Ensuring high-quality water resources is crucial for sustainable urban development, public health, and resilient city infrastructure, yet traditional anomaly detection methods struggle with the highly variable, non-stationary, and concept-drifting nature of urban water quality data streams. This study proposes a Physics-Informed Dynamic Ensemble Learning (PIDEL) framework, an artificial intelligence approach that combines diverse classical and deep learning models with Physics-Informed Neural Networks (PINNs) embedding convection–diffusion constraints, a Genetic Algorithm (GA) for ensemble optimization, and a Jensen–Shannon Divergence (JSD) based mechanism for dynamic model switching. Applied to a real-world urban water quality dataset, PIDEL achieves an F1-score of 0.95, representing a 59% improvement over the best static ensemble, while reducing false alarms by 73% compared to traditional methods and maintaining F1-scores above 0.9 across all sliding windows. The framework processes each 60-minute window in approximately 2.3 s on standard hardware, demonstrating its suitability for real-time deployment in smart city water systems. These results highlight that integrating physics-informed constraints with dynamic ensemble learning can substantially enhance the reliability, interpretability, and operational value of automated water quality anomaly detection for urban utilities. • Novel LSTM-PINN integrates physics constraints with deep learning for anomaly detection. • Dynamic ensemble adapts to concept drift via Jensen–Shannon divergence-based switching. • Genetic algorithm optimizes ensemble, achieving 95% F1-score and 59% improvement. • Optimal physics loss coefficient ( λ = 0 . 35 ) balances physical and data-driven learning. • Real-time processing (2.3 s per window) enables practical smart city water monitoring.

Incomplete time series data is a common problem in real-world application scenarios. Recent research has taken the approach of separating interpolation and anomaly detection, which is not interactive and performs poorly. On the other hand, interpolation using traditional methods relies on a large amount of a priori knowledge, and using deep learning methods takes up a large amount of computational resources and is inefficient. In this study, we propose a correlation-aware diffusion model that successfully bypasses the above problems. Our approach focuses on capturing deep multivariate correlations from limited incomplete data and use low-frequency component to guide generation. Experiments on four realistic scenario datasets covering three domains show that our method achieves better anomaly detection results than existing methods for various missing rates.

This paper presents SPDM, an innovative spatiotemporal-periodic diffusion model for multivariate time series interpolation, which addresses the challenges of spatiotemporal dependency and periodicity inherent in real-world datasets. Unlike prior methods, SPDM integrates conditional features capturing spatiotemporal correlations and geographic relationships, enhancing the model's ability to account for complex interdependencies within time series data. A noise prediction module, leveraging Fast Fourier Transform, decomposes time series into periodic components, thereby enabling the model to capture both intra- and inter-period dynamics and inter-channel correlations effectively. Experimental results on multiple industrial datasets show that SPDM outperforms state-of-the-art methods across various missing data scenarios, highlighting its robustness and effectiveness. This work establishes a new approach to time series interpolation by combining conditional information construction with periodicity-aware diffusion modeling, offering promising insights for further applications in time series analysis.

This paper presents a summary of the 2nd Latent in the Wild Fingerprint Recognition Competition held at the 2025 International Joint Conference on Biometrics. The competition has two tracks: latent fingerprint 1) recognition, and 2) quality assessment. It attracted a total of 12 participating teams from academia and industry for both tracks, representing 10 countries. In total, 8 valid submissions were evaluated by the organizers. The competition aimed to advance the state-of-the-art in latent fingerprint recognition and quality assessment by providing a challenging dataset of latent fingerprints collected in natural, non-ideal conditions. This paper summarizes the dataset, evaluation protocols, submitted methods, and the competition results.

Accurate wind power forecasting is critical for maintaining stable power grids, yet the inherent variability of wind and limited data availability for new wind farms present significant challenges. To address these issues, we present a novel artificial intelligence framework that integrates a self-attention enhanced Spatiotemporal Long Short-Term Memory (ST-LSTM) network with Model-Agnostic Meta-Learning (MAML), termed as the Meta-Learning Spatiotemporal Attention Long Short-Term Memory framework (MAML-STALSTM). This deep learning combination enables the model to effectively capture long-range spatiotemporal dependencies while rapidly adapting to new wind farm configurations or changing wind conditions with minimal training data. By employing rigorous data preprocessing techniques and ensuring temporal separation in data splitting, we mitigate potential data leakage and enhance the model’s generalizability. Extensive experiments conducted on both onshore and offshore wind farm datasets demonstrate that our artificial intelligence approach outperforms established baseline models, particularly excelling in data-scarce environments. Ablation studies highlight the crucial roles of the self-attention mechanism and meta-learning in improving forecasting accuracy, adaptation speed, and model robustness. These results emphasize the practical benefits of our approach in enhancing grid stability and supporting the seamless integration of wind energy, thereby contributing significantly to the advancement of sustainable energy solutions.

Hyperspectral images (HSIs) offer rich spectral–spatial information but pose challenges due to a high dimensionality and noise. This letter proposes a shape-adaptive high-order tensor decomposition (SAHTD) framework for hyperspectral feature extraction. The SAHTD employs a shape-adaptive sampling strategy based on a simplified local polynomial approximation and intersection confidence interval (LPA-ICI) to enhance edge segmentation. It integrates tensor Tucker decomposition and a nuclear norm-constrained regression model to extract discriminative features while preserving low-rank structures. Experiments on the Pavia University and Houston 2013 datasets demonstrate that the SAHTD achieves superior classification performance compared to relevant methods, validating its effectiveness for HSI analysis.

Deep learning is a powerful technique for semantic change detection (SCD) of bitemporal remote sensing images. In this work, we propose to improve SCD accuracy using deep learning with frequency feature enhancement. Specifically, we develop a frequency feature enhancement module that aims to enhance the performance of both binary change detection and semantic segmentation, two main key components for obtaining high SCD accuracy, by integrating the Fourier transform and attention mechanisms. Experimental results on the SECOND and LandSat-SCD datasets demonstrate the effectiveness of the proposed method, and it achieves high resolution for change boundaries.

Wind power prediction is a challenging task due to the high variability and uncertainty of wind generation and weather conditions. Accurate and timely wind power prediction is essential for optimal power system operation and planning. In this paper, we propose a novel Adaptive Expert Fusion Model (EFM+) for online wind power prediction. EFM+ is an innovative ensemble model that integrates the strengths of XGBoost and self-attention LSTM models using dynamic weights. EFM+ can adapt to real-time changes in wind conditions and data distribution by updating the weights based on the performance and error of the models on recent similar samples. EFM+ enables Bayesian inference and real-time uncertainty updates with new data. We conduct extensive experiments on a real-world wind farm dataset to evaluate EFM+. The results show that EFM+ outperforms existing models in prediction accuracy and error, and demonstrates high robustness and stability across various scenarios. We also conduct sensitivity and ablation analyses to assess the effects of different components and parameters on EFM+. EFM+ is a promising technique for online wind power prediction that can handle nonstationarity and uncertainty in wind power generation.

High-dimensional and incomplete (HDI) data are frequently encountered in big date-related applications for describing restricted observed interactions among large node sets. How to perform accurate and efficient representation learning on such HDI data is a hot yet thorny issue. A latent factor (LF) model has proven to be efficient in addressing it. However, the objective function of an LF model is nonconvex. Commonly adopted first-order methods cannot approach its second-order stationary point, thereby resulting in accuracy loss. On the other hand, traditional second-order methods are impractical for LF models since they suffer from high computational costs due to the required operations on the objective's huge Hessian matrix. In order to address this issue, this study proposes a generalized Nesterov-accelerated second-order LF (GNSLF) model that integrates twofold conceptions: 1) acquiring proper second-order step efficiently by adopting a Hessian-vector algorithm and 2) embedding the second-order step into a generalized Nesterov's acceleration (GNA) method for speeding up its linear search process. The analysis focuses on the local convergence for GNSLF's nonconvex cost function instead of the global convergence has been taken; its local convergence properties have been provided with theoretical proofs. Experimental results on six HDI data cases demonstrate that GNSLF performs better than state-of-the-art LF models in accuracy for missing data estimation with high efficiency, i.e., a second-order model can be accelerated by incorporating GNA without accuracy loss.

An online streaming feature selection (OSFS) model is highly efficient in processing the high-dimensional streaming features. In practical big data-related applications, streaming features are mostly highly-incomplete due to various unpredictable reasons like the privacy protection, leading to the issue of online sparse streaming feature selection (OS2FS). The incomplete streaming features can lead to the uncertain relationship between the labels and sparse features during the feature selection process, yet existing OSFS and OS2FS models focus on the certain relationships, resulting in accuracy loss by improperly-selected features. To address this critical issue, this article presents a three(3)-way decision-incorporated OS2FS (3WDO) model with the following two-fold ideas: 1) utilizing the latent factor analysis (LFA) approach to pre-estimate the missing data of the concerned sparse streaming features and 2) integrating the three-way decision (3WD) into the streaming features selection process for appropriately modeling the uncertainty within the label-feature interactions. By doing so, the uncertain relationships between labels and sparse features are characterized by more information and looser tolerance, thereby minimizing the decision risk of feature selection. Experimental results on twelve real-world datasets demonstrate that the proposed 3WDO model significantly outperforms seven state-of-the-art OSFS and OS2FS models, which strongly supports its ability of addressing practical issues.

This paper presents a summary of the Latent in the Wild Fingerprint Recognition Competition held at the 2024 International Joint Conference on Biometrics (IJCB 2024). The competition attracted a total of 6 participating teams from academia and industry, representing 6 countries. In total, 3 valid submissions were evaluated by the organizers. The competition aimed to advance the state-of-the-art in latent fingerprint recognition by providing a challenging dataset of latent fingerprints collected in natural, non-ideal conditions. This paper summarizes the dataset, evaluation criteria, participant methods, and the competition results.

With the prosperity of deep learning techniques, salient object detection in remote sensing images (RSI-SOD) is concomitantly in full flourishing. However, due to the inherent challenges such as uncertainty in object quantities and scales, cluttered backgrounds, and blurred edges arising from shadows, most current approaches struggle for salient feature learning with the aid of heavy model architecture, yet often result in barely satisfactory performance. Some methods compromise model complexity to improve efficiency, albeit with significantly degraded results. To earn a satisfactory balance of efficacy and efficiency, we propose a new network for RSI-SOD, namely SFANet, based on progressive semantic flow and uncertainty-aware refinement. Specifically, we design a global semantic enhancement block (GSEB) to reduce background interference and accurately localize salient objects of varying quantities and scales, which further consists of three modularized components, i.e., semantic extraction module (SEM), interscale fusion module (IFM), and deep semantic graph-inference module (DSGM). SEM together with IFM contributes to the effective aggregation of multi-scale contexts by extracting fused and progressive semantic cues. DSGM performs semantic inference to better localize salient objects with irregularities in scale and topological structure. Furthermore, we present an uncertainty-aware refinement module (URM) to recognize salient objects in cluttered backgrounds and effectively suppress shadows. Extensive experiments are conducted on three RSI-SOD datasets, from which superior results can be achieved by our SFANet, outperforming the other cutting-edge methods. The code is available at https://github.com/ZhengJianwei2/SFANet.

Deep learning techniques have been used to build velocity models (VMs) for seismic traveltime tomography and have shown encouraging performance in recent years. However, they need to generate labeled samples (i.e., pairs of input and label) to train the deep neural network (NN) with end-to-end learning, and the real labels for field data inversion are usually missing or very expensive. Some traditional tomographic methods can be implemented quickly, but their effectiveness is often limited by prior assumptions. To avoid generating and/or collecting labeled samples, we propose a novel method by integrating deep learning and dictionary learning to enhance the VMs with low resolution by using the traditional tomography-least square method (LSQR). We first design a type of shallow and simple NN to reduce computational cost followed by proposing a two-step strategy to enhance the VMs with low resolution: (1) Warming up. An initial dictionary is trained from the estimation by LSQR through dictionary learning method; (2) Dictionary optimization. The initial dictionary obtained in the warming-up step will be optimized by the NN, and then it will be used to reconstruct high-resolution VMs with the reference slowness and the estimation by LSQR. Furthermore, we design a loss function to minimize traveltime misfit to ensure that NN training is label-free, and the optimized dictionary can be obtained after each epoch of NN training. We demonstrate the effectiveness of the proposed method through the numerical tests on both synthetic and field data.

Latent fingerprints are among the most important and widely used evidence in crime scenes, digital forensics and law enforcement worldwide. Despite the number of advancements reported in recent works, we note that significant open issues such as independent benchmarking and lack of large-scale evaluation databases for improving the algorithms are inadequately addressed. The available databases are mostly of semi-public nature, lack of acquisition in the wild environment, and post-processing pipelines. Moreover, they do not represent a realistic capture scenario similar to real crime scenes, to benchmark the robustness of the algorithms. Further, existing databases for latent fingerprint recognition do not have a large number of unique subjects/fingerprint instances or do not provide ground truth/reference fingerprint images to conduct a cross-comparison against the latent. In this paper, we introduce a new wild large-scale latent fingerprint database that includes five different acquisition scenarios: reference fingerprints from (1) optical and (2) capacitive sensors, (3) smartphone fingerprints, latent fingerprints captured from (4) wall surface, (5) Ipad surface, and (6) aluminium foil surface. The new database consists of 1,318 unique fingerprint instances captured in all above mentioned settings. A total of 2,636 reference fingerprints from optical and capacitive sensors, 1,318 fingerphotos from smartphones, and 9,224 latent fingerprints from each of the 132 subjects were provided in this work. The dataset is constructed considering various age groups, equal representations of genders and backgrounds. In addition, we provide an extensive set of analysis of various subset evaluations to highlight open challenges for future directions in latent fingerprint recognition research.

Blade icing detection is vital for wind turbines in cold climates, as it can prevent revenue loss and power degradation. Many machine learning models have been proposed to improve the detection of blade icing; however, earlier studies do not adequately address these issues due to the dynamics of sensor correlations and the imbalance of blade icing data, resulting in low precision and a high false alarm rate. In this study, we aim to address both of these challenges in order to identify blade icing more accurately. On this premise, we develop a spatial–temporal graph convolutional network (SGCN) that leverages the graph convolutional network for adaptively analyzing the dynamics of sensor correlations and a distance-based classifier to improve imbalanced learning. Experiments on the public UEA time series classification datasets and the real-world wind turbine datasets indicate that SGCN is capable of state-of-the-art accuracy, especially in the case of extremely imbalanced data.

With high spectral resolution, hyperspectral image(HSI) data will result in the Hughes phenomenon, which brings a huge challenge to hyperspectral image classification(HIC). Feature extraction can be applied to address this problem. But several traditional methods often ignore the spatial structure information of HSI data. In this paper, we propose a tensor nuclear norm based matrix regression based projections(TNMRP) for feature extraction of hyperspectral images. Firstly, TNMRP preprocesses the data by a filling method. Then, it automatically builds the graph of block-tensor samples and uses the optimal sparse coding coefficients to obtain the weight matrix. Finally, based on tensor representation, TNMRP calculates the optimal projection matrix. Experiments of classification on Indian Pines and Pavia University databases demonstrate the effectiveness of our proposed method.

Change detection is an important branch in remote sensing image processing. Deep learning has been widely used in this field. In particular, a wide variety of attention mechanisms have made great achievements. However, some models have become increasingly complex and large, often unfeasible for edge applications. This poses a major obstacle to industrial applications. In this paper, to solve the above challenges, we propose a Lightweight network structure to improve results while taking into account efficiency. Specifically, first, the shallow features are extracted by using the spatial exchange and change exchange of the down-sampling bi-temporal channel of the three-layer EfficientNet backbone network, and then the shallow features are used for low-dimensional skip-connection. After that, a hybrid dual-temporal data module is designed to mix the dual-temporal phase into a single image, then the high-dimensional low-pixel image is restored through the up-sampling. Finally the final change map is generated through the pixel-level classifier. Our method was evaluated on public datasets by evaluation indicators such as OA, IoU, F1, Recall, Precision.

Deep learning techniques have been widely used for semantic change detection (SCD) of remote sensing images (RSIs) and have shown encouraging performance. In this paper, we propose a novel neural network by embedding the difference enhancement (DE) module into the adjacent layers of ResNet for SCD of RSIs (DESNet), which can pay more attention to the changes of bi-temporal RSIs. Furthermore, we deploy the module of multi-scale parallel sampling spatial-spectral non-local (SSN) after feature extraction, which can effectively improve the robustness to large-scale changes and the integrity of the changed objects by fusing global features that sampled from the multi-scale feature space. The experimental tests demonstrate that our DESNet can achieve state-of-the-art accuracy on the SECOND dataset and the LandSat-SCD dataset.

A High-dimensional and incomplete (HDI) matrix is regularly adopted to portray the inherent non-negativity of interactions among numerous nodes, which is involved in countless industrial applications driven by big data. An inherently non-negative latent factor (LF) model can take out the intrinsical features from such data conveniently and effectually due to its unimpeded training process. However, it constructs the learning objective relying on a standard Euclidean distance, thereby seriously restricting its representative ability to HDI data generated by different domains. To address this issue, this work proposes an adaptive divergence-based non-negative LF (ADNLF) model following: 1) constructing a generalized objective function based on$\alpha - \beta $-divergence to inflate its ability to represent various HDI data; 2) connecting the optimization variables with output LFs by a smooth and single LF-dependent bridging function to satisfy the non-negativity constraints constantly; and 3) facilitating adaptive divergence in the learning objective through particle swarm optimization for high scalability. Empirical studies on eight HDI matrices validate that an ADNLF model evidently outstrips state-of-the-art models in terms of estimation accuracy as well as computational efficiency for missing data of an HDI dataset.

The accurate extraction of surface rivers is of great significance to ecology, residence and so on. In view of the incomplete recognition of river edge contour in the surface river extraction of remote sensing image in the classical deep learning network U-Net, the ability of the network to learn and retain the detailed information of feature map is enhanced by strengthening the attention mechanism and introducing the densely connected Atrous Spatial Pyramid Pooling on the basis of U-Net. The experimental results show that the Pixel Accuracy of water extraction results by this method is 92.1%, and the Mean Intersection Over Union is up to 90.3%, the improved algorithm can effectively extract accurate surface river information.

Two deep learning networks, patch-based and fully convolutional networks, are employed for automated detection and segmentation of shoulder joint structure on MRI. First, four segmentation models are build including three U-Net based models (glenoid segmentation model, humeral head segmentation model, glenoid and humeral head as a whole segmentation model) and one patch-based adjusted AlexNet (AANet) segmentation model. Then the four segmentation models are used to get the candidate bone regions from which the correct locations and regions of glenoid and humeral head are obtained by voting. Last, AANet model is further used to segment the edge of the bone with accuracy at the pixel level. From the experimental results, Dice Coefficient, Positive Predicted Value (PPV) and Sensitivity average accuracy are 0.92 ± 0.02,0.96 ± 0.03 and 0.94 ± 0.02 respectively. Note that our framework is also generic enough to be applied to the precise segmentation of specific organs and tissues in CT and MRI under small sample data.