Search Results (27,493)

Reconfigurable Intelligent Surface (RIS) has emerged as a promising enabling technology for wireless communications, which significantly enhances system performance through real-time manipulation of electromagnetic wave reflection characteristics. In RIS-assisted communication systems, existing deep learning-based channel state information (CSI) feedback methods often suffer from excessive parameter requirements and high computational complexity. To address this challenge, this paper proposes LwCSI-Net, a lightweight autoencoder network specifically designed for RIS-assisted multiple-input single-output (MISO) systems, aiming to achieve efficient and low-complexity CSI feedback. The core contribution of this work lies in an innovative lightweight feedback architecture that deeply integrates multi-layer convolutional neural networks (CNNs) with attention mechanisms. Specifically, the network employs 1D convolutional operations with unidirectional kernel sliding, which effectively reduces trainable parameters while maintaining robust feature-extraction capabilities. Furthermore, by incorporating an efficient channel attention (ECA) mechanism, the model dynamically allocates weights to different feature channels, thereby enhancing the capture of critical features. This approach not only improves network representational efficiency but also reduces redundant computations, leading to optimized computational complexity. Additionally, the proposed cross-channel residual block (CRBlock) establishes inter-channel information-exchange paths, strengthening feature fusion and ensuring outstanding stability and robustness under high compression ratio (CR) conditions. Our experimental results show that for CRs of 16, 32, and 64, LwCSI-Net significantly improves CSI reconstruction performance while maintaining fewer parameters and lower computational complexity, achieving an average complexity reduction of 35.63% compared to state-of-the-art (SOTA) CSI feedback autoencoder architectures. Full article

Climate concerns are driving the automotive industry to adopt advanced manufacturing technologies that aim to improve energy efficiency and reduce vehicle weight. In this context, lightweight structural materials such as aluminium alloys have gained significant attention due to their favorable strength-to-weight ratio. Laser welding plays a crucial role in assembling such materials, offering high flexibility and fast joining capabilities for thin aluminium sheets. However, welding these materials presents specific challenges, particularly in controlling heat input to minimize distortions and ensure consistent weld quality. As a result, numerical simulations based on the Finite Element Method (FEM) are essential for predicting weld-induced phenomena and optimizing process performance. This study investigates welding-induced distortions in laser butt welding of 1.5 mm-thick Al 6061 samples through FEM simulations performed in the SYSWELD 2024.0 environment. The methodology provided by the software is based on the Moving Heat Source (MHS) model, which simulates the physical movement of the heat source and typically requires extensive calibration through destructive metallographic testing. This transient approach enables the detailed prediction of thermal, metallurgical, and mechanical behavior, but it is computationally demanding. To improve efficiency, the Imposed Thermal Cycle (ITC) model is often used. In this technique, a thermal cycle, extracted from an MHS simulation or experimental data, is imposed on predefined subregions of the model, allowing only mechanical behavior to be simulated while reducing computation time. To avoid MHS-based calibration, this work proposes using thermal cycles acquired in-line during welding via infrared thermography as direct input for the ITC model. The method was validated experimentally and numerically, showing good agreement in the prediction of distortions and a significant reduction in workflow time. The distortion values from simulations differ from the real experiment by less than 0.3%. Our method exhibits a slight decrease in performance, resulting in an increase in estimation error of 0.03% compared to classic approaches, but more than 85% saving in computation time. The integration of real process data into the simulation enables a virtual representation of the process, supporting future developments toward Digital Twin applications. Full article

Aiming to enable intelligent vehicles to achieve autonomous charging under low-battery conditions, this paper presents a navigation system for autonomous charging that integrates an improved bidirectional A* algorithm for path planning and an optimized YOLOv11n model for visual recognition. The system utilizes the improved bidirectional A* algorithm to generate collision-free paths from the starting point to the charging area, dynamically adjusting the heuristic function by combining node–target distance and search iterations to optimize bidirectional search weights, pruning expanded nodes via a greedy strategy and smoothing paths into cubic Bézier curves for practical vehicle motion. For precise localization of charging areas and piles, the YOLOv11n model is enhanced with a CAFMFusion mechanism to bridge semantic gaps between shallow and deep features, enabling effective local–global feature fusion and improving detection accuracy. Experimental evaluations in long corridors and complex indoor environments showed that the improved bidirectional A* algorithm outperforms the traditional improved A* algorithm in all metrics, particularly in that it reduces computation time significantly while maintaining robustness in symmetric/non-symmetric and dynamic/non-dynamic scenarios. The optimized YOLOv11n model achieves state-of-the-art precision (P) and mAP@0.5 compared to YOLOv5, YOLOv8n, and the baseline model, with a minor 0.9% recall (R) deficit compared to YOLOv5 but more balanced overall performance and superior capability for small-object detection. By fusing the two improved modules, the proposed system successfully realizes autonomous charging navigation, providing an efficient solution for energy management in intelligent vehicles in real-world environments. Full article

Objectives: Spinal diseases are commonly encountered health problems with a wide spectrum. In addition to degenerative changes, other common spinal pathologies include metastases and compression fractures. Benign tumors like hemangiomas and infections such as spondylodiscitis are also frequently observed. Although magnetic resonance imaging (MRI) is considered the gold standard in diagnostic imaging, the morphological similarities of lesions can pose significant challenges in differential diagnoses. In recent years, the use of artificial intelligence applications in medical imaging has become increasingly widespread. In this study, we aim to detect and classify vertebral body lesions using the YOLO-v8 (You Only Look Once, version 8) deep learning architecture. Materials and Methods: This study included MRI data from 235 patients with vertebral body lesions. The dataset comprised sagittal T1- and T2-weighted sequences. The diagnostic categories consisted of acute compression fractures, metastases, hemangiomas, atypical hemangiomas, and spondylodiscitis. For automated detection and classification of vertebral lesions, the YOLOv8 deep learning model was employed. Following image standardization and data augmentation, a total of 4179 images were generated. The dataset was randomly split into training (80%) and validation (20%) subsets. Additionally, an independent test set was constructed using MRI images from 54 patients who were not included in the training or validation phases to evaluate the model’s performance. Results: In the test, the YOLOv8 model achieved classification accuracies of 0.84 and 0.85 for T1- and T2-weighted MRI sequences, respectively. Among the diagnostic categories, spondylodiscitis had the highest accuracy in the T1 dataset (0.94), while acute compression fractures were most accurately detected in the T2 dataset (0.93). Hemangiomas exhibited the lowest classification accuracy in both modalities (0.73). The F1 scores were calculated as 0.83 for T1-weighted and 0.82 for T2-weighted sequences at optimal confidence thresholds. The model’s mean average precision (mAP) 0.5 values were 0.82 for T1 and 0.86 for T2 datasets, indicating high precision in lesion detection. Conclusions: The YOLO-v8 deep learning model we used demonstrates effective performance in distinguishing vertebral body metastases from different groups of benign pathologies. Full article

The sustainable integration of Distributed Energy Resources (DER) with the next-generation distribution networks requires robust, adaptive, and accurate hosting capacity (HC) forecasting. Dynamic Operating Envelopes (DOE) provide real-time constraints for power import/export to the grid, ensuring dynamic DER integration and efficient network operation. However, conventional HC analysis and forecasting approaches struggle to capture temporal dependencies, the impact of DOE constraints on network operation, and uncertainty in DER output. This study introduces a dynamic optimization framework that leverages the benefits of the sensitivity gate of the Sensitivity-Enhanced Recurrent Neural Network (SERNN) forecasting model, Particle Swarm Optimization (PSO), and Bayesian Optimization (BO) for HC forecasting. The PSO determines the optimal weights and biases, and BO fine-tunes hyperparameters of the SERNN forecasting model to minimize the prediction error. This approach dynamically adjusts the import/export of the DER output to the grid by integrating the DOE constraints into the SG-PSO-BO architecture. Performance evaluation on the IEEE-123 test network and a real Australian distribution network demonstrates superior HC forecasting accuracy, with an $R^2$ score of 0.97 and 0.98, Mean Absolute Error (MAE) of 0.21 and 0.16, and Root Mean Square Error (RMSE) of 0.38 and 0.31, respectively. The study shows that the model effectively captures the non-linear and time-sensitive interactions between network parameters, DER variables, and weather information. This study offers valuable insights into advancing dynamic HC forecasting under real-time DOE constraints in sustainable DER integration, contributing to the global transition towards net-zero emissions. Full article

Interpreting diagnostic imaging and identifying clinically relevant features remain challenging tasks, particularly for novice radiologists who often lack structured guidance and expert feedback. To bridge this gap, we propose an Eye-Gaze Guided Multimodal Fusion framework that leverages expert eye-tracking data to enhance learning and decision-making in medical image interpretation. By integrating chest X-ray (CXR) images with expert fixation maps, our approach captures radiologists’ visual attention patterns and highlights regions of interest (ROIs) critical for accurate diagnosis. The fusion model utilizes a shared backbone architecture to jointly process image and gaze modalities, thereby minimizing the impact of noise in fixation data. We validate the system’s interpretability using Gradient-weighted Class Activation Mapping (Grad-CAM) and assess both classification performance and explanation alignment with expert annotations. Comprehensive evaluations, including robustness under gaze noise and expert clinical review, demonstrate the framework’s effectiveness in improving model reliability and interpretability. This work offers a promising pathway toward intelligent, human-centered AI systems that support both diagnostic accuracy and medical training. Full article

Objective: Shoulder dystocia (ShD) is a rare but serious obstetric emergency associated with significant neonatal morbidity. This study aimed to evaluate the predictive performance of machine learning (ML) models based on fetal biometric ratios and clinical characteristics for the identification of ShD in pregnancies without clinical suspicion of macrosomia. Methods: We conducted a retrospective case-control study including 284 women (84 ShD cases and 200 controls) who underwent spontaneous vaginal delivery between 37 and 42 weeks of gestation. All participants had an estimated fetal weight (EFW) below the 90th percentile according to Hadlock reference curves. Univariate and multivariate logistic regression analyses were performed on maternal and neonatal parameters, and statistically significant variables (p < 0.05) were used to construct adjusted odds ratio (aOR) models. Supervised ML models—Logistic Regression (LR), Random Forest (RF), and Extreme Gradient Boosting (XGB)—were trained and tested to assess predictive accuracy. Performance metrics included AUC-ROC, sensitivity, specificity, accuracy, and F1-score. Results: The BPD/AC ratio and AC/FL ratio markedly enhanced the prediction of ShD. When added to other features in RF models, the BPD/AC ratio got an AUC of 0.884 (95% CI: 0.802–0.957), a sensitivity of 68%, and a specificity of 83%. On the other hand, the AC/FL ratio, along with other factors, led to an AUC of 0.896 (95% CI: 0.805–0.972), 68% sensitivity, and 90% specificity. Conclusions: In pregnancies without clinical suspicion of macrosomia, ML models integrating fetal biometric ratios with maternal and labor-related factors significantly improved the prediction of ShD. These models may support clinical decision-making in low-risk deliveries where ShD is often unexpected. Full article

Continuous blood pressure (BP) monitoring provides valuable insight into the body’s dynamic cardiovascular regulation across various physiological states such as physical activity, emotional stress, postural changes, and sleep. Continuous BP monitoring captures different variations in systolic and diastolic pressures, reflecting autonomic nervous system activity, vascular compliance, and circadian rhythms. This enables early identification of abnormal BP trends and allows for timely diagnosis and interventions to reduce the risk of cardiovascular diseases (CVDs) such as hypertension, stroke, heart failure, and chronic kidney disease as well as chronic stress or anxiety disorders. To facilitate continuous BP monitoring, we propose an AI-powered estimation framework. The proposed framework first uses an expert-driven feature engineering approach that systematically extracts physiological features from photoplethysmogram (PPG)-based arterial pulse waveforms (APWs). Extracted features include pulse rate, ascending/descending times, pulse width, slopes, intensity variations, and waveform areas. These features are fused with demographic data (age, gender, height, weight, BMI) to enhance model robustness and accuracy across diverse populations. The framework utilizes a Tab-Transformer to learn rich feature embeddings, which are then processed through an ensemble machine learning framework consisting of CatBoost, XGBoost, and LightGBM. Evaluated on a dataset of 1000 subjects, the model achieves Mean Absolute Errors (MAE) of 3.87 mmHg (SBP) and 2.50 mmHg (DBP), meeting British Hypertension Society (BHS) Grade A and Association for the Advancement of Medical Instrumentation (AAMI) standards. The proposed architecture advances non-invasive, AI-driven solutions for dynamic cardiovascular health monitoring. Full article

Exploring the impact of national park construction on county-level livelihood and well-being holds significant implications for enhancing social livelihood. This study treats Wuyishan National Park Construction (WNPC) as a quasi-natural experiment, utilizing panel data from 138 counties (2011–2023) to construct a county-level livelihood and well-being index through the CRITIC weighting method. Kernel density estimation and the Theil index are applied to depict the spatiotemporal dynamics of WNPC. Moreover, the difference-in-differences model and mediating effect model are employed to assess the impact and mechanisms of WNPC on livelihood and well-being. The results reveal that, in the period 2011–2023, livelihood and well-being scores ranged from 0.1329 to 0.4565, indicating considerable scope for improvement. Over time, inter-county disparities narrowed, displaying a spatial pattern of “higher in the east and west, lower in the middle.” Overall disparities remained pronounced, driven chiefly by within-region variation, and Jiangxi displayed notably larger internal gaps than Fujian and Zhejiang. Benchmark regressions confirm that WNPC significantly improved livelihood and well-being, with robust results according to multiple tests. Mechanism analysis indicates that WNPC enhances livelihood and well-being by promoting population mobility and improving infrastructure. Heterogeneity analysis suggests that compared to industrial counties, WNPC has a stronger positive effect on the livelihood and well-being of agricultural counties. Based on this, it is suggested that WNPC promotes population mobility and improves infrastructure construction. This study provides a scientific basis and decision-making reference for achieving high-quality construction of national parks and enhancing livelihood and well-being. Full article

Large language models (LLMs), built upon Transformer architectures, have demonstrated remarkable performance in a wide range of natural language processing tasks. However, their practical deployment—especially in long-context scenarios—is often hindered by the computational and memory costs associated with managing the key–value (KV) cache during inference. Optimizing this process is therefore crucial for improving LLM efficiency and scalability. In this study, we propose a novel entropy-guided KV caching strategy that leverages the distribution characteristics of attention scores within each Transformer layer. Specifically, we compute the entropy of attention weights for each head and use the average entropy of all heads within a layer to assess the layer’s contextual importance. Higher-entropy layers—those exhibiting broader attention dispersion—are allocated larger KV cache budgets, while lower-entropy (sink-like) layers are assigned smaller budgets. Instead of selecting different key–value tokens per head, our method selects a common set of important tokens per layer, based on aggregated attention scores, and caches them uniformly across all heads within the same layer. This design preserves the structural integrity of multi-head attention while enabling efficient token selection during the prefilling phase. The experimental results demonstrate that our approach improves cache utilization and inference speed without compromising generation quality. For example, on the Qwen3 4B model, our method reduces memory usage by 4.18% while preserving ROUGE score, and on Mistral 0.1v 7B, it reduces decoding time by 46.6%, highlighting entropy-guided layer analysis as a principled mechanism for scalable long-context language modeling. Full article

Speaker profiling systems are often evaluated on a single corpus, which complicates reliable comparison. We present a fully reproducible evaluation pipeline that trains Convolutional Neural Networks (CNNs) and Long-Short Term Memory (LSTM) models independently on three speech corpora representing distinct recording conditions—studio-quality TIMIT, crowdsourced Mozilla Common Voice, and in-the-wild VoxCeleb1. All models share the same architecture, optimizer, and data preprocessing; no corpus-specific hyperparameter tuning is applied. We perform a detailed preprocessing and feature extraction procedure, evaluating multiple configurations and validating their applicability and effectiveness in improving the obtained results. A feature analysis shows that Mel spectrograms benefit CNNs, whereas Mel Frequency Cepstral Coefficients (MFCCs) suit LSTMs, and that the optimal Mel-bin count grows with corpus Signal Noise Rate (SNR). With this fixed recipe, EfficientNet achieves 99.82% gender accuracy on Common Voice (+1.25 pp over the previous best) and 98.86% on VoxCeleb1 (+0.57 pp). MobileNet attains 99.86% age-group accuracy on Common Voice (+2.86 pp) and a 5.35-year MAE for age estimation on TIMIT using a lightweight configuration. The consistent, near-state-of-the-art results across three acoustically diverse datasets substantiate the robustness and versatility of the proposed pipeline. Code and pre-trained weights are released to facilitate downstream research. Full article

Developing agricultural new productive forces (ANPF) according to local conditions is a key strategy for agricultural modernization. Using panel data from 16 prefecture-level cities in Anhui Province from 2010 to 2022, this study constructed indicator systems for ANPF and the construction of a strong agricultural province (CSAP). The entropy-weight TOPSIS method was used to calculate the levels of ANPF and the SAP index. This study employed a modified gravity model and social network analysis (SNA) to investigate the spatial correlation and evolutionary characteristics of these networks. Geographical detectors were also used to identify the driving factors behind agricultural transformation. The findings indicate that both ANPF and CSAP showed an upward trend during the study period, with significant regional heterogeneity, with Central Anhui being the most prominent. This study revealed spatial spillover effects and strong network correlations between ANPF and CSAP, with the spatial network structure exhibiting characteristics of multi-core, multi-association, and multidimensional connections. The entities within the network are tightly connected, with no “isolated island” phenomenon, and Hefei, as the central hub, showed the highest number of connections. Laborer quality, tangible means of production, and new-quality industries emerged as the core driving forces, working in synergy to propel CSAP. This study contributes new insights into the spatial network dynamics of agricultural development and offers actionable recommendations for policymakers to enhance agricultural modernization globally. Full article

Ergonomic risk classification during manual lifting tasks is crucial for the prevention of workplace injuries. This study addresses the challenge of classifying lifting task risk levels (low, medium, and high risk, labeled as 0, 1, and 2) using multi-modal time-series data comprising raw facial landmarks and bio-signals (electrocardiography [ECG] and electrodermal activity [EDA]). Classifying such data presents inherent challenges due to multi-source information, temporal dynamics, and class imbalance. To overcome these challenges, this paper proposes a Multi-Adaptive Functional Neural Network (Multi-AdaFNN), a novel method that integrates functional data analysis with deep learning techniques. The proposed model introduces a novel adaptive basis layer composed of micro-networks tailored to each individual time-series feature, enabling end-to-end learning of discriminative temporal patterns directly from raw data. The Multi-AdaFNN approach was evaluated across five distinct dataset configurations: (1) facial landmarks only, (2) bio-signals only, (3) full fusion of all available features, (4) a reduced-dimensionality set of 12 selected facial landmark trajectories, and (5) the same reduced set combined with bio-signals. Performance was rigorously assessed using 100 independent stratified splits (70% training and 30% testing) and optimized via a weighted cross-entropy loss function to manage class imbalance effectively. The results demonstrated that the integrated approach, fusing facial landmarks and bio-signals, achieved the highest classification accuracy and robustness. Furthermore, the adaptive basis functions revealed specific phases within lifting tasks critical for risk prediction. These findings underscore the efficacy and transparency of the Multi-AdaFNN framework for multi-modal ergonomic risk assessment, highlighting its potential for real-time monitoring and proactive injury prevention in industrial environments. Full article

In the context of globalization and the Belt and Road Initiative, maintaining the stability and security of the China Railway Express network (CRN) is critical for international logistics operations. However, unexpected events can lead to node and edge failures within the CRN, potentially triggering cascading failures that critically compromise network performance. This study introduces a Coupled Map Lattice model that incorporates cargo flow dynamics, distributing cargo based on distance and the residual capacity of neighboring nodes. We analyze cascading failures in the CRN under three scenarios, isolated node failure, isolated edge disruption, and simultaneous node and edge failure, to assess the network’s vulnerability during emergencies. Our findings show that deliberate attacks targeting cities with high node strength result in more significant damage than attacks on cities with a high node degree or betweenness. Additionally, when edges are disrupted by unexpected events, the impact of edge removals on cascading failures depends on their strategic position and connections within the network, not just their betweenness and weight. The study further reveals that removing collinear edges can effectively slow the propagation of cascading failures in response to deliberate attacks. Furthermore, a single-factor cargo flow allocation method significantly enhances the network’s resilience against edge failures compared to node failures. These insights provide practical guidance and strategic support for the CR Express in mitigating the effects of both unforeseen events and intentional attacks. Full article

In this paper, a linearized mixed finite volume element (MFVE) scheme is proposed to solve the nonlinear time fractional fourth-order reaction–diffusion models with the Riemann–Liouville time fractional derivative. By introducing an auxiliary variable $\sigma =\Delta u$, the original fourth-order model is reformulated into a lower-order coupled system. The first-order time derivative and the time fractional derivative are discretized by using the BDF2 formula and the weighted and shifted Grünwald difference (WSGD) formula, respectively. Then, a fully discrete MFVE scheme is constructed by using the primal and dual grids. The existence and uniqueness of a solution for the MFVE scheme are proven based on the matrix theories. The scheme’s unconditional stability is rigorously derived by using the Gronwall inequality in detail. Moreover, the optimal error estimates for u in the discrete $L^{\infty }\left(L^2(\Omega )\right)$ and $L^2\left(H^1(\Omega )\right)$ norms and for $\sigma$ in the discrete $L^2\left(L^2(\Omega )\right)$ norm are obtained. Finally, three numerical examples are given to confirm its feasibility and effectiveness. Full article