Search Results (47,587)

Building damage and collapse are the main sources of serious casualties and financial losses during earthquakes, which are among the most destructive natural disasters that endanger human life and property. Therefore, quick and precise post-earthquake building damage assessment is essential for risk assessment and emergency action. Convolutional neural networks (CNNs) primarily concentrate on local features and frequently ignore global contextual information within and across buildings, despite the fact that deep learning-based techniques allow automated damage identification. Transformer-based approaches, on the other hand, are good at capturing global dependencies, but their large memory and processing costs restrict their usefulness. As a result, existing networks still struggle to achieve an effective balance between accuracy and efficiency. To address this issue, this study proposes a lightweight and efficient network for post-earthquake building damage assessment. Specifically, we develop a two-stage method based on EfficientViM with an encoder–decoder architecture. In the encoder, Mamba is introduced to extract multi-scale change features with long-range dependencies, leveraging the state space model to preserve global modeling capability while significantly reducing computational complexity. In the decoder, two lightweight modules are designed to further enhance discriminative capability and computational efficiency. The network finally outputs building localization and pixel-level building damage, respectively. Experiments were conducted on four earthquake events from the BRIGHT dataset using a three-for-training and one-for-testing cross-event rotation evaluation strategy. The results demonstrate that LEViM-Net requires only 30.94 M parameters and 27.10 G FLOPs. In addition, for the Türkiye earthquake event, the proposed method achieves an F1 score of 80.49%, an overall accuracy (OA) of 88.17%, and a mean intersection over union (mIoU) of 49.73%. The proposed model enables efficient remote-sensing-based mapping of macroscopic and image-visible building damage, providing timely support for early-stage emergency response. Full article

Effective laboratory training is essential in engineering education, yet conventional on-site instruction is often constrained by time, accessibility, and safety considerations. To address these challenges, this study presents the design, implementation, and evaluation of a web-based virtual reality (WebVR) representation of a large-scale engineering laboratory constructed from massive colorized point cloud data. This study proposes a novel WebVR approach that integrates Unity and Potree for high-fidelity point-cloud visualization combined with advanced interactive capabilities in a browser-based virtual laboratory. It supports immersive first-person exploration, guided navigation, interactive hotspots conveying equipment and safety information, and emergency evacuation simulations. The usability, usefulness, and acceptance of the virtual laboratory were evaluated through an anonymous questionnaire administered to students and laboratory staff. User evaluation results indicated consistently positive feedback, with 100% of respondents rating the interface/navigation and visual/interactive content as good or excellent, 88.6% identifying scene realism as the biggest system strength (the most frequently selected), 74.3% reporting significantly higher engagement compared with traditional online laboratory training, and 82.9% indicating they would definitely recommend the system as a learning resource. In addition, a thematic analysis of qualitative feedback was performed to inform future enhancements of the WebVR environment. Overall, the findings demonstrate that the WebVR-based virtual laboratory can effectively complement conventional on-site laboratory instruction, offering a scalable, accessible, and low-risk platform that enhances learning experiences in engineering education. Full article

Enterprise bankruptcy prediction is a critical research issue in financial risk early warning, credit evaluation, and investment decision-making. To address the limitations of traditional methods in handling high-dimensional, nonlinear, and complex financial data, including parameter sensitivity, susceptibility to local optima, and insufficient prediction stability, this study proposes a multi-strategy enhanced Beaver Behavior Optimizer and applies it to optimize kernel extreme learning machines, constructing the MEBBO KELM prediction model. Three improvement mechanisms are introduced, including an elite pool enhanced exploration strategy, a stochastic centroid reverse learning strategy, and a leader guided boundary control strategy, which improve population diversity, global search capability, boundary handling capacity, and convergence accuracy. The proposed algorithm is evaluated on CEC2017 and CEC2022 benchmark datasets and compared with EWOA, HPHHO, MELGWO, TACPSO, CFOA, ALA, AOO, RIME, and BBO. Statistical analyses are conducted using the Wilcoxon rank sum test and the Friedman test. The results demonstrate that MEBBO achieves superior solution accuracy and stability, indicating strong global optimization capability and robustness. Further experiments on the Wieslaw Corporate Bankruptcy Dataset show that MEBBO-KELM achieves strong and robust performance across multiple evaluation metrics, including ACC, MCC, Sensitivity, Specificity, Precision, Recall, and F1 score. Specifically, ACC reaches 79.7578, MCC reaches 0.6050, and F1 score reaches 78.8504, confirming its effectiveness. Full article

The rapid expansion of e-commerce has transformed market access in developing economies, yet its impact on the spatial structure of market participation remains insufficiently understood. While existing studies primarily examine welfare outcomes such as income growth and consumption smoothing, few investigate how digital platforms reshape the balance of market entry between urban and rural areas. Drawing on New Economic Geography and platform economics theory, this study proposes that e-commerce development rebalances urban–rural market vitality through three associative pathways: alleviating rural capital constraints, improving rural innovation environments, and promoting agricultural-industry agglomeration. Using county-level panel data covering 2725 Chinese counties from 2011 to 2022, we employ a Double Machine Learning (DML) framework to examine the association between designation as an “E-commerce into Rural Comprehensive Demonstration County” and changes in the urban–rural market vitality balance (URMAR). The results indicate that demonstration county designation is associated with a statistically significant reduction in urban–rural market disparity, as measured by both the Theil index and the absolute difference in new enterprise registrations. The directional URMAR indicator further reveals that this convergence is driven primarily by accelerated rural enterprise formation. Subsample analysis confirms that the rebalancing interpretation holds across counties with different baseline market structures. Mechanism analysis provides suggestive evidence consistent with all three proposed associative pathways. Heterogeneity analysis further reveals that these effects are stronger in economically developed eastern regions, in counties linked to higher-tier cities, and in secondary and tertiary industries. These findings advance a market-structure perspective on digital development that complements existing welfare-based approaches and offer policy insights for fostering balanced regional development through targeted digital and complementary investments. Full article

To improve the efficiency of comparative process-window screening in selective laser melting (SLM), this study developed a finite element simulation-driven machine learning framework for 316L stainless steel. A simulation dataset covering laser power (LP), scanning speed (SS), heat-source diameter (HSD), and substrate preheating temperature (SPH) was generated using ANSYS and used to train nine regression models. In the present work, the primary machine learning target was defined as the simulated average thermal stress, σ_avg, which is used as a simulation-derived comparative thermal stress indicator for ranking process conditions within the investigated parameter window rather than as a direct prediction of the final residual-stress field. Among the evaluated models, the Backpropagation Neural Network (BPNN) showed the best predictive performance and was selected as the representative surrogate model because of its strong predictive accuracy, stable behavior, and direct applicability to the present structured tabular dataset. Shapley additive explanations (SHAP) and partial dependence plots (PDPs) indicated that LP is the dominant variable governing the σ_avg-based response, followed by SPH, whereas SS and HSD mainly affect the response through secondary or coupled effects. Within the investigated parameter window, conditions near 180–200 W corresponded to a relatively lower predicted σ_avg level. Experimental observations provided limited but meaningful trend-level support for the simulation-guided screening results: metallographic examination showed improved forming quality near 200 W, while XRD-derived macroscopic stress estimates exhibited a similar variation trend to the simulated σ_avg values under the tested LP–SS conditions. These results suggest that the proposed framework can serve as an efficient surrogate-based tool for comparative parameter screening in SLM-fabricated 316L stainless steel within the assumptions and parameter range of the present model. Full article

Although Graph Neural Networks (GNNs) are widely regarded as an ideal tool for capturing spatial dependencies in river basins, their effectiveness in hydrological forecasting is severely challenged by a topology paradox: under a purely data-driven paradigm, GNNs fail to spontaneously learn physical laws, generating predictions that lack physical interpretability and frequently violate mass conservation. To address this fundamental problem, this paper proposes a physics-informed graph learning framework integrated with an explicit, differentiable water balance constraint (WB-GNN). By reconstructing the continuity equation into a differentiable loss function, we directly embed physical conservation as a strong inductive bias into the neural network’s training objective. We comprehensively evaluated the model on two large-sample datasets (LamaH-CE and CAMELS) against state-of-the-art baselines, including EA-LSTM and unconstrained Pure-GNN. Quantitative results demonstrate that the proposed physical constraint successfully awakens the potential of river network topology. On the LamaH-CE dataset, WB-GNN achieved a Nash-Sutcliffe Efficiency (NSE) of 0.86 and a Root Mean Square Error (RMSE) of 9.2 ${\mathrm{m}}^3/\mathrm{s}$, outperforming both the domain-specific EA-LSTM (NSE: 0.83) and the unconstrained Pure-GNN (NSE: 0.74). Crucially, the introduction of the differentiable constraint reduced the Physical Inconsistency Ratio (PIR) by an order of magnitude-from 39.8% in the unconstrained model to just 4.3%. Similar robust improvements were validated across the highly heterogeneous CAMELS dataset. These quantifiable results confirm that the proposed method not only achieves superior forecasting accuracy but also fundamentally guarantees physical trustworthiness, making it highly robust for critical decision-making in extreme flood events. Full article

The pancreatic pseudocyst is a collection of pancreatic fluid surrounded by a non-epithelialized wall comprising granulation tissue and fibrosis, occurring in approximately 10% of patients diagnosed with acute pancreatitis and in 20–38% of those with chronic pancreatitis. Most pseudocysts are situated in the pancreatic head and pancreatic body, but about 20% develop in extrapancreatic locations. We present the case of a 46-year-old male patient diagnosed with chronic alcohol pancreatitis with acute exacerbation, who developed a large pancreatic pseudocyst with subcapsular location in the right hepatic lobe; this was successfully treated by laparoscopic surgical drainage, with no postoperative complications and no recurrence of the pseudocyst. The computed tomography scan and postoperative biochemical analysis of the intracystic fluid played a key role in establishing the diagnosis of this rare condition. An intrahepatic pancreatic pseudocyst is a rare location for pancreatic pseudocysts, but one located in the right hepatic lobe is extremely rare. The treatment of intrahepatic pancreatic pseudocysts may be conservative, though endoscopic, percutaneous, or surgical drainage may be necessary. The presence of symptoms, signs of extrinsic compression, or complications require drainage of the pseudocyst. The “take-away” lesson learned from this case: surgical treatment for pancreatic pseudocysts located subcapsularly in the liver may be considered when they are very large, or when minimally invasive treatment has not been effective. Full article

Accurately classifying gait abnormalities is crucial for the effective monitoring and rehabilitation of stroke patients. This study proposed a lightweight machine learning framework for distinguishing healthy from abnormal gait patterns using statistical features extracted from wearable gyroscope data. Statistical z-axis angular velocity values from both limbs were derived and used to evaluate the performance of multiple classifiers, including logistic regression, support vector machines, and ensemble methods. A leave-one-out cross-validation strategy was employed to enhance generalizability across subjects. The results indicated that several classifiers achieve accuracy and area under the curve (AUC) values exceeding 0.95, with random forest and support vector machine-based models demonstrating near-perfect class separability, with an AUC of 0.98. These findings highlighted the effectiveness of using minimal set of biomechanically relevant gyroscope features for gait classification in real-world healthcare applications. The proposed pipeline is computationally efficient, making it well suited for implementing in wearable and remote monitoring systems. Full article

Purpose: Access to assistive technology (AT) is a fundamental human right and a critical component of Universal Health Coverage (UHC). In the Middle East and North Africa (MENA) region, the scarcity of trained professionals remains a significant barrier to AT service provision. This study evaluates the effectiveness and perceived usefulness of the Assistive Technology Training Program (ATTP), a specialized continuing education initiative designed to build workforce capacity among rehabilitation and healthcare professionals. Methods: A convergent mixed methods design was used to analyze quantitative pre/post-test scores and qualitative focus group open-ended responses. Quantitative data were gathered from 386 participants across 11 MENA countries using a pre- and post-test assessment of AT knowledge. Qualitative utility and participant satisfaction were assessed through a 5-point Likert scale survey evaluating content relevance, trainer expertise, and facilities. Association tests (ANOVA and t-tests) were conducted to identify factors influencing knowledge gain. Results: Participants demonstrated a statistically significant improvement in AT knowledge, with the overall mean score increasing from 3.67 ± 1.13 to 7.50 ± 1.25 (p < 0.001). High levels of satisfaction were reported, with 92% of participants rating the training as “Very Good” or “Excellent” regarding its relevance to clinical needs. Association tests revealed that professional background (p < 0.001), employment status (p = 0.0017), level of education (p = 0.011), and prior training experience (p = 0.026) were significant factors in the magnitude of improvement, although all subgroups achieved significant learning gains. Qualitative thematic analysis per the focus group discussions using the WHO-GATE 5 P framework identified three major themes: (1) Structural Challenges: Issues with Products and Provision point toward a need for better infrastructure and localized supply chains. (2) Human Capital: Personnel barriers emphasize that training shouldn’t just be for professionals, but should extend to caregivers as well. (3) Systemic and Social Change: Policy and People focus on the “soft” side of AT moving toward user-involved guidelines and fighting social stigma to ensure rights are upheld. Conclusions: The ATTP is an impactful educational intervention that significantly enhances the foundational competencies of healthcare professionals in the MENA region. By addressing knowledge gaps and fostering practical skills, the program serves as a preliminary model that demonstrates potential for building regional capacity and supporting the United Nations’ Sustainable Development Goal (SDG) #3 related to health and wellbeing and SDG #4 related to quality education and lifelong learning opportunities for all. Further research is required to evaluate its long-term scalability and clinical impact. Full article

As a critical component of sustainable energy systems, the efficient maintenance of photovoltaic (PV) panels is essential. While deep learning is an important approach for PV panel defect detection, the high complexity of existing models and their substantial computational demand make deployment on edge platforms difficult. This paper studies an acceleration method for photovoltaic panel defect detection on the Zynq-7020 heterogeneous platform. We design DARE-YOLO, a lightweight network for photovoltaic panel defect detection, together with a Zynq-based accelerator. In DARE-YOLO, we introduce RepConv and a lightweight single-path backbone to reduce the memory bandwidth overhead caused by multi-branch structures. We further design a Dilated Context Block (DCB) and a Dual-scale Decoupled Head (DDH), which effectively improve the detection accuracy of DARE-YOLO. On the Zynq platform, we develop the accelerator through a mixed fixed-point quantization strategy, a custom convolution IP core, and pipeline unrolling. These optimizations reduce data access latency, improve computational parallelism, and increase computational throughput. Experimental results show that DARE-YOLO achieves 93.84% mAP@0.5 with only 6.4 M parameters. The accelerator has a total on-board power consumption of only 1.95 W, while delivering a throughput of 37.5 GOPS, an energy efficiency of 19.23 GOPS/W. The image inference latency is 661.3 ms. This low-power, high-efficiency co-design paradigm ensures the long-term reliability of renewable energy facilities. Full article

Alzheimer’s disease (AD) is a serious neurodegenerative disorder that can severely affect behavior and thinking patterns, and is accompanied by frequent memory loss. The early diagnosis of AD is essential, as this can benefit the patient, but detecting AD is a complex process due to the nature of its associated clinical data. Electroencephalography (EEG) serves as a promising and cost-effective technique for analyzing AD-related brain activity patterns. In this work, a consolidated framework for detecting AD using EEG signals and hybrid models is proposed that uses a dataset that is available online. For the feature extraction module, five efficient techniques—Principal Component Analysis (PCA), Kernel Partial Least Squares (KPLS), Kriging Model, Isomap, and K-means clustering—are used. For feature selection, with the help of biomimetics-based concepts, three efficient algorithms are used: hybrid Cuckoo Search Optimization–Rat Swarm Optimization (CSO-RSO), Zebra Optimization (ZOA), and hybrid Gravitational Search Algorithm–Particle Swarm Optimization (GSA-PSO). Four interesting hybrid classifiers are utilized here to detect AD using EEG signals—hybrid Extreme Learning Machine–Adaboost (ELM–Adaboost), hybrid Classification and Regression Trees–Adaboost (CART–Adaboost), and hybrid weighted broad learning system-based Adaboost (HWBLSA), followed by a hybrid machine learning classification model with a soft voting technique—and, finally, these are compared with other standard machine learning classifiers. The highest classification accuracy of 98.71% is found when the Kriging Model feature extraction concept is combined with the hybrid GSA-PSO feature selection method and classified with the ELM–Adaboost classifier. Full article

A substantial proportion of construction accidents is associated with unsafe worker behavior. Identifying their underlying mechanism is vital for designing effective interventions. As prior studies could not capture complex nonlinear interactions among organizational and individual factors, this study leverages machine learning (ML) techniques, which can capture complex relationships by handling large datasets, and can identify patterns in worker behavior. The study proposes an explainable ML model to interpret key determinants of safe behavior. The data were collected from 425 construction workers in Saudi Arabia. Multiple ensemble and benchmark ML algorithms—including random forest (RF), categorical boosting, decision jungle, light gradient boosting machine, support vector machine, and adaptive boosting—were implemented and compared. The results indicate that the RF model achieved the best predictive performance, outperforming several competing models. To enhance the model’s interpretability, explainable artificial intelligence (XAI) techniques were applied to reveal the interaction of key predictors influencing workers’ behaviors. The results demonstrate that safety communication, risk perception, and supportive work environment are the most influential determinants shaping safety behavior. As a key novelty, this study introduces an ML-based approach for predicting construction workers’ safety behavior and applies XAI techniques to systematically interpret the key determinants of safety behavior. The results also provide valuable insights for safety managers and offer data-driven guidance to enhance the effectiveness of safety interventions. Full article

Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) is a complex, multi-system disease characterized by a multitude of symptoms across various organ systems. Diagnosis has relied heavily on heterogeneous clinical symptom presentation and evolving case definitions, with treatment focused on addressing presenting symptoms due to the paucity of validated biomarkers. Meanwhile, advances have been made in understanding the underlying pathophysiology through strong epidemiologic, clinical, and basic science studies. This narrative review synthesizes recent advances that are likely to drive a shift in understanding from symptom-based classification toward a molecularly defined understanding of the disease. This shift in understanding will likely provide the foundation for future research efforts focused on targeting diagnosis and treatment more effectively. Specifically, we reference the identification of rare genetic risk variants through the HEAL2 deep learning framework, the large-scale DecodeME genome-wide association study, and dynamic epigenetic markers of disease state. In addition, the findings revealed the downstream consequences of this genetic and epigenetic priming: chronic innate immune activation, CD8+ T cell exhaustion characterized by upregulation of the exhaustion-driving transcription factors Thymocyte Selection-Associated HMG Box (TOX) and Eomesodermin (EOMES), and a cellular energy crisis centered on mitochondrial dysfunction. Furthermore, results of recent studies have revealed sex-specific transcriptomic and proteomic signatures of maladaptive recovery. We also highlight the role of machine learning and artificial intelligence integrations in translating high-dimensional multi-omics data into actionable biological insights, including the identification of monocyte subsets via Positive Unlabeled Learning, circulating cell-free RNA diagnostic signatures, and integrated multi-modal disease models such as BioMapAI. The combination of these findings, which highlight multiple identifiable mechanisms of molecular activity, support the feasibility of molecular subtyping, precision diagnostics, and targeted therapeutic strategies for ME/CFS. Full article

To achieve low-carbon and cost-effective operation of multi-park electricity–heat–hydrogen integrated energy systems (EHHSs), this paper proposes a low-carbon dispatch framework based on federated safe reinforcement learning. First, a multi-park EHHS dispatch model is established by considering heterogeneous park characteristics, electricity–heat–hydrogen coupling, stepped carbon trading, and peer-to-peer (P2P) energy trading. Then, to address the coupled challenges of privacy preservation, operational coupling, and safety constraints, the dispatch problem is formulated as a constrained Markov decision process (CMDP). On this basis, a federated safe proximal policy optimization algorithm (FedSafePPO) is developed by integrating PPO, Lagrangian-based safety constraint handling, and federated parameter aggregation. The proposed method enables each park to learn a local dispatch policy from private data while sharing global knowledge without exchanging raw operational data. In addition, an actor–dual-critic architecture is adopted to jointly evaluate economic returns and constraint costs, thereby improving convergence stability and dispatch feasibility. Case studies involving three heterogeneous parks—industrial, commercial, and residential—demonstrate that the proposed method effectively reduces total operating costs and carbon emissions while satisfying system constraints. Compared with PPO, FedPPO, and SafePPO, the proposed FedSafePPO achieves superior low-carbon economic performance, greater training stability, and better adaptability to heterogeneous operating conditions. The results verify the effectiveness and engineering applicability of the proposed method for the low-carbon dispatch of multi-park EHHSs. Full article

Scene graph generation (SGG) provides a structured representation for visual understanding. However, most existing methods are designed to optimize global triplet recall rather than retrieve relation instances specified by a user query. In query-driven visual relation retrieval, two major challenges arise: the target relation must compete with a highly redundant candidate space, and query semantics are not incorporated before relation classification. To address these challenges, we propose a Query-Driven Candidate Relation Screening (QCRS) module, which injects query semantics into the candidate screening process. Specifically, QCRS encodes the query and candidate visual relation features, and then filters query-relevant candidates through relevance scoring. By reducing interference from irrelevant candidates and avoiding redundant computation, QCRS improves the final exact triplet hit performance and enhances the interpretability of query-specific relations, thereby facilitating query-driven visual relation retrieval. Built upon the strong EGTR baseline, QCRS learns query relevance to prioritize relation instances matching the target query, enabling precise triplet retrieval. Extensive ablation studies and analyses on the VG150 benchmark validate the effectiveness of the proposed approach: when integrated with EGTR, QCRS improves PairR@50 from 61.52% to 80.06% and ETR@50 from 30.54% to 47.07%, achieving absolute gains of over 16 percentage points in both correct object pair retention and end-to-end target relation retrieval performance. Full article