Rough conical surfaces are widely encountered in solar-thermal collectors, heat shields, aircraft nose cones, chemical reactors, conical diffusers, nozzles, compact heat exchangers, and thermal management systems, where surface texture can strongly influence boundary-layer transport. Motivated by these applications, this study investigates steady mixed-convective Ag–Al–Au–Cu/H2O tetra-hybrid nanofluid flow over a vertically oriented rough cone subjected to a spatially periodic magnetic field. The wall roughness is modelled through a deterministic sinusoidal profile and incorporated via a slip-velocity condition. Using Mangler’s transformations, the governing boundary-layer equations are reduced to a non-similar dimensionless system and solved by quasilinearisation with an implicit finite-difference scheme. The effects of Richardson number, magnetic parameter, Eckert number, roughness frequency, slip velocity ratio, nanoparticle volume fraction, and particle sphericity are analysed for mono-, hybrid-, ternary-, and tetra-hybrid nanofluids. In particular, the comparison with and without viscous dissipation is performed for different particle shapes, with each nanoparticle volume fraction fixed at 5%, while the remaining governing parameters are kept constant. The results reveal that deterministic roughness induces pronounced sinusoidal variations in both skin friction and heat-transfer rate along the cone surface. Increasing the roughness frequency and slip velocity ratio intensifies these oscillations; for instance, at a slip velocity ratio of 0.5, the heat-transfer-rate measure increases from 0.11030 for a smooth cone to 4.22853 at n = 30. The periodic magnetic field suppresses velocity through Lorentz-force resistance, whereas viscous dissipation enhances thermal transport. Among the tested configurations, the Ag + Au + Cu/H2O ternary nanofluid provides a 35.50% reduction in skin friction under viscous dissipation. The numerical results are validated against published benchmark data, confirming the reliability of the present computational approach.
A rich array of dynamical structures, including localized solitons, periodic wave trains, and breather-type oscillations, arises from the nonlinear partial differential equations governing the propagation of waves in multiple spatial dimensions. The current work presents and comparatively evaluates two neural surrogate frameworks, labeled PINN and E-PINN for continuity with their solution-informed physical grounding even though neither enforces the governing equation as an explicit residual term. These frameworks are designed to approximate solutions of the (2+1)-dimensional Zakharov–Kuznetsov (ZK) equation: a baseline neural approximator trained directly on analytical target data, and an exponential-augmented counterpart (E-PINN) in which a structured exponential basis function, derived from the traveling-wave geometry of the ZK equation, is embedded as a fixed auxiliary predictor alongside the trainable network. To study the effects of solution complexity on approximation quality, three representative, physically motivated, exponentially structured benchmark waveforms – soliton, periodic, and breather-type – are adopted as benchmark targets, allowing a systematic assessment of accuracy, training convergence, and model reliability across a range of nonlinearity and oscillatory behavior. The computed results reveal a distinctly solution-specific performance profile. For localized soliton solutions, whose spatial structure is naturally exponential, the augmented framework attains a peak-region error reduction of up to 24% relative to the baseline, demonstrating that incorporating a structurally compatible basis function simplifies the residual learning task. For periodic and breather-type solutions, however, the standard neural approximator proves more effective across all domain-averaged error measures – and the augmented variants’ peak error rises to between roughly 1.3 and 2.0 times the baseline value depending on the case – as the exponential basis introduces structural incompatibility rather than benefit in these regimes. Taken together, the findings indicate that architectural decisions driven by the mathematical form of the target solution yield selective gains rather than across-the-board improvements, and that principled model selection should be informed by prior knowledge of the dominant solution features. The proposed computational pipeline provides a clear and fully reproducible base for benchmarking neural approximation strategies in nonlinear dispersive wave systems.
Dispersive soils used in earth-rockfill dam clay cores are highly susceptible to internal erosion, while conventional mechanical mixing methods suffer from limitations in achieving uniform improvement. To address these issues, a seepage-driven in situ improvement method employing a lime filter layer was proposed. Laboratory seepage tests were conducted to investigate the effects of lime filter layer thickness, hydraulic gradient, and seepage duration on the improvement performance of dispersive soils. The results indicate that the thickness of the lime filter layer is the primary factor governing the improvement performance. Under the experimental conditions investigated, the original dispersive soil was successfully transformed into non-dispersive soil when the lime filter layer thickness was 15 mm, the hydraulic gradient was 3.75, and the seepage duration was 5 d, corresponding to the best overall improvement performance. Microscopic analyses revealed that seepage promoted the migration of Ca2⁺ into the soil, inducing Na⁺/Ca2⁺ ion exchange, diffuse double-layer compression, and particle rearrangement, thereby driving the pore structure to evolve from a loose to a denser state. After treatment, the mass fraction of Na decreased from 3.84% to 1.01%, while the porosity decreased from 21.60% to 5.94%.The proposed method adopts a non-contact in situ improvement approach and shows potential for integration with clay core construction practices, providing an experimental basis for the in situ treatment of dispersive soils in earth-rockfill dam clay cores.
The transformations in architectural design have led to new developments in commercial settings that have affected perceptions, experience, and memories of individuals. In this research, the old Qaisery Bazaar and modern shopping malls in Erbil were considered to assess the relationship between spatial layout and perception and nostalgia in commercial environments. A mixed-method approach was adopted in this study utilizing space syntax and a survey with 133 participants. The results obtained through this research indicate that the shopping malls offer greater legibility and ease of wayfinding, but the bazaar gives greater attachment and nostalgic experience for users. Regression analysis indicated that sensory richness, symbolic identity, temporal continuity, and social interaction are important factors influencing nostalgia but not spatial layout and wayfinding. Space syntax analysis indicates that the shopping mall has high spatial integration, but the complexity of the bazaar promotes exploration and interaction.
In the present investigation, the overfall is employed as a practical tool for measuring discharges in stone bed rectangular channels with overfalls. Fifteen stone bed rectangular channel models with a length of 240 cm and an end drop fall of 20 cm are constructed and examined. These models assess different stone sizes Ds = 0.5 cm, 1.0 cm, and 1.5 cm in horizontal bed channels as well as channels with different bed slopes Sr = 0.0033, 0.005, 0.01, and 0.02. The examination of laboratory data for ratios of brink flow depth Ye to critical depth Yc (EDR) and to normal flow depth Yn for different values of Sr and Ds showed straight line relationships. The average Ye/Yc values varied with Sr according to a quadratic formula. For the ranges of Sr from 0.0033 to 0.02, Froude number Frn from 0.41 to 2.78, and Ye/Ds from 0.88 to 15.55, a direct empirical power correlation is presented for estimating the dimensionless discharge qr/(gr1/2Ye3/2) in terms of Ye/Yc, Ye/Ds and Sr with determination coefficient R2 of 0.960 and Mean Percentage Error MPE = ±0.45%. Another empirical power expression is presented for the dimensionless discharge in stone bed rectangular channels of horizontal beds with R2 of 0.965. The correlations identified in this study shown good consistency when compared to the findings of many other researchers.
Aiming at three bottlenecks of multi-source remote sensing eco-health inversion: cross-modal feature redundancy, decoupled single-task modeling, and detail distortion of fragmented urban land covers, this paper proposes CAFF-Net for coupled eco-health prediction of urban green spaces and water bodies using Sentinel-1 SAR and Sentinel-2 optical data. A dual-branch Swin-Transformer separately extracts heterogeneous features, and serial CBAM realizes adaptive feature filtering to suppress redundant information. An improved multi-task UNet synchronously predicts green space, water body and coupled ecological health indices with joint loss to model aquatic-terrestrial ecological linkages. Experiments show CAFF-Net achieves an R2 of 0.928, 3.3% higher than SOTA, with strong robustness under heavy cloud interference, supporting sponge city and watershed governance.
This study examines how Aleppo’s long-term transformation of urban centrality can inform resilient post-war recovery. Using a qualitative historical-spatial approach based on maps, planning documents, heritage studies, statistics, and visual records, it identifies four overlapping stages: historic core dominance before the 1860s, dual centrality from the 1860s to 1970, district-level differentiation from 1970 to 1990, and peripheral road-based centrality from 1990 to 2010. The findings show that the historic core retained strong functional and symbolic importance while modern, district, and peripheral centers acquired differentiated roles. However, the emergence of multiple centers did not automatically produce balanced polycentricity or reduce pressure on the core, as some transport and redevelopment policies continued to reinforce central access. The study’s originality lies in interpreting Aleppo’s evolution as cumulative differentiation rather than center replacement and in distinguishing multiple nodes from demonstrated functional polycentricity. It translates these findings into a centrality-sensitive resilience framework for deciding which functions should remain in the historic core, which may be redistributed to secondary centres, and how peripheral investment should be conditioned by public transport, functional integration, and updated post-war evidence.
Modern mission-critical applications, such as autonomous vehicle networks and aerospace systems, require extreme reliability and energy efficiency. This paper proposes a hybrid informatics framework integrating Orthogonal Filter Bank Time Frequency Space (OFBTFS) modulation with Reversible Majority Logic Enhanced Polar Codes (RML-PC). The OFBTFS layer utilizes multi-dimensional orthogonality to mitigate multi-path fading and high-Doppler shifts inherent in high-mobility environments. Simultaneously, the RML-PC architecture leverages reversible logic gates and majority-rule decision mechanisms to rectify hardware-level transient and permanent faults during decoding. Experimental results demonstrate that this integrated approach significantly reduces Bit Error Rate (BER) and Mean Square Error (MSE) compared to traditional polar-coded systems. Additionally, the reversible computing implementation minimizes power dissipation, aligning with green computing paradigms. By balancing spectral efficiency with hardware fault tolerance, this framework provides a robust foundation for Ultra-Reliable Low-Latency Communication (URLLC) in 6G and autonomous engineering ecosystems.
This paper describes a power-aware adaptive security architecture for resource-constrained FPGA edge devices. Always on cryptographic modules waste energy during normal operation, so this design activates a 16-bit PRESENT lightweight block cipher only when the input data looks anomalous. A 4-tap symmetric FIR filter smooths the signal in real time, and a dedicated anomaly detector watches the output. Once filtered data crosses an absolute threshold, a BUFGCE integrated clock-gating cell routes the system clock to the secure module and encryption begins. The design was synthesized and implemented on a Xilinx Spartan-6 FPGA at 50 MHz. It’s compact 62 LUTs and 64 flip-flops and the BUFGCE-based clock gating cuts unnecessary switching activity, which should improve dynamic power efficiency. Together, the results show that this event-driven, clock-gated approach can deliver solid cryptographic protection without breaking the energy budget that embedded hardware demands.
Identifying landscape functions is one of the fundamental components of ecological planning, as it reveals the spatial distribution of natural and cultural processes and supports the development of planning decisions aimed at achieving a balance between conservation and land use. However, studies that comprehensively evaluate multiple landscape functions at the district scale in Mediterranean landscapes remain limited. The aim of this study was to identify the water, soil protection, biodiversity, and cultural landscape functions of Selcuk District, İzmir Province, Türkiye, using a Geographic Information Systems (GIS)-based landscape function analysis approach, to reveal their spatial distribution, and to provide a scientific basis for landscape planning. Landsat satellite imagery, ASTER GDEM data, digital geological and soil maps, forest management plan data, and biodiversity data were integrated within a GIS environment. Water, soil protection, biodiversity, and cultural function maps were produced using Overlay, Reclassification, and decision matrix analyses. The results showed that the water function was concentrated in agricultural areas characterized by low slopes and high infiltration capacity, whereas the biodiversity function exhibited high values in wetlands, protected natural areas, and forest ecosystems. The cultural function was primarily concentrated in and around the archaeological and historical sites of the Ancient City of Ephesus and Sirince. Furthermore, the water and soil protection functions displayed similar spatial distribution patterns, whereas the biodiversity and cultural functions were concentrated in different landscape units. This study aims to provide a GIS-based decision-support approach that can be used to identify conservation priorities and support landscape planning processes for sustainable land use.
Diffusion models have established themselves as the leading paradigm for high-fidelity generative modeling, yet persistent challenges including training instability, mode collapse, and computational inefficiency continue to limit their practical deployment across diverse real-world applications. We present Self-Regularizing Diffusion Transformers (SRDT), a novel generative framework that addresses these limitations through three core technical innovations: (1) a hierarchical cross-modal contrastive alignment module that jointly optimizes diffusion dynamics and multimodal semantic consistency within a single end-to-end objective, enforcing structural coherence between visual and textual latent trajectories at every denoising timestep; (2) an uncertainty-aware attention mechanism that dynamically suppresses unreliable activations through feature-level variance and entropy estimation, enabling robust generation under varying noise conditions; and (3) a self-regularization term that enforces hidden-state trajectory smoothness across consecutive denoising steps, providing an explicit inductive bias against training instability with theoretical convergence guarantees. SRDT was rigorously evaluated on six benchmark datasets spanning three task domains: text-to-image synthesis (MS-COCO 330 K, LAION-400 M), medical image reconstruction (BraTS 2023, ISIC 2020), and audio generation (AudioSet, VoxCeleb2). Experimental results demonstrate that SRDT achieves an 18.3% reduction in Fréchet Inception Distance (FID = 11.42), a 23.7% improvement in computational efficiency (58.3 GFLOPs, 7.1 GB memory), and superior zero-shot generalization (61.9% average accuracy) over state-of-the-art baselines including Stable Diffusion, DALL·E 2, and Imagen, while using 385 M parameters — fewer than all compared methods. Comprehensive ablation studies confirm the individual contribution of each proposed component, with cross-modal alignment yielding the largest performance gain (FID reduction of 3.31). Statistical significance testing across five independent runs confirms all improvements at p < 0.001 with large effect sizes (Cohen’s d > 1.4). These results establish SRDT as an effective, efficient, and theoretically grounded framework for stable multimodal generative AI.
Urban metro networks are increasingly vulnerable to flood-induced disruptions, necessitating resilience assessment frameworks that explicitly account for uncertainty while integrating both expert judgment and empirical evidence. This study develops an uncertainty-aware expert-data fusion framework for multi-stage resilience assessment of metro networks, structured along the Prepare–React–Recover–Adapt cycle. Using 23 indicators spanning natural, social and operational dimensions, the framework captures the stage-specific resilience characteristics across flood-management phases. Methodologically, the framework integrates hybrid weighting and uncertainty-aware evaluation within a unified architecture. The Comprehensive Weight Determination (CWD) scheme combines entropy-based weighting, causal interaction modeling and Pythagorean fuzzy hierarchical analysis to jointly represent data variability, indicator interdependencies and expert-informed importance. This is further coupled with a Cloud Matter-Element (CME) model, which incorporates probabilistic–fuzzy representation to account for both stochastic and epistemic uncertainties. Application to the Shanghai metro network demonstrates the practical utility of the framework for stage-specific, uncertainty-aware resilience classification. Under a 50% data-disturbance scenario, weight variability is reduced by 59.8% compared with conventional methods. The results reveal pronounced spatial heterogeneity, with higher resilience in central districts and persistent vulnerabilities in peripheral corridors. Causal analysis indicates that resilience is governed by the interaction between demand pressure and infrastructural capacity, with population density as a key driver and station load capacity as a critical bottleneck. These findings highlight the capability of the proposed framework to provide interpretable and uncertainty-aware support for resilience-oriented decision-making, enabling targeted intervention and stage-adaptive management in flood-prone urban metro networks.
Differential evolution is an effective tool in numerical and engineering optimization. Although LSHADE-SPACMA performs excellently, it often struggles to converge to high-precision solutions in complex multimodal and hybrid search spaces and is prone to premature convergence. To address this, we propose ELSHADE-SPACMA, which integrates a fitness-distance-based non-greedy selection update mechanism, an elimination mechanism based on individual density, and a hybrid update escape mechanism. This combination enhances the exploration–exploitation balance and the ability to escape local optima without increasing algorithmic complexity. ELSHADE-SPACMA is compared with three representative categories of algorithms (widely cited classical methods, recently proposed optimizers, and high-performance competitors)to ensure a comprehensive evaluation. Results show that, compared with LSHADE-SPACMA, ELSHADE-SPACMA performs better on 65.52%–68.97% of the CEC2017 functions and on 23.33% of the CEC2020 problems. Finally, ELSHADE-SPACMA is applied to the real-world problem of 3D UAV path planning, which involves complex constraints, to test its effectiveness in solving complex optimization problems. Simulations demonstrate that under three static terrain scenarios, ELSHADE-SPACMA reduces path length by 2.44%–5.21% relative to LSHADE-SPACMA, convincingly proving its strong competitiveness in both numerical optimization and engineering tasks.
Accurate and real-time detection of objects is important in the field of autonomous vehicle (AV) safety, but the current detection systems have numerous issues. Conventional Region Proposal Networks (RPNs) do not typically succeed in the technical urban setting with tiny or hidden features, whereas Fast R-CNN-based techniques do not possess the capability to use the posture of vulnerable road individuals, like pedestrians and cyclists. The research gaps exist in the ability to incorporate the posture information directly into RPNs, real-time performance on edge devices, and performance in various conditions of the environment. The paper will deal with these drawbacks by proposing a posture-sensitive object detection model that combines Automated Posture Indexing (API) with a RPN and Fast R-CNN with HRNet. Posture and orientation gross will contribute to the following objectives: Ensuring posture and orientation characteristics are incorporated into region proposals; Keeping the feature extraction resolution up while assuring good performance for detecting small areas and occlusions and optimizing the inference operation for edge devices; Furthermore, the kitchen team is testing the human-robot interaction in real space and getting better detection results of the animals. Its methodology consisted of seven execution steps, namely, data acquisition and preprocessing; API; posture-aware RPN; HRNet feature extraction; Fast R-CNN classification and localization; real-time optimization; and decision-making with navigation support. It was tested with each stage by means of properly defined test cases and real-life situations, such as the pedestrian running, the cyclist turning, and multi-object interactions. An urban intersection with a running pedestrian, a leaning-to-the-left cyclist, and a distant, obscured traffic sign was used to validate a sample dataset. Sample performance indicated that sensor fusion could achieve sensor fusion latency of 8 ms, posture key points detection of 96 percent, and posture-aware RPN of 12.4 percent higher recall than baselines. HRNet achieved an improvement of 9.1% mean average precision (mAP) for small objects with only 12% latency overhead, and Fast R-CNN achieved 91.7% classification accuracy and a 0.87 posture F1-score. On an NVIDIA Jetson AGX Xavier, inference operated at 31 FPS with 46 ms latency to frame firmware according to real-time constraints. In the scenarios of navigation, situations of early braking between running pedestrians were activated in 428 ms with a minimum time to collision of 2.3 s, whereas a cyclist turning intention with an F1 of 0.89 was predicted, while a 100% collision avoidance rate in a single scenario was achieved as of the sample run. The results prove that the proposed posture-conscious framework not only addresses the existing research gaps but also guarantees the robust perception, effective edge deployment, and predictive navigation safety. This work provides an authenticated methodology for improving the safety and reliability of autonomous vehicles in dynamic city settings by consistently attaining more than 90% detection accuracy, more than 95% collision avoidance, and real-time performance.
Securing decentralized identity and control in 6G OpenRAN requires a framework that preserves protocol integrity under heterogeneous trust domains, variable edge resources, and adversarial traffic. This study presents a hybrid post-quantum trust architecture that combines a permissioned blockchain, Self-Sovereign Identity, Learning-With-Errors encryption, ML-DSA signatures, and classically secure non-interactive zero-knowledge verification over Ristretto255. The Proof-of-Edge Participation mechanism selects validator committees using authenticated measurements of compute throughput, memory availability, link bandwidth, node availability, and protocol reliability. A deep Q-network performs load-aware task allocation over srsRAN-derived RAN behavior represented within NS-3. Across the evaluated conditions, the framework achieved an overall authentication success rate of 98.3%, a normal-condition authentication latency of 3.91 ms, and a mean latency of 5.33 ms across L1–L8. Mean consensus delay was 105.5 ms. Mean entropy-based privacy leakage was 0.0188, with scenario values ranging from 0.013 to 0.026. Throughput retention was 91.4% under the 10,000-node SC8 condition. Comparative evaluation against DZTF, QIDM, PQDID, BDRM, SDDTV, MDNS, and DROA identified consistent improvements in authentication reliability, latency, consensus delay, throughput retention, and privacy leakage. The physical evaluation further confirmed operation within the processing, memory, power, and thermal limits of the tested Raspberry Pi 4B platform.
This paper proposes a residual multi-layer perceptron (MLP)-based approach for phase calibration in phased array antennas, aiming to improve computational efficiency while maintaining high calibration accuracy. Conventional calibration methods, such as the Greedy algorithm, achieve accurate results by exhaustively searching all possible phase states of phase shifters; however, the conventional method suffers from severe computational inefficiency, particularly for large-scale arrays or systems that require frequent calibration. To address this limitation, the proposed residual MLP is trained on the optimal phase vectors obtained from the conventional method, which eliminates the need for exhaustive phase searches. The residual MLP-based calibration framework achieves nearly the same accuracy as the conventional method while reducing calibration time by more than an order of magnitude. The proposed approach is therefore highly suitable for practical phased array applications that demand fast and frequent calibration.
High-pressure laboratory splitting grouting tests are essential for validating grouting theories and assessing the reliability of numerical simulation results, thereby providing a fundamental basis for practical engineering applications. However, existing experimental facilities are limited in their ability to simulate the coupled effects of high in-situ stress and high injection pressure, accommodate large-scale specimens, adapt to various grouting materials, and perform multi-dimensional synchronous monitoring. To investigate the mechanisms of splitting grouting in deep rock masses under the coupled effects of high confining stress and high splitting pressure, a high-pressure splitting grouting testing apparatus for deep rock masses was independently developed. The system consists of three core subsystems: a high-precision grouting control subsystem, a true triaxial stress loading subsystem, and a multi-dimensional synchronous monitoring subsystem. The true triaxial stress loading subsystem employs a KDSH1000 high-pressure, high-speed servo pump to achieve servo-controlled loading in the three principal stress directions. Grouting pressure is provided by a KDHB100S dual-cylinder high-precision injection pump, which is connected to the specimen through a high-pressure grouting mixing tank. The apparatus can generate a maximum injection pressure of 70 MPa and a maximum loading capacity of 4,500 kN along each of the X-, Y-, and Z-axes, while accommodating cubic specimens with edge lengths of up to 300 mm. It also enables the synchronous acquisition of multi-dimensional data, including conventional mechanical parameters, fluid pressure and flow characteristics, acoustic emission (AE) signals, and ultrasonic wave responses. To verify the reliability and stability of the testing apparatus, splitting grouting experiments were conducted on specimens subjected to three cyclic loading paths: constant-amplitude cyclic loading, stepwise cyclic loading, and variable upper-limit stepwise cyclic loading. Multi-dimensional data, including injection pressure histories, failure morphologies, AE signals, and ultrasonic wave data, were synchronously collected and analyzed. The results demonstrate that the apparatus can accurately reproduce complex stress paths representative of deep rock formations while providing stable and reliable data acquisition. Under all three cyclic loading paths, the cumulative AE ring-down counts exhibited a pronounced stepwise increase throughout the splitting grouting process. During the stable seepage stage, AE activity remained at a high level because of the combined effects of compressive damage along the fracture surfaces and shear deformation on the rough fracture interfaces. During the initial stress loading and cyclic loading–unloading stages, only low-energy, sparsely distributed AE events were detected, whereas the splitting grouting stage generated high-energy, densely distributed AE events. The spatial distribution of AE events corresponded well with the propagation characteristics of splitting fractures. Moreover, different cyclic loading modes produced significantly different levels of internal damage in the specimens, following the order: variable upper-limit stepwise cyclic loading > constant-amplitude cyclic loading > stepwise cyclic loading. Furthermore, splitting grouting increased the ultrasonic frequency-domain energy of specimens subjected to variable upper-limit stepwise cyclic loading and constant-amplitude cyclic loading, whereas it decreased the ultrasonic frequency-domain energy of specimens subjected to stepwise cyclic loading.
The rapid growth of digital financial services has significantly increased the complexity and volume of financial transactions, making the accurate detection of fraudulent activities a critical challenge for conventional machine learning approaches. This study proposes a hybrid intelligent fraud detection framework that integrates data preprocessing, feature selection, deep learning classification, and adaptive hyperparameter optimization to improve fraud detection accuracy while maintaining computational efficiency. Initially, transaction records are preprocessed through missing value treatment, feature normalization, outlier analysis, and Tomek Links undersampling to enhance data quality and address class imbalance. Subsequently, the Perfumer Optimization Algorithm (POA) is employed to identify the most informative feature subset, whereas a Capsule Network performs hierarchical feature representation learning to capture complex relationships among transaction attributes. The Fennec Fox Optimization Algorithm (FFOA) is further utilized to optimize the network hyperparameters, improving convergence and overall classification performance. The proposed framework was evaluated using a publicly available Kaggle financial fraud benchmark dataset and compared with several state-of-the-art deep learning models. Experimental results achieved an accuracy of 99.67%, precision of 99.52%, recall of 99.74%, F1-score of 99.65%, and an MCC of 0.993, while maintaining low computational complexity suitable for real-time applications. Furthermore, SHAP and permutation importance analyses demonstrated that behavioural attributes, particularly failed transaction attempts and transaction frequency, contributed most significantly to fraud prediction. The proposed framework advances financial fraud detection by combining optimization-driven feature selection, hierarchical representation learning, and adaptive hyperparameter tuning into a unified architecture, providing an accurate, interpretable, and computationally efficient solution suitable for next-generation intelligent financial security systems.
Wide band gap (WBG) materials have emerged as promising candidates for high-frequency power electronics applications, including fast chargers for electric vehicles, RF generation, and high-power microwave applications. However, intrinsic material limitations of SiC constrain further improvements in the high voltage blocking capability and high-frequency switching performance. In this work, a Planar-Gate (PG) vertical diamond Power MOSFET is investigated for high voltage and high-frequency applications. A systematic investigation of static and dynamic characteristics of the proposed diamond PG-MOSFET is carried out and benchmarked against 1200 V vertical 4H-SiC-MOSFET with Planar-Gate (PG) and Split-Gate (SG) structures using COMSOL Multiphysics. All devices are designed with identical geometry and comparable doping concentration to enable material focused comparison. The static characteristics, switching characteristics and device capacitances are compared and analyzed. The Diamond-PG-MOSFET achieves forward blocking voltage of 2686 V, which is up to 97.07% higher than SiC-MOSFET, mainly due to higher critical electric field. Furthermore, the total switching losses of the diamond MOSFETs are reduced by up to 47.06% compared to SiC-MOSFET, owing to reduced reverse transfer capacitances (Crss) and lower specific on-state resistance (Ron.sp).
Monocular Simultaneous Localization and Mapping (SLAM) has gained increasing attention for low-cost and lightweight deployment in robotics and autonomous systems. However, conventional monocular SLAM pipelines often lack metric scale and rely on depth networks that are trained on a single dataset, limiting their generalization and applicability to resource-constrained platforms. In this study, we present an extension of ORB-SLAM3 with a depth-agnostic integration framework initially tested with several modern monocular depth predictors, including MiDaS, Depth Anything V2, RT-MonoDepth, and Lite-Mono. Among these, Depth Anything V2 showed the most robust performance on our collected dataset, benefiting from its training on diverse data and state-of-the-art design. To address the mismatch between relative depth outputs and metric requirements, we propose a four-parameter disparity-to-depth mapping optimized using Particle Swarm Optimization (PSO), which we apply to Depth Anything V2 in our final configuration. The PSO-based calibration is performed once per deployment environment using a single representative sequence; the resulting parameters are then fixed and applied to all remaining sequences without re-optimization. Experiments were conducted on the KITTI benchmark, a publicly available IUST car dataset, and newly collected phone-mounted driving sequences. Using Depth Anything V2 as the backbone, our system achieved real-time performance on a modest GTX 1050 laptop, demonstrating feasibility for Jetson-class embedded hardware. In our method, the trajectory scale is fixed to 1.0 for all sequences, whereas original ORB-SLAM3 and LFFS require per-sequence scale estimation before error computation. On the KITTI odometry benchmark, the proposed PSO-calibrated framework significantly outperforms baselines. For instance, on sequence 00, ATE is reduced from 128.0740 m (ORB-SLAM3) and 69.1908 m (LFFS) to 4.5642 m (Ours). Similar improvements are observed across other sequences: sequence 02 (155.5040 m → 12.4065 m), sequence 05 (91.8820 m → 7.0894 m), sequence 07 (14.6667 m → 2.6175 m), and sequence 10 (30.2473 m → 1.2800 m). On the IUST driving dataset, ATE drops from 48.5212 m (ORB-SLAM3) to 9.4257 m. For the newly collected smartphone-mounted dataset under challenging real-world conditions (motion blur, dynamic traffic, suburban driving), the system maintains stable tracking with ATE values of 9.2483 m (sequence 00) and 13.5612 m (sequence 01) against GPS reference, while baselines fail to produce reliable trajectories. These results highlight improved scale stability and accuracy compared to fixed mappings, with latency compatible with real-world deployment.
