Search Results (126)

Existing multi-model fusion methods generally adopt a global, fixed fusion strategy, applying all base models and weighting rules uniformly to all samples to be classified and all categories, thereby lacking adaptability to specific samples and category-specific targeting. Most traditional dynamic selection and dynamic weighting fusion methods only implement global parameter adjustments at the sample level, without considering the significant differences in category-specific capabilities among the base models. When a single base model exhibits superior recognition capabilities for only certain categories, its prediction accuracy and confidence for the remaining categories are low. If all base models are fused directly, poor-quality class predictions can cause negative interference and even dominate the final decision, leading to classification errors. Furthermore, traditional dynamic fusion suffers from computational redundancy, difficulty in suppressing interference from low-confidence samples, and the challenge of balancing dynamic optimization with inference efficiency. To address these issues, this paper proposes an improved multi-model dynamic fusion classification technique that differs from the traditional global dynamic fusion paradigm. By constructing a voting matrix, contribution weights, and a matrix of effective category voting weights, this method establishes a category-level model performance evaluation and differentiated weighting mechanism. This enables the precise selection of superior base models for each sample and category, thereby filtering out interference from low-confidence and suboptimal category predictions. At the same time, in the network architecture design, lightweight models are organized into a branch structure, and effective branches are dynamically activated as needed to participate in decision-making, significantly reducing the computational overhead of inference. To validate the fusion classification technique proposed in this paper, for the experiments, we selected mainstream lightweight models such as MobileNetV2, EfficientNetB0, ShuffleNetv2, MNASNet 0.75, and MobileNetV3_Small for evaluation on the NEU dataset and NASA’s Milling Data Set. The experimental results demonstrate that the fusion method proposed in this paper can fully aggregate the category-specific strengths of different lightweight models, effectively mitigate the risk of misclassification associated with traditional fusion methods, and enhance model robustness while ensuring high classification accuracy. It achieves classification performance comparable to that of large deep models with extremely low computational overhead. This method is not only suitable for application in multiple-criteria decision-making but can also be implemented and extended to multi-source/multi-modal data fusion and deep neural networks, making it of practical value. Full article

Wearable inertial sensing and markerless vision are increasingly integrated to enable objective assessment of locomotor and postural function for sports telerehabilitation, intelligent physiotherapy, and athlete performance monitoring. Before such multimodal systems can be translated to clinical practice, their fusion, optimization, and temporal modeling strategies require validation under controlled conditions with reliable ground truth. This study presents a unified multimodal framework that hierarchically integrates inertial measurement unit (IMU) signals and RGB visual information through kernelized representation learning, adaptive multimodal fusion, evolutionary feature optimization, and deep temporal classification. The IMU branch employs Kernelized Extreme Learning Machine (KELM) denoising, Kernelized Canonical Correlation Fusion (KCCF), entropy-guided adaptive windowing, and complementary time-series descriptors (MINIROCKET, TS-CHIEF, and r-STSF). Concurrently, the RGB branch combines anisotropic diffusion filtering, HRNet-based silhouette extraction, DensePose R-CNN, Mesh Graphormer, and Multi-Model Pose-Flow Fusion (MPFF) to learn robust visual representations. Both modalities are integrated through Weighted Canonical Feature Fusion (WCFF) and optimized using a Genetic Algorithm for feature selection and adaptive modality weighting before temporal modeling with cluster-based alignment, Gaussian Process Sequence Modeling, and DeepConvLSTM. As the selected benchmarks do not provide complete inertial recordings, the inertial modality is established according to the adopted experimental protocol to support multimodal fusion analysis. Under 5-fold subject-independent cross-validation, the framework achieves accuracies of 86.56 ± 0.31% on SoccerDiffusion and 88.04 ± 0.25% on HumanoidRobotPose. Although evaluated on humanoid robotic benchmarks, the proposed framework provides a methodological basis for future wearable-enabled clinical movement assessment, remote rehabilitation, and athlete monitoring, while validation on synchronized human inertial-visual datasets remains an important direction for future research. Full article

Cyanobacterial blooms occur frequently in lakes worldwide, disrupting aquatic ecosystem balance and directly threatening drinking water safety and fisheries production. Establishing a reliable bloom early-warning system has therefore become an urgent priority for lake water management. This study adopts a structured narrative review approach to synthesize the major early-warning methods, including indicator threshold methods, statistical and empirical models, mechanistic models, machine learning, and remote sensing monitoring. These methods are compared in terms of their fundamental principles, data requirements, predictive capabilities, applicability, interpretability, and computational and maintenance requirements. Emerging trends in multi-source data fusion, multi-model integration, and the development of integrated early-warning systems are also summarized. The findings indicate that each method has distinct strengths and limitations with respect to forecasting lead time, spatial coverage, process interpretation, and operational costs, and that no single method can simultaneously meet the requirements of multiscale bloom monitoring and forecasting. Integrating multi-source data from in situ monitoring, remote sensing observations, and meteorological and hydrological measurements, while coordinating statistical models, mechanistic models, and artificial intelligence algorithms according to specific forecasting objectives, represents an important pathway for improving the robustness and operational applicability of early-warning systems. Given the pronounced seasonal ice cover, substantial hydrological variability, limited monitoring data, and marked regional heterogeneity of some cold and arid region lakes, future research should strengthen high-frequency monitoring during critical periods, promote coordination between remote sensing and in situ observations, and conduct local calibration of early-warning thresholds and model parameters. Season-specific models should also be developed to account for environmental differences among ice-covered, ice-off transition, and open-water periods. Overall, early warning of cyanobacterial blooms in lakes is evolving from the application of individual methods toward the integration of multi-source monitoring, multi-model integration, and decision support, thereby providing a reference for bloom risk prevention and water environment management across different types of lakes. Full article

This paper investigates state estimation for strongly nonlinear systems with unknown inputs under non-Gaussian heavy-tailed impulsive noise. Conventional recursive three-step filters (RTSF) based on the minimum-variance criterion are sensitive to outliers, while a single local linearization is often inadequate for strongly nonlinear dynamics. To overcome these limitations, a multi-model minimum error entropy recursive three-step filter (MMMEERTSF) is proposed. The minimum error entropy criterion is embedded into the RTSF framework to enhance robustness against abnormal disturbances, and iterative reweighted solutions are developed for unknown-input estimation and state correction by combining residual whitening with entropy-based optimization. Meanwhile, multiple local linear submodels are constructed to approximate the nonlinear system, and compatibility-based posterior fusion is employed to obtain the final estimate. The proposed method shows improved robustness and competitive estimation accuracy under non-Gaussian mixture and impulsive noise, especially in the nonlinear multi-model case. Full article

Against the backdrop of global warming and the ‘warming and wetting’ trend in north-western China, changes in seasonal snowpack and glacial ice in high-altitude cold regions directly impact water security in inland river basins. At present, there is a paucity of systematic research concerning the long-term evolution of snow and ice cover, multi-scale climate responses and future trends in the source region of the Keriya River on the northern slope of the Kunlun Mountains. To address this, this study utilised Landsat remote sensing imagery and meteorological station data from 2005 to 2024. Employing a multi-model fusion framework that integrates various machine learning and time-series models—including random forests, gradient boosting trees and ARIMA—the research incorporated trend factors, climate cycle identification and probabilistic modelling of extreme events to systematically analyse the spatiotemporal variability of snow/ice coverage and its multiscale coupling relationships with air temperature and precipitation. Given the inherent limitations of optical remote sensing methods in distinguishing between seasonal snow and glacial ice, this study defines the extracted coverage type as snow/ice coverage. Given the inherent limitations of optical remote sensing methods in distinguishing between seasonal snow and glacial ice, this study defines the extracted coverage type as snow/ice coverage. The results indicate that: (1) the annual average snow/ice cover percentage in the study area shows a non-significant decreasing trend (−0.69%/year, p > 0.1); within the year, it exhibits a pattern of accumulation in winter and melting in summer, with a peak in January (average 63.2%) and a trough in August (average 11.6%); (2) snow/ice cover percentage increases significantly with altitude; the annual average SICP in the <2000 m elevation zone is 5.2%; in the 2000–3000 m and 3000–4000 m altitude ranges, this rises to 5.7% and 8.3%, respectively, representing the primary seasonal snow/ice distribution zones; in areas above 6000 m, the annual average reaches 70.3%, constituting a zone of perennial stable snow/ice cover; (3) the relationship between snow/ice and temperature and precipitation exhibits significant time-scale dependence: correlations are weak on an annual scale (temperature R = −0.25, precipitation R = −0.14), but significantly strengthen on a monthly scale and exhibit seasonal differentiation; during the melting season, temperature exerts a dominant negative influence (August R = −0.35), whilst during the accumulation season, solid precipitation provides a positive supplement (February R = 0.34), with the strongest correlation with temperature occurring in September (R = −0.50); (4) it is projected that between 2025 and 2044, snow and ice cover will follow a fluctuating downward trend (averaging an annual decrease of roughly −0.12%), falling to approximately 29% by 2044; at the same time, temperatures are expected to continue rising (+0.035 °C per year), whilst precipitation will increase slightly (+0.4% per year). The results of this study provide a sound scientific basis for formulating sustainable water resource management strategies for the northern flank of the Kunlun Mountains and optimising measures to regulate snowmelt runoff. They are of great importance for safeguarding the stability of the oasis ecological systems in the Keriya River basin and ensuring the sustainable development and utilisation of water resources. Full article

In unmanned aerial vehicle (UAV)-based power line inspection, multi-scale defects and complex backgrounds challenge the balance between detection accuracy, speed, and model lightweighting, limiting automated grid inspection. This paper proposes a Multi-Scale Mamba Framework (MS-Mamba) for efficient and accurate defect perception. A drone inspection dataset containing 5137 images from 14 defect categories was constructed and divided into training and validation sets with an 8:2 split. To address the large scale variation among defects, the categories are decoupled into macroscopic, mesoscopic, and microscopic groups according to physical attributes and visual scales. As the core perception engine, a lightweight state-space mechanism is designed to balance accuracy and deployability. A spatial resolution-aware hierarchical reconstruction strategy and a dynamic feature selection mechanism are integrated to enhance feature extraction, reduce background redundancy, and improve small-target representation. Compared with the YOLOv5s baseline, MS-Mamba achieves an mAP@0.5 of 0.749, corresponding to a 15.6 percentage-point improvement, while reducing parameters by 0.13 M and computational cost by 1.7 GFLOPs. Ablation studies and visual analyses further confirm fewer missed and false detections in complex backgrounds. The developed end-to-end inspection system was validated through closed-loop engineering tests, demonstrating strong potential for industrial deployment. Full article

The battery management system (BMS) is the core of ensuring the safety and performance of new energy vehicles, and real-time high-precision estimation of battery state of charge (SOC) is its key function, which directly affects battery safety, endurance, and service life. Faced with the challenges brought by high energy density and ultra-fast charging technology, lithium-ion batteries exhibit strong nonlinear and time-varying characteristics, making it difficult for existing SOC estimation methods to balance computational efficiency and accuracy. This study proposes a Bayesian-based Hammerstein multi-model (MM) fusion algorithm for accurate lithium battery SOC estimation across a wide temperature range, especially under low-temperature conditions. First, two Hammerstein SOC submodels are constructed: a traditional polynomial Hammerstein model and a TPA-Hammerstein model incorporating the temporal pattern attention mechanism. Second, KV-ADAM is employed for parameter training and identification of the submodels. Finally, a Bayesian weighted fusion strategy is used to dynamically integrate the outputs of the two submodels. The experimental results show that this method significantly improves the accuracy and robustness of SOC estimation, overcomes the limitations of a single model under complex dynamic conditions, provides an effective solution for lithium battery SOC estimation, and helps the safe operation of electric vehicles and the sustainable development of the industry. Full article

In the current energy context, enhancing the precision of wind power prediction serves as a key enabler for the stable development of the power grid. In the existing wind power prediction models, there are often problems of modal aliasing and noise residue, or the prediction accuracy of the model is not high. In an effort to solve the problem of short-term wind power forecasting, a wind power series decomposition and reconstruction method based on improved complete ensemble empirical mode decomposition with adaptive noise-variational modal decomposition (ICEEMDAN-VMD) secondary decomposition is proposed. Using ICEEMDAN, wind power data (wind direction, wind speed, temperature, humidity, air pressure, etc.) is decomposed into several IMF sub-series, and these IMF sub-series are categorized into three different frequency components by combining sample entropy, Q statistics and sequence frequency. Secondly, the gray wolf optimization (GWO) is improved by using the empirical exchange strategy (EES), and the optimization performance of the EES-GWO proposed in this paper is verified by using 10 test functions. Finally, the EES-GWO-convolutional neural network–bidirectional gated recurrent unit–global attention (EES-GWO-CNN-BiGRU–Global attention) high-frequency component prediction model is constructed. Finally, we employ the XGBoost model to forecast the mid- and low-frequency components, thereby generating the corresponding forecasting results. The support vector machine (SVM) model nonlinearly integrates all the forecasting results to produce the final forecasting results. Through example analysis and comparison, the performance of the proposed model is verified from two perspectives. Full article

In millimeter wave V2I communication systems, accurate beam prediction is crucial for optimizing network performance and improving signal transmission efficiency. Traditional beam prediction methods mainly rely on single-modal data, which often fails to capture the comprehensive environmental information required for high accuracy prediction. In contrast, multi-modal approaches leverage complementary information from different data sources and offer a more promising solution. However, many existing fusion methods primarily depend on real-time sensory inputs and do not fully exploit stable environmental features in V2I scenarios, limiting the effective use of each modality. To address these limitations, this paper proposes a environment-aware proactive beam prediction method based on a multi-modal prior mask map (MMPMM), which integrates offline mapping with an online beam prediction network. Specifically, the method fuses information from images, point clouds, positions, and the MMPMM to predict the optimal beam index. The MMPMM provides channel-related prior information by extracting static V2I scene features offline without incurring any additional online measurement overhead. Experimental results on real-world datasets demonstrate that the proposed method achieves a Top-3 beam prediction accuracy of up to 71.23% while maintaining stable performance under the evaluated dynamic and degraded conditions, demonstrating its effectiveness in the considered scenarios. Full article

Water quality prediction in non-stationary environmental systems requires not only high predictive accuracy but also structural robustness under physical, ecological, and operational constraints. This study reframes multi-model fusion as a constraint-governed inference architecture and synthesizes advances in rule-based adjudication, reliability-aware aggregation, post-fusion projection, dual-track adaptation, and hierarchical backoff control. By establishing a taxonomy of boundary constraints—specifically mass conservation, reaction kinetics, hydraulic transport, and ecological tipping points—an admissible prediction manifold identifies key structural limitations in existing paradigms, particularly their vulnerability to physical inconsistency and diminished reliability during non-stationary distribution shifts. A unified end-to-end robust framework is proposed in which candidate predictions are separated from admissibility validation, uncertainty is directly coupled to aggregation logic, and degradation pathways are explicitly defined under distribution shift. Furthermore, a multidimensional robustness evaluation matrix is introduced, incorporating structural consistency, ecological compliance, calibration quality, and adaptive stability alongside conventional accuracy metrics. The study advances water quality forecasting from model-centric optimization toward architecture-level governance, demonstrating that constraint-aware designs improve structural consistency, robustness under distribution shifts, and early warning reliability, providing a systematic reference for developing resilient, transparent, and operationally deployable environmental prediction systems. Full article

With accelerating global urbanization and increasingly diverse demands for public spaces, promoting urban low-carbon transitions and enhancing residents’ quality of life have become central missions of modern urban development. As one of the city’s primary arteries, streets—through their green landscapes, slow-moving transportation systems, and public facilities—play an indispensable role in reducing carbon emissions, promoting healthy living, and improving residents’ well-being. In this study, the Yubei District of Chongqing was selected as the research area, and an automated evaluation framework was proposed for street visual quality, based on multi-source street view data and ensemble learning. PSP-Net semantic segmentation model was employed to extract eight key visual indicators from street view images, including green view index, Visual Entropy (Entropy), sky view factor (SVF), drivable space, sidewalk, safety facilities, buildings, and enclosure. Based on these features, a Stacking-based ensemble learning model was constructed, integrating multiple base models such as Random Forest, XGBoost, and LightGBM, with Linear Regression as the meta-learner, to predict street visual quality. The results demonstrate that the ensemble model significantly outperforms any single model, achieving a correlation coefficient (r) of 0.77 and effectively capturing the complex perceptual features of street environments. This study provides a reliable, intelligent, and quantitative method for large-scale evaluation of urban street visual quality, while supplying data support and decision-making references for street renewal and spatial optimization. Full article

In karst mountainous areas where high-dimensional features coexist with extremely limited sample sizes, accurate landslide susceptibility mapping remains challenging. To address this issue, we propose an ensemble framework termed the Triple-Source Probabilistic Fusion Bidirectional Long Short-Term Memory network (TSPF-BiLSTM). The approach was tested in Sangzhi County, Hunan Province, by integrating three base learners—Random Forest (RF), LightGBM, and AdaBoost. Their raw outputs were first calibrated using five-fold Platt scaling to generate posterior probabilities on a unified scale. A bidirectional LSTM was then employed to perform deep nonlinear fusion of these cross-model probability features. Using a total of 618 landslide and 618 non-landslide samples (split into training and testing sets), the TSPF-BiLSTM model achieved a mean AUC of 0.9525 (±0.0115) under ten-fold cross-validation, outperforming not only the individual base learners but also standalone deep learning models (CNN and Transformer). The frequency ratio in the very high susceptibility zone reached 3.97, significantly exceeding all benchmark models and confirming its superior capability in high-risk area identification. Multi-model importance analysis identified NDVI, elevation, and annual rainfall as the dominant regional landslide predisposing factors. Within the specific ranges of NDVI 0–0.686, elevation 155–462 m, and annual rainfall 1273.6–1301 mm, landslide frequency ratios consistently exceeded 1.96. The proposed framework, with its probability-level fusion and embedded regularization mechanisms, effectively mitigated overfitting despite the small sample size, providing a robust technical solution for geological hazard risk identification and prevention in the data-scarce karst terrain of the Wuling Mountains. Full article

Obstacle-avoidance risk threshold control and global discrete keypoint re-entry are critical factors influencing the smooth dynamic obstacle avoidance of unmanned vessels. For underactuated USVs, which operate in planar motion with three degrees of freedom (surge, sway, and yaw) but only two independent control inputs (surge velocity and yaw rate), this paper designs a layered obstacle-avoidance strategy featuring adaptive global path re-entry points, combined with short- and long-term obstacle trajectory prediction and risk perception. This method employs an Interactive Multiple Model (IMM) integrating Constant Velocity (CV), Constant Acceleration (CA), and Constant Turn Rate and Acceleration (CTRA) models to perform long-term spatiotemporal trajectory prediction for dynamic obstacles, constructing a spatiotemporal risk cost map. Long-term dynamic obstacle-avoidance trajectory planning is achieved through optimized adaptive global trajectory re-entry points and an improved A* algorithm. This long-term avoidance trajectory replaces the global path from the avoidance start to the re-entry point, providing a smooth, continuous long-term avoidance prediction. To ensure real-time collision avoidance effectiveness, an improved Dynamic Window Approach (DWA) algorithm uses the long-term avoidance trajectory as a foundation. It integrates the IMM’s short-term spatiotemporal obstacle trajectory prediction, sampling in the velocity and steering angle space to generate short-term avoidance control commands. Finally, the long-term and short-term obstacle-avoidance planning are executed in a receding-horizon manner, where the local DWA planner updates control inputs over a short rolling window without solving a full constrained optimization problem. This establishes a hierarchical avoidance strategy: long-term prediction enables smooth avoidance, while short-term prediction enables real-time avoidance, ensuring the continuity and timeliness of dynamic obstacle avoidance. Simulation results demonstrate that compared with traditional A* planning, the proposed risk-aware A* reduces cumulative collision risk by 62% and increases the minimum obstacle clearance distance by over 32.1%, while maintaining acceptable path length growth. This approach effectively reduces collision risks during navigation, enhances path smoothness, and improves navigation safety. Full article

To address the limitations of static and coarse-grained analysis in mining online tourism reviews, this study proposes a gated attention-based multi-model fusion framework for dynamic topic evolution and complaint-driven latent issue pattern mining. Using 300,000 reviews from Ctrip and Meituan, we fuse global semantics from Sentence-BERT with attention (SBERT-Attention), local features from Bidirectional Encoder Representations from Transformers–Text Convolutional Neural Network (BERT-TextCNN), and topic distributions from the Biterm Topic Model (BTM) via a learnable gating mechanism. The fused model achieves an F1-score of 92.3% in review classification. We partition the corpus quarterly and apply Uniform Manifold Approximation and Projection (UMAP) followed by K-means++ clustering to the fused vectors, yielding interpretable topics, including Scenery, Transportation, Amenities, Management, Culture, and Value for Money, and enabling dynamic topic discovery over time. River map visualizations and negative review analysis reveal seasonal evolution patterns and recurring complaint patterns associated with specific topics. The framework enables dynamic, interpretable semantic mining, advancing intelligent processing of short-text user content and offering a generalizable approach for temporal knowledge discovery in smart tourism and beyond. Full article

Marine target classification is a key technology for unmanned surface vehicles (USVs) to perform ocean surveillance. Traditional maritime target classification methods require improvements in both accuracy and processing speed when handling classification tasks. In this paper, a distributed maritime target classification (DMTC) method based on broad learning and MobilityFirst is proposed. Firstly, a multi-model collaborative classification and fusion framework is proposed to achieve feature consistency fusion. Secondly, to enhance the security and privacy of communication in autonomous surface vehicles, the MobilityFirst approach is employed to improve information complementarity among multiple models within the distributed framework. Finally, the broad learning system, as the model’s classification layer, reduces the training complexity. Extensive experimental results demonstrate that this proposed approach surpasses single-model and distributed methods in accuracy, F1 score, and the area under the precision–recall curve (AUPR). This approach offers a clear advantage in multi-ship classification tasks while simultaneously enhancing the model’s generalization capability. Full article