frequency-domain structured Hankel low-rank transformation coupled with sparse structuring to amplify the intrinsic low-rank characteristics of magnetic signals. Second, incorporating a nonlocal similarity-based iterative clustering strategy that synergizes with adaptive singular value thresholding to achieve simultaneous noise suppression and anomaly detail retention. Third, developing a prototype detection system for comparative validation against four state-of-the-art techniques. Experimental results demonstrate SLRD-IMC’s superior performance across SNR from -5 dB to 20 dB, exhibiting average improvements of 32% in SNR, 100% in structural similarity (SSIM), and 42% reduction in mean square error (MSE) versus benchmark methods. In particular, when the initial SNR plunges to -10 dB, a scenario in which conventional methods fail, SLRD-IMC maintains robust detection capability. These findings conclusively validate the method’s exceptional noise resilience and precision in extreme conditions.