Search Results (74,866)

This study proposes a lightweight crack-segmentation model optimized for industrial and edge-computing environments, where both high accuracy and real-time inference are required. Conventional convolution-based and U-Net-based crack segmentation models offer relatively simple architectural designs, but often suffer from limited boundary precision or an unfavorable accuracy–efficiency trade-off. Swin Transformer-based approaches can model broader contextual information but may still show poor segmentation quality relative to their computational cost in fine crack analysis. To address these limitations, we propose the Stabilized Crack Network++ (SCN++), a U-Net backbone crack segmentation network that integrates edge fusion, hybrid loss with deep supervision, exponential moving average (EMA)-based stabilization, and lightweight post-processing. The model was trained and evaluated on 40,000 concrete surface images, including 20,000 crack images and 20,000 non-crack images, using quantitative metrics such as intersection over union (IoU), Dice coefficient, frames per second (FPS), giga floating-point operations (GFLOPs), and the number of parameters, together with overlay-based qualitative analysis. Compared with the CNN, U-Net, and Swin Transformer baselines, SCN++ achieved the best overall balance between segmentation accuracy and computational efficiency, with an IoU of 0.7346, a Dice coefficient of 0.8457, 35.09 FPS, 8.45 GFLOPs, and only 2.22 M parameters. These results demonstrate that SCN++ effectively mitigates the conventional accuracy–efficiency trade-off and is a strong candidate for practical structural crack segmentation in edge-computing environments. Full article

Multimodal Named Entity Recognition (MNER) aims to integrate textual and visual information to identify entities with specific semantic categories. However, existing methods often suffer from insufficient intra-modal semantic modeling, coarse cross-modal alignment, and vulnerability to noisy or ambiguous expressions in social media. To address these challenges, we propose a Visual–Semantic Guided Interaction Network (VSGN), which improves multimodal representation learning from both semantic and structural perspectives. Specifically, we first design an adaptive visual–semantic fusion module that incorporates visual descriptions as semantic guidance, enabling more informative cross-modal interactions. To further enhance feature quality, we introduce a deviation-aware channel-wise inhibitory routing (CIR) mechanism, which jointly models channel importance and distributional deviation to suppress noisy or redundant visual signals. In addition, we propose a visual–semantic guided graph structure learning module (VSG), which explicitly captures structural dependencies across modalities. By enforcing distribution-level alignment between textual and visual graph representations, the model achieves structure-aware cross-modal interaction and reduces modality inconsistency. Extensive experiments on the Twitter-2015 and Twitter-2017 datasets demonstrate the effectiveness of the proposed method, achieving F1 scores of 76.72% and 87.86%, respectively. The results show that jointly modeling semantic enhancement and structural alignment leads to more robust and discriminative multimodal representations. Full article

To address issues existing in current multi-media pollution assessment, such as data mismatch, parameter conflicts, and inadequate characterization of nonlinear effects, this study developed a multi-factor synergistic assessment methodological system encompassing “data preprocessing-parameter calibration-damage quantification-model coupling”. A three-stage parameter calibration system of “inheritance-linkage-sensitivity screening” was established to achieve cross-media parameter synergy; an Environmental Damage Entropy (EDE) model was constructed based on information entropy to quantify the nonlinear coupled damage of multiple factors; and the optimal governance threshold was determined by combining the coupling theory of marginal damage and governance cost. Taking a multi-media pollution incident (atmosphere-soil-surface water-groundwater) caused by a chemical plant explosion as a case study, pollution chain identification, damage quantification, ecological risk cascading effect analysis, and health risk assessment were conducted. The results show that this method can accurately identify key pollution pathways. Based on the calculation of Environmental Damage Entropy (EDE = 0.604) and the synergy coefficient (δ = 1.32), the comprehensive damage value was quantified as 8.21 million yuan. Additionally, the threshold exceedance characteristics of various media were identified, reflecting the cumulative and lagging nature of ecological risk cascading effects. The method proposed in this study can accurately identify key pollution pathways and quantify comprehensive damage as well as ecological risks, providing scientific support for the allocation of multi-media pollution governance responsibilities and precise prevention and control. Full article

Introduction: Idiopathic pulmonary fibrosis (IPF) is associated with high morbidity and mortality and a substantial burden of comorbidities, which may influence prognosis and survival. This study aimed to evaluate the burden of comorbidity in patients with IPF receiving antifibrotic therapy using the Charlson, TORVAN, and GAP indices and to analyse their relationships and prognostic impact on survival. Methods: Retrospective observational study including patients with IPF diagnosed according to ATS/ERS/JRS/ALAT criteria. Patients receiving antifibrotic therapy between June 2010 and September 2025 were included. Baseline comorbidities were recorded, and the Charlson, TORVAN, and GAP indices were calculated. Associations between indices were assessed using chi-square tests and kappa statistics. Survival was analysed using Kaplan–Meier curves and compared with the log-rank test. Cox proportional hazards regression and model comparison metrics (Harrell’s C-index and Akaike Information Criterion) were also performed to assess the independent prognostic value of each index. Results: Seventy-two patients were included (76.7% male; mean age 73.8 ± 7.4 years). Pirfenidone was prescribed in 63.9% and nintedanib in 36.1%. The most frequent comorbidities were gastro-oesophageal reflux disease (62.5%), arterial hypertension (57.5%), pulmonary hypertension (32.9%), diabetes mellitus (24.7%), and non-metastatic solid tumours (17.6%), including lung cancer. Survival differed significantly according to GAP stage (p = 0.020) and Charlson categories (p = 0.006). The TORVAN stage was associated with the GAP stage (p < 0.001; kappa = 0.246), whereas the Charlson index showed no association with GAP or TORVAN. Conclusions: In this cohort of patients with IPF receiving antifibrotic therapy, both the GAP and Charlson indices were associated with survival. These findings suggest that combining disease-specific and comorbidity indices may provide a more comprehensive prognostic assessment, although further validation in larger cohorts is required. Full article

Estimating the state of charge (SoC) is a critical task for effective management of electric vehicle batteries. Simple machine learning methods (LR, KNN, etc.) often suffer from limited prediction accuracy, while deep learning approaches (LSTM, CNN, etc.) generally require high computational resources and behave as black-box models with limited explainability. To overcome these limitations, the present work proposes a SoC estimation approach based on the Light Gradient Boosting Machine (LightGBM). The proposed model provides a balanced trade-off between prediction accuracy and computational efficiency. Furthermore, feature engineering is performed to derive additional informative features, improving the model’s ability to learn driving conditions and battery dynamics. In addition, the study incorporates a seasonal analysis by evaluating the model under both summer and winter conditions, allowing the impact of environmental variations on SoC estimation performance to be investigated. Moreover, Explainable Artificial Intelligence (XAI) techniques are employed to interpret the model predictions. Evaluation on real-world on-road data demonstrated that the proposed model achieved substantial improvements in estimation performance compared to recent studies. Full article

This paper introduces a policy-search-based reinforcement learning algorithm aimed at generating optimal set-points of wind turbines in wind farms. The proposed approach addresses the problem of multivariable optimization in systems where the objective function is unknown or difficult to model. The algorithm is a model-free framework and relies solely on measured performance of the system. Namely, it does not require gradient information of the objective function or an explicit model of the aerodynamic interaction between wind turbines. The proposed scheme utilizes stochastic policy perturbations to explore the search space and update the policy parameters directly based on the observed reward signal. In this way, the algorithm progressively drives the control variables toward optimal operating conditions. The proposed policy-search reinforcement learning framework is analyzed to establish its connection with gradient-free optimization methods. The proposed method is applied to wind farm power optimization, where multiple turbine control variables must be adjusted in the presence of wake interactions cooperatively. The performance of the proposed approach is evaluated through extensive simulations under both steady-state and time-varying wind conditions. The proposed algorithm is compared with an extremum-seeking control method that was previously suggested for the same problem. The results demonstrate that the proposed approach is able to effectively maximize power production in wind farms while maintaining a simple and model-free optimization structure. Full article

In top-N recommendation, incorporating item-side information can improve ranking quality under sparse user–item interactions; however, common flat concatenation strategies may weaken the structural correspondence between user ratings and item attributes while simultaneously increasing model size. To address this issue, this study proposes DAERec-GCA, a deep autoencoder-based collaborative filtering framework that organizes rating signals and genre information in a genre-channel-aligned two-dimensional representation. The model applies shared weights across genre channels and aggregates channel outputs to generate item scores, enabling side-information integration without the parameter growth associated with flattened genre-aware formulations. The framework was evaluated on MovieLens-100K, 1M, and 10M under a warm-start five-fold cross-validation protocol using ranking-based metrics. In addition, a structured ablation study was conducted against ROnly, Flat1D, GenreProfile, GenreEmbed, and GenreGated, together with a controlled train-side sparsity analysis and a computational profiling analysis covering trainable parameters, epoch time, inference latency, and peak GPU memory. The results show that DAERec-GCA remains competitive across all three datasets and exhibits its clearest advantage under sparse and moderately sparse training conditions. The findings suggest that genre-channel alignment provides a practical trade-off between structural expressiveness, parameter efficiency, and recommendation quality in sparse recommendation settings. Full article

The ability to continuously learn over time by incorporating new information while holding onto previously acquired expertise is known as incremental learning (IL). Although this concept is fundamental to human learning, existing machine learning techniques have a significant propensity to forget prior experience by overwriting previously learned patterns from classes. The continuous learning of new information in K-nearest neighbor (KNN) with lazy learning strategies compounds to loss of old knowledge upon learning new information and stability-plasticity dilemma. The change in new data points and data distributions in unforeseen ways impacts KNN’s ability to adapt to changes in class label distribution, leading to concept drift. This experiment models a hybrid 3WDKNN-based incremental learning algorithm (ILA) designed for application in a heterogeneous and dynamically changing environment. This model addresses the limitations of KNN by overcoming computational costs and inefficiencies associated with loss of information in classes, while facilitating incremental learning to attain high predictive accuracy in crop yield datasets. The algorithm employs weighted voting to identify optimal assigned classes for the nearest neighbor and uses memory reconstruction strategy for class incremental learning until the memory is full without forgetting. Using weighted voting for the best assigned classes for the nearest neighbor, the algorithm uses a local mean vector to determine the best distances for the shortest-term incremental learning to achieve the highest performance accuracy in a concept drift environment. The hybrid 3WDKNN_ILA was developed and evaluated alongside advanced algorithms within the same dataset context. The model improves performance in incremental learning contexts by utilizing current concepts and minimizing errors on both current and recent data to avoid parameterization. The model achieves optimal efficient incremental learning by mitigating intentional loss and minimizing errors associated with valuable class information derived from aggregated mean values through class rectification and transfer. The hybrid model achieves the best efficient performance accuracy in all the tested weighted averages of 200W, 500W, and 1000W with tested set K values of 5, 9, and 13K. This hybrid model demonstrates performance accuracy of 97% at a value of 13K, whereas 3WD_KNN achieves 96% at 9K, HoKNN attains 89% at 13K, and 1IKNN reaches 88% at 9K accuracy, respectively. The integrated novelty in the hybrid 3WDKNN_ILA proves superior in terms of computational efficiency, accuracy, and high-level incremental performance and learning in comparison with other tested models of algorithms. Full article

This study addresses the challenge of simplifying massive multibeam seafloor topographic point cloud datasets featuring significant spatial heterogeneity. We propose a feature joint entropy-based quantification method for seafloor terrain complexity, which provides a foundation for the adaptive and differentiated simplification of point clouds. In this method, the elevation and slope features of point clouds are treated as two-dimensional random variables that describe terrain morphology; we estimate the Shannon entropy of their joint distribution by constructing a two-dimensional adaptive histogram and use the entropy value to quantify the topographic information content and complexity of local regions. To overcome the parameter sensitivity and subjective dependence inherent in traditional fixed-bin methods, we incorporate the Minimum Description Length (MDL) principle to guide binning optimization, taking the sum of stochastic complexity and model coding length as the evaluation criterion. A dimension-alternating optimization strategy combining dynamic programming and an iterative greedy algorithm is adopted to solve for the optimal binning structure, thus achieving data-driven adaptive binning. To ensure the fairness and reliability of quantification, we adopt a fixed-point number partitioning strategy to decompose the point cloud into several independent analysis nodes and determine the minimum sample size supporting the stable estimation of entropy values through convergence analysis. Experimental results demonstrate that the proposed method, as a consistent and data-driven complexity metric, can reliably reflect the relative complexity of different seafloor terrain regions, thereby providing an objective quantitative basis for subsequent differentiated point cloud simplification. Full article

Delivery services have drawn much attention and become of topmost significance in urban areas by presenting online food delivery selections for a diversity of dishes from a wide range of restaurants, decreasing both travel and waiting times. Customer data analysis acts as a cornerstone in corporate strategy, allowing enterprises to gather and interpret user feedback and helping them to make informed decisions that drive future business development. However, major knowledge gaps remain due to the scarcity of literature review studies on these delivery services, hindering a complete understanding of customer satisfaction in this sector. Furthermore, there has been little systematic research on managerial response tactics to online consumer complaints and negative reviews. Researchers have contributed by applying artificial intelligence, including deep learning and machine learning models, to analyze customer sentiment and understand customer brand perceptions. This study presents a Hybrid Deep Neural Network Model for Customer Satisfaction Analysis (HDNNM-CSA), with the aim of developing an efficient model which is capable of accurately classifying customer satisfaction levels in delivery apps based on textual responses provided by customer experience managers. To achieve this, the model initially pre-processes text data using text cleaning, emoji removal, normalization, tokenization, stop word removal, and stemming to clean and standardize the unstructured text data for further analysis. Following this, term frequency–inverse document frequency-based word embedding is utilized to transform the pre-processed text into meaningful feature representations. Lastly, an ensemble architecture involving bidirectional long short-term memory, temporal convolutional, and graph convolutional networks is deployed to classify customer satisfaction levels with managers’ responses. A series of experimental analyses are performed, and the results are examined for numerous features. A comparative analysis demonstrates the enhanced performance of the HDNNM-CSA technique with respect to existing approaches. Full article

Background: The AptoDetect™-Lung assay is an aptamer-based test designed for risk assessment in patients with pulmonary nodules, but its potential role in predicting lung cancer stage has not been evaluated. We investigated whether the assay could predict advanced-stage disease beyond conventional diagnostic modalities. Methods: This multicenter prospective cohort study enrolled 1672 patients with Lung-RADS 3–4 pulmonary nodules across ten university-affiliated hospitals in South Korea between June 2023 and December 2024. Among them, 934 patients with histologically confirmed lung cancer were retrospectively selected, and 852 patients were included in the final analysis after exclusions. The AptoDetect™-Lung assay was performed before invasive diagnostic procedures. Results: Among the 852 patients, 450 (52.8%) had advanced-stage disease. The AptoDetect™-Lung score was significantly higher in advanced-stage than in early-stage lung cancer (median, 6.2 vs. 2.8, p < 0.001). In a multivariable logistic regression analysis, a high AptoDetect™-Lung score (≥5) was independently associated with advanced disease (odds ratio 1.99, 95% confidence interval 1.35–2.95, p < 0.001). The AptoDetect™-Lung assay showed moderate discrimination of advanced-stage disease (area under the curve [AUC] 0.696) and in non–small cell lung cancer (AUC 0.720), whereas its discriminative ability was limited in small cell lung cancer (AUC 0.561). A combined prediction model incorporating the AptoDetect™-Lung assay, serum CEA, and radiologic findings demonstrated improved discriminative performance (AUC 0.821). Conclusions: The AptoDetect™-Lung assay score was independently associated with advanced-stage lung cancer and could provide clinically useful information for early risk stratification before definitive diagnosis and staging are available. Full article

Accurate short-term load forecasting at disaggregated levels is critical for energy management in microgrids and institutional environments, yet it remains a challenge due to high consumption variability and limited contextual information. This paper proposes a hybrid model that combines Fuzzy ARTMAP neural networks with K-means clustering to improve hourly load forecasting using real data from a university microgrid. The methodology includes key preprocessing steps such as filtering low-load records, removing holidays, interpolating missing values, and applying cyclic encoding to standardize the data into 96 time intervals per day (15-min resolution). For each prediction, the average load profile of the five most recent weekdays is computed and compared to cluster centroids to identify the most similar group, which is then used to train the neural network. Results demonstrate consistent improvements in MAPE, RMSE, and MAE compared to the non-clustered baseline. The model showed robustness to non-stationary behavior and atypical patterns, even when relying solely on timestamp and load data. The proposed strategy outperformed conventional approaches and proved suitable for complex, data-limited environments. Full article

In the field of drone navigation systems, a high-precision positioning solution can be provided by an integrated strapdown inertial navigation system (SINS)/global navigation satellite system (GNSS). But when satellite signals are interfered with or blocked by tall buildings, the errors of SINS will disperse rapidly due to the complex air and mechanical vibrations, leading to a serious degradation of navigation accuracy. To enhance the positioning performance in this situation, this paper proposes a hybrid information fusion method based on attention-augmented temporal convolutional network (TCN) for drone SINS/GNSS navigation system. A feature integration and prediction model is constructed to provide a pseudo-positioning reference for the integrated navigation filter during GNSS-denied periods, in which TCN is used to establish a predictive positioning error correction model based on inertial measurements and SINS data, while a self-attention model is incorporated to extract complex global drone motion features. The performance of the proposed method has been experimentally verified using Global Positioning System (GPS) and SINS data collected from real drone flight test. Comparison results among the proposed model, SINS with TCN, SINS with convergent Kalman filter (KF) prediction section and SINS-only indicate that the proposed method can effectively improve the drone positioning accuracy in specific GNSS-denied environments. Full article

This paper examines whether a residual structural state extracted from cross-asset downside-risk dependence contains incremental information for forecasting next-day market downside risk beyond a strong heterogeneous autoregressive (HAR) benchmark. The empirical analysis uses Binance intraday data from September 2019 to December 2025 and a fixed sample of 24 liquid cryptocurrencies obtained through explicit data-quality screening and sample diagnostics. The forecasting target is the log of an equal-weight cross-sectional downside-risk index constructed from strictly valid asset-level realized downside semivariance measures. The empirical design is deliberately conservative: the market sample is fixed ex ante, the target is evaluated against Bitcoin (BTC) and Ethereum (ETH) dominance diagnostics, and asset-level HAR-type models are estimated recursively to generate ex-ante one-step-ahead residuals, from which rolling residual-dependence matrices and structural signatures are constructed. The selected residual state contains four components: average residual correlation, Frobenius-type deformation, influence concentration, and influential-set turnover. The evidence supports three qualified conclusions. First, the full residual state attains the lowest average QLIKE loss relative to the HAR benchmark, although the corresponding Diebold–Mariano test under the primary QLIKE loss does not reject equal predictive accuracy at conventional levels. Complementary Clark–West evidence on the nested log-scale comparison supports incremental predictive content for the level-state and full-state augmentations. Second, the strongest forecasting evidence comes from the full state rather than from deformation-only specifications. Third, event-window diagnostics show that structural reorganization is most pronounced around stress-entry and extreme-risk episodes, supporting an onset-sensitive rather than a long-lead early-warning interpretation. Overall, the evidence supports a cautious and statistically qualified predictive conclusion: residual market structure may contain incremental information for short-horizon downside-risk forecasting in cryptocurrency markets, especially around stress onset, but the result should not be interpreted as a decisive primary-loss improvement or as evidence that deformation alone dominates a strong benchmark. Full article

Semantic segmentation of high-resolution remote-sensing images is critical for urban planning, land-cover mapping, and ecological monitoring. However, existing methods face limitations in handling complex land-cover types, multi-scale objects, and modeling long-range dependencies. To address these challenges, we propose RFA2Net, an enhanced semantic segmentation model based on the DeepLabv3+ framework. The key innovations include the integration of the RFCSA-Conv module into the ResNet101 backbone to enhance feature representation and long-range dependency modeling, the design of the RFA-DASPP structure built upon the Dense ASPP framework with the novel RFCA-DConv dilated convolution module to reduce information loss during multi-scale feature fusion and enhance the model’s ability to perceive long-range directional structures, and the introduction of a Dual-Branch Fusion Network to improve segmentation accuracy for small-scale objects. Experimental results on the ISPRS Potsdam and LoveDA datasets demonstrate that RFA2Net outperforms several CNN and Transformer-based models, achieving 78.94% and 59.46% mean intersection over union (mIoU) on the ISPRS Potsdam and LoveDA datasets, respectively, with improvements of 3.19% and 3.08% over the original DeepLabv3+. Ablation studies and comparative experiments further confirm the model’s effectiveness, robustness, and practical applicability in high-resolution remote-sensing image segmentation, with particular relevance to environmental monitoring and sustainable energy applications. Full article