Distributed acoustic sensing (DAS), which can provide dense spatial and temporal measurements using optical fibers, is quickly becoming critical for large-scale infrastructure monitoring. However, anomaly detection in DAS data is still challenging owing to the spatial correlations between sensing channels and nonlinear temporal dynamics. Recent approaches often disregard the explicit sensor layout and instead handle DAS data as two-dimensional images or flattened sequences, eliminating the spatial topology. This work proposes GraphDiffusion, a novel generative anomaly-detection model that combines a conditional denoising diffusion probabilistic model (DDPM) and a graph neural network (GNN) to overcome these limitations. By treating each channel as a graph node and building edges based on Euclidean proximity, the GNN explicitly models the spatial arrangement of DAS sensors, allowing the network to capture local interchannel dependencies. The conditional DDPM uses iterative denoising to model the temporal dynamics of standard signals, enabling the system to detect deviations without the need for anomalies. The performance evaluations based on real-world DAS datasets reveal that GraphDiffusion achieves 98.2% and 98.0% based on the area under the curve (AUC) of the F1-score at K different levels (F1_K-AUC), an AUC of receiver operating characteristic (ROC) at K different levels (ROC_K-AUC), outperforming other comparative models.

Distributed acoustic sensing (DAS) systems capture complex, nonlinear wave propagation across fiber-optic cables. Conventional event classification architectures, constrained by static physical topologies or isolated spatial grids, fail to effectively adapt to the dynamic feature relationships associated with such events, particularly when modeling complex spatiotemporal interactions across sensor arrays. To resolve these structural limitations, we introduce the Spectro-Spatial Dynamic Graph Neural Network (S2-DyGNN), whose architecture couples a two-dimensional frequency–time convolutional front-end with a dual-matrix graph neural network (GNN). First, the convolutional module extracts spectro-temporal features, explicitly capturing localized acoustic dynamics independent of inter-sensor interference. Subsequently, the graph module constructs a dual-matrix topology, fusing a static physical distance prior with a data-driven adjacency matrix that recalculates spatial connections frame by frame from input signals. When evaluated on a highly skewed nine-class DAS field dataset, S2-DyGNN outperformed other conventional models by achieving a peak macro-averaged F1-score of 86.6% and an overall accuracy of 94.0%. The dual-matrix graph topology prevented dominant background features from washing out sparse transient events, improving the minority “openclose” class F1-score to 55.7% compared to the 48.0% ceiling of a static graph topology. These results demonstrate that explicitly coupling localized spectro-temporal representations with physically anchored spatial topologies consistently outperforms models that process these domains in isolation, providing a highly robust and scalable solution for real-world continuous monitoring systems.