Aug 2026· Communications in Transportation Research· 0 citations
TL;DR
The results show that the proposed PINN framework effectively improves the accuracy of DAS-based TSP compared to the solely data-driven baselines, and few-shot transfer learning can enhance model performance in previously unseen scenarios, demonstrating the potential of the proposed framework for adaptation to site-specific deployment conditions.
Abstract
This paper introduces distributed acoustic sensing (DAS) as an emerging sensing technique for large-scale traffic state perception (TSP) on expressways. By enabling optical fibers to operate as dense sensing arrays, DAS offers a promising solution for continuous traffic monitoring along expressway corridors. However, the raw traffic information extracted from DAS is inaccurate, which limits its direct application to large-scale TSP. This study proposes a physics-informed neural network (PINN) framework for DAS-based TSP. A physics-based DAS simulation platform is developed and validated against field observations which provides a flexible testbed for generating data under diverse traffic scenarios. On this basis, two network architectures, ResUNet and Fourier Neural Operator, are investigated in combination with three types of traffic flow constraints, namely the Lighthill-Whitham-Richards (LWR) model, LWR with a fundamental diagram, and the Aw-Rascle-Zhang model. Totally, 16 PINN models are constructed and evaluated. The results show that the proposed PINN framework effectively improves the accuracy of DAS-based TSP compared to the solely data-driven baselines. Among the physical constraints considered, the LWR-based models show the best overall performance. In addition, few-shot transfer learning can enhance model performance in previously unseen scenarios, demonstrating the potential of the proposed framework for adaptation to site-specific deployment conditions.
Continuous traffic monitoring is critical for accurate structural load assessment and fatigue life estimation of highway bridges. However, conventional vision-based methods suffer from limitations such as line-of-sight restrictions, susceptibility to adverse weather and lighting conditions, and limited spatial coverage. Distributed acoustic sensing (DAS) offers a robust alternative by repurposing existing telecommunications dark fiber into dense, kilometer-scale sensor arrays. Nevertheless, interpreting the complex DAS signals generated by vehicular traffic remains challenging due to overlapping dynamic signatures and the scarcity of ground-truth data for model training. To overcome this, we present a cross-modal (vision-to-optic) supervision framework that transforms existing fiber infrastructure into a traffic monitoring system. We deployed a synchronized camera–DAS testbed along a roadway segment served by dark fiber. Video data is processed using modern computer vision foundation models SAM3 to automatically extract vehicle trajectories and classifications. These camera-derived labels supervise a deep sequence learning model trained on the corresponding DAS strain data. Once trained, the fiber-optic system independently achieves accurate vehicle detection, classification, localization, and speed estimation. Finally, we demonstrate how these continuous, DAS-derived traffic metrics can be directly translated into dynamic load profiles, providing a scalable, continuous monitoring solution for bridge fatigue and structural health assessment.
Cong Chen, Shenghan Zhang· e-Journal of Nondestructive...· 0 citations
Distributed acoustic sensing (DAS), together with broader distributed fibre-optic sensing (DFOS), has emerged over the past decade as a candidate technology for large-scale infrastructure monitoring, while machine learning (ML) is increasingly used to interpret its high-throughput data. This study presents a PRISMA 2020-guided bibliometric and systematic-mapping review of this convergence. A structured search of Web of Science, Scopus, and IEEE Xplore returned 291 records published between 2015 and 2025; after duplicate removal and two stages of screening, a final analytical corpus of 31 publications was established, of which 16 are primary DAS-with-ML infrastructure studies. Publication output rises sharply to seven papers in 2024 and twelve in 2025. Among the primary studies, convolutional neural networks (CNNs), recurrent models, and CNN-LSTM hybrids are prominent, with more recent use of few-shot learning, graph neural networks (GNNs), generative models, and self-supervised learning. Perimeter security is the most represented application domain, followed by traffic, railway, and structural-health monitoring. Critical synthesis shows that the apparent dominance of CNNs is partly explained by the natural two-dimensional spatiotemporal representation of DAS data and by mature, reusable CNN pipelines, rather than evidence of universal superiority. Cross-study comparison remains limited by heterogeneous sensing environments, inconsistent acquisition and label reporting, non-uniform evaluation protocols, sparse computational reporting, and minimal long-term robustness evaluation. Five research gaps are identified: cross-infrastructure generalisation, standardised benchmarks and open datasets, physics-data-driven integration, interpretability and uncertainty, and operationally robust edge deployment.
Abd Rahim Saidin, I. M. Nawi· International Journal of Adv...· 0 citations
Fibre optic sensing, such as distributed acoustic sensing (DAS), has become a widespread technology for geophysical studies. To process the large-scale datasets produced by DAS, several machine learning methods have been proposed. However, without standardization of data and models, these methods lack comparability and interoperability. This introduces a gap between model developers and practitioners analyzing DAS data and inhibits adoption of deep learning for DAS. To address these limitations, here we present SeisBench DAS, an extension to the SeisBench library for machine learning in seismology. SeisBench DAS defines standard formats for DAS benchmark datasets, including standardised metadata and labels, and DAS models. It builds on the xdas framework for data ingestion and virtual array handling, and on PyTorch for reading and applying the machine learning models. Importantly, SeisBench provides an engine to efficiently apply deep learning models to diverse formats of DAS data, bridging the gap between model developers and practitioners. SeisBench DAS is designed as an open and extensible framework, allowing to easily incorporate future developments in deep learning for DAS.
Jannes Münchmeyer, Han Xiao, Frederik Tilmann· 0 citations
In distributed acoustic sensing (DAS), a 100 km-scale optical fiber is transformed into ∼100 000 independent, synchronized, and meter-spaced acoustic sensors, each having the combined bandwidth of geophones, accelerometers, and microphones. Recently, there has been significant interest in applying machine learning to DAS. Machine learning can extract actionable information from otherwise esoteric DAS data. However, real-time, commercial implementation remains a challenge due to the intensive computation required for ∼100 000 sensors sampling at kHz rates (Gbps-scale data). This work overviews the real-time, commercial implementation of machine learning for DAS. This includes novel, domain-knowledge-based representations of DAS data in multiple dimensions (e.g., space, time, and frequency), their use in deep learning-based artificial neural networks, and real-time deployment via edge-based GPUs. We overview ∼100 000 km of cumulative commercial and academic DAS deployments that are disruptively innovating situational awareness in physical security, physical monitoring of the natural environment, and physical monitoring of critical infrastructure, such as telecommunications, transportation, and energy.
Giovanni Milione, D. Hill· Journal of the Acoustical So...· 0 citations
Integrated sensing and communication (ISAC) is envisioned as a key functionality of sixth-generation wireless systems, enabling spatial perception of the environment through the same hardware, spectrum, and waveforms used for data transmission. Realizing this vision is challenged by modeling mismatch between assumed signal models and real operating conditions, which conventional model-based signal processing struggles to address. This thesis studies two such forms of mismatch: (i) hardware impairments and (ii) complex propagation conditions such as non-line-of-sight. This thesis investigates machine learning as a data-driven complement to model-based signal processing in ISAC, with contributions spanning supervised, semi-supervised, unsupervised, and self-supervised learning regimes across four appended papers.Paper A proposes a model-based end-to-end learning framework that jointly optimizes the ISAC transmitter and sensing receiver under antenna-array hardware impairments, enabling supervised calibration through differentiable ISAC algorithms. The results show that learning the parameterized impairments outperforms standard model-based calibration while generalizing to unseen scenarios. Paper B reduces the labeling cost of this framework, showing that semi-supervised learning matches fully supervised performance with far less labeled data, while purely unsupervised learning falls short. Paper C closes this gap in the multi-target case, calibrating transmitter and receiver impairments with no labeled data via an approximation of the channel gradient, performing close to supervised calibration and generalizing better across signal-to-noise ratios.Paper D removes the reliance on parametric line-of-sight channel models, addressing active positioning through a self-supervised channel-charting (CC) framework augmented with a digital twin. Matching large-scale channel state information features, the proposed method outperforms the CC state of the art and remains robust to modeling mismatch and distribution shifts.
Distributed acoustic sensing (DAS) adds a spatial component to conventional acoustic recordings, providing positional insights for baleen whale monitoring. Compared to traditional PAM methods, DAS drastically increases data volumes (many TBs/day), requiring real-time analysis solutions. Deep learning methods are particularly well-suited for efficient data analysis of these datasets and ultimately, the generation of actionable data products, but require labeled data for training and evaluation. Here, we present a dedicated and efficient workflow for creating strong labels from DAS data, building on the DAS4Whales Python package. Our process includes: filtering and compressing datasets using sparse matrices in the f–k domain (ranging from 66 to 99% compression); standardizing the data structure and storage format by interpolating datasets to a sampling rate of 200 Hz and an element spacing of 8 m; and a user-interface prototype which allows for fast and intuitive labeling. We use this workflow to strongly label data from Svalbard, Norway, and offshore of Oregon, USA, with fin and blue whale calls, in multiple linked representations that include time, space, and frequency. This dataset is then used to train and evaluate a deep learning model to detect baleen whale calls in DAS data.
Eric R. Snyder, Holger Klinck, Léa Bouffaut· Journal of the Acoustical So...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
