Underwater Acoustic Monitoring Networks (UAMNs) are vital for maritime situational awareness but pose significant security risks when deployed by non-cooperative entities for covert reconnaissance. Current detection methods focus primarily on signal-level analysis of individual targets, failing to account for the tactical coordination within "swarmed and networked" underwater threats. This paper proposes a detection framework based on spatiotemporal graph representation and sequential reasoning (ST-GRSR) to identify network existence from a structural perspective. By introducing connectivity and scale constraints, the detection task is formulated as a blind inference problem of unknown topologies. A multi-dimensional feature space integrating physical, protocol, spatial, and behavioral attributes is constructed to characterize the sparse and heterogeneous nature of non-cooperative targets. We then develop an inductive spatiotemporal graph neural network that combines Graph Sample and Aggregate (GraphSAGE) for spatial neighborhood aggregation with Gated Recurrent Units for capturing long-term dependencies in uncertain observation sequences. This architecture enables feature-to-link mapping to determine network existence. Experimental results using a Network Simulator-3 (NS-3, AquaSim) simulated dataset demonstrate that the proposed method achieves over 90% accuracy in dynamic adversarial scenarios. The framework significantly outperforms benchmark models in precision and robustness, providing a theoretical foundation for identifying non-cooperative entities in complex maritime environments.
Sun et al. (Fri,) studied this question.