8 October 2026 in SSRN Electronic Journal
Robust surveillance within resilient cyber-physical systems (CPS) demands continuous tracking of spatially distributed physical intensity sources alongside uncooperative moving obstacles. Conventional multi-sensor systems rely on loosely-coupled decision-level fusion, processing macroscopic spatial topology and micro-level physical fields independently. However, moving obstacles act as dense, heterogeneous geometric attenuators, drastically warping the local physical field and causing severe intensity fluctuations. Consequently, decoupled algorithmic pipelines frequently misclassify dynamic physical shielding as abnormal source displacements. To address this fundamental limitation, we propose a tightly-coupled physics-informed cross-modal observation fusion framework. Specifically, discrete geometric point clouds are math-ematically transformed into a continuous dynamic occupancy grid. We introduce a fast voxel traversal operator that explicitly models the physical attenuation of spatial fields traversing through dynamic geometric obstacles. This geometric-physical constraint is seamlessly inte-grated into a physics-informed particle filtering framework to formulate a highly robust fusion likelihood. Crucially, to realize anomaly decoupling under severe masking, the system intrin-sically integrates kinematic tracking priors, employing a Kullback-Leibler divergence metric between the decoupled source posterior 𝑃𝑆 and authorized carrier posterior 𝑃𝐶 for sub-second causal inference. Utilizing nuclear safeguarding (gamma-ray tracking) as a rigorous, high-stakes case study, experiments with a sparse 4-node sensing network validate the architecture. Results demonstrate that our framework substantially reduces multi-source localization errors and compresses anomaly decoupling delays to sub-second levels, establishing a generalized physics-informed observation paradigm for intelligent CPS surveillance.
30 August 2026 in SSRN Electronic Journal
Reconstructing three-dimensional volumetric patterns from sparse, near-field two-dimensional projections remains a fundamental challenge in computational imaging and multi-modal perception, where depth ambiguity and occlusion-induced phantom artefacts routinely undermine reconstruction fidelity. To address this ill-posed inverse problem, we propose a Hierarchical Physics-Constrained Information Fusion architecture that progressively collapses spatial uncertainty through sequential physical, statistical, and geometric constraints. Rather than relying on late-stage heuristic overlay, the proposed architecture operates across three inference tiers. In Tier 1, the Physical Constraint Prior, a multi-channel convolutional neural network surrogate acts as a physics-informed encoder, mapping discrete hardware parameters to spatial tensors and circumventing computationally prohibitive Monte Carlo simulations while tightly bounding the physical solution space. In Tier 2, the Statistical Uncertainty Estimation, a dynamic-response generalized expectation-maximization algorithm synthesizes distance-correct response matrices, adaptively resolving near-field depth mismatch and restoring focal sharpness with provable monotonic log-likelihood ascent. In Tier 3, the Geometric Uncertainty Collapse, a distributed Logarithmic Opinion Pool consensus protocol cross-validates two-dimensional probability heatmaps against three-dimensional exteroceptive structural constraints, enforcing Boolean intersection across orthogonal views to deterministically eliminate phantom artefacts. Extensive physical experiments in heavily occluded environments demonstrate that the proposed paradigm reduces the mean three-dimensional localization error from 0.86 ± 0.29 metres to 0.05 ± 0.01 metres. This architecture establishes a generalizable probabilistic framework for sparse-view tomographic pattern recognition, bridging physics-enhanced learning and multi-agent geometric consensus.
2 September 2026 in SSRN Electronic Journal
In this paper, an intelligent mobile distributed measurement system is proposed to locate multiple unknown radioactive sources in complex environments. To overcome traditional measurement bottlenecks such as multi-source interference, particle degeneracy in zero-signal areas, and epistemic uncertainty, the system integrates Dynamic Voronoi Partitioning (DVP) with a novel dual-layer data fusion mechanism. Firstly, a DVP mechanism dynamically allocates exclusive sensing sub-regions based on real-time node topologies, ensuring concurrent and conflict-free data acquisition. Secondly, a dual-layer map-driven architecture—incorporating a Probability of Detection (POD) map for hypothesis erasing and a Gaussian situational map for continuous intensity field reconstruction—is designed to guide mobile nodes toward highly uncertain areas. Crucially, the system is explicitly initialized with rough priors containing severe position deviations and false alarms. By rigorously fusing real-time counts per second (CPS) measurements, the nodes autonomously filter out initial errors and reconstruct a global radiation field intensity map in real time. Simulation results in a 5-node, 1-source (5V1) scenario confirm that the proposed framework compresses localization measurement errors by 66.7\% (from 0.30 m to 0.10 m) compared to standard non-cooperative tracking. Furthermore, rigorous real-world physical experiments involving three robots and two sources (3V2) validate the structural superiority of the framework. Compared to an uncoordinated three-robot baseline swarm, the proposed cooperative measurement approach eliminates spatial aggregation, drastically reducing redundant total travel distance from 56.8 m to 21.5 m and compressing search time from 35 s to 19 s, while maintaining mean localization errors of 0.20 m for both sources.