Search Results (312)

Continouous unmanned aerial vehicle (UAV) missions are fundamentally limited by Lithium-Polymer (LiPo) battery endurance under intermittent and power-constrained renewable energy conditions. This paper proposes an integrated energy management and charging framework for a photovoltaic (PV)-powered mobile station equipped with a hybrid energy storage system (HESS) and an automated battery replacement (ABR) mechanism. A lexicographic priority-based allocator sequentially serves ABR actuation, multi-slot LiPo charging, and Brushless DC (BLDC) propulsion, while the HESS compensates for PV intermittency. At the charging level, a constraint-aware constant current–constant voltage (CC–CV) strategy is enhanced by an adaptive neuro-fuzzy inference system (ANFIS) trained on optimization-derived labels using battery temperature and its rate of change, thus enabling anticipatory thermal current derating with smooth, discontinuity-free control action. Anti-windup proportional–integral (PI) regulation and bumpless mode transfer ensure stable CC-to-CV transitions. An event-triggered emergency mode accelerates battery readiness via a max-first selection policy. Comparative simulations against a PSO/DE-optimized PID benchmark over a full diurnal PV cycle demonstrate that the ANFIS controller reduces the CC-mode current tracking root-mean-square error (RMSE) by up to 96.9%, delivers higher charge throughput, and lowers battery degradation proxies, including SOC-weighted thermal dose and equivalent full cycles (EFC). The proposed framework reliably sustains continuous charge–swap–recharge logistics under fluctuating renewable generation. Full article

Accurate arrival picking of acoustic emission (AE) data is essential for AE event localization and hydraulic fracture characterization in true triaxial hydraulic fracturing experiments. However, conventional arrival picking methods are highly sensitive to manually defined thresholds, whereas existing deep learning models are constrained by low signal-to-noise ratios (SNRs) and limited AE dataset sizes. To address these challenges, this study proposes an attention-based deep learning method for AE arrival picking. The proposed method introduces an attention mechanism into the PhaseNet framework to suppress noise feature transmission in the skip connections. In addition, a kernel density estimation (KDE)-based label smoothing strategy was adopted to alleviate label imbalance and account for arrival-time uncertainty. The results demonstrate that the proposed method reduced the mean absolute error (MAE) by 10.58%, 92.92%, and 98.25% compared with PhaseNet, STA/LTA, and AR-AIC, respectively. The proposed method exhibited superior picking accuracy, robustness, and computational efficiency relative to the other methods, providing a reliable foundation for AE event localization and high-precision AE monitoring in hydraulic fracturing experiments. Full article

The Gaizi River Basin, an alpine region in China crossed by the Karakoram Highway, is highly prone to glacier-related debris flows (GDF). Accurate debris flow susceptibility assessment in this high-altitude area remains challenging due to complex terrain, active tectonics, and dynamic glacial processes. This study develops a hybrid model integrating statistical methods and machine learning-based anomaly detection for debris flow susceptibility mapping. To address data noise, certainty factor (CF) distributions of debris flow predisposing factors (DFPFs) were derived via Locally Weighted Scatterplot Smoothing (LOWESS). The strength of the association between DFPFs and GDF susceptibility was evaluated using the mean residual between the raw and LOWESS-smoothed CF values. Multiple anomaly detection algorithms, including distance-based (L2 Norm), density-based (One-Class SVM), ensemble (Isolation Forest, RandNet), and GAN-based (WBiGAN-GP) methods, were tested on raw and CF-transformed data, using only the GDF inventory as the label. The CF-WBiGAN-GP model delivers the most balanced performance, excelling at identifying both high- and low-susceptibility zones. Results show that distance to stream, slope, and the topographic roughness and wetness indices are strongly associated with GDF susceptibility. Distance to glacier and precipitation appear less informative for direct susceptibility inference under our specific dataset and analytical setup. Full article

To address the limited adaptability of fixed-parameter PID controllers under dynamically varying reference signals and the strong dependence of data-driven PID methods on training-data quality, this paper proposes a GRU-based online PID gain-scheduling framework supported by high-quality dataset construction. Diverse reference excitations are first designed, and sequential quadratic programming (SQP) is used as an expert label generator to produce trajectory-level PID gain labels. A region-of-interest (ROI)-based dynamic sample selection strategy is then introduced to retain informative transient samples and reduce the dominance of redundant steady-state data. The gated recurrent unit (GRU) network learns a temporal mapping from error-state sequences to PID gains and is deployed online with closed-loop safeguards, including filtered derivative information, gain denormalization, smoothing, and actuator constraints. In a representative nominal neural-controller benchmark, GRU-PID achieves a rise time of 0.59 s, a settling time of 0.97 s, ISE = 2.10, ITAE = 39.35, and TV = 394.48, showing a favourable balance between tracking accuracy and control-signal smoothness. Five-seed tests further indicate that GRU-PID provides stable nominal performance comparable to competitive neural schedulers, while simulation-based robustness evaluations suggest lower tracking errors than the tested neural baselines under measurement noise, step disturbance, actuator saturation, and combined uncertainty scenarios within the considered benchmark setting. Full article

This paper develops a unified asymptotic theory for conditional single-index U-statistics and the associated conditional U-processes in the setting of locally stationary functional time series subject to missing-at-random response mechanisms. The proposed framework addresses, within a single nonparametric inferential architecture, three major sources of complexity in modern functional data analysis: infinite-dimensional covariates, smoothly time-varying stochastic dynamics, and incomplete response observations. The methodology is based on a class of kernel-type estimators combining temporal localization, functional single-index smoothing, and inverse-propensity correction. Temporal localization captures the gradual evolution of the underlying regression structure, the single-index projection provides an effective dimension-reduction mechanism for functional covariates, and the propensity adjustment restores the target conditional functional under the MAR sampling scheme. The principal contribution of the paper is the establishment of weak convergence, in a suitable space of bounded functions, for the resulting propensity-adjusted conditional U-process indexed by a general class of measurable kernels. Under absolute regularity conditions, local stationarity assumptions, small-ball probability requirements, entropy restrictions of VC type, and uniform consistency of the propensity-score estimator, the normalized process is shown to converge weakly to a tight centered Gaussian process. The limiting covariance structure explicitly reflects the interaction between temporal smoothing, functional concentration, dependence, and the random loss of responses. In parallel, uniform convergence rates are derived for the associated conditional single-index U-statistic estimators, thereby quantifying the respective contributions of smoothing bias, stochastic fluctuation, local-stationarity approximation error, and missingness-induced variance inflation. A substantial part of the analysis is devoted to the technical difficulties created by the simultaneous presence of dependence, nonstationarity, functional covariates, and incomplete observations. The proofs combine Hoeffding-type decompositions adapted to weighted incomplete data, blocking and coupling arguments for absolutely regular triangular arrays, refined entropy bounds for kernel-indexed function classes, and small-ball probability techniques for functional covariates. The MAR mechanism is incorporated via inverse-propensity weighting, and its effects on the effective sample size, asymptotic variance, and bias structure are made explicit. The theory also provides a rigorous foundation for bandwidth selection through blocked, propensity-adjusted cross-validation and clarifies its relation to the corresponding oracle risk. The proposed framework encompasses a broad class of statistical learning and inference problems involving pairwise or higher-order functionals of functional time series. In particular, it applies to conditional Kendall-type functionals, discrimination problems, metric learning with incomplete labels, and conditional independence testing under local stationarity. A simulation study illustrates the finite-sample behavior of the proposed estimators and supports the theoretical findings across varying regimes of temporal nonstationarity, serial dependence, functional concentration, and response missingness. Overall, the results provide a mathematically rigorous and methodologically flexible foundation for inference from evolving functional data when dependence, infinite dimensionality, and incomplete observation are present simultaneously. Full article

Background/Objectives: Brain magnetic resonance imaging (MRI) is an important imaging modality for assessing neurological disorders. However, automatic multi-class MRI classification remains challenging because of visual similarity between disease categories, heterogeneous pathological patterns, class imbalance, and the need for reliable confidence estimation. This study aims to develop a comprehensive and well-calibrated deep learning framework for image-level brain MRI classification across multiple neurological categories. Methods: This paper introduces a new deep learning framework, MCND-ComputeNet++, for brain MRI classification into eight image-level categories using the MCND dataset, which comprises 16,400 two-dimensional brain MRI images belonging to eight diagnostic categories: AD-MildDemented, AD-ModerateDemented, AD-VeryMildDemented, BT-glioma, BT-meningioma, BT-pituitary, MS, and Normal. The proposed model uses a single pretrained EfficientNetV2-S backbone to extract hierarchical feature maps from three intermediate stages. These multi-level features are projected into a common latent space, spatially aligned, adaptively fused through learnable gated multi-scale fusion, further refined using convolutional processing, and aggregated using spatial attention pooling before classification. The training strategy combines class-balanced focal loss with label smoothing, MixUp/CutMix regularization, exponential moving average weight smoothing, warmup cosine learning-rate scheduling, temperature scaling, and test-time augmentation to improve generalization and calibration. The framework was evaluated using accuracy, precision, recall, macro-F1, macro-AUC, macro-average precision, expected calibration error, Brier score, bootstrap confidence intervals, ablation analysis, McNemar testing, and comparisons against standard pretrained baseline models. Results: MCND-ComputeNet++ achieved mean accuracy, macro-F1, macro-AUC, and macro-average precision values of 0.9738, 0.9771, 0.9993, and 0.9971, respectively, with narrow bootstrap confidence intervals indicating stable image-level performance. These findings should be interpreted as image-level/slice-level performance on MCND, because patient-level identifiers and subject-wise splitting were not available. These results outperformed most evaluated baselines, including ResNet50, DenseNet121, EfficientNetB0, EfficientNetV2-S with a standard classifier, Swin-Tiny, and ConvNeXt-Tiny, across several discrimination and calibration metrics. Compared with ConvNeXt-Tiny, the proposed model achieved higher macro-AUC and macro-average precision, together with a lower ECE and Brier score, suggesting improved image-level discrimination and confidence reliability. Compared with the EfficientNetV2-S standard classifier, accuracy increased from 0.9308 to 0.9738, while the Brier score decreased from 0.1045 to 0.0400. Conclusions: The results suggest that MCND-ComputeNet++ is a promising image-level brain MRI classification framework for the eight MCND categories. The proposed model integrates hierarchical feature extraction, shared latent projection, gated multi-scale fusion, convolutional refinement, spatial attention pooling, and calibrated inference within a unified architecture. However, because the current evaluation was conducted at the image/slice level without available patient-level identifiers, the findings should not be interpreted as patient-level clinical diagnostic validation. Further studies using subject-wise splitting, external multi-center datasets, 3D volumetric modeling, and multimodal clinical information are required to assess generalizability and potential clinical decision-support applicability. Full article

This study compares deep learning models for point-level anomaly detection in multichannel satellite telemetry data. Raw event-based telemetry was converted into segment-based multivariate time series without windowing or feature extraction, allowing models to learn system behavior at each time step. Preprocessing included channel alignment, training set-based normalization, missing value imputation, and temporal label smoothing, while Focal Loss and segment-level oversampling addressed class imbalance. Five architectures, BiLSTM, BiGRU, Transformer, Hybrid BiLSTM–Transformer, and Hybrid BiGRU–Transformer, were evaluated, with thresholds optimized on a validation set. The results show that hybrid models combining recurrent networks and attention mechanisms effectively capture both short- and long-term dependencies. The standalone BiGRU model achieves the highest overall classification performance in terms of F1 score and accuracy. In contrast, the Hybrid BiGRU–Transformer architecture does not outperform BiGRU in classification metrics but provides superior temporal stability, improved boundary sensitivity, and better interpretability in anomaly detection tasks. Full article

Kernel-combination procedures used in classification often return only a combined kernel matrix on the training sample, rather than a kernel function that can be evaluated consistently at new points. This limitation is especially important for supervised or label-aware combinations, whose entries may depend on training labels and therefore have no immediate out-of-sample meaning. We study the problem of constructing an inductive, finite-rank kernel extension from such empirical matrices. The proposed framework makes the non-uniqueness of this extension explicit: it is determined by empirical coordinates, a positive-semidefinite coefficient matrix, and a continuation model for the coordinates. Experiments on vector, tabular, and relational classification problems give a deliberately diagnostic picture. Smooth direct combinations are stable: on Synthetic, the direct mean gives error $0.0793\pm 0.0227$, essentially matching the best individual RBF kernel ($0.0809\pm 0.0231$), and on Telco it remains close to the best individual polynomial kernel ($0.2061\pm 0.0154$ versus $0.2045\pm 0.0154$). In the controlled Synthetic oracle diagnostic, reconstructing a smooth sum/mean gives relative Frobenius error $4.13\times {10}^{-6}\pm 9.41\times {10}^{-6}$ and functional MSE at numerical scale. By contrast, abrupt label-aware matrix-only rules are less robust: the Synthetic percentile_inout_auto rule has error $0.1404\pm 0.1198$, Telco matrix-only supervised rules are around $0.307$–$0.326$ error, and the Chickenpieces pickout_auto rule fails under strict out-of-sample reconstruction ($0.3545\pm 0.2666$ error), whereas direct relational combinations match the best individual relational kernel within ${10}^{-3}$. Overall, the empirical evidence supports the method as a bridge from finite matrix-level information fusion to deployable kernels, while also identifying abrupt label-aware geometries as the main limitation for stable generalization. Full article

Background/Objectives: Retinal layer segmentation in optical coherence tomography (OCT) is essential for the diagnosis and monitoring of retinal diseases such as age-related macular degeneration (AMD) and diabetic macular edema (DME). Although deep learning methods have achieved strong performance, most rely on dense pixel-wise predictions and often struggle to preserve anatomical consistency, particularly in regions with low contrast or structural deformation. This study aims to address these limitations by introducing a query-based segmentation framework that explicitly models retinal layer structure. Methods: In this paper, we propose the RetiQueryNet architecture that employs encoding of retinal layers in the form of query embeddings with the use of cross attention to interact with pixel level features encoded by a transformer based encoder. The architecture integrates multi-scale features through a compact query-driven decoder with modest additional computational overhead. Normalization and resizing of OCT images preceded their usage as inputs, while the layer labels were converted to multi-class segmentation maps. In the training process, we used loss function with combination of cross entropy loss and Dice loss. Our model performance was compared with multiple state-of-the-art models such as U-Net, DeepLabV3, FPN, MANet and SegFormer, while performance metrics were Dice, IoU and mean surface distance (MSD). Results: RetiQueryNet was able to attain a mean Dice score of 0.934 ± 0.0046 and outperformed all baseline models on the main performance measures. Improvements were particularly evident in challenging retinal layers such as IBRPE and OBRPE, where boundary ambiguity is high. It should be noted that RetiQueryNet had a relatively lower MSD value, meaning that the predicted boundaries were more accurate. Furthermore, visual observations suggest that the approach generated smooth and coherent segmentations. Conclusions: The findings demonstrate that query-based modeling offers a viable approach to pixel-wise segmentation. In particular, by making use of structural priors in the form of learnable queries, RetiQueryNet improves not only segmentation accuracy but also anatomical consistency. Query-based modeling appears to be an exciting area for retinal image segmentation that could potentially be applied to other applications in medical image segmentation. Full article

Change detection (CD) refers to the analysis of changes in the utilization of land, buildings, and other targets in the same surface environment using relevant technologies and remote sensing images. Although deep learning-based change detection methods have achieved excellent results, they remain highly dependent on extensive labeled data. High-resolution remote sensing imagery typically encompasses an abundance of details and a greater quantity of pixels compared to low-resolution datasets. Therefore, data annotation costs are significantly higher. Currently, within the context of semi-supervised change detection (SSCD) driven by consistency learning, pseudo-labels are usually selected only by threshold screening, but this ignores the spatial relationships among pixels and does not fully utilize unlabeled data, thereby affecting the model’s performance. Consequently, we propose a semi-supervised high-resolution remote sensing image change detection method based on label expansion. First, a “one weak, two strong” (OW-TS) consistency regularization (CR) framework is introduced to constrain the overall consistency between the prediction results of weak and strong augmentations, as well as between the two strong augmentations. At the same time, the location interaction map (LIM) is introduced to utilize the global–local relationship between pixels and mine the consistency of pseudo-labels, thereby improving the model’s accuracy. Empirical findings indicate that when the model is trained utilizing 20% labeled data and 80% unlabeled data on the LEVIR-CD dataset, the IoU^c index reaches 83.38%. The model performs well in smoothing the boundary between changed and unchanged areas and is comparable in performance to some fully supervised methods. Full article

The Poisson exponentially weighted moving average (PEWMA) control chart rests upon the equidispersion assumption of the pure Poisson distribution, a structural symmetry condition stipulating that the process mean and variance are equal. In manufacturing environments characterized by latent process heterogeneity, this assumption is systematically violated: the resulting distributions are inherently asymmetric, heavily right-skewed, and overdispersed. This structural asymmetry renders standard PEWMA control limits artificially narrow, inducing a substantial inflation of false alarm rates. This paper introduces the Poisson mixture EWMA (PM-EWMA) control chart, which models the latent heterogeneous structure of count data as a finite Poisson mixture distribution, with parameters estimated via the Expectation–Maximization (EM) algorithm without requiring prior labeling of process states. The optimal number of components is determined via the Bayesian Information Criterion (BIC) as the primary criterion, supplemented by the Akaike Information Criterion (AIC), its bias-corrected variant (AICc), and the log-likelihood ratio diagnostic. The PM-EWMA chart incorporates the exact mixture variance, accounting for both within-component and between-component variability, into the EWMA control limit structure, thereby providing a theoretically justified correction under the fitted Poisson mixture assumption. A Monte Carlo simulation study comprising 495 factorial configurations benchmarks the PM-EWMA chart against both the standard PEWMA chart and the negative binomial EWMA (NB-EWMA) chart with oracle dispersion calibration, confirming stable in-control ARL performance and demonstrating improved discrimination relative to the misspecified PEWMA baseline. Empirical validation using fabric defect count data from two textile manufacturers in Türkiye, with Overdispersion Indices of 6.01 and 2.74, respectively, demonstrates false alarm reductions ranging from 40.9% to 89.2% relative to the standard PEWMA chart, depending on the smoothing parameter and degree of overdispersion. Full article

Rolling bearing fault diagnosis is essential for ensuring the safety and reliability of rotating machinery. Although deep learning-based methods have achieved promising performance, they usually require sufficient labeled data, which is difficult to obtain in practical industrial scenarios where fault samples are scarce and data sharing across sites is restricted by privacy and confidentiality constraints. Federated learning enables collaborative model training without transmitting raw data, but existing federated fault diagnosis methods often degrade under few-shot conditions. Moreover, current federated meta-learning approaches mainly focus on model-level adaptation and lack explicit class-level representation alignment, leading to prototype drift across heterogeneous operating conditions. To address these challenges, this paper proposes ProtoFed, a prototype-enhanced federated meta-learning framework for few-shot rolling bearing fault diagnosis. ProtoFed converts raw vibration signals into time–frequency representations using continuous wavelet transform and performs local episodic learning with prototypical networks. A Global Prototype Calibration mechanism aggregates local class prototypes into stable global prototypes with exponential moving average smoothing, while a Prototype-Distance Aware Aggregation strategy adaptively adjusts client aggregation weights according to local–global prototype divergence. Experiments on the CWRU and Paderborn University bearing datasets under non-IID 5-shot and 10-shot settings show that ProtoFed consistently outperforms standard federated learning, prototype-based federated learning, and federated meta-learning baselines. Under the 5-shot setting, ProtoFed achieves 95.63% and 91.35% accuracy on CWRU and PU, respectively, approaching centralized few-shot upper-bound performance while preserving the federated training paradigm. Full article

Background: Tumor–stroma interactions play a critical role in renal cell carcinoma (RCC) progression. Cancer-associated fibroblasts (CAFs) are considered key components of the tumor microenvironment; however, their origin remains controversial. This study aimed to determine whether bone marrow-derived mesenchymal stem cells (MSCs) contribute to CAF-like stromal changes and RCC progression. Methods: An orthotopic xenograft mouse model was established using luciferase- and GFP-labeled Caki-1 cells. MSCs labeled with PKH26 were administered intravenously. Tumor growth was evaluated using an in vivo imaging system and tumor volume measurements. Immunohistochemical analyses were performed to assess MSC localization and α-smooth muscle actin (α-SMA) expression. In vitro proliferation and migration assays were conducted using direct and indirect co-culture systems. Results: The intravenous administration of MSCs significantly increased tumor growth and bioluminescence intensity in an orthotopic model. The tumor volumes were significantly larger in the MSC-treated versus control group. An immunofluorescence analysis demonstrated partial co-localization of PKH26-labeled MSCs with α-SMA-positive fibroblast-like cells, suggesting acquisition of CAF-like features. Direct co-culture with MSCs significantly enhanced RCC cell proliferation and migration in vitro, whereas culturing in conditioned medium alone did not produce similar effects. Conclusions: Exogenously administered bone marrow-derived MSCs may be recruited into RCC tissues and acquire CAF-like features through interactions with tumor cells. These findings suggest that stromal–tumor cell interactions within the tumor microenvironment may contribute to RCC progression and represent a potential therapeutic target. Full article

Attributed graphs have recently emerged as a powerful tool for representing diverse data in numerous real-world sensors. Among various applications, unsupervised graph anomaly detection (UGAD) aims to identify abnormal data that significantly deviate from the majority of normal nodes without label annotations. Hence, UGAD can provide crucial assistance in enhancing the reliability of IoT, intelligent sensors and so on. Under the class-imbalanced reality caused by anomaly scarcity, the common paradigm of UGAD focuses on learning a model that primarily captures normal patterns. However, the traditional Graph Neural Network (GNN) paradigm suffers from local-aggregation limitations and over-smoothing, constraining their discrimination capacity. To address these issues, we introduce Graph Transformers (GTs) into UGAD task, termed as unsupervised attributed graph Anomaly detectioN wIth Masked grAph TransformErs (ANIMATE). Leveraging the global receptive field of Transformers, we can capture graph information that preserves the distinguishable characteristics of abnormalities from a global perspective. Furthermore, we employ masked auto-encoders to reconstruct node features and prompt our model to focus more on learning normal patterns. Additionally, we enhance the performance through a self-paced enhancement scheme specifically for UGAD tasks. Experiments conducted on various real-world benchmark datasets with organic anomalies validate the effectiveness of our proposed method compared to state-of-the-art competitors. Full article

Accurate day-ahead estimation of residential demand response (DR) potential is essential for load aggregators participating in electricity markets. It is also difficult to estimate because public residential datasets rarely contain observed DR event labels and household flexibility is shaped by heterogeneous, weather-sensitive consumption behavior. This paper proposes an appliance-agnostic two-stage framework for forecasting residential DR potential from aggregate hourly load and weather data. In the first stage, a thermal-response model estimates household heating and cooling sensitivities and converts thermostat-setback assumptions into synthetic DR-potential targets. Because these targets are model-derived proxies rather than measured DR events, the reported forecasting errors should be interpreted in terms of accuracy against a physically motivated synthetic target. In the second stage, the synthetic target sequence is forecast using a mixture of KAN experts (MoKE). The architecture combines Wavelet-KAN, Fourier-KAN, and RBF-KAN experts through sparse top-k routing with reversible instance normalization, allowing the model to represent local irregularities, recurrent daily/seasonal structure, and smooth nonlinear response regimes in the same forecasting layer and these forecasting characteristics are absent from traditional deep learning forecasting models. The framework is evaluated on the UMass residential dataset, which contains hourly electricity and meteorological measurements from 114 apartments collected during 2015 and 2016, using a 24 h day-ahead forecasting horizon. Across both winter and summer evaluation windows, the proposed model achieves the lowest error among all benchmark methods, outperforming TimesNet, Informer, N-HiTS, FEDformer, PatchTST, and TCN across MAE, MAPE, RMSE, and sMAPE. In particular, MoKE attains MAE values of 3.19 in winter and 3.18 in summer, demonstrating stable predictive accuracy under seasonally distinct operating conditions. These results show that heterogeneous KAN experts offer a feasible method for residential DR forecasting when appliance-level metering and observed event-level DR measurements are unavailable. Full article