The full outbreak of various major security incidents has presented considerable challenges to many fields, including public crisis management, health and epidemic prevention, and high-risk place management. The objective of this study was to enhance the dynamic coordination of the emergency process in large venues, strengthen the prevention and prevention in advance, and improve the performance of emergency response management. The present case study was conducted in accordance with the provisions stipulated in the DB 32/T 3761.11-2020 standard. The software program SPSS 28 was utilised for the purpose of testing the reliability and validity of the questionnaire and 2-tuple linguistic method for language integrity, with the aim of an assessment system that would more accurately reflect the comprehensive management effect of large venues in the emergency response process. The findings of the study suggest that the membership degree of U-2 Personnel health management (2.35), U-3 High-risk site management (1.88) and U-4 Emergency response requirements (2.41) are suboptimal. The present study aims to address the shortcomings of large-scale event venues by identifying the weak links in management and establishing targeted improvement plans. This will result in a gradual improvement and continuous enhancement of the emergency management capabilities of large-scale event venues. Concurrently, the challenge lies in the further management of dynamic changes in semantic information.
A dynamic distributed dominant information system makes it challenging to achieve efficient reduction in data mining applications as it changes with time. This study aimed to focus on the incremental attribute reduction approach from the perspective of adding an attribute set. First, the dominant matrix and dominant conditional entropy of the distributed, dominant dataset were defined. Second, an incremental learning mechanism and a fusion mechanism of the dominant matrix were proposed. Then, a matrix-based incremental attribute reduction approach was presented. Finally, experiments were conducted on four University of California Irvine (UCI) datasets to validate the efficiency of the incremental attribute reduction approach. The results revealed that the running time of the incremental approach was reduced by an average of 87.9 % compared with the nonincremental approach. The software platform used was a Java environment in the Eclipse Platform.
Detecting tomato maturity is complicated by illumination changes, gradual color shifts, intra-class variability, and class imbalance, where single-modal approaches often struggle in intricate settings. To tackle this, we introduce TMMFNet (Tomato Multi-Modal Fusion Network), which combines a lightweight framework with cross-modal modeling. The architecture features a three-branch MobileNetV2 backbone for RGB, NIR, and Depth images to retain modality-specific characteristics. A tensor-level cross-attention module (EMOBlock) captures global correlations and ensures semantic consistency across modalities, improving discrimination in maturity transitions and reducing single-modal noise. Additionally, a modality credibility gating mechanism adjusts weights using global statistics for optimal fusion. Experiments on a multi-modal tomato dataset under four lighting conditions demonstrate that TMMFNet outperforms comparison methods in accuracy and F1-score, while maintaining low parameters and computational demands, facilitating real-time agricultural deployment.
Major emergencies not only cause enormous harm to society, but also have a stronger driving force for organizational crisis learning. Guiding universities to integrate social crisis resources has become an important component of the national major public safety strategy. In order to evaluate the satisfaction with the impact of major emergencies on crisis learning in universities. This study takes H University as a case and uses Grey Relational Analysis (GRA) to evaluate the impact of crisis event investment, crisis event response timeliness, individual and departmental learning duration, and the effectiveness of transformational improvement measures on the satisfaction of crisis learning in universities. The results indicate that the investment in crisis events has the highest relation on crisis learning in universities. The results imply that the best measure for handling major crisis events in universities is to prevent them beforehand rather than deal with them afterwards.
Addressing conservation challenges posed by deteriorating ancient buildings and mitigating risks from traditional surveying, this study proposes a digital, authenticity-preserving methodology that utilizes multisource data fusion (aerial, terrestrial, handheld). Integrating terrestrial laser scanning, handheld fine-scanning, and UAV oblique photogrammetry enabled the construction of comprehensive, millimeter-accurate 3D building models, overcoming technical barriers to non-destructive data acquisition for fragile, geometrically complex heritage. Furthermore, to counter high misclassification rates and opacity in traditional semantic recognition of carved components, explainable deep learning was innovatively integrated into point cloud analysis. This framework facilitates automatic, accurate identification of high-value decorative elements and structural damage (e.g., dislodged rafters, wood decay), while providing clear visualizations of the key features driving decisions. This capability directly addresses the core "verifiability" requirement of the restoration-as-original principle, delivering intuitive, reliable support for restoration planning. Validation on multi-scale architectural complexes demonstrates efficacy: carved component recognition accuracy exceeds 94.9%, multisource registration achieves 8 mm accuracy, and small-component measurement error is below 6.5 mm. The research provides a robust data foundation and intelligent analytical tools for the rescue of historic buildings, offering a novel pathway for refined conservation of complex heritage.
This paper introduces a concept-based approach for classification in group decision-making scenarios. We propose an improved algorithm to extract a compact concept-based from a given set of sample points. To address the complexity of multi-factor classification and the inconsistencies among multiple decision-makers, we define a new metric called preference entropy to quantify preference inconsistency. Based on this metric, we formulate an integer programming model to select an optimal set of representative cases that minimizes preference entropy while satisfying overall preference consistency. Using this optimal case set as training data, we extract a concept base for each decision category. We then develop a multi-valued classification principle by analyzing the relationship between a new scheme and the convex hulls of the concept bases for all categories. Finally, a case study on MBA program classification illustrates the feasibility and advantages of the proposed approach.
Spatial co-location pattern mining aims to discover sets of objects that are geographically adjacent and frequently co-occur. Recent enhancements in this field have introduced the fuzzy participation index (FPI) to quantify the prevalence of co-location patterns, thereby improving pattern discovery accuracy. However, several limitations remain: (1) When defining fuzzy row instances (clique instances), the current approach only considers pairwise fuzzy membership degrees, overlooking the joint membership degree among all instances within a clique. (2) The method calculates the fuzzy participation ratio (FPR) of an object using only the minimum membership degree between an instance and its neighbors in the clique as the instance's contribution index. (3) The calculation of fuzzy participation index (FPI) relies solely on the smallest FPR among all objects, ignoring the contributions of other objects. This paper proposes two new metrics based on the alpha-fuzzy cliques: the joint participation ratio (JPR) and the joint participation index (JPI). A novel algorithm is proposed based on JPR-JPI metrics, accompanied by theoretical proofs of its correctness and completeness. Finally, experiments on real-world datasets validate the effectiveness and performance of the proposed algorithm.
Green finance primarily channels funds toward green industries, while transition finance focuses on supporting high-carbon firms in their shift toward low-carbon development. Building on a heuristic model, this study theoretically analyzes the mechanisms through which transition finance instruments facilitate the low-carbon transformation of emission-intensive firms, and empirically tests the effects of multiple policy tools and their combinations using firm-level data from China's high-pollution industries during 2017 to 2022. The results show that both green bonds and risk-compensation mechanisms significantly promote the low-carbon transition of brown firms, and their combination generates strong synergistic effects. In particular, the sequencing of green bonds first, followed by risk compensation proves more effective in reducing transition costs and fostering low-carbon development. Heterogeneity analysis further reveals that large firms, with stronger technological reserves and risk-bearing capacity, are better able to leverage transition finance tools, while firms in highly competitive markets respond more actively to policy incentives due to market pressures. The contribution of this paper lies in uncovering the synergistic effects and sequencing mechanisms of transition finance instruments and in proposing differentiated policy pathways, thereby offering insights to improve policy efficiency and reduce firms' transition costs.
This study introduces a novel framework called HDL-MODTFO tuation Optimization), which is designed to enhance real-time virtual reality (VR) presentations of architectural heritage environments. Existing RGB- and LiDAR-based methods suffer from low accuracy, noise interference, and slow processing speed. To address these issues, HDL-MODTFO integrates solid-state Light Detection and Ranging (ss-LiDAR), pseudo-LiDAR, and Red-Green-Blue Depth (RGB-D) inputs through sensor fusion. It also involves adaptive fuzzy filtering and contrast enhancement through Enhanced Moth Swarm Optimization (EMSO) for noise removal. The preprocessed data are combined and then segmented using a Lightweight Generative Adversarial Network (LGAN) with channel and position attention mechanisms to improve accuracy. For object detection and classification, VGG-16 performs the extraction of features, while You Only Look Once (YOLOv4) performs classification. Finally, the Improved Unscented Kalman Filter (IUKF) is applied for accurate object tracking and timebased mapping. The simulation of the model is performed under a comprehensive dataset consisting of LiDAR and stereo RGB images. The outcomes demonstrate the model's performance, achieving a notable accuracy of 98%, precision of 95%, a recall of 90%, an F-score of 93%, and a reduction in computation time of up to 30 seconds compared to existing methods. These results confirm the effectiveness of HDL-MODTFO in enhancing the real-time detection and tracking of architectural elements in digital heritage VR environments.
The proliferation of Linux-based IoT malware poses a severe threat to critical infrastructure. Existing detection methods, however, suffer from two critical limitations: they lack the fine-grained resolution to distinguish between variants and the interpretability for traceable attribution. To address these challenges, we present MalHCExp, an interpretable and fine-grained classification framework for IoT malware. Our framework utilizes a hierarchical decision model built upon 42 core static features from a diverse dataset of 52,917 samples (spanning MIPS, ARM, and x86 architectures across 10 families and 114 variants). To ensure transparency, we integrate a SHAP-based module that provides clear and traceable decision pathways. Experimentally, MalHCExp achieves a variant-level F1-score of 0.866 and an accuracy of 0.873, significantly outperforming standard flat-classification baselines. Crucially, the model is highly efficient, with a low inference latency of 14.5 ms and a small memory footprint of 45 MB, underscoring its suitability for deployment on resource-constrained devices. The framework also demonstrates its practical utility by successfully attributing 5,882 unlabeled samples and revealing latent evolutionary relationships between major families. These capabilities highlight its potential to support automated security defense in real-world environments.
As cast-in-place piles are increasingly adopted in engineering practice, their concealed underground nature renders them vulnerable to quality defects such as necking, bulging, soil inclusion, and concrete segregation, any of which can compromise structural stability and service safety. This study proposes a non-destructive assessment framework that leverages Distributed Temperature Sensing (DTS) to evaluate borehole quality by real-time tracking of hydration-heat evolution during concrete curing. By integrating temperature-sensing optical cables with the finite-element method (FEM), a comprehensive detection scheme is established, with particular applicability to piles in high-groundwater environments. Indoor experiments confirmed the feasibility of DTS for capturing hydration-heat evolution, revealing positive correlations between both the rate and the peak of temperature rise and the thickness of the concrete placement. Outdoor model tests further indicated that deeper concrete layers exhibit lower temperature-rise rates because of diminished environmental interference, whereas shallow layers are more susceptible to ambient temperature fluctuations. Field application at a 500 kV substation in Suzhou demonstrated reliable pile-radius estimation using DTS-monitored data combined with FEM simulations, and the results closely matched mechanical caliper measurements, with a maximum deviation of less than 5%. This work advances the application of distributed fiber-optic sensing in geotechnical engineering by offering a real-time, non-destructive alternative to conventional detection methods while maintaining high fidelity in defect identification.
Large-amplitude vortex-induced vibration (VIV) in long-span bridges can compromise safety and accelerate structural fatigue, underscoring the need for real-time monitoring and suppression. This study introduces a Classification algorithm for the VIV Development Process (CVIVDP), which monitors vibration states and issues early warnings via a trained classification model. The CVIVDP framework involves four steps: (1) categorizing the VIV process into five classes and annotating vertical response signals under various wind velocities; (2) converting VIV signals into RGB images using continuous wavelet transform (CWT) to reveal latent features through color-based geometric patterns; (3) training a GoogLeNet-based classifier adapted for the five-class task; and (4) outputting warnings based on classified real-world signals. Experimental results show that CVIVDP achieves precision, recall, and F1-scores of 92.86%, 94.44%, and 92.94%, respectively, outperforming existing methods. The proposed approach offers a viable solution for IoT-enabled bridge health monitoring.
The authentication of Chinese paintings presents unique challenges due to the subtle variations in artistic styles, techniques, and historical contexts. Traditional methods rely heavily on expert knowledge, which is often subjective and time-consuming. Recent advances in deep learning offer promising solutions, but most existing approaches focus on Western art or generic image classification, failing to address the distinctive characteristics of Chinese ink wash paintings. To bridge this gap, we proposed a novel deep learning framework for automated Chinese painting analysis by integrating EfficientNet and ResNet through hierarchical feature fusion. To tackle the complexities of style and technique analysis, we constructed two datasets. Our framework leverages ResNet with transfer learning to capture complementary features and EfficientNet as a backbone for multi-scale representation. Cross-dataset evaluation demonstrated the model ' s robustness, achieving 97.34% accuracy in painter identification and 98.3% accuracy in technique classification & horbar;surpassing existing methods. These results validate the potential of hybrid deep learning architectures for art historical research and forgery detection, offering practical tools for museums, collectors, and digital humanities scholars.
Transportation ranks as the third-largest contributor to carbon emissions, following the industrial and construction sectors. An accurate and comprehensive framework for calculating project-specific carbon emissions is crucial for implementing effective carbon reduction strategies. This study clarifies the production flow and operational processes of highway bridge reconstruction and expansion projects, delineating the carbon source boundaries for the demolition, new construction, operation, and maintenance phases, considering the specific characteristics of these projects. Based on these project features, a carbon emission calculation methodology is proposed that integrates the life cycle assessment (LCA) framework with the carbon flow diagram approach. The carbon reduction potential is subsequently evaluated through case studies. A comparative analysis of carbon emissions in an actual bridge reconstruction and expansion project reveals that the proposed method outperforms traditional approaches in terms of both accuracy and computational efficiency. Therefore, this methodology offers valuable theoretical and practical guidance for achieving carbon reduction targets in similar reconstruction and expansion projects.
Key core technology innovation (KCTI) in enterprises is linked to achieving high-level scientific and technological self-reliance and self-improvement at the national level. Using panel data from Chinese A-share listed companies between 2015 and 2023, and employing the quasi-natural experiment of the establishment of industrial intellectual property operation centers (IIPOCs), this study applies a double machine learning (DML) model to examine the impact of industrial intellectual property operations on KCTI in Chinese enterprises. The results indicate that industrial intellectual property operations significantly promote KCTI in enterprises. The impact of IIPOCs is more pronounced in the eastern regions and among technology-intensive enterprises. Mechanism tests further show that IIPOCs stimulate KCTI by increasing R&D investment, enhancing senior management's attention to innovation, and fostering corporate participation in collaborative R&D alliances. The study offers insightful insights that government authorities could capitalize upon to encourage the development of industrial intellectual property operation centers, which will help enterprises achieve significant technical advancements.
This study addresses the demand for crop disease recognition in edge computing environments by proposing an ONNX-optimized lightweight ResNet50D-CBAM model. The model features three main contributions: (1) introducing the CBAM attention mechanism to enhance feature extraction capability; (2) designing a model compression strategy that combines structured pruning with dynamic FP16 quantization; (3) implementing efficient edge deployment optimization based on ONNX Runtime. Experimental results demonstrate that the optimized model maintains 95.5% recognition accuracy while reducing parameters by 38%, inference latency by 30%, and memory usage significantly. In simulated edge environments, inference latency is only 85ms with GPU utilization below 37%. The model maintains recognition accuracy above 91% under various interference conditions, providing a practical solution for edge deployment in agricultural intelligence systems.
Ailerons, initially designed for roll control, hold promise as secondary actuators for lateral control in jetliners. This study examines the influence of integrating these redundant control surfaces (i.e., ailerons) on lateral control performance through the derivation of novel analytical bounds. Specifically, the least upper bound and greatest lower bound for the control cost are established within a linear-quadratic regulator framework modified for redundant inputs. These bounds are shown to correspond to the Rayleigh quotient and generalized eigenvalues derived from the unique positive definite solutions of the associated Riccati equations. Application of these theoretical bounds demonstrates that incorporating ailerons yields a marginal reduction in optimal lateral control cost - i.e., less than 1% - relative to the non-redundant system. Such a minimal reduction signifies limited practical utility for lateral control, and consequently, ailerons exhibit insufficient efficacy to justify their deployment as active components in primary lateral control systems.
This study proposed an effective valuation method for cultural and media enterprises, combining the Free Cash Flow to the Firm (FCFF) model with the Black-Scholes (B-S) option pricing model under a sub-mixed fractional Brownian motion (fBm) model, with an aim to enhance the scientific rigor and accuracy of valuation for such enterprises. The study addressed the deficiencies in the standard valuation methods for cultural and media enterprises through the proposed hybrid model. The FCFF model was used to calculate their existing value, and the B-S and sub-mixed fBm B-S models were used to estimate their potential value. Finally, companies listed in China's cultural and media industry were selected as research objects, and the aforementioned two combined valuation methods were used separately to conduct the empirical analysis. Comparison of the valuation results of the models before and after optimization revealed that the improved combined model demonstrated higher accuracy and reliability in valuing cultural and media enterprises. This indicated that the combined FCFF and sub-mixed fBm B-S model was more aligned with the unique characteristics of cultural and media enterprises, such as distinct operation mode, high growth potential, and a large number of intangible assets, thus helping to comprehensively and accurately capture their true value.
Tensor decomposition has established itself as a powerful analytical framework for processing multimedia data, such as computer vision and signal processing. The tubal nuclear norm (TNN) has demonstrated remarkable effectiveness in addressing challenging tensor-related tasks, such as tensor-robust principal component analysis (TRPCA). This formulation can be rigorously interpreted as the nuclear norm of an underlying circulant matrix, which simultaneously captures tube-direction correlations while maintaining the intrinsic row and column structural relationships. Nonetheless, video data exhibits far more substantial (row and column) spatial redundancy compared to(tube) temporal redundancy. To harness the rich spatial redundancy in video data, we introduce a novel norm known as the double twist tubal nuclear norm (Dt-TNN), designed to jointly capture both row and column correlations. Building upon this, we propose a Dt-TNN-based model for TRPCA, which holds significant applications in video processing, including video restoration from severe corruptions. These optimization problems are subsequently addressed using the ADMM framework. Comprehensive experimental results derived from video restoration provide compelling evidence for the efficacy of the proposed model.
On-site loudspeaker announcements and video broadcasts are key tools used to control crowds during large-scale incidents. This study aimed to develop a model illustrating a simplified process by which an agent receives audiovisual media stimuli, experiences emotional contagion and transformation, activates working memory, and ultimately exhibits behavioral changes. The simulation experiments were then conducted. The model was simulated with 100 agents configured into 3 roles: police officers, core agitators, and bystanders. The simulation examined various strategies, particularly focusing on electronic screen placement and narrative content design, to assess their correlation with crowd dispersion rates. The results indicated that the placement of electronic screens was more critical than their size or quantity, narrative strategies emphasizing punitive consequences contributed more significantly to dispersal rates, and loudspeakers demonstrated more pronounced effects when used in combination with electronic screens.
