Unsupervised Magnetic Anomaly Detection Method Based on Granular Ball One-Class Classification

In complex marine environments, underwater magnetic anomaly detection is challenging because target magnetic anomaly signals are typically weak and easily overwhelmed by background magnetic noise. Although deep learning-based methods have significantly improved detection capability, most existing approaches still rely on abundant labeled target data, which is difficult to obtain in practical applications. To address this challenge, this paper proposes an unsupervised underwater magnetic anomaly detection method based on Gaussian granular ball one-class classification (GBOC). A density-guided hierarchical partitioning strategy is introduced to divide the latent space into multiple compact high-density regions and construct corresponding Gaussian granular ball representations. This strategy enables more effective modeling of complex background magnetic noise and improves anomaly detection under low signal-to-noise ratio (SNR) conditions. Experimental results show that the proposed method achieves robust performance across different SNR levels in the unsupervised setting. Compared with other methods, it yields a higher detection rate and more stable results under a fixed false alarm rate. Furthermore, a semi-supervised magnetic anomaly detection method is developed by introducing a small amount of prior information on magnetic anomalies. Experimental results demonstrate that the proposed semi-supervised method can further improve detection accuracy while maintaining good robustness and stability.