The widespread use of intelligent algorithms in motion recognition has driven the continuous evolution of technology for recognizing dance performances. The goal of this study is to improve the accuracy, robustness, and temporal modeling capabilities of dance motion recognition using micro-electro-mechanical systems sensors. This objective is achieved by developing a hybrid analysis framework that integrated a calibrated and denoised MEMS acquisition process, a TimeGAN-based sequence augmentation module, and a 1DCNN-ResBi-LSTM-Attention recognition network. The proposed method first uses window-based segmentation to extract multichannel motion representations. Then, the TimeGAN model generates high-fidelity synthetic samples to address data imbalance and improve intra-class variability. Subsequently, spatial-temporal features are learned through deep convolutional encoding and bidirectional residual recurrent layers. An attention mechanism then highlights key, discriminative motion frames. The experimental results demonstrated that the enhanced dataset improved recognition accuracy by 6.8%. The full model achieved peak accuracy of 98.7%, reduced RMSE, and an improved response time compared to the CNN-LSTM and CNN-ResBi-LSTM baselines. The main novelty lies in the combination of MEMS data augmentation and hierarchical, spatial-temporal modeling. This combination simultaneously addresses the issues of insufficient data diversity, weak long-sequence dependency representation, and suboptimal focus on salient motion details. The result is a more reliable solution for real-time dance movement analysis.
Industrial centrifugal pumps operating under complex conditions frequently encounter failures, yet existing diagnostic methods face challenges in effectively fusing multi-source sensor data with physical mechanisms. To address these limitations, this paper proposes a multidimensional digital twin modeling approach that establishes a deep bidirectional synergy between data-driven perception and mechanism-based simulation. First, a Dynamic Sparse Spatio-Temporal Graph Attention Network (DS-STGAT) is designed to capture dynamic local and global dependencies among multi-source signals. Second, unlike conventional unidirectional methods, a novel data-mechanism collaborative adaptive mechanism is introduced. This creates a closed-loop pathway of “data-guided → simulation-refined → data-enhanced,” where the perception model retroactively optimizes simulation parameters (e.g., stiffness coefficients) via consistency constraints, while the simulation model provides physical priors to guide graph construction. Experimental results on multiple bearing datasets and real-world pump conditions demonstrate that the proposed method outperforms baseline models in accuracy, physical consistency, and robustness. Notably, the approach achieves high fault identification rates even in zero-shot scenarios, validating its effectiveness and scalability for the intelligent fault diagnosis of complex rotating machinery.
This study aims to investigate how social media embedding influences micro and small enterprises’ (MSEs) adaptability through resource integration mechanisms in agricultural enterprises. The present research theoretical framework is based on resource-based, vocational ability development, and social media embedding theories. Its objective is to determine the influence of social media embedding on MSEs’ adaptability. A convenience sampling survey gathered data from 327 small and micro agriculture-related entrepreneurs in China. According to the study results, both online community embedding and web media embedding had a significant and positive impact on MSEs. While web media embedding significantly affected resource integration, online community integration did not. Resource integration was identified as a mediating factor between the two dimensions of social media embedding (i.e., online community embedding and web media embedding) and the adaptability of entrepreneurial enterprises. This study advances existing literature by decomposing social media embedding into distinct dimensions. This decomposition offers a nuanced understanding of the differential impacts across various dimensions. The research establishes resource integration as a critical mechanism linking social media use to organizational adaptability. Furthermore, it extends resource-based theory application in the digital transformation context of agricultural MSEs. These findings offer practical implications for enhancing enterprise adaptability through social media utilization.
The purpose of this research was to create a fully automatic deep-learning method to accurately classify different types of cardiac arrhythmias based on patterns observed in electrocardiograms (ECGs). There are several challenges that arise when manually interpreting cardiac arrhythmias. These include variability in how different observers interpret arrhythmias and variability in the diagnostic results obtained by different clinicians. To address these issues, we propose an architecture that includes both residual transfer learning and transformer encoder blocks. In our proposed architecture, the residual learning block uses Conv1D layers with skip connections to learn hierarchical representations of features in the ECG signal. The transformer block uses multi-head self-attention to identify longer-range dependencies in the ECG sequence. Our proposed model is tested on two publicly available benchmark databases of ECG recordings, namely the MIT-BIH Arrhythmia Database and the PTBDB Diagnostic ECG Database. We evaluate the performance of our model using a stratified 10-fold cross-validation procedure as well as Receiver Operating Characteristic (ROC) analysis. The proposed model achieved a classification accuracy of 99% on both datasets, with precision scores of 0.88 to 0.99 and recall scores of 0.87 to 0.99 for each arrhythmia class. Furthermore, the AUC values of the ROC analysis ranged from 0.98 to 1.0, indicating high levels of discrimination against minority classes, including supraventricular ectopic beats and fusion beats.
Accurate macro-administrative water distribution forecasting is essential for infrastructure planning and regional governance. While most prior studies focus on short-term urban demand, provincial-scale systems exhibit aggregated heterogeneity, structural shifts, and limited annual observations that challenge conventional forecasting approaches. This study proposes a Particle Swarm Optimization (PSO)-optimized hybrid ARIMA-LSTM framework designed for cross-domain robustness in macro-scale infrastructure forecasting. Linear components are modeled using ARIMA, nonlinear residuals are learned via LSTM, and PSO jointly optimizes statistical and neural hyperparameters under validation constraints. The framework is evaluated across 34 Indonesian provinces using sequential train–validation–test splits and recursive multi-step forecasting. Beyond conventional accuracy metrics (RMSE, MAE, MAPE, R²), this study introduces cross-provincial RMSE variance and multi-domain Diebold-Mariano statistical testing as robustness indicators. Results show that the PSO-optimized hybrid model achieves the lowest average MAPE (21.79%), reduces inter-provincial RMSE variance by approximately 15%, dominates in 52.9% of provinces, and demonstrates statistically significant improvement in 76.5% of provinces. These findings confirm that optimization-enhanced hybrid decomposition improves structural stability and cross-domain generalization in heterogeneous macro-administrative infrastructure systems.
Artificial Intelligence (AI), Blockchain, and Cloud Computing (CC) are key technologies to improve financial automation and risk management of developing countries in particular. The purpose of this research is to explore the combined impact of AI, Blockchain, and CC on financial automation and how it, in turn, affects improved risk management in financial institutions in Oman. This study adopted a quantitative method by conducting questionnaires with 201 financial and IT experts in banks and fintech firms in Oman using convenience and snowball sampling methods. The respondents were analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM) based on Technology-Organization-Environment (TOE) and Innovation Diffusion Theory (IDT). Results indicate that Blockchain (β = 0.561) and AI (β = 0.264) have a strong impact on financial automation, with a moderate positive influence observed for CC (β = 0.173). Financial automation has a significant positive effect on improved risk management (β = 0.827, R² = 0.81) and regulatory compliance, market responsiveness, and technology access. This research is novel in its integrated empirical model, which brings together three frontier technologies in one model and empirically validates their impact on financial performance and risk management in a developing country, offering practical insights for policymakers and financial practitioners.
From 2010 to 2025, this research looks at the Western Balkans to see how artificial intelligence (AI) helped women become more economically independent and entrepreneurs. This study aims to examine the effects of artificial intelligence (AI), digital transformation (DT), innovation (I), and human capital (HC) on the success and longevity of businesses run by women in emerging markets. To investigate the direct and indirect links among the major variables, the study uses an integrated empirical framework that incorporates Fixed Effects (FE), Hausman-Taylor estimation, Difference-in-Differences (DID), Structural Equation Modeling (SEM), and interaction models. Adopting AI has a favorable and statistically significant impact on innovative ability, company success, and the empowerment of women entrepreneurs, according to the research. The results show that women-led enterprises are more likely to adopt and use AI technologies when they have digital skills, education, and access to finance. Furthermore, the DID analysis verifies that the region's female entrepreneurs have flourished and will continue to thrive thanks to policies that encourage digital transformation and entrepreneurship after 2019. Using a variety of advanced econometric methods that have seldom been combined in prior research on transition economies, this study builds an integrated empirical model that connects the adoption of artificial intelligence (AI), innovation, company performance, and the empowerment of women entrepreneurs in the Western Balkans.
Objectives: This study aims to design, implement, and evaluate a VR Metaverse-Based Digital Signage Game for Tsunami Evacuation as an interactive medium to support disaster mitigation, particularly in improving user understanding and decision-making during evacuation scenarios. Methods/Analysis: The system was developed using a prototyping approach, enabling iterative refinement through continuous feedback. The evaluation employed multiple methods, including questionnaire-based assessment for digital signage effectiveness, Cognitive Walkthrough during the prototyping and refinement stages, and usability testing using VRSUQ on the final system. Furthermore, MSSQ was used to assess users’ susceptibility to motion sickness in immersive environments. Findings: The results indicate that the proposed system effectively supports users in understanding evacuation routes and making appropriate decisions in simulated disaster situations. The usability evaluation indicates that the system achieves an acceptable level of usability, particularly for users with limited prior VR experience. Novelty/Improvement: This study introduces a novel integration of digital signage within a VR metaverse environment for disaster mitigation training. It contributes a prototyping-based development framework and provides empirical insights into user interaction in immersive systems. The study also highlights opportunities for improvement, including simplifying interactions, improving navigation support, enhancing user onboarding, and reducing excessive simulation effects to improve the overall user experience.
A comprehensive grasp of the vertical distribution profiles of soil organic carbon (SOC) and its underlying governing mechanisms is fundamental to assessing terrestrial carbon stocks. Research efforts in mountain ecosystems, however, frequently fall short in providing depth-resolved analysis. This study investigated the spatial patterns and dominant drivers of SOC content across the full 0–200 cm soil profile in the Dabie Mountain Area (DMA), a subtropical montane region in central China. By utilizing high-resolution (250 m) SOC data and multi-source environmental variables for topography, climate, and soil properties, we employed hots pot analysis (Getis-Ord Gi*) and the geographical detector model to quantify spatial clustering and identify key influencing factors across six depth intervals in the DMA. The results showed that SOC content decreases exponentially with depth, with the surface 0–5 cm layer containing the highest concentration. A distinct shift in spatial organization occurred at an approximate depth of 60 cm: surface layers (0–60 cm) exhibited strong, clustered patterns (hot spots and cold spots), whereas deeper layers (>60 cm) transitioned to a more dispersed and spatially homogeneous distribution. Factor detection identified elevation and soil bulk density (SBD) as the most influential factors overall. More importantly, interaction detection revealed a depth-dependent transition in the dominant controlling complexes. In surface soils (0–15 cm), SOC heterogeneity was primarily governed by the interaction between topography (elevation) and soil properties (pH and SBD). In contrast, within the subsoil (15–200 cm), the interaction between climatic factors (temperature, precipitation) and soil properties (pH, SBD) became dominant. These findings demonstrated a fundamental shift from a topo-edaphic control regime in surface layers to a climate-edaphic control regime in deeper layers. This study provided a novel, three-dimensional perspective on SOC storage in mountains, highlighting the necessity of depth-resolved analyses for accurate carbon accounting and for formulating stratified land management strategies aimed at soil carbon conservation.
To address the accuracy bottleneck in the naturalness and rhythm synchronization of music-driven dance generation, an enhanced dance generation model integrating sequence-to-sequence modeling and human pose recognition was developed to improve the synchronization, naturalness, and structural consistency of generated movements. The model uses multi-scale music features as input, extracts temporal music semantics through a bidirectional long short-term memory network and an attention mechanism, and optimizes motion structure by incorporating skeleton keypoint feedback, thereby achieving joint modeling of music semantics and human motion. Experimental results on the AIST++ and DanceTrack datasets demonstrate that the proposed model achieves a beat alignment error as low as 0.12 s, a joint point error of 11.2 px, and a motion smoothness score of 2.41. In the generation of a 90-second dance sequence, the beat error is reduced by more than 32% compared with mainstream models, and the model achieves a high score of 0.97 in the evaluation of complex dance symmetries such as “arm-lifting rotation.” These results indicate that the joint modeling of music semantics and skeletal structure effectively improves movement coordination and rhythm matching in dance generation, enabling the production of natural and coordinated dance movements adaptable to different dance styles.
This study addresses the detection of illicit digital payments, specifically online gambling and child exploitation, which are frequently hidden within legitimate transaction streams. The primary objective is to overcome the limitations of traditional rule-based systems and unimodal models that struggle with class imbalance and sophisticated evasion. It is proposed that the Temporal-Semantic Fusion Network (TSFN), a novel architecture integrating Temporal Convolutional Networks (TCN) for numerical sequences and FinBERT for semantic textual encoding. The key novelty is a bidirectional cross-modal attention mechanism that enables dynamic information exchange between behavioral patterns and transaction descriptions. Evaluated on 10,000 synthetic transactions, TSFN achieved a macro F1-score of 0.847, outperforming concatenation-based fusion by 6.5 percentage points (p < 0.001). Significant improvements were noted in minority classes, with F1-scores of 0.823 for gambling and 0.741 for exploitation, while maintaining a 99.4% precision rate on legitimate data. Ablation studies confirm that bidirectional attention allows the model to adaptively prioritize temporal features for gambling and semantic cues for exploitation. This research provides a robust framework for multimodal financial crime detection, offering a significant improvement in identifying complex illicit patterns compared to existing benchmarks.
Healthcare organizations increasingly require secure, governed and AI-ready data pipelines capable of handling heterogeneous and sensitive data sources. This study aims to design and evaluate a unified DataOps reference architecture that operationalizes the full data lifecycle through a governed Medallion Lakehouse model. Methodologically, the proposed architecture integrates data-centric CI/CD, Infrastructure as Code, workflow orchestration, governance and metadata management, monitoring, and explicit promotion contracts across Bronze, Silver, and Gold layers. The framework was implemented and evaluated in a controlled healthcare testbed using approximately 3.5 GB of multi-source clinical data over a 25-day workload. The findings show that the proposed architecture achieved a DataOps Operational Excellence Index (DOEI) of 0.92, an ingestion throughput of approximately 100 MB/s, a data quality score of 97.87% and a 72% reduction in infrastructure provisioning time, from 3 hours to 50 minutes. The main novelty of this work lies in combining a governed Lakehouse-based DataOps architecture with explicit promotion contracts and a composite benchmarking index for assessing operational maturity. This improvement provides a reproducible, auditable, and scalable framework for secure data operations in regulated environments such as healthcare.
Faults in high-voltage transmission lines (HVTLs) represent a significant challenge to the stability and reliability of power grids. Therefore, fault detection (FD) and fault classification (FC) in HVTLs are crucial for ensuring rapid power supply restoration and preventing electrical energy losses. In this paper, a novel hybrid deep learning model combining Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) techniques with the Harris Hawks Optimization (HHO) algorithm is proposed for accurate fault detection and classification. Three power grids, rated at 132 kV and 50 MW and extending over a total length of 150 km, were designed in the MATLAB environment to generate the training data. The data were then processed using Python to train the proposed model. Subsequently, HHO was employed to determine the optimal parameters of the LSTM-GRU model, including the number of neurons, learning rate, dropout rate, and mean squared error. To evaluate the model’s performance, four statistical measures were used: the confusion matrix, precision, recall, and F1-score. The results show that the LSTM-GRU-HHO model achieved the highest FD and FC accuracies of 99.90% and 99.84%, respectively, outperforming the LSTM, GRU, and LSTM-GRU models, as well as related models reported in previous studies.
Electromagnetic signal propagation is particularly complex in hot regions because it is strongly affected by extreme weather conditions, including temperature, humidity, and dust concentration. This poses a significant challenge for telecommunications operators in network planning and management. The Middle East is an important region for such investigations because of its extreme temperatures, and Iraq provides a representative example. This study aims to evaluate the effects of climatic parameters on signal propagation across different frequency bands, including those used in 4G, 5G, and 6G systems. The analysis employs combined calculations based on signal attenuation and measurements of the most influential weather parameters, particularly temperature, humidity, and dust concentration. This study represents the first multivariable investigation in Iraq covering these three frequency bands while considering the Signal-to-Noise Ratio (SNR), Bit Error Rate (BER), and the feasibility of using Reconfigurable Intelligent Surface (RIS) technology to mitigate climatic effects. The results reveal that temperature has a major effect on path loss, producing attenuation ranging from 0.6 to 2.4 dB/km. The findings also indicate that operation in the sub-terahertz band can be effective over short distances and in indoor environments. During the summer, the SNR may deteriorate significantly, highlighting the need for a climate-aware network management system.
This study investigates the hydraulic and structural performance of a Cross-flow turbine runner manufactured from recycled HDPE reinforced with 40% wood fibers, with the objective of assessing its suitability as a sustainable and mechanically reliable alternative to steel and pure HDPE for micro-hydropower applications. A comprehensive multiphysics methodology was implemented, combining CAD-based design, thermoforming fabrication, Computational Fluid Dynamics (CFD), Finite Element Analysis (FEA), and experimental validation on a calibrated hydraulic test bench. Mesh sensitivity analysis, rotor-stator coupling, and turbulence-model verification were performed to ensure numerical robustness, while structural predictions were benchmarked against tensile testing data for the biocomposite. Results show that the 40% wood‑fiber reinforcement significantly enhances structural rigidity, reducing maximum blade deformation by 74% compared to pure HDPE (from 0.16 mm to 0.042 mm). The biocomposite runner achieves a peak hydraulic efficiency of approximately 57%, corresponding to 87-92% of the efficiency of a geometrically identical steel runner. Structurally, the maximum von Mises stress reaches 14.64 MPa, yielding a safety factor of 2.85 relative to the material’s tensile yield strength of 41.72 MPa. Overall, this research demonstrates that high‑fraction wood‑fiber biocomposites can deliver near‑metallic performance while offering a cost‑effective, corrosion‑resistant, and environmentally sustainable solution for rural and remote electrification.
In the process of China's economic transformation toward high-quality development, green development has become the core path to breaking through resource constraints and environmental pressures. Against this backdrop, computer technology has demonstrated significant advantages in addressing the complex network data in areas such as digital trade and the efficiency bottlenecks of traditional models in big data scenarios. This study develops a digital trade scale measurement system based on data envelopment analysis (DEA) and slacks-based measure (SBM) in an attempt to investigate the coordinated development path of digital trade and the green economy. Digital trade is measured using the DEA-SBM-GA model embedded with genetic algorithm (GA). Each province's degree of development in digital trade is assessed using the entropy value technique. The level of green economic development is gauged by the super-efficiency SBM model. Spatial spillover effects are examined using the spatial Durbin model. The results indicated that DEA-SBM-GA was significantly more efficient than traditional models p<0.001), reducing single iteration time by 87.3%. For every 1-unit increase in artificial intelligence-enabled digital trade, local green economic growth increased by 0.129 (p<0.05), with an indirect effect of 0.112 (p<0.05) and a total effect of 0.246 (p<0.05). Technological innovation had a dual threshold (0.314/0.721), and after crossing the high threshold, the impact coefficient increased from 0.063 to 0.247 (p<0.01). There was a positive geographical spillover impact and notable regional heterogeneity. The Moran's I index for the green economy was 0.304, with a Z-value of 4.82 (p<0.001), indicating significant spatial clustering. The study addresses the shortcomings of earlier research by offering a theoretical framework and policy suggestions for the coordinated growth of the green economy and digital trade. This work has significant academic and practical value.
To realize accurate extraction and digital design of lacquerware color and decorative patterns for traditional craft inheritance and innovation, this study proposes an improved technical method. The improved K-means++ clustering algorithm is used for color extraction. First, singular value decomposition (SVD) reduces the dimensionality of compressed images to retain core color information. Then, a quadratic clustering strategy optimizes the initial centers. For pattern detection, hybrid adaptive median filtering and bootstrap filtering denoise images, combined with adaptive linear interpolation suppression, improve edge positioning accuracy. Experiments on typical lacquerware images showed that when extracting 16 feature colors, the improved K-means++ had a mean peak signal-to-noise ratio (PSNR) of 31.47 dB and a structural similarity index (SSIM) of 0.95. When extracting 8 colors, it still maintained a PSNR of 28.83 dB. The improved Canny algorithm achieved a PSNR of 24.43 dB at 1.0% noise level, with 64.88% sensitivity and 93.04% specificity. It accurately restored color proportions of crafts like needle carving with a pulling knife and generates patterns with clear edges. This method synergistically optimizes color and edge extraction, enhances digital design precision and efficiency, and provides reliable support for traditional craft digital preservation.
Objective: To reveal the intrinsic relationship between agricultural economic growth and ecological environment quality in the northwest region, the study took Shaanxi, Qinghai, and Gansu as the research areas and conducted an empirical analysis of the panel data of the three provinces from 2020 to 2023. Method: This study utilizes the Autoregressive Integrated Moving Average model to analyze the dynamic trends of the agricultural economy, handles the cross-sectional heterogeneity of the data through the bidirectional fixed effects model, and addresses spatial dependence using the Spatial Durbin model. Result: The development of the agricultural economy in the northwest region has a positive effect on the quality of the ecological environment, and the degree of this effect is influenced by factors such as time and geographic location. However, the extensive economic development model will instead reduce the quality of the ecological environment. Innovativeness: By focusing precisely on the characteristics of arid and semi-arid regions, the reliability of the conclusion has been enhanced through the collaborative analysis of multiple models. This has verified the bidirectional influence relationship between agricultural economic development and ecological environment quality in the northwest region, providing empirical evidence for the coordinated development of the ecological economy in this type of region.
Surface defect detection on microdevices is challenged by extremely small defect sizes, complex background interference, and strict requirements on both detection accuracy and computational efficiency. The objective of this study is to develop a lightweight yet high-precision detection framework suitable for resource-constrained industrial deployment. To this end, this paper proposes LiteKANformer, a multi-module lightweight Transformer-based architecture. Based on LiteKANformer, a novel detection framework named LKF-YOLO is constructed by embedding global contextual modeling into the backbone and optimizing multi-scale feature fusion in the neck network. Experimental analysis is conducted on a PCB surface defect dataset and a semiconductor chip defect dataset. Compared with the C3TR module, LiteKANformer achieves comparable detection accuracy while reducing the parameter count by approximately 3.1% and improving the inference frame rate by 2.3%. Furthermore, the proposed LKF-YOLO framework outperforms other mainstream detection models on the PCB dataset in terms of accuracy, recall, and real-time performance. The main novelty of this work lies in the co-design of activation representation, normalization strategy, and computation primitives within a unified lightweight Transformer block, providing an effective solution that balances detection precision and deployment efficiency for microdevice surface defect detection.
This study aims to examine the effects of system quality, information quality, and service quality on individual performance, which in turn affects the organizational performance by extending DeLone and McLean success model in the Jordanian context. To achieve the study aims, a quantitative research method was applied to collect data from 119 accounting managers in listed Jordanian firms in the Amman Stock Exchange (ASE). The Partial Least Squares-Structural Equation Modeling (PLS-SEM) procedure was used for data analysis. Six direct and indirect relationships were tested, whereby five were supported as expected. Specifically, the empirical results reveal that individual performance is significantly influenced by system quality, information quality, and service quality. Besides, the results also show that organizational performance is influenced by individual performance and IT infrastructure. Regarding the indirect relationship of IT infrastructure, the results show that IT infrastructure has not moderated the relationship between individual performance and organizational performance, and hence, the moderating hypothesis was not supported. Ultimately, the current study contributes to the understanding of the essential success factors underlying the usage of Digital Accounting Information Systems (DAIS) in listed Jordanian firms in ASE, which can help policymakers in those enterprises.
