Search Results (12,656)

Overfitting remains a major difficulty in training deep neural networks, especially when attempting to achieve good generalization in complex classification tasks. Standard dropout is often employed to address this issue; however, its uniform random inactivation of neurons typically leads to instability and insufficient performance increases. This paper proposes an upgraded regularization technique merging adaptive sigmoidal dropout with weight amplification, seeking to dynamically adjust neuron deactivation depending on weight statistics, activation patterns, and neuron history. The proposed dropout process uses a sigmoid function driven by a temperature parameter to determine deactivation likelihood and incorporates a “neuron recovery” step to restore important activations. Simultaneously, the method amplifies high-magnitude weights to select crucial traits during learning. The proposed method is tested on CIFAR-10, and CIFAR-100 datasets using four unique CNN architectures, including deep and residual-based models, to evaluate the approach. Results demonstrate that the suggested technique consistently outperforms both standard dropout and baseline models without dropout, yielding higher validation accuracy and lower, more stable validation loss across all datasets. In particular, it demonstrated superior convergence and generalization performance on challenging datasets such as CIFAR-100. These findings demonstrate the potential of the proposed technique to improve model robustness and training efficiency and provide an alternative in complex classification tasks. Full article

Accurate coastal bathymetry is critical for navigation, environmental monitoring, and marine resource management. This study presents a deep learning-based approach that fuses Sentinel-2 multispectral imagery with airborne hyperspectral-derived reference data to generate high-resolution satellite-derived bathymetry (SDB). To address the spatial resolution mismatch between Sentinel-2 (10 m) and LiDAR reference data (1 m), three interpolation methods—Inverse Distance Weighting (IDW), Natural Neighbor (NN), and Spline—were employed to resample spectral reflectance data to a 1 m grid. Two spectral input configurations were evaluated: the log-ratio of Bands 2 and 3, and raw RGB composite reflectance (Bands 2, 3, and 4). A Fully Convolutional Neural Network (FCNN) was trained under each configuration and validated using LiDAR-based depth. The RGB + NN combination yielded the best performance, achieving an RMSE of 1.2320 m, MAE of 0.9381 m, bias of +0.0315 m, and R² of 0.6261, while the log-ratio + IDW configuration showed lower accuracy. Visual and statistical analyses confirmed the advantage of the RGB + NN approach in preserving spatial continuity and spectral-depth relationships. This study demonstrates that both interpolation strategy and input configuration critically affect SDB model accuracy and generalizability. The integration of spatially adaptive interpolation with airborne hyperspectral reference data represents a scalable and efficient solution for high-resolution coastal bathymetry mapping. Full article

In the context of intelligent manufacturing, deep learning-based remaining useful life (RUL) prediction has become a research hotspot in the field of Prognostics and Health Management (PHM). The traditional approaches often require strong programming skills and repeated model building, posing a high entry barrier. To address this, in this study, we propose and implement a visualization tool that supports multiple model selections and result visualization and eliminates the need for complex coding and mathematical derivations, helping users to efficiently conduct RUL prediction with lower technical requirements. This study introduces and summarizes various novel neural network models for DL-based RUL prediction. The models are validated using the NASA and HNEI datasets, and among the validated models, the LSTM model best met the requirements for remaining useful life (RUL) prediction. In order to achieve the low-code usage of deep learning for RUL prediction, the following tasks were performed: (1) multiple models were developed using the Python (3.9.18) language and were implemented on the PyTorch (1.12.1) framework, providing users with the freedom to choose their desired model; (2) a user-friendly and low-code RUL prediction interface was built using Streamlit, enabling users to easily make predictions; (3) the visualization of prediction results was implemented using Matplotlib (3.8.2), allowing users to better understand and analyze the results. In addition, the tool offers functionalities such as automatic hyperparameter tuning to optimize the performance of the prediction model and reduce the complexity of operations. Full article

Pile foundations are critical structural elements, and their integrity is essential for ensuring the stability and safety of construction projects. Low-strain pile integrity testing (LSPIT) is widely used for defect detection; however, conventional manual interpretation of reflectograms is both time-consuming and susceptible to human error. This study presents a deep learning-driven approach utilizing transfer learning with convolutional neural networks (CNNs) to automate pile defect detection. A dataset of 328 reflectograms collected from real construction sites, including 246 intact and 82 defective samples, was used to train and evaluate the model. To address class imbalance, oversampling techniques were applied. Several state-of-the-art pretrained CNN architectures were compared, with ConvNeXtLarge achieving the highest accuracy of 98.2%. The accuracy reported was achieved on a dedicated test set using real reflectogram data from actual construction sites, distinguishing this study from prior work relying primarily on synthetic data. The proposed novelty includes adapting pre-trained CNN architectures specifically for real-world pile integrity testing, addressing practical challenges such as data imbalance and limited dataset size through targeted oversampling techniques. The proposed approach demonstrates significant improvements in accuracy and efficiency compared to manual interpretation methods, making it a promising solution for practical applications in the construction industry. The proposed method demonstrates potential for generalization across varying pile lengths and geological conditions, though further validation with broader datasets is recommended. Full article

This comprehensive review maps the fast-evolving landscape in which artificial intelligence (AI) and deep-learning (DL) techniques converge with the Internet of Things (IoT) to manage energy, comfort, and sustainability across smart environments. A PRISMA-guided search of four databases retrieved 1358 records; after applying inclusion criteria, 143 peer-reviewed studies published between January 2019 and April 2025 were analyzed. This review shows that AI-driven controllers—especially deep-reinforcement-learning agents—deliver median energy savings of 18–35% for HVAC and other major loads, consistently outperforming rule-based and model-predictive baselines. The evidence further reveals a rapid diversification of methods: graph-neural-network models now capture spatial interdependencies in dense sensor grids, federated-learning pilots address data-privacy constraints, and early integrations of large language models hint at natural-language analytics and control interfaces for heterogeneous IoT devices. Yet large-scale deployment remains hindered by fragmented and proprietary datasets, unresolved privacy and cybersecurity risks associated with continuous IoT telemetry, the growing carbon and compute footprints of ever-larger models, and poor interoperability among legacy equipment and modern edge nodes. The authors of researches therefore converges on several priorities: open, high-fidelity benchmarks that marry multivariate IoT sensor data with standardized metadata and occupant feedback; energy-aware, edge-optimized architectures that lower latency and power draw; privacy-centric learning frameworks that satisfy tightening regulations; hybrid physics-informed and explainable models that shorten commissioning time; and digital-twin platforms enriched by language-model reasoning to translate raw telemetry into actionable insights for facility managers and end users. Addressing these gaps will be pivotal to transforming isolated pilots into ubiquitous, trustworthy, and human-centered IoT ecosystems capable of delivering measurable gains in efficiency, resilience, and occupant wellbeing at scale. Full article

The four-way shuttle storage and retrieval system (FWSS/RS) is an advanced automated warehousing solution for achieving green and intelligent logistics, and task allocation is crucial to its logistics efficiency. However, current research on task allocation in three-dimensional storage environments is mostly conducted in the single-operation mode that handles inbound or outbound tasks individually, with limited attention paid to the more prevalent composite operation mode where inbound and outbound tasks coexist. To bridge this gap, this study investigates the task allocation problem in an FWSS/RS under the composite operation mode, and deep reinforcement learning (DRL) is introduced to solve it. Initially, the FWSS/RS operational workflows and equipment motion characteristics are analyzed, and a task allocation model with the total task completion time as the optimization objective is established. Furthermore, the task allocation problem is transformed into a partially observable Markov decision process corresponding to reinforcement learning. Each shuttle is regarded as an independent agent that receives localized observations, including shuttle position information and task completion status, as inputs, and a deep neural network is employed to fit value functions to output action selections. Correspondingly, all agents are trained within an independent deep Q-network (IDQN) framework that facilitates collaborative learning through experience sharing while maintaining decentralized decision-making based on individual observations. Moreover, to validate the efficiency and effectiveness of the proposed model and method, experiments were conducted across various problem scales and transport resource configurations. The experimental results demonstrate that the DRL-based approach outperforms conventional task allocation methods, including the auction algorithm and the genetic algorithm. Specifically, the proposed IDQN-based method reduces the task completion time by up to 12.88% compared to the auction algorithm, and up to 8.64% compared to the genetic algorithm across multiple scenarios. Moreover, task-related factors are found to have a more significant impact on the optimization objectives of task allocation than transport resource-related factors. Full article

Traffic crashes represent a major challenge for road safety, public health, and mobility management in complex urban environments, particularly in metropolitan areas characterized by intense traffic flows, high population density, and strong commuter dynamics. The development of short-term traffic crash prediction models represents a fundamental line of analysis in road safety research within the scientific community. Among these efforts, macro-level modeling plays a key role by enabling the analysis of the spatiotemporal relationships between diverse factors at an aggregated zonal scale. However, in cities like Bogotá, predicting short-term traffic crashes remains challenging due to the complexity of these spatiotemporal dynamics, underscoring the need for models that more effectively integrate spatial and temporal data. This paper presents a strategy based on deep learning techniques to predict short-term spatiotemporal traffic crashes in Bogotá using 2019 data on socioeconomic, land use, mobility, weather, lighting, and crash records across TMAU and TAZ zones. The results showed that the strategy performed with a model called SpatioConvGru-Net with top performance at the TMAU level, achieving $R^2$ = 0.983, MSE = 0.017, and MAPE = 5.5%. Its hybrid design captured spatiotemporal patterns better than CNN, LSTM, and others. Performance improved at the TAZ level using transfer learning. Full article

In nature, organisms often rely on the integration of local sensory information and prior experience to flexibly adapt to complex and dynamic environments, enabling efficient path selection. This bio-inspired mechanism of perception and behavioral adjustment provides important insights for path planning in mobile robots operating under uncertainty. In recent years, the introduction of deep reinforcement learning (DRL) has empowered mobile robots to autonomously learn navigation strategies through interaction with the environment, allowing them to identify obstacle distributions and perform path planning even in unknown scenarios. To further enhance the adaptability and path planning performance of robots in complex environments, this paper develops a deep reinforcement learning framework based on the Soft Actor–Critic (SAC) algorithm. First, to address the limited adaptability of existing transfer learning methods, we propose an action-level fusion mechanism that dynamically integrates prior and current policies during inference, enabling more flexible knowledge transfer. Second, a bio-inspired radar perception optimization method is introduced, which mimics the biological mechanism of focusing on key regions while ignoring redundant information, thereby enhancing the expressiveness of sensory inputs. Finally, a reward function based on ineffective behavior recognition is designed to reduce unnecessary exploration during training. The proposed method is validated in both the Gazebo simulation environment and real-world scenarios. Experimental results demonstrate that the approach achieves faster convergence and superior obstacle avoidance performance in path planning tasks, exhibiting strong transferability and generalization across various obstacle configurations. Full article

Automatic Speaker Verification (ASV) systems are increasingly employed to secure access to services and facilities. However, recent advances in speech deepfake generation pose serious threats to their reliability. Modern speech synthesis models can convincingly imitate a target speaker’s voice and generate realistic synthetic audio, potentially enabling unauthorized access through ASV systems. To counter these threats, forensic detectors have been developed to distinguish between real and fake speech. Although these models achieve strong performance, their deep learning nature makes them susceptible to adversarial attacks, i.e., carefully crafted, imperceptible perturbations in the audio signal that make the model unable to classify correctly. In this paper, we explore adversarial attacks targeting speech deepfake detectors. Specifically, we analyze the effectiveness of Basic Iterative Method (BIM) attacks applied in both time and frequency domains under white- and black-box conditions. Additionally, we propose an ensemble-based attack strategy designed to simultaneously target multiple detection models. This approach generates adversarial examples with balanced effectiveness across the ensemble, enhancing transferability to unseen models. Our experimental results show that, although crafting universally transferable attacks remains challenging, it is possible to fool state-of-the-art detectors using minimal, imperceptible perturbations, highlighting the need for more robust defenses in speech deepfake detection. Full article

Change detection (CD) in remote sensing (RS) is a fundamental task that seeks to identify changes in land cover by analyzing bitemporal images. In recent years, deep learning (DL)-based approaches have demonstrated remarkable success in a wide range of CD applications. However, most existing methods have limitations in detecting building edges and addressing pseudo-changes, and lack the ability to model feature context. In this paper, we introduce DFANet—a Deep Feature Attention Network specifically designed for building CD in RS imagery. First, we devise a spatial-channel attention module to strengthen the network’s capacity to extract change cues from bitemporal feature maps and reduce the occurrence of pseudo-changes. Second, we introduce a GatedConv module to improve the network’s capability for building edge detection. Finally, Transformer is introduced to capture long-range dependencies across bitemporal images, enabling the network to better understand feature change patterns and the relationships between different regions and land cover categories. We carried out comprehensive experiments on two publicly available building CD datasets—LEVIR-CD and WHU-CD. The results demonstrate that DFANet achieves exceptional performance in evaluation metrics such as precision, F1 score, and IoU, consistently outperforming existing state-of-the-art approaches. Full article

Estimating heterogeneous treatment effects (HTEs) across multiple correlated outcomes poses significant challenges due to complex dependency structures and diverse data types. In this study, we propose a novel deep learning framework integrating empirical copula transformations with a CNN-LSTM (Convolutional Neural Networks and Long Short-Term Memory networks) architecture to capture nonlinear dependencies and temporal dynamics in multivariate treatment effect estimation. The empirical copula transformation, a rank-based nonparametric approach, preprocesses input covariates to better represent the underlying joint distributions before modeling. We compare this method with a baseline CNN-LSTM model lacking copula preprocessing and a nonparametric tree-based approach, the Causal Forest, grounded in generalized random forests for HTE estimation. Our framework accommodates continuous, count, and censored survival outcomes simultaneously through a multitask learning setup with customized loss functions, including Cox partial likelihood for survival data. We evaluate model performance under varying treatment perturbation rates via extensive simulation studies, demonstrating that the Empirical Copula CNN-LSTM achieves superior accuracy and robustness in average treatment effect (ATE) and conditional average treatment effect (CATE) estimation. These results highlight the potential of copula-based deep learning models for causal inference in complex multivariate settings, offering valuable insights for personalized treatment strategies. Full article

Malware continues to develop, posing significant challenges for traditional signature-based detection systems. Visual malware classification, which transforms malware binaries into grayscale images, has emerged as a promising alternative for recognizing patterns in malicious code. This study presents a hybrid deep learning architecture that combines the local feature extraction capabilities of ConvNeXt-Tiny (a CNN-based model) with the global context modeling of the Swin Transformer. The proposed model is evaluated using three benchmark datasets—Malimg, MaleVis, VirusMNIST—encompassing 61 malware classes. Experimental results show that the hybrid model achieved a validation accuracy of 94.04%, outperforming both the ConvNeXt-Tiny-only model (92.45%) and the Swin Transformer-only model (90.44%). Additionally, we extended our validation dataset to two more datasets—Maldeb and Dumpware-10—to strengthen the empirical foundation of our work. The proposed hybrid model achieved competitive accuracy on both, with 98% on Maldeb and 97% on Dumpware-10. To enhance model interpretability, we employed Gradient-weighted Class Activation Mapping (Grad-CAM), which visualizes the learned representations and reveals the complementary nature of CNN and Transformer modules. The hybrid architecture, combined with explainable AI, offers an effective and interpretable approach for malware classification, facilitating better understanding and trust in automated detection systems. In addition, a real-time deployment scenario is demonstrated to validate the model’s practical applicability in dynamic environments. Full article

Air pollution, particularly fine (PM_2.5) and coarse (PM₁₀) particulate matter, poses significant risks to public health and environmental sustainability. This study aims to develop robust predictive and forecasting models for hourly PM concentrations in Craiova, Romania, using advanced hybrid Artificial Intelligence (AI) approaches. A five-year dataset (2020–2024), comprising 20 meteorological and pollution-related variables recorded by four air quality monitoring stations, was analyzed. The methodology consists of three main phases: (i) data preprocessing, including anomaly detection and missing value handling; (ii) exploratory analysis to identify trends and correlations between PM concentrations (PMs) and predictor variables; and (iii) model development using 23 machine learning and deep learning algorithms, enhanced by 50 feature selection techniques. A deep Nonlinear AutoRegressive Moving Average with eXogenous inputs (Deep-NARMAX) model was employed for multi-step-ahead forecasting. The best-performing models achieved R² values of 0.85 for PM_2.5 and 0.89 for PM₁₀, with low RMSE and MAPE scores, demonstrating high accuracy and generalizability. The GEO-based feature selection method effectively identified the most relevant predictors, while the Deep-NARMAX model captured temporal dynamics for accurate forecasting. These results highlight the potential of hybrid AI models for air quality management and provide a scalable framework for urban pollution monitoring, predicting, and forecasting. Full article

Wi-Fi sensing leverages variations in CSI (channel state information) to infer human activities in a contactless and low-cost manner, with growing applications in smart homes, healthcare, and security. While deep learning has advanced macro-motion sensing tasks, micro-motion sensing such as keypad stroke recognition remains underexplored due to subtle inter-class CSI variations and significant intra-class variance. These challenges make it difficult for existing deep learning models typically relying on fully connected MLPs to accurately recognize keypad inputs. To address the issue, we propose a novel approach that combines a discriminative feature extractor with a Kolmogorov–Arnold Network (KAN)-based classifier. The combined model is trained to reduce intra-class variability by clustering features around class-specific centers. The KAN classifier learns nonlinear spline functions to efficiently delineate the complex decision boundaries between different keypad inputs with fewer parameters. To validate our method, we collect a CSI dataset with low-cost Wi-Fi devices (ESP8266 and Raspberry Pi 4) in a real-world keypad sensing environment. Experimental results verify the effectiveness and practicality of our method for keypad input sensing applications in that it outperforms existing approaches in sensing accuracy while requiring fewer parameters. Full article

The growing adoption of IoT applications has led to increased use of low-power microcontroller units (MCUs) for energy-efficient, local data processing. However, deploying deep neural networks (DNNs) on these constrained devices is challenging due to limitations in memory, computational power, and energy. Traditional methods like cloud-based inference and model compression often incur bandwidth, privacy, and accuracy trade-offs. This paper introduces a novel Decentralized Distributed Sequential Neural Network (DDSNN) designed for low-power MCUs in Tiny Machine Learning (TinyML) applications. Unlike the existing methods that rely on centralized cluster-based approaches, DDSNN partitions a pre-trained LeNet across multiple MCUs, enabling fully decentralized inference in wireless sensor networks (WSNs). We validate DDSNN in a real-world predictive maintenance scenario, where vibration data from an industrial pump is analyzed in real-time. The experimental results demonstrate that DDSNN achieves 99.01% accuracy, explicitly maintaining the accuracy of the non-distributed baseline model and reducing inference latency by approximately 50%, highlighting its significant enhancement over traditional, non-distributed approaches, demonstrating its practical feasibility under realistic operating conditions. Full article