Search Results (2,741)

Satellite Synthetic Aperture Radar (SAR) is an established technology for studying and monitoring archaeological landscapes, providing insights into surface morphology and the presence of near subsurface features. However, its application in large-scale archaeological prospection is limited by the lack of robust, automated methods for SAR data analysis. This study introduces a novel Deep Learning pipeline to automatically detect and segment archaeological settlement mounds, known as tells, in central Iraq on satellite SAR data. The pipeline leverages a state-of-the-art supervised method for instance segmentation, YOLOv8-Seg, and medium-resolution satellite SAR products, specifically the Copernicus Sentinel-1 Interferometric Wide Swath Mode Ground Range Detected and Copernicus Global 30-m Digital Elevation Model products. The model identifies tell sites with an Average Precision of $0.495\pm 0.010$ and a pixel-wise Intersection over Union of $0.361\pm 0.048$ over the test areas. Archaeological interpretation of the model’s inferences confirms its reliability in locating and segmenting archaeological sites, leading also to the identification of previously unmapped potential sites. After a main test in central Iraq, the proposed workflow demonstrates promising transferability to a nearby test area in Iran, although with a need for regional fine-tuning to account for inherent variations in feature morphology and environmental context. This research establishes a baseline for future Deep Learning applications in Synthetic Aperture Radar-based archaeological prospection. Full article

Fixed-bottom or floating offshore structures are the foundations, platforms, and associated infrastructure that allow for oil and gas production systems, offshore wind turbines, and cabling. The remote nature of these structures and the harsh environment with high variability in wind, waves, currents, and weather make construction activity very difficult and unpredictable; the cost of variation in the schedule can lead to high construction vessel and personnel costs. The adoption of artificial intelligence using trends observed in historical data can help achieve more accurate construction costs and schedule predictions, reducing the capital expenditure cost of installation. A resource-significant activity, sometimes called a resource-critical activity or high-resource-demand activity, is an activity on a construction or project schedule that consumes a disproportionately large share of one or more resources compared with others. Plant Design Modelling (PDM) is a digital process that creates and manages a detailed 3D model of a building’s physical and functional characteristics and semantic information, such as cost and schedule. PDM serves as a single source of truth for multidisciplinary activities and, therefore, serves as a rich data source for various construction applications, including project scheduling and cost estimation. Neural networks (NNs), a subset of machine learning algorithms inspired by the human brain, excel at identifying patterns in complex datasets and making predictions, such as forecasting costs based on non-linear relationships and historical trends. Data from an offshore structure modification project were extracted from Aveva’s Everything PDM, focusing on installation activities to create a dataset for machine learning model training. The structured data extracted exhibit non-linear patterns; therefore, linear, regularised linear, robust linear, and the ensemble (tree-based) models and supervised neural network models with varied architecture and hyperparameter values were evaluated and compared. The best performance was obtained using the deep-optimised ANN model. The result obtained is consistent with previous studies. The neural network models show a superior ability to predict the non-linear nature of offshore construction activities’ time. Full article

This study compared unsupervised and supervised machine learning, and deep learning (U-Net) classifiers on Unmanned Aerial Vehicle (UAV) multispectral imagery to identify nitrogen status in potato crops under nitrogen (N) fertilization treatments, irrigation (I), and their interaction (N × I). The U-Net model outperformed all other methods, achieving accuracies for crop nitrogen status of 65–99% in N, 84–100% in I, and 41–82% in N × I treatments, with variation due to different input data. Supervised machine learning also performed well, with Support Vector Machine achieving 53–87, 66–86, and 32–66% respectively, and Random Forest 61–96, 70–81, and 33–65%. Unsupervised K-means yielded the lowest accuracies (47–58, 9–65, and 8–34%), demonstrating necessity of substantial supervision to delineate crop nitrogen and water status. These findings were confirmed by repeated analyses of UAV imagery acquired later in the growing season with consistent results. Comparable classification performance was observed for crop water status and leaf area index at both time points. Despite being demonstrated in a single-field, single-crop framework, the results provide proof of concept for applying deep learning classifiers to detect subtle nitrogen and water stress under field conditions in precision agriculture. Future research could test diverse agroecosystems and growing seasons, alternative deep learning algorithms, and sensor data fusion to improve classification accuracies. Full article

Infrared (IR) and visible image fusion aims to synthesize a composite representation that integrates the thermal target saliency of IR imagery with the textural richness of visible imagery. Existing deep learning-based methods have achieved promising progress in this field. However, they either operate at the pixel level without semantic priors, or rely on segmentation supervision to obtain such priors. Both approaches limit their practicality and performance in complex scenes. To design a lightweight fusion network that leverages semantic priors without segmentation supervision, we propose SPE2Fusion, a semantic prior-driven fusion network that operates through a dual-stage semantic injection paradigm. Specifically, a lightweight semantic encoder is designed to extract multi-scale scene priors in an end-to-end manner optimized solely by the fusion loss, without requiring segmentation mask annotations. Then, these priors are injected at two complementary stages: the Efficient Semantic Feature Awareness (ESFA) module applies spatially adaptive attention at the encoding stage to amplify semantically salient regions, while the Efficient Semantic Feature Embedding (ESFE) module applies semantically conditioned spatial normalization at the decoding stage to ensure coherent texture reconstruction. Finally, a bidirectional cross-attention fusion block is introduced to integrate complementary cross-modal features under this dual semantic guidance. The network is supervised by a multi-constraint loss combining gradient fidelity, intensity preservation, and structural similarity terms. Comprehensive experiments on the MSRS, LLVIP, and RoadScene benchmarks demonstrate that SPE2Fusion achieves state-of-the-art performance against representative methods (e.g., CrossFuse and DDBFusion), ranking first on four of six metrics on the MSRS test set, specifically EN (6.70), $Q^{AB/F}$ (0.86), AG (6.06), and SD (43.44), while maintaining strong generalization on unseen datasets without domain adaptation. Full article

Traffic signal control increasingly faces disturbed operating conditions such as incidents, abrupt demand surges, sensing degradation, and abnormal driving patterns. Under these nonstationary regimes, classical fixed-time and actuated strategies may exhibit slow recovery, while purely data-driven controllers can be brittle when disturbance characteristics shift. This paper proposes an LLM-in-the-loop architecture for disturbed traffic signal control that integrates (i) deep learning for disturbance detection and short-horizon traffic forecasting, (ii) a disturbance-aware candidate generation and scoring layer (template/retrieval-based), and (iii) a constrained large language model (LLM) that selects or minimally repairs signal plans only within constraint-screened action templates. A deterministic validator enforces safety and operational constraints, including minimum/maximum greens, cycle feasibility, and clearance rules, by checking action feasibility before execution. The method is formulated as constrained decision making under uncertainty, where disturbance estimates and predictive confidence shape both retrieval/scoring and LLM supervision. The originally reported SUMO evaluation considered multiple disturbance categories, including capacity drops, demand shocks, and sensing dropouts as well as reported network delay, queue spillback, recovery time, and switching stability. Within the originally reported SUMO scenarios, descriptive results suggest that among the selected baselines, the proposed DL + LLM framework reported lower mean values of delay, spillback frequency, and recovery time than the fixed-time, actuated, and retrieval-only baselines. The reported validator-detected action-feasibility violations were zero; this result concerns timing-action feasibility rather than an absence of traffic-state risks such as spillback. Full article

Pig vocalization recognition can support non-invasive monitoring in precision livestock farming, but labelled pig-sound recordings are often limited for specific behaviours or physiological states. Under few-shot conditions, deep models may overfit, whereas traditional acoustic features may not fully describe class-specific time-frequency patterns. This study proposed PSA-AP, a pig-sound adaptation pipeline that uses log-Mel spectrograms and integrates SpecAugment-based domain expansion, ImageNet-pretrained ResNet18 knowledge transfer, and ArcFace-based feature alignment. The method was designed to reduce dependence on limited labelled samples, improve task-adapted representation learning, and enhance inter-class separability in the embedding space. Experiments were conducted on a five-class few-shot pig vocalization classification task, including eat, estrous, farrowing (fap), howl, and oink sounds collected from 10 adult Landrace pigs. Using $K=\{5,10,15,20,25,30\}$ labelled wav files per class and five random seeds, each selected training wav file and each held-out test wav file was converted into one 1.0 s log-Mel spectrogram for model training or evaluation. Final evaluation was based on the last checkpoint of each training run. PSA-AP achieved the best mean Accuracy, Macro-F1, and UAR at every K-shot setting. At $K=30$, PSA-AP reached 90.60% Accuracy, 90.49% Macro-F1, and 90.60% UAR, exceeding Raw by 7.80, 7.82, and 7.80 percentage points, respectively. These results indicate that the proposed integration of domain expansion, knowledge transfer, and feature alignment provides a feasible supervised adaptation strategy for few-shot pig vocalization recognition within the current protocol. Full article

Accurate forecasting of high-dimensional meteorological fields remains challenging due to the complex spatio-temporal dynamics of atmospheric systems and the presence of heterogeneous training difficulty across space and lead time. Existing deep forecasting approaches usually optimize all prediction units uniformly, which may overemphasize low-benefit or weakly generalizable supervision signals. To address this issue, we propose Spatio-Temporal Selective Learning (ST-SL), an online training framework that estimates the learnability of each prediction unit by comparing the main model with a frozen reference model and computes the loss only over selected high-benefit spatio-temporal units. To provide an effective forecasting backbone, we further introduce VASTFormer, a variable-aware spatio-temporal Transformer that models cross-variable dependencies, incorporates physics-enhanced Solar Positional Encoding, and captures atmospheric trajectories with an efficient temporal translator. Experiments on the ERA5 reanalysis dataset show that VASTFormer outperforms representative spatio-temporal baselines, while ST-SL further improves accuracy without adding inference-time parameters or computational cost. Compared with the strongest baseline, VASTFormer+ST-SL reduces MSE, MAE, and RMSE by 8.84%, 6.70%, and 4.54%, respectively. Meteorological skill evaluation further shows an average ACC of 0.9801 and RMSESS of 0.8104, and percentile-based extreme-condition evaluations confirm consistent improvements across standard and high-impact forecasting scenarios. These results indicate that selective supervision can improve generalization in dense meteorological forecasting. Full article

Vibration-based fault diagnosis is widely used for rotating machinery health monitoring, but practical diagnosis is often limited by scarce fault labels and uncertain measurement length. Longer vibration records can improve decision reliability but increase sensing and computational cost, whereas overly short records may yield unreliable predictions under noise and measurement corruptions. This paper studies few-shot fault diagnosis as a measurement-constrained decision task, in which the model identifies the fault class and determines when sufficient vibration evidence has been acquired. We propose a measurement-efficient diagnosis framework that combines prior knowledge from unlabeled healthy signals, physically constrained augmentation of scarce labeled samples, and adaptive early stopping in a shared one-dimensional feature extractor. The framework is evaluated on the UORED-VAFCLS and Paderborn University bearing datasets under 6-, 8-, and 10-shot settings with controlled corruption levels. Results show robust diagnostic performance with fewer acquired vibration windows than with fixed-length inference. In the representative PU-Hard 8-shot setting, the proposed method achieves 80.26% accuracy with an average of 1.2432 acquired windows and reduces the evaluation cost J from 0.3929 to 0.2596 compared with fixed four-window inference. These results indicate that adaptive measurement improves the accuracy–cost trade-off in few-shot vibration diagnosis. Full article

In electronic training scenarios, the safety risk of a soldering iron cannot be determined by object detection alone, as its state must be further distinguished among hand-held, stand-supported, desk-exposed, and uncertain interactions. To address this problem, this paper proposes RISNet, the Relation-aware Interaction State Network, which establishes a two-stage instance-level interaction understanding framework for soldering iron safety monitoring. In the first stage, YOLO is used to generate candidate instances of soldering irons and related environmental objects, and dual-layer feature fusion is adopted to jointly exploit shallow details and deep semantics. In the second stage, the soldering iron is treated as the interaction subject. The Pointer-Head models associations between the subject and contextual objects, and the State-Head predicts the safety state conditioned on subject-object relational constraints. To reduce false alarms from false detections and weak interactions, RISNet introduces a Quality-Head that estimates the reliability of each interaction conclusion and filters low-quality predictions during inference. The unknown label is used during training as conservative supervision for weak, unreliable, or indeterminate interaction evidence, with semantics close to the no-interaction label in HOI. This paper also constructs the Soldering Iron Safety Interaction Dataset (SISID) to support detection, interaction modeling, and state evaluation of slender metallic tools in training scenarios. On the SISID validation split, RISNet achieves an Overall F1 of 95.38%, an Overall Precision of 96.73%, and an inference speed of 57.1 FPS, satisfying the centralized single-frame polling requirement considered in this work. Full article

Clouds frequently degrade optical satellite imagery, limiting the reliability of remote sensing models. However, in the literature, cloud-free analyses are often performed by excluding cloudy images from machine learning datasets and methods. This restricts their usefulness in time-critical scenarios such as disaster response, where waiting for cloud-free imagery is impractical. Cloud removal can mitigate this issue, but methods remain imperfect and may introduce visual artifacts. Therefore, it is desirable to develop cloud-robust methods by combining optical imagery with radar data, a modality unaffected by clouds. While datasets for machine learning combine optical and radar data, most researchers exclude cloudy images from training and evaluation. We identify this exclusion as a limitation that reduces applicability to cloudy scenarios and address it by introducing CloudyBigEarthNet (CBEN), a dataset of paired optical and radar images containing cloud occlusions for land-use and land-cover classification. Using average precision (AP), we show that state-of-the-art methods trained on clear-sky optical and radar data suffer performance drops of between 23.8 and 33.4 AP points when tested on cloudy imagery. We adapt these methods using cloudy images during training and improve AP on cloudy test cases by 17.2 to 28.7 AP points. Code and dataset have been published. Full article

This study develops a cooperative multi-agent deep reinforcement learning (MARL) framework for simulation-based cost-overrun mitigation in smart grid construction projects under dynamic engineering uncertainty. Modern smart grid construction involves digital substations, renewable-energy-connected facilities, flexible transmission assets, intelligent monitoring systems, and geographically distributed contractors; therefore, cost escalation is driven by sequential interactions among procurement, schedule execution, equipment deployment, supervision, weather, logistics, and price volatility. The proposed framework models procurement management, construction scheduling, equipment allocation, and supervision-control units as decentralized agents embedded in a calibrated construction simulation environment. The environment is parameterized from 42 smart grid construction projects in Henan Province, China and generates disturbance scenarios involving weather efficiency loss, transportation delay, market-price volatility, labor shortage, and supply-chain interruption. A hybrid DQN–PPO mechanism represents mixed decision structures: value-based DQN modules handle discrete managerial choices such as task acceleration, supplier switching, and procurement timing, whereas PPO modules adjust continuous resource-allocation and recovery-intensity decisions. A hierarchical reward function combines local departmental objectives with project-level penalties for cost overrun, schedule delay, idle resources, recovery expenditure, safety risk, and environmental impact. The experimental protocol uses 30 paired random seeds, nonparametric bootstrap confidence intervals, Holm-adjusted Wilcoxon signed-rank tests, and comparison with deterministic optimization, rolling-horizon MPC, stochastic/robust optimization, single-agent DRL, MAPPO, MADDPG/MATD3, QMIX, and HAPPO baselines. The proposed framework achieves a mean cost-overrun rate of 6.83% and a mean schedule deviation of 16.82 days, reducing cost overrun by 18.7% and schedule deviation by 21.4% relative to rule-based construction management under the reported disturbance settings. The calibrated simulation evidence establishes a statistically evaluated decision-support framework for coordinated construction cost control and provides an artifact-level reproducibility pathway through configuration files, random-seed lists, anonymized synthetic benchmarks, and aggregated logs. Full article

Background/Objectives: Conjunctival hyperemia is a common clinical finding in clinical practice; however there are significant differences between graders. Vessel detection using deep-learning approaches could enable more objective measures. We aimed to evaluate vascular metrics derived from automated vessel detection and compare these metrics with manual severity gradings. Methods: Slit lamp images from 139 glaucoma patients were included. Images from 103 participants were used as the primary development dataset and the remaining as a validation subset. The images were independently graded by two graders for conjunctival hyperemia using the Efron Grading Scheme. Conjunctival vessels were detected using an automated vessel detection pipeline based on semi-supervised learning. Vessel density, fractal dimension and tortuosity were calculated and compared with the manual Efron grades. Results: Grading of conjunctival hyperemia between the two graders were consistent (Spearman’s rho: 0.79; ICC: 0.79 [95%CI: 0.72–0.84]) but showed significant differences with a higher proportion of differences in the moderate grades. Of the vascular metrics, vessel density showed significant associations with the individual Efron grading and against the mean Efron grading (0.78, p < 0.001). Fractal dimension was significantly associated with the mean Efron grading (0.55, p < 0.001). Agreements were similar in the subset (vessel density, 0.80, p < 0.001; fractal dimension 0.62, p < 0.001). Vessel tortuosity showed lower agreements (<0.23). Conclusions: Vessel density and fractal dimension showed significant associations with manual Efron gradings. These metrics could be potentially used to enable more objective and interpretable measures of conjunctival hyperemia severity. Full article

Machine learning (ML) has become a central enabler of data-driven decision-making across smart manufacturing, healthcare, finance, computer vision, robotics, and environmental sustainability. Despite the rapid growth of ML applications, existing review studies remain largely domain-specific and provide limited cross-domain synthesis of methodological trends, deployment challenges, and emerging research directions. This systematic literature review aims to provide a comprehensive and comparative analysis of ML applications across seven high-impact domains while identifying dominant learning paradigms, implementation challenges, and future research opportunities. Following the PRISMA 2020 guidelines, peer-reviewed studies published between 2015 and 2025 were systematically collected from major scientific databases, including ScienceDirect, IEEE Xplore, SpringerLink, Wiley Online Library, MDPI, and Web of Science. Studies were screened using predefined inclusion and exclusion criteria and categorized according to application domain, ML paradigm, algorithm type, data characteristics, and deployment context. The findings indicate that supervised learning and deep learning dominate most application areas, with convolutional neural networks emerging as the primary approach for image-based and perception-driven tasks. Reinforcement learning, although highly promising for sequential decision-making and adaptive control, remains comparatively underutilized due to safety, computational, and deployment constraints. Across domains, recurring challenges include data quality, interpretability, scalability, model robustness, computational requirements, and ethical considerations. Overall, this review provides a structured cross-domain synthesis of ML applications and highlights the growing importance of explainable, trustworthy, and deployable AI systems for future intelligent and sustainable technologies. Full article

Depth estimation is an active area of research in computer vision. When restricted to passive imaging, it is frequently approached as a stereo matching problem. Due to significant advancements in deep learning, stereo matching models have seen substantial development. Most stereo matching models utilize supervised learning, relying on disparity maps as ground truth, typically obtained via active imaging systems. In this work, we introduce a passive multi-view framework for generating ground-truth depth data for stereo matching models using Computational Integral Imaging (CII). Using CII, we extract object depths from multi-view images obtained via a passive camera array and use these depths as training targets, eliminating the need for active depth acquisition. To improve the robustness and accuracy of CII-based depth extraction, we propose an object-aware formulation that incorporates pretrained object segmentation into the depth extraction process. This enables more reliable depth estimation for objects with complex appearances and challenging scene contexts such as occlusions. Furthermore, we exploit the camera array as a multi-stereo acquisition system, generating diverse stereo pairs with varying baselines and viewing orientations. The resulting training data expose stereo matching models to a broader range of geometric configurations than conventional stereo datasets. Our results demonstrate that this approach enhances the generalization capabilities of stereo matching models. Full article

Intrusion detection in IoT and IIoT networks must operate under tight resource constraints, yet most published machine learning-based IDS solutions report accuracy on held-out data without addressing whether the trained model can actually run on the target hardware. We address this gap with an end-to-end study spanning dataset preprocessing, model training, INT8 quantisation, and physical execution on two real microcontrollers. Five supervised classifiers—Logistic Regression, Decision Tree (depth 5), Random Forest, XGBoost, and LightGBM—plus an MLP deep learning baseline are evaluated on binary and ten-class intrusion detection tasks using the TON_IoT network dataset. A 5-fold stratified cross-validation confirms stable performance across splits, with LightGBM reaching $F_1=0.9993\pm 0.0001$. Models are then exported through three quantisation pipelines: m2cgen C code generation for the two lightest classifiers, TensorFlow Lite Micro full-integer INT8 for the MLP (9.34× size reduction to 13.03 KB), and a custom post-training INT8 binary format for XGBoost and LightGBM (18.91× compression for LightGBM to 73.85 KB). All five quantised models are deployed to an Arduino Mega 2560 (ATmega2560, 16 MHz, 8 KB SRAM) and an ESP32-C3 SuperMini (RISC-V, 160 MHz, 400 KB SRAM) and benchmarked on physical hardware across 500 timed inferences per model (250 per input class), with firmware predictions confirmed to match the Python 3.11 float model on both test vectors. The Decision Tree achieves 5.6 µs inference on the ESP32-C3; LightGBM INT8 ($F_1=0.9992$) provides the best accuracy–size trade-off among ensemble models. Cross-platform comparison reveals that the RISC-V device is 5.8–7.8× faster than the 8-bit AVR for identical model code. A cross-domain evaluation using CIC-IoT-Dataset2023 identifies large normalised distribution shifts (up to $\delta =5.95$ in packet asymmetry), quantifying the generalisation gap that remains an open challenge. Full article