Search Results (6,265)

Structured radiology reporting can improve clinical decision support by standardizing clinical findings into hierarchical formats. However, thousands of questions in structured report templates about clinical findings are prohibitively time-consuming, which can limit clinical adoption. Furthermore, early medical VQA datasets primarily focused on free-text and independent question–answer pairs while a recent dataset, Rad-ReStruct, introduced a hierarchical VQA, but the accompanying model still relies heavily on flattened embedding representations and single-path text–image fusion mechanisms that inadequately handle complex hierarchical dependencies in responses. In this paper, we propose DPA-HiVQA (Dual-Path Cross-Attention for Hierarchical VQA), addressing these limitations through two key contributions: (1) multi-scale image embedding representing global semantic embeddings with patch-level spatial features from domain-specific BioViL encoder; (2) dual-path cross-attention mechanism enabling simultaneous holistic semantic understanding and fine-grained spatial reasoning. Evaluated on the Rad-ReStruct benchmark, the model substantially outperforms the established benchmark baseline with an overall F1-score and Level 3 F1-score improvement by 21.2% and 31.9%, respectively. The proposed model demonstrates that dual-path cross-attention architectures can effectively connect holistic semantic understanding and fine-grained spatial detail, paving the way for practical AI-assisted structured reporting systems that reduce radiologist burden while maintaining diagnostic accuracy. Full article

Power systems with high renewable penetration impose multi-dimensional demands on energy storage (ES) regulation. Short-duration ES is required for power balance and frequency support, while medium- and long-duration ES is essential for daily, weekly, and seasonal peak shaving and energy time-shifting. Aiming at the challenge of multi-timescale configuration of hybrid energy storage (HES) in the initial planning stage of carbon-neutral transition, this paper proposes an optimal configuration strategy combining STL-PDM-VMD. First, the seasonal and trend decomposition using Loess (STL) is used to extract quarterly trends of annual net power for seasonal ES configuration. Then, the Past Decomposable Mixing (PDM) module in the time-mixer model is applied to decouple and mix multi-scale features of the detrended power curve for monthly and weekly configurations. Finally, an improved Variational Mode Decomposition (VMD) is adopted to decompose daily net power fluctuations and optimize intra-day energy storage schemes. Based on actual data from a carbon-neutral transition region, simulations are carried out and compared with the VMD method with decomposition layers optimized by Gurobi. The results show that the proposed STL-PDM-VMD multi-timescale hybrid energy storage configuration strategy can effectively capture the multi-timescale fluctuation characteristics of net load, significantly improve the Renewable Energy (RE) penetration rate, and ensure the power and energy balance of the new power system at multiple timescales. penetration, and maintain power and energy balance in the new-type power system. Full article

In modern intelligent food processing, the unpredictable variability in pomelo orientation on high-speed conveyors poses a significant challenge to automated grading and precision peeling operations. To address this, a deep learning-based method is proposed for the real-time detection of pomelo posture. Firstly, a pomelo posture dataset was constructed to support model training and validation. Secondly, to balance the extraction of posture features from uniform fruits with the low-power constraints of edge deployment, a domain-specific architectural optimization is presented. Building on the YOLOv8n framework, the proposed model synergistically integrates specialized modules. A lightweight GhostHGNetV2 foundation is utilized to significantly reduce computational redundancy while maintaining the resolution required to detect key anatomical landmarks. To overcome spatial confusion and capture multi-scale global appearance information, a multi-path coordinate attention (MPCA) module is introduced. Furthermore, the SlimNeck architecture and VoVGSCSP module streamline multi-scale feature fusion via one-time aggregation, effectively preventing computational bottlenecks. This design optimizes the computational efficiency of the model while maintaining detection accuracy. Experimental results demonstrate that compared with the baseline YOLOv8n model, the proposed method increased the mAP50 accuracy by 3.67% while reducing parameter count and computational load by 17.5% and 23.3%, respectively. Additionally, it achieved a processing speed of 19.3 FPS on the Jetson Orin Nano 6G edge platform. This research provides a critical technical foundation for the recognition of pomelo posture, enabling subsequent orientation rectification and fostering the development of streamlined, automated pomelo processing lines. Full article

Accurate crop classification in complex and heterogeneous agricultural landscapes is often challenged by mixed-pixel effects and spatial autocorrelation. This study proposes a prior-guided crop classification framework that integrates accessible Moderate Resolution Imaging Spectroradiometer (MODIS) optical and Sentinel-1 synthetic aperture radar (SAR) time-series data with explicit phenological and structural priors. By embedding physically meaningful constraints into temporal feature learning, the model shifts from purely data-driven learning toward biophysically interpretable discrimination between crop types and background classes. Performance was rigorously evaluated using spatial cross-validation (SCV) to ensure geographic independence. Results demonstrate that the prior-guided CNN achieves an overall accuracy (OA) of 98.66% and a Kappa of 0.9832, outperforming unguided deep learning and conventional machine learning models. Notably, the framework exhibits high spatial robustness, with a minimal performance gap between random and spatial validation (ΔOA = 0.0049). In addition to improving classification accuracy, integrating phenological features with SAR-based prior information enhances the stability of non-crop categories in fragmented scenarios, while leveraging readily available medium-resolution data to support large-scale applications. These findings demonstrate that embedding physically meaningful prior knowledge into multi-source time-series learning improves classification accuracy while enhancing spatial generalizability and interpretability. More broadly, the proposed framework offers a transferable paradigm for integrating domain knowledge with deep learning, providing a practical and scalable solution for crop mapping in heterogeneous agricultural landscapes using widely accessible medium-resolution data. Full article

Peanut leaf disease detection in complex field environments faces two major challenges: distinguishing visually similar symptoms and identifying severely occluded lesions. This study presents YOLOv8-MSDH, a lightweight detection model built upon an improved YOLOv8 framework to address these issues. Four architectural enhancements are introduced. First, the MHSA attention mechanism is integrated to enhance sequential feature dependency modeling and suppress background noise. Second, the Slim-Neck module is adopted for neck reconstruction, which lowers computational cost and facilitates multi-scale feature fusion. Third, the original C2f module is replaced with the C2f-Dual module to further reduce computational load. Last, the HWD downsampling module is incorporated into the backbone to improve the retention of disease-specific features while promoting lightweight design. Evaluated on a peanut leaf disease dataset, YOLOv8-MSDH achieves 90.14% precision, 82.16% recall, 90.71% mAP50 and 72.14% mAP50-95 under complex conditions—surpassing the baseline YOLOv8 by 3.5, 0.7, 1.5 and 2.89 percentage points, respectively. Parameter count and computational complexity are reduced by 12.9% and 20.7%, confirming effective lightweighting. Operating at 509.95 FPS, the model maintains strong real-time performance and exhibits high robustness across varying lighting conditions. Full article

Transformers serve as crucial hubs for power transmission, but during no-load energization, the nonlinear magnetization of their cores frequently induces extreme magnetizing inrush currents. Current suppression methods encounter challenges regarding transient feature extraction and excessive circuit complexity. To overcome these limitations, this study develops a high-fidelity model of a 100 kVA transformer using MATLAB/Simulink to investigate the interaction between residual flux and the closing angle. Extensive simulations were executed across a closing phase angle range of 0° to 360° and a residual flux domain of −0.8 p.u. to 0.8 p.u. Furthermore, this study utilizes Wavelet and Clarke transforms to extract characteristic parameters and quantitatively analyze the transients within the energy domain, enabling a multi-scale assessment of the mitigation efficacy based on these extracted features. The analytical results demonstrate that an optimal pre-magnetization distribution of −0.8 p.u. for Phase A, 0 p.u. for Phase B, and 0.8 p.u. for Phase C, coupled with a target closing angle of 330°, achieves the best suppression. This strategy strictly clamps the peak inrush current to 1.5 times the rated current, significantly outperforming conventional demagnetization alone. Consequently, this highly pronounced mitigation effect provides robust support for reliable transformer protection and overall power grid security. Full article

Background: Multi-center resting-state functional magnetic resonance imaging (rs-fMRI) is important for neurodevelopmental disorder diagnosis, but cross-site differences in repetition time (TR) can cause temporal feature misalignment. In addition, blood-oxygen-level-dependent (BOLD) signals are non-stationary, so disease-related information may be distributed across multiple time scales. Existing methods usually do not explicitly model physical sampling intervals or coordinate temporal and spectral information across scales, which may limit cross-site generalization in heterogeneous multi-center settings. Methods: We propose Spec-RWKV, a spectrum-guided linear recurrent framework for multi-site rs-fMRI diagnosis. It includes three components: PrismTimeMix, which models temporal dynamics using decay rates derived from physical half-lives and converts them adaptively across TRs; a TR-adaptive continuous wavelet transform, which aligns spectral representations across sites by adjusting frequency coverage; and spectrum-guided adaptive temporal aggregation, which uses spectral context to weight temporal features. Results: On ABIDE-I and ADHD-200, Spec-RWKV achieved AUCs of 75.86% and 76.31%, respectively. Under leave-one-site-out validation, it achieved the best mean AUC on ABIDE-I and the best mean accuracy and AUC on ADHD-200. Conclusions: Spec-RWKV explicitly models sampling-rate differences and multi-scale spectral structure, with results supporting strong cross-site generalizability. Full article

The cooperative airdrop of UAV swarms by multiple transport aircraft creates a large-scale multi-agent planning problem. The mission involves heterogeneous aircraft, multi-visit airdrop areas, strict time windows, and threat-aware flight paths. To address these challenges, this work develops an integrated framework for both global task allocation and real-time replanning in complex three-dimensional operational environments. First, for the combinatorial optimization of task execution sequences across multiple aircraft, a static task assignment method is proposed. This method employs a Hybrid-encoding Constrained Black-winged Kite Algorithm (HCBKA), which incorporates optimization metrics such as mission execution time, completion rate, and load-balancing symmetry among aircraft. The HCBKA aims to find a task assignment scheme that achieves a comprehensive optimum across multiple objectives through efficient model solving. Second, to handle potential real-time dynamic changes during mission execution, a rapid-response and generalizable replanning mechanism is developed. This mechanism utilizes an event-triggered strategy based on a Time-window aware Dynamic Auction Algorithm (TDAA). It ensures that the system can promptly initiate and execute online task reallocation in response to contingencies such as changing mission requirements or losses within its own drone swarm, thus maintaining the adaptability and robustness of the overall plan. Simulation results show that the proposed framework produces high-quality global solutions and maintains strong robustness under dynamic changes. The approach provides an effective and scalable solution for coordinated multi-aircraft swarm airdrop missions. Full article

For complex systems with complex structures and limited fault data, expert knowledge-based methods are often preferred. The multi-signal flow graph model is a representative qualitative approach, but generating a minimum-cost diagnostic strategy under this model becomes increasingly difficult as system scale grows, leading to excessive diagnostic cost or unacceptable computation time. To address this issue, this paper reformulates the minimum-cost diagnostic strategy generation problem (MDP) as a functional extremum value problem (FEVP) through a multipartite graph model, which decomposes the recursive strategy-generation process into deterministic combination and nondeterministic elimination. This reformulation is motivated by the observation that only a small fraction of generated fault subsets contributes to the optimal strategy. Based on this insight, a fault elimination neural network with an intelligent algorithm (ENI) structure is proposed to learn an elimination function that retains promising subsets while discarding redundant ones. The neural network scores candidate fault subsets from fixed-dimensional statistical features, while the intelligent algorithm optimizes the non-differentiable objective. Simulation and real-case results show that the proposed method reduces diagnostic cost by more than 30% while maintaining acceptable computation time, with greater advantages on larger-scale systems. Full article

To address the difficulty of detecting demagnetization severity in permanent magnet synchronous motors (PMSMs), this paper proposes a demagnetization severity detection method based on temperature signal and Convolutional Neural Network (CNN). First, the differences between local demagnetization and eccentricity fault in stator current harmonics are analyzed from an electromagnetic perspective, and fast Fourier transform (FFT) is used for frequency-domain analysis of the stator current to identify local demagnetization faults. On this basis, an electromagnetic–thermal coupling model is established by considering motor losses and heat dissipation boundary conditions to obtain the winding temperatures under different demagnetization severities and operating conditions. Furthermore, the temperature time series, together with speed and load torque, is constructed into a three-dimensional state space, and the proposed Conditionally Modulated Multi-Scale Convolutional Neural Network (CMSCNN) is introduced for feature learning to achieve demagnetization severity detection. Experimental results show that the proposed method achieves an average detection accuracy of 98.06% on the simulation test set and outperforms the baseline CNN model. On measured data collected from the faulty prototype, the average detection accuracy reaches 93.34%, verifying the effectiveness of the proposed method for demagnetization severity detection. Full article

Accurate, high-resolution emission inventories are essential for air quality modeling and policy evaluation, yet national-scale inventories for Thailand remain limited in spatial and temporal detail. This study develops a comprehensive national emission inventory for Thailand in 2019 (EI–TH 2019), covering 12 major air pollutants and greenhouse gases across key sectors, including energy, transport, industry, agriculture, waste, and residential activities. The inventory is constructed using country-specific activity data from official statistics and sectoral surveys, combined with GAINS-consistent emission factors and control assumptions. Emissions are resolved at 1 × 1 km spatial resolution and monthly temporal resolution to capture Thailand-specific emission dynamics. The results show that emissions across major pollutants are dominated by a limited number of source groups, with biomass burning and residential solid-fuel use driving particulate matter, transport dominating NO_x and CO emissions, large-scale combustion and industry controlling SO₂ emissions, and agriculture contributing the majority of NH₃ emissions. Strong seasonal variability is observed in PM_2.5, CO, and NH₃, primarily driven by dry-season biomass burning, whereas NO_x and SO₂ exhibit relatively stable temporal patterns. The reliability of EI–TH 2019 is supported by a multi-dimensional evaluation framework. Temporal consistency is demonstrated through strong agreement between modeled PM_2.5 emissions and ground-based observations, as well as between NO_x emissions and satellite-derived TROPOMI NO₂ (r = 0.93; ρ = 0.96). Biomass burning timing is further validated using satellite fire activity (VIIRS), showing consistent seasonal patterns. Comparisons with global inventories (EDGAR v8.1, HTAP v3.2, and GFED5.1) reveal systematic differences in sectoral contributions, temporal profiles, and emission magnitudes, particularly for biomass burning, reflecting the importance of country-specific data and assumptions. Overall, EI–TH 2019 provides a robust, high-resolution, and policy-relevant emission dataset that improves the representation of emission processes in Thailand. The results highlight key priority sectors—biomass burning, transport, industry, and agriculture—for targeted emission-reduction strategies and support applications in chemical transport modeling, exposure assessment, and integrated air-quality and climate-policy analysis. Full article

Photovoltaic (PV) power stations are pivotal for the renewable energy transition, yet their operational efficiency is often compromised by defects such as surface dust accumulation and cracks. Traditional manual inspections are labor-intensive and subjective, while conventional monitoring methods struggle with environmental interference. This study proposes YOLO-MSG, a lightweight framework specifically designed for the automated detection of PV module defects during system operation, including normal panels as well as defective conditions such as dusty and cracked panels. The methodology integrates a Multi-Scale Grouped Convolution (MSGC) module for enhanced feature extraction and a Group-Stem Decoupled Head (GSD-Head) to reduce parameter redundancy. Furthermore, a joint optimization strategy involving LAMP and logits-based knowledge distillation is employed to facilitate edge deployment. Experimental results on a specialized PV defect dataset demonstrate that YOLO-MSG achieves a superior balance between detection accuracy and computational cost. Compared to state-of-the-art models like YOLO11 and YOLOv12, YOLO-MSG significantly reduces GFLOPs and parameter count while maintaining highly competitive mean Average Precision (mAP), with improvements of 1.35% in mAP and 2.37% in mAP50-95 over the baseline models. Specifically, the model achieves an average inference speed of 90.30 FPS on the NVIDIA Jetson AGX platform. These findings confirm the algorithm’s industrial viability, providing a robust and efficient solution for the real-time automated maintenance of photovoltaic infrastructures. Full article

Accurate sleep stage classification is essential for evaluating sleep quality, yet clinical polysomnography is impractical for continuous home-based monitoring. Ballistocardiography (BCG) enables unobtrusive sleep monitoring through sensors embedded in sleep furniture; however, existing BCG-based approaches either rely on complex physiological feature extraction or employ fixed-parameter signal-to-image transformations that cannot adapt to inter-subject variability. This study proposes a learnable recurrence plot (RP) framework for three-stage sleep classification (Wake, NREM, REM) from single-channel BCG signals. The Learnable RP introduces three innovations: multi-scale phase-space reconstruction at physiologically motivated time delays (τ = 5, 10, 20), differentiable per-scale thresholds optimized end-to-end, and attention-based spatial fusion of multi-scale recurrence maps. The framework was evaluated through 10-fold stratified cross-validation across six backbone architectures using 50 overnight recordings. The Learnable RP consistently outperformed four baseline transformation methods (GAF, MTF, Classical RP, Modified RP), achieving an aggregate mean accuracy of 73.60%, with EfficientNet-B5 reaching 78.91%. and 78.91%. Statistical validation across all 24 pairwise comparisons (4 baselines × 6 backbones) confirmed consistent superiority (all p < 0.001). The proposed framework achieves competitive performance without explicit physiological feature engineering, offering a viable path toward end-to-end unobtrusive sleep monitoring. Full article

The output power of photovoltaic (PV) systems is influenced by various environmental factors, exhibiting strong nonlinearity and non-stationarity, which poses significant challenges for accurate forecasting. To address these issues, this paper proposes a short-term PV power forecasting method based on wavelet convolutional neural networks and an improved Transformer. First, the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) is employed to decompose the original PV power sequence into several intrinsic mode functions (IMFs). Fuzzy entropy is then utilized to evaluate the complexity of each component, and subsequences with similar entropy values are reconstructed to reduce the non-stationarity of the original series. Subsequently, Pearson correlation coefficients and the maximal information coefficient (MIC) are applied to capture both linear and nonlinear relationships between each reconstructed component and meteorological features, enabling the selection of strongly correlated variables. On this basis, a wavelet convolutional network (WTConv) is introduced to perform multi-scale decomposition and frequency-band feature extraction on the reconstructed components by integrating wavelet transform with convolution operations, effectively expanding the receptive field and extracting deep-seated features of the sequences. Finally, an improved iTransformer model is adopted for time-series modeling, leveraging its inverted encoding structure and self-attention mechanism to fully capture long-term dependencies among multivariate variables. The proposed model is validated using actual power data from a PV plant in Ningxia, China, across four seasons. Comprehensive experiments, including ablation studies, comparative analyses, loss function convergence evaluation, and Diebold–Mariano significance tests, are conducted to thoroughly assess the model’s effectiveness and superiority. Experimental results demonstrate that the proposed model achieves excellent prediction accuracy and stability in spring, summer, autumn, and winter, showing strong potential for engineering applications. Full article

Aiming at the significant variation in clay content within the orebody of the Letpadaung copper mine in Monywa, Myanmar, this study conducted comprehensive research on ore classification based on clay content and its impact on bio-heap leaching performance. Pressure filtration tests confirmed that clay content is a critical factor affecting ore permeability and copper leaching efficiency. Accordingly, a classification standard centered on clay content was established based on the ore properties of the Letpadaung Copper Mine, categorizing the ore into four types: low-clay ore (<3%), medium-clay ore Type 1 (3%–10%), medium-clay ore Type 2 (10%–25%), and high-clay ore (>25%). Corresponding differentiated stacking strategies were proposed and applied for the first time in industrial operations at a scale of several hundred thousand tons. Industrial practice results demonstrated that, compared with the unclassified mixed ore with a leaching efficiency of 45.92%, the implementation of classified heap leaching increased the copper leaching rates of low-clay ore and medium-to-high-clay ore to 68.07% and 63.41%, respectively. Moreover, multi-layer heap leaching within the same leaching unit showed consistent leaching efficiency across different layers after ore classification. These results further validate that ore classification based on clay content combined with differentiated heap leaching processes serves as a vital technical approach for ensuring efficient and stable heap leaching operations at the Letpadaung copper mine. Full article