Search Results (346)

Accurate lens distortion correction is important for calibration, registration, image stitching, and 3D reconstruction, especially in low-data device-specific settings where disposable or specialised cameras cannot provide large calibration datasets. We address distortion correction for cameras with highly irregular or non-stationary distortion fields, where fixed polynomial models and generic learning-based rectification methods can struggle. We propose a framework based on Deep Gaussian Processes (DGPs) to model the non-linear mapping required for undistortion. The key motivation is that conventional single-layer GPs with stationary kernels must use one global notion of smoothness, whereas DGPs can represent spatially varying behaviour through composed latent mappings while preserving per-pixel predictive uncertainty. This uncertainty can be used to identify or downweight unreliable corrected regions in downstream tasks. We evaluate the method on three real camera datasets with increasing distortion complexity. The full structured acquisitions contain 512 horizontal and 512 vertical line images per camera. These are not thousands of natural calibration images, but they yield up to 29,532, 11,311, and 31,686 detected intersection correspondences for the RPI, Theta, and Pillcam datasets, respectively. This distinction is important for cameras where acquiring many independent images is impractical. The results are assessed using qualitative rectification, uncertainty maps, normalised collinearity errors, and total training time. Polynomial calibration remains strongest for the regular radial RPI distortion, while DGP and DGP2 models show lower normalised collinearity-error distributions than the standard GP and lightweight MLP baselines on the more distorted Theta and Pillcam datasets. For the full datasets, total DGP/DGP2 training times ranged from 2383.50 s to 10092.50 s, reflecting the additional computational cost of probabilistic non-stationary modelling. Full article

Deformable image registration (DIR) is central to medical image analysis, supporting spatial alignment for longitudinal studies and multi-modal fusion. Learning-based methods such as CNNs and transformers provide rapid inference but often require large training datasets and can underperform classical iterative methods for specific anatomies or modalities. Implicit neural representations (INRs) offer a data-efficient alternative by modeling deformation fields as continuous coordinate-to-displacement mappings, yet their per-pair optimization makes runtime efficiency and robustness to initialization essential. We introduce KAN-IDIR and RandKAN-IDIR, the first Kolmogorov–Arnold network (KAN)-based INR framework for pairwise-optimized, resolution-independent DIR, designed to improve seed stability and resource efficiency without requiring a large training dataset. KANs use learnable activation functions that are well suited to continuous, physically structured deformation fields. RandKAN-IDIR further reduces cost through randomized basis sampling, preserving registration quality with fewer basis functions. We evaluate the methods on lung CT, brain MRI, and cardiac MRI datasets against pairwise-optimized neural approaches, dataset-trained deep models, and classical baselines. KAN-IDIR and RandKAN-IDIR provide the strongest overall performance among pairwise-optimized neural registration methods across all three datasets, with low computational overhead and superior stability across random initializations. On ACDC, KAN-IDIR also achieves the highest DSC and best deformation regularity among all compared methods. RandKAN-IDIR slightly outperforms adaptive basis selection variants while avoiding their additional training-time complexity. This makes the approach practical for reproducible clinical research use. Source code is publicly available. Full article

Mongolian rangelands face interacting ecological and livelihood pressures, including livestock pressure, vegetation change, soil-carbon dynamics, household income variability, and inefficiencies in livestock by-product recovery. This paper examines whether observed administrative and household data, field-observed pilot-area audit evidence, satellite-derived/backcast vegetation indicators, model-reconstructed ecological trajectories, econometric associations, machine-learning diagnostics, Monte Carlo uncertainty outputs, and scenario-based carbon-finance calculations are consistent with a study-specific ecological–economic feedback framework in Mongolian pastoral rangelands. The analysis combines observed livestock and household data, satellite-derived vegetation indicators, field-anchored soil organic carbon (SOC) information, climate controls, and pilot-area by-product audit evidence in a dual-scale framework comprising nine pasture-user groups in Öndörshireet Soum, Töv Aimag, and a national soum-level panel for 2002–2024. SOC, above-ground biomass (AGB), and below-ground biomass (BGB) trajectories are treated as model-reconstructed series rather than independently observed annual field measurements. Fixed-effects panel models are used to estimate conditional associations, while machine-learning models assess predictive consistency within reconstructed data structures. Under the fitted full specification, the best-performing national-panel model reports an out-of-sample R² of 0.942 for model-reconstructed SOC; this value is interpreted as high internal predictive consistency within the reconstructed SOC panel, not as independent validation of observed annual SOC change. Because the SU/SOC ratio mechanically contains SOC, the full-specification predictive results are subject to leakage risk, and leakage-free validation is needed for a more conservative assessment of predictive performance. Panel estimates suggest that vegetation condition is positively associated with ln(household income), while the by-product waste ratio is negatively associated with ln(income), conditional on fixed effects and model specification. Scenario-based carbon-finance outputs, framed with reference to Verra’s VM0042 Improved Agricultural Land Management methodology, vary materially with compliance, carbon price, weighted average cost of capital, and revenue-sharing assumptions; these outputs are illustrative sensitivity calculations and do not demonstrate VM0042 compliance, project eligibility, project-registration readiness, verified emission reductions, or credit-issuance readiness. The findings are associational, reconstruction-dependent, and scenario-based. They support an analytical framework rather than establish a closed causal loop. Full article

Accurate attitude measurement of non-cooperative space targets is essential for on-orbit servicing, active debris removal, and autonomous rendezvous missions. To address the challenges associated with unknown geometry, rapid tumbling motion, and the limited observability of single-view systems, this study proposes a cooperative multi-view measurement framework for three-dimensional reconstruction and attitude estimation. Multiple spacecraft are deployed to form a stable observation configuration, and multi-view image sequences are acquired to strengthen geometric constraints. A learning-based multi-view stereo reconstruction module is used to estimate depth information and reconstruct point clouds, which are further processed through iterative closest point (ICP) registration to derive inter-frame attitude variations. An extended Kalman filter (EKF) is then introduced to improve temporal consistency and suppress measurement noise. Validation is conducted in a numerical simulation using a simplified Fengyun-1 (FY-1) satellite model under a three-spacecraft cooperative fly-around scenario. The simulation results demonstrate that the proposed method achieves high-precision attitude estimation, with attitude errors below $0.3°$ and positional errors within $0.05\mathrm{m}$. Comparative experiments show that the method maintains stable measurement performance under varying observation distances and viewing configurations. The proposed framework provides a reliable and robust measurement solution for dynamic attitude determination of free-tumbling space targets. Full article

Background/Objectives: Postoperative remodeling and positional deviations of the mandibular coronoid process (CP) after orthognathic surgery remain insufficiently characterized, particularly in three-dimensional analyses. The aim of this study was to evaluate qualitative and quantitative CP changes following bimaxillary orthognathic surgery using a deep learning-assisted three-dimensional workflow. Methods: This retrospective study included 75 patients treated with combined orthodontic–surgical therapy, including 25 patients with skeletal Class II malocclusion and 50 patients with skeletal Class III malocclusion. Preoperative and 6-month postoperative computed tomography scans were analyzed. Automatic segmentation and three-dimensional reconstruction were performed using a convolutional neural network based on the nnU-Net architecture. Qualitative assessment included evaluation of CP displacement patterns and visualization of local surface differences using heat maps. Quantitative analysis included volumetric assessment of preoperative and postoperative CP models, calculation of apposition-compatible (Vapo) and resorption-compatible (Vres) volumetric changes, and mixed-effects modeling accounting for within-patient correlations. Results: Medial displacement of the CP predominated in both skeletal classes and was more frequent in Class III patients. Qualitative surface analysis demonstrated a consistent location-dependent remodeling pattern, characterized by predominant apposition-compatible changes on the lateral and medial surfaces and predominant resorption-compatible changes along the anterior border. Quantitative analyses revealed an overall positive remodeling balance, although substantial inter-individual variability was observed. Mixed-effects analyses demonstrated no significant overall effects of side or skeletal class on volumetric remodeling; however, a significant interaction between side and skeletal class was identified for net remodeling balance. A significant random patient effect indicated considerable variability in remodeling response among individuals. Conclusions: AI-assisted three-dimensional analysis enables a reproducible assessment of postoperative CP remodeling following orthognathic surgery. Coronoid process remodeling is characterized by heterogeneous, location-dependent surface changes and substantial inter-individual variability. The observed remodeling patterns are compatible with adaptive responses to altered postoperative biomechanical conditions, although the underlying biological mechanisms remain to be clarified. Full article

Transformer-based deformable brain MRI registration achieves high overlap accuracy, but predicted displacement fields can contain voxels with a non-positive Jacobian determinant—local foldings that violate the diffeomorphism assumption required by tensor-based morphometry and atlas-fusion segmentation workflows. We introduce HypEReg, a non-linear hyperelastic regularizer that acts directly on the Jacobian determinant of the predicted displacement field. HypEReg couples a clamped-rational volume-distortion penalty ${(det{J}_{\varphi }-1)}^2/max(det{J}_{\varphi },ϵ)$ with an explicit per-voxel anti-folding hinge ${[max(0,ϵ-det{J}_{\varphi })]}^2$, integrated as a purely loss-side module into a TransMorph backbone with no inference-graph modifications. On the IXI atlas-to-subject benchmark (115 test subjects), HypEReg-TransMorph maintains grouped Dice (0.7537) while reducing the $det\left(J_{\varphi }\right)\le 0$ voxel ratio from $1.502\times {10}^{-2}$ (TransMorph) to $1.5\times {10}^{-5}$, with identical per-case runtime and parameter count to the unregularized baseline. In strict zero-shot transfer to OASIS Learn2Reg test pairs (no fine-tuning), HypEReg-TransMorph achieves Dice 0.7756 with a $det\left(J_{\varphi }\right)\le 0$ ratio of $7.6\times {10}^{-5}$, roughly two orders of magnitude below plain TransMorph zero-shot (Dice 0.7691; ratio $9.6\times {10}^{-3}$); downstream multi-atlas label fusion further confirms the practical benefit of fold suppression (fused Dice 0.8271 vs. 0.8201 for TransMorph). OASIS-2 longitudinal and ROI analyses support deformation plausibility (lower folding/SDlogJ and stronger ventricular ROI agreement), while clinical-covariate associations remain exploratory rather than biomarker-validating. Determinant-level, non-linear hyperelastic regularization substantially suppresses folding in Transformer dense-flow brain MRI registration while preserving alignment accuracy and adding zero inference cost, providing a practical drop-in regularization strategy that improves the reliability of deformation fields for morphometry-oriented deformable registration. Full article

Wheat is a major staple crop, and storage mold growth poses a severe threat to grain safety and quality stability. Natural mold development in stored wheat exhibits subtle, localized, and highly heterogeneous characteristics. Existing unimodal methods and global fusion approaches generally suffer from insufficient local feature sensitivity, hindering fine-grained mold severity grading. To address this limitation, we propose a Mask-Guided Fine-Grained Fusion Network, a weakly supervised framework based on local RGB–HSI fusion. This framework employs a dynamic parallel A/B experimental design to construct time-matched proxy labels via weakly supervised learning. A standardized preprocessing pipeline including single-kernel extraction, foreground segmentation, and cross-modal registration is established to resolve RGB–HSI spatial misalignment, ensuring physical-level spatial consistency of multimodal features. The model incorporates a Foreground-Aware Spectral Recalibration (FASR) module to suppress background noise, a Mask-Guided Dilated Cross-modal Local Attention (MDCLA) mechanism to establish fine-grained local mappings between RGB visual phenotypes and hyperspectral responses, and a sample-level adaptive fusion strategy to dynamically weight features by modal reliability, enhancing representation of complex samples across all mold stages. Experiments show that the Mask-Guided Fine-Grained Fusion Network achieves 0.9689 classification accuracy, 0.9698 Macro-F1 score, and 0.0593 Mean Absolute Error (MAE), significantly outperforming state-of-the-art unimodal deep models and global attention fusion baselines. This work provides a proof-of-principle framework for fine-grained non-destructive mold risk assessment in stored wheat. Full article

The growth of Intelligent Transportation Systems (ITS) has made high-quality perception data from multi-sensor setups essential. Pairing millimeter-wave (mmW) radar with a monocular camera is a common way to recover three-dimensional information about the environment, but aligning the two is difficult because sparse radar point clouds and dense camera images differ sharply in how they sense a scene. The problem grows more severe in roadside infrastructure, where the high mounting elevation introduces perspective distortion that vehicle-mounted systems rarely face. This paper presents a systematic review, conducted under the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, of radar-camera extrinsic calibration for fixed roadside infrastructure, organizing existing work into a taxonomy that separates traditional two-stage pipelines from recent end-to-end learning frameworks. Because methods designed specifically for roadside units remain scarce, the review also covers vehicle- and robot-mounted methods whose static-sensor formulation carries over to fixed roadside deployment. For the two-stage pipeline, the analysis covers target-based and targetless correspondence registration along with the optimization techniques and algorithmic assumptions behind parameter estimation. The end-to-end learning literature shows a clear shift toward self-supervised and fusion-based models, some of which report real-time performance. The review also compares the metrics and procedures used to quantify calibration accuracy. Progress is evident, but robustness in cluttered urban environments remains an open challenge, and the paper closes by outlining future directions, arguing that standardized roadside benchmarks are needed before scalable, targetless calibration can mature. Full article

Web service classification plays an important role in software reuse, service discovery, and automatic metadata organization. Although recent deep learning approaches have improved classification performance by using service names and natural-language descriptions, most existing methods still operate as black-box models and offer limited insight into how different metadata sources influence classification decisions. This lack of transparency reduces their practical usefulness for developers who need to verify predicted categories, analyze incorrect classifications, and improve service metadata quality. A well-trained interpretable model can not only help developers choose more appropriate and reliable categories for each web service, but also help write a more reasonable service name and description. In this paper, we present xServeNet, an explainability-oriented extension of ServeNet for transparent web service classification. xServeNet preserves the BERT-based representation and CNN–BiLSTM feature extractor of ServeNet and introduces (i) an instance-wise dynamic source-fusion mechanism that adaptively combines service-name and service-description features according to their semantic contribution, and (ii) model-internal importance indicators at both the source and word levels that support inspection of classification decisions without introducing additional trainable parameters. We benchmark xServeNet against eleven machine learning baselines on two real-world ProgrammableWeb datasets of 10,943 and 14,086 services covering 50 categories. xServeNet reaches 71.08% Top-1/91.35% Top-5 accuracy on the original dataset and 74.10% Top-1/92.95% Top-5 accuracy on the updated dataset, consistently improving Top-1 accuracy over ServeNet while remaining competitive on Top-5, and achieving the lowest per-category Top-5 standard deviation among all twelve compared methods. In practice, the importance indicators support three concrete activities at the service registry: helping developers verify predicted categories at registration time, iterating on description wording when the predicted category looks wrong, and supporting registry curators in flagging likely mislabelled services for review. Full article

In UAV-based remote sensing, accurate and efficient image mosaicing is crucial for achieving real-time monitoring. Traditional cloud-centric processing paradigms, however, face core scientific challenges such as high latency, bandwidth bottlenecks, and limited autonomy, making them inadequate for dynamic, real-time scenarios. To address these issues, this paper proposes an edge-computing-enabled UAV image mosaicing system. The system consists of a UAV remote sensing platform and an edge computing terminal, with the core being our novel B-SIFT-ILS algorithm. The algorithm first uses geographic coordinates for unified registration, constructs a Gaussian scale space for multi-resolution representation, and then precisely locates extrema in the Difference of Gaussian (DoG) space using a 3D quadratic function. A BANSAC algorithm is subsequently employed to refine feature points and extract stable SIFT features, and finally, Iterative Least Squares (ILS) are used to achieve seamless mosaicing. Experimental results demonstrate that, compared with classical RANSAC, the proposed method achieves superior feature sampling accuracy (rotation: 0.879, translation: 0.877) and lower latency. The ILS-based smoothing stage effectively eliminates noise and ghosting without introducing gradient reversal, performing comparably to deep learning methods while significantly outperforming direct averaging and Gaussian approaches. On the NVIDIA Jetson Orin NX edge terminal, a single processing instance requires only 1124 ms, highlighting its strong potential for real-time, low-latency, and autonomous mosaicing tasks. Future research will focus on extending the approach to non-planar terrains and implementing adaptive parameter tuning for the BANSAC algorithm. Full article

This article examines the medium to long-term enrollment patterns of an AI-based platform designed to support children in learning robotics and participating in a national robotics competition in Romania. Drawing on registration and participation data covering students and teachers across urban and rural schools between 2020 and 2025, the study documents a consistent pattern: an initial period of high enrollment and rapid adoption followed by a steady decline over time. A key feature of the initiative is that hardware, platform access, and learning resources were provided entirely free of charge, allowing cost-related explanations for the decline to be set aside and structural and human factors to be examined directly. The paper makes two primary contributions. First, it proposes a System Dynamics framework grounded in innovation diffusion theory as a first-generation calibration model for understanding AI-enabled educational robotics adoption in a resource-constrained national context. The model is designed to be progressively tested and refined as anonymized aggregate data accumulates, and it relies exclusively on anonymized aggregated public data in accordance with GDPR requirements. Second, it advances the hypothesis that an AI-based educational platform, even one from which all financial barriers have been removed, will experience sustained enrollment decline in the absence of adequate human teacher involvement. The empirical trajectory and model outputs are consistent with this hypothesis and motivate further investigation. This represents a hypothesis-generating and framework-building paper. The framework reveals pronounced urban-rural disparities and differential outcomes by age of entry. All findings are presented as model-generated hypotheses rather than empirically demonstrated conclusions. The paper invites researchers gathering comparable data from similar initiatives in other countries to collaborate in testing and refining the model. The central conclusion is cautiously optimistic: AI may support robotics education adoption, but it is not a substitute for dedicated teachers, and without sustained investment in human capital, even a financially accessible platform is insufficient to maintain long-term enrollments. Full article

Background and Aim: The gluten-free diet (GFD) can have a huge impact on the quality of life of people with celiac disease (CD), especially on a social level. The objective of this work is to evaluate a structured nutrition education program focused on CD and GFD that aims to increase knowledge and improve inclusion attitudes about the disease in children. Methods: This is a one-month intervention for school children aged 10–12 years called Zeliakide (8 sessions). It was carried out through a STEAM methodology, using inquiry-based learning. The participants responses were evaluated through questionnaires before and after the intervention, and participants were also followed up one month later. The control group was a similar group of students who followed their regular school curriculum. Results: 299 children from one school of Vitoria-Gasteiz took part in the study (155 intervention group; 144 control group). Zeliakide significantly improved knowledge about CD and GFD in children, and this knowledge was retained for one month. Concretely, students increased their ability to explain what CD is, to assess gluten, and to classify food groups according to gluten content. The intervention contributed to augmenting the selection of behaviors to overcome differences between individuals, assessed one month after the intervention. In addition, the program allowed students to understand the work of scientists. Conclusions: Zeliakide can contribute to nutrition education initiatives that aim to improve knowledge of CD and GFD in the general population, while promoting empathetic behavior towards people with CD. Registration: clinicaltrials.gov, NCT05467865 on 21 July 2022. Full article

Background/Objectives: This study evaluated the daily delivered dose in prostate cancer patients using the automated artificial intelligence (AI)-based software Adaptbox (v2.3.2, Therapanacea). The aim was to assess target coverage and organ-at-risk (OAR) exposure. Methods: Twenty patients were included. All received 80 Gy in 40 fractions to the prostate and 56 Gy simultaneously to the seminal vesicles using two-arc VMAT on a TrueBeam STx, with daily CBCT for setup. For each fraction, CBCT images were imported into Adaptbox. A synthetic CT (sCT) was generated using a deep learning algorithm. OARs were automatically segmented, while targets were propagated from the planning CT (pCT) using rigid registration. Dose calculation was performed using Adaptbox’s collapse-cone algorithm. Dose parameters were extracted for each session and compared with planned values. Results: All 800 fractions were analyzed. The planning target volume (PTV) remained consistent with planning, with a maximum deviation of 0.1% for both PTVs. For the rectum, 78.38%, 77.75%, and 78.13% of fractions exceeded planned doses for V_70Gy, V_76Gy and V_80Gy, respectively. One patient had five consecutive fractions with >5% deviation across all rectal metrics. For the bladder, 52.34% of fractions exceeded the planned V_80Gy, and two patients had ≥5 consecutive fractions with >5% deviation; however, this was attributed to contouring inaccuracies. Conclusions: This AI-based workflow enables reliable daily dose reconstruction and can identify clinically relevant OAR dose deviations that may support adaptive interventions, although accurate contouring remains essential. Full article

Accurate LiDAR point cloud registration on resource-constrained edge platforms is a prerequisite for intelligent robotics and industrial automation, yet it remains challenging because low-overlap matching, false correspondences, and fine alignment must be handled under limited computing budgets without GPU acceleration. While learning-based methods have advanced the field, their heavy hardware dependency and training requirements often hinder their practical deployment on mobile edge devices. To bridge this gap, this paper proposes GeoRescue, a training-free geometric registration framework designed for high-precision perception under stringent hardware limits. The method consists of three modular stages: Asymmetric Correspondence Expansion (ACE), which enlarges the candidate correspondence set to reduce the loss of true matches; Dynamic Geometric Topology Gating (DGTG), which suppresses false matches through distance-consistency-based hypothesis filtering; and Uncertainty-Aware Manifold Refinement (UAMR), which improves fine alignment by explicitly modeling local anisotropic noise via covariance-guided optimization. Experiments on 3DMatch, 3DLoMatch, and KITTI show that GeoRescue achieves registration recall rates of 84.84% and 41.27%, respectively, and a 94.95% success rate on KITTI. Remarkably, the framework matches the accuracy of high-capacity learning models while running on a GPU-free, 15 W edge CPU platform (Intel Core i5-8265U). These results indicate that GeoRescue provides a deployment-ready solution with an optimal efficiency–accuracy trade-off for LiDAR sensing and robotics perception in complex, real-world scenarios. Full article

Accurate and efficient 3D reconstruction of trees is of paramount importance for studying forest spatial structures and dynamic resource patterns, optimizing forest management, protecting environments, and analyzing carbon cycles. Currently, Light Detection and Ranging (LiDAR) remains the dominant method for generating 3D models of forest scenes. However, with advancements in computer vision, photogrammetry has emerged as a crucial tool for forest inventory and 3D reconstruction due to its cost-effectiveness. Nevertheless, in practical forestry applications, traditional photogrammetry often suffers from low reconstruction efficiency and poor quality during feature extraction and matching. These issues stem from the complex structure of forest scenes, severe occlusion, and repetitive texture patterns. To address these challenges, this paper proposes an improved 3D tree reconstruction approach based on images, integrating deep learning-based methods. In the sparse reconstruction stage, we utilize the ALIKED (A LIghter Keypoint and descriptor Extraction network with Deformable transformation) algorithm and construct an image pyramid to extract multi-scale robust features. Furthermore, by combining the LightGlue matching algorithm with a neighborhood search constraint strategy, we enhance the stability of camera pose recovery while reducing redundant computations. Experimental results demonstrate that our method outperforms traditional algorithms in both accuracy and robustness regarding image matching. Compared to baseline models, the proposed approach increases the number of feature points by approximately 50% with a more widespread distribution, improves matching accuracy by 4% to 8%, and achieves a 100% image registration rate. Consequently, under the condition of maintaining equivalent re-projection errors, the subsequent sparse point clouds exhibit an average track length increase of 0.6 to 1.4 and a density increase of up to 1.2 times. Notably, this method effectively mitigates artifacts and spurious reconstructions caused by pose drift in forest photogrammetry. Full article