In this article, the problem of detecting undesirable transitions in deterministic state transition systems in the absence of complete and reliable anomaly labels is considered. In contrast to classical methodologies of anomaly detection in time series, which model the distribution of trajectories as a stochastic process, the analyzed system is deterministic - each state is uniquely determined by the previous state and the operation performed. The proposed framework, designated as USTAD (Uncertainty-aware State Transition Anomaly Detection), integrates deterministic state-trajectory reconstruction, deep sequential modeling, and a threshold-based decomposition of the model score into confidence components. This approach facilitates the identification of redundancies and structural returns in the trajectory, without manual annotation or domain-specific rules. The reconstructed trajectories are used only to generate training labels. The trained model works on a window of past states and moves, so it can indicate a redundant path before the return to an earlier state occurs. The score is decomposed into the certainty of the normal class, the certainty of the anomalous class and a residual uncertainty, which indicates borderline decisions. This is of particular importance in conditions of strong class imbalance and limited supervision. The method has been experimentally verified in the context of speedcubing, a complex yet deterministic system characterised by state transitions. The findings suggest that the proposed solution exhibits superior performance on metrics sensitive to data imbalance, and that the confidence decomposition supports selective decisions on borderline windows.
Object detection in unmanned aerial vehicle (UAV) imagery is challenging because foreground targets occupy only a few pixels, vary widely in scale, and crowd against cluttered backgrounds. Parameter-efficient YOLO detectors meet on-board memory constraints but suffer from progressive information degradation across backbone, neck, and head stages. This paper proposes HCP-YOLO, a hierarchical context propagation detector that places one targeted lightweight module at each degradation point: DirC2f preserves directional structure in the backbone, GSPI injects compact semantic priors into shallow features, and CrossHead couples the classification and localization branches through sequential asymmetric gating. On VisDrone2019-DET, HCP-YOLO attains 40.40% mAP50 and 24.02% mAP50:95 with only 3.70M parameters, exceeding YOLOv8s by 0.53% mAP50 at one-third of its parameters and adding 1.54% mAP50 over a same-structure P2–P4 baseline. The gains are reproducible, with the standard deviation across three independent trials at or below 0.10%. Edge deployment on an NVIDIA Jetson Nano with TensorRT FP16 achieves 38.32% mAP50 at 10.54 FPS with 17.4 MB memory, outperforming YOLOv8s by +0.64% while using 67% fewer parameters, confirming that the accuracy advantage and parameter efficiency transfer to resource-constrained embedded hardware.
Impulse-radio ultra-wideband (IR-UWB) radar can detect physiological chest-wall micro-motion without skin contact and can operate through clothing or bedding under suitable conditions. This PRISMA-informed review with narrative synthesis examines its healthcare applications, supporting evidence, and barriers to clinical use. Vital-sign, sleep, activity, rehabilitation, and exploratory diagnostic applications are organized by the type and limitations of the reported evidence. Related UWB ranging and microwave-imaging work is distinguished from contactless radar sensing. IR-UWB is compared with contact sensors, cameras, Wi-Fi sensing, FMCW radar, and conventional imaging along application-dependent dimensions. Multimodal fusion with inertial, visual, and audio sensing is examined as a potential response to motion artifacts and limited scene context. The review proposes a conceptual audio–video–UWB framework for patient monitoring; it has not been experimentally implemented or benchmarked. The literature examined did not establish a clinically validated integrated framework of this kind or a dedicated patient-monitoring benchmark containing all three streams. Incomplete selection records limit the coverage of this finding. Commercial platforms and reported datasets show hardware concentration, small cohorts, heterogeneous protocols, and limited synchronized healthcare recordings. Proposed acquisition and benchmarking recommendations, together with a distinction between radio authorization, exposure assessment, and medical-device requirements, identify the evidence needed for further clinical translation.
Given an input triplet (A, A′, B), can a model infer the visual transformation encoded by (A, A′) and apply it faithfully to B—without any textual supervision or task-specific fine-tuning? Existing approaches either constrain generation to specific architectures, or encode the transformation as learned tokens that inadvertently absorb source content, corrupting the output B′ with unintended artifacts. More fundamentally, existing methods often face a trade-off between faithfully applying the transformation and preserving the structural integrity of B. In this paper, we present AnalogWeave, a training-free visual analogy framework that addresses these limitations through three tightly coupled components: 1) Multi-Agent Visual Analysis, which extracts semantic transformation and spatial structure information, 2) Delta Extraction and Aggregation (DEA), which fuses image-domain and text-domain transformation directions into a unified guidance signal, and 3) Zonal PNP Generation, which balances structural preservation and generative freedom during denoising. AnalogWeave operates as a plug-and-play extension to standard Stable Diffusion pipelines, and generalizes across object-level edits, attribute changes, and style transfers. Extensive experiments show higher mean directional-alignment and structural-preservation scores under the evaluated protocol, together with favorable perceptual evaluations, although the observed differences vary across evaluators and evaluation criteria.
Hybrid Low Earth Orbit (LEO)–Medium Earth Orbit (MEO) satellite networks are increasingly being considered for upcoming non-terrestrial networks (NTNs), combining the low propagation delay of LEO constellations with the wider coverage and greater orbital stability of MEO systems. Motivated by emerging multilayer initiatives, this paper evaluates the impact of integrating a MEO transport layer into an already connected LEO-based network. An IRIS2-like architecture is modeled, composed of a Walker-Star LEO constellation, a complementary Walker-Delta MEO layer, European gateways, and a reference persistent traffic-flow matrix. LEO-only and hybrid LEO–MEO architectures are compared under different inter-satellite link (ISL) formation policies, user-satellite link (USL) selection policies, and cross-layer connectivity levels, using time-varying snapshot-based graphs and both unloaded Dijkstra shortest-path and capacity-aware routing. The analysis considers end-to-end propagation delay, hop count, LEO congestion, traffic routed through MEO, served demand, sensitivity to traffic conditions and admission order, and link-capacity dimensioning. Results showthat theMEOlayer provides additional path diversity, relieves recurrent LEO bottlenecks, and can reduce the ISL capacity required to meet target service levels while maintaining competitive path-level performance. Overall, the study highlights the role of MEO backbones as a capacity-relief and routing-diversity enabler in multilayer non-geostationary satellite orbit (NGSO) systems and provides quantitative guidance for future NTN infrastructure design.
