Search Results (734)

Efficient parking space management in urban settings represents a growing challenge owing to the sustained increase in the vehicle fleet. This study presents a comparative evaluation of five object detection architectures —YOLOv8s, YOLOv11s, YOLOv12s, RT-DETR-L and Faster R-CNN—applied to real-time intelligent vehicle occupancy monitoring under variable lighting conditions. The models were trained via transfer learning on a custom dataset of 1463 source images (21,944 annotated instances; expanded to 3511 files and 52,664 instances through offline augmentation of the training subset; three classes: free, occupied and unavailable) captured on a university campus located in Juliaca (Puno region), Peru, at 3824 m a.s.l. under daytime and nighttime clear-sky conditions from a single fixed-camera viewpoint. Each architecture was evaluated in ten independent experiments. Six dataset partitioning schemes of increasing strictness—a random control (R0) plus five leakage-controlled partitions—were evaluated. Under the strictest scheme (D3), simultaneously disjoint in acquisition date and camera viewpoint and therefore the most rigorous generalization estimate obtained in this study, accuracy ranges from mAP@0.5:0.95 of 0.9325 for Faster R-CNN to 0.8763 for YOLOv11s. Under the random partitioning conventionally applied to fixed-camera datasets, the same five architectures fell within 0.0055 of one another, all above 0.985, and their ranking was essentially inverted (Spearman $\rho =-0.80$). The differences in computational efficiency across architectures were statistically significant ($H=47.06$, $p<0.001$). YOLOv8s was the fastest of the four non-dominated architectures under the disjoint partition and was selected in 73.3% of weightings, although it ranked fourth in accuracy; its recommendation therefore rests on computational efficiency under a real-time constraint, whereas deployments that prioritize accuracy are better served by Faster R-CNN. The integrated system YOLOv8s + ByteTrack + FastAPI + Next.js 14 achieved per-slot accuracies of 87.5% and 91.8% under daytime and nighttime clear-sky conditions, respectively, using 1395 observations collected in a single university parking lot. For YOLOv8s, the transition from random to disjoint partitioning costs 0.1085 in mAP@0.5:0.95 (0.9913 to 0.8828), indicating that the near-saturated performance obtained under random partitioning substantially reflects the memorization of a fixed spatial configuration rather than generalization. The results support the feasibility of single-stage CNN architectures for intelligent parking monitoring in high-altitude Andean university environments under the evaluated acquisition conditions. Full article

By analyzing the semantic information of Direct Current (DC, the zeroth-order spherical harmonic coefficient) gradients during 3D Gaussian Splatting (3DGS) optimization, this paper achieves unsupervised state classification in scenes with discrete appearance states under the proposed State-Discovery Gaussian Splatting (SD-GS) framework via SVD dimensionality reduction and K-means clustering. To improve the stability of the clustering results, an appearance-difference-weighted refinement mechanism is further proposed to confirm high-confidence labels. To address the difficulty of distinguishing similar states when the number of states exceeds two, a sequential peeling strategy is proposed that decomposes a multi-class partition into several two-class separations. On four real-world scene datasets, SD-GS achieves 100% classification accuracy with reconstruction quality of 31.98–38.83 dB PSNR. Ablation studies validate the effectiveness of the gradient direction mode and the SVD dimensionality reduction strategy. Full article

We present ESNformer, a hybrid architecture that couples an Echo State Network (ESN) reservoir with a Transformer encoder for classification of structured, multi-indicator assessment data: a fixed-order vector of complementary indicators per assessment instance rather than a repeated-measures time series. The reservoir acts as a fixed nonlinear feature map over the indicator vector, while self-attention, made position-aware over the fixed column order, learns how each indicator’s evidence contributes to the final decision, so the two components, together, capture local, indicator-level detail and global, cross-indicator interactions within a single, end-to-end trainable model. Interpretability is treated as a first-class design requirement rather than an afterthought: the architecture is paired with an explainability layer combining SHAP feature attribution (reported both globally and per class), the model’s own attention weights, a deletion/insertion faithfulness test that quantitatively verifies which inputs the model actually relies on, and counterfactual maps that translate a prediction into an actionable, inspectable recommendation. We evaluate the architecture on a concrete case study, classifying Braille-literacy instructional recommendations from 15 pedagogical indicators grouped into three categories (Mangold’s, ABKL, and Progresar), using a benchmark of 900 real assessment instances (630 used, together with a class-conditional augmentation procedure, to build a 2100-instance training set) with validation and test partitions (135 instances each) kept exclusively real. On this benchmark, the tuned model reached 85.33% accuracy, 85.90% macro-precision, 85.33% macro-recall, an F1 score of 85.25%, and an AUC of 0.95 on the real test set. SHAP attribution, attention weights, and the faithfulness test converge on the same two dominant indicators (response time and error count): removing them alone collapses accuracy to chance, while retaining only them recovers most of the model’s accuracy. We report this transparently alongside a comparison against ESN-only, Transformer-only, and tabular baselines (logistic regression, decision tree, random forest, XGBoost, and an MLP) on the same data and discuss what the hybrid architecture and its explainability pipeline add beyond what the two dominant indicators already explain and how the approach generalizes to other tabular and mixed-granularity assessment settings that require both predictive accuracy and a verifiable account of what drove each decision. Full article

Soybean production commonly faces challenges including severe pod abscission, incomplete seed filling, and uncoordinated source–sink relationships that limit yield potential. Plant growth regulator application may influence source–sink relationships and thereby affect soybean (Glycine max (L.) Merr.) yield formation. This three-year field study (2023–2025) evaluated the effects of four spray programs on soybean yield, quality, total above-ground biomass partitioning, and endogenous hormone dynamics under Xinjiang’s irrigated production conditions. The four treatments were an untreated control (CK), naphthaleneacetic acid alone (NAA; 300 g ha⁻¹), naphthaleneacetic acid plus prohexadione-calcium (NPC; 300 + 450 g ha⁻¹), and naphthaleneacetic acid plus prohexadione-calcium and iron chlorin e6 (NCE; 300 + 450 + 45 g ha⁻¹). The spray programs were applied at the fourth-trifoliolate and full-pod stages. Results showed that NCE treatment consistently produced the greatest yield increases (13.1–14.4%) compared with the control. This response was associated with greater middle-node pod retention, increased 100-seed weight (3.6–6.4%), and improved reproductive organ biomass allocation (44.7–53.3% at maturity vs. 40.9–45.7% in controls). NCE significantly elevated leaf trans-zeatin content (37.1–91.9%) within 24 h after application and improved seed protein concentration by up to 11.2%, while seed residues of all applied compounds remained well below safety thresholds (<0.05 mg kg⁻¹). Correlation analysis revealed strong positive relationships between trans-zeatin levels and both reproductive biomass allocation (r = 0.74–0.90, p < 0.01) and grain yield (r = 0.86–0.96, p < 0.01). These findings indicate that NCE was the best-performing of the four tested spray programs under Xinjiang’s irrigated production conditions, although the individual contributions and possible interactions of the three compounds require further factorial evaluation. Full article

Residual creases generated during the leather spreading process exhibit highly variable morphologies and irregular feature distributions, causing significant challenges for feature extraction and leading to low localization accuracy. To address these issues, this paper proposes the Dynamic, Cross-Calibrated, and Edge-Enhanced YOLOv8n-Pose (DCE-YOLOv8n-Pose) algorithm. Instead of providing regional approximations, this framework outputs precise spatial coordinates for robotic grasping by integrating three synergistic components in a progressive network flow. First, dynamic snake convolution adaptively perceives the continuous geometric features of elongated creases; subsequently, an efficient multi-scale attention mechanism provides cross-dimensional weight calibration to suppress highly homochromatic background interference and correct spatial misalignments; finally, an edge-enhanced content-aware reassembly of features module preserves high-frequency gradients and prevents feature fracturing during multi-scale fusion. For comprehensive evaluation, a dataset comprising 700 original laboratory images was constructed. To prevent data leakage, the dataset was partitioned into training and validation sets based on individual leather specimens, ensuring that images of the same leather piece do not appear in both sets. Additionally, an independent test set of 500 images collected from an actual processing plant was designed for industrial validation. Experimental results indicate that, at an Intersection over Union (IoU) threshold of 0.5, the DCE-YOLOv8n-Pose model achieves a bounding box mean average precision (mAP@0.5) of 91.8% and a keypoint mAP@0.5 of 85.1%, with a keypoint precision of 87.9%. The computational load is maintained at 9.2 GFLOPs, alongside an inference speed of 114.3 FPS. Furthermore, consistent convergence across four independent training runs substantiates the model’s reliability in reducing missed detection rates and localization deviations. In conclusion, the proposed algorithm demonstrates practical applicability for the visual guidance of automated leather spreading equipment by balancing detection precision and inference speed, thereby offering an effective coordinate reference for subsequent robotic stretching operations. Full article

This paper develops a Banach space-valued approximation framework based on symmetrized neural network (SNN) operators generated by a deformation-dependent sigmoidal activation function. Symmetry is introduced directly at the activation level through a reciprocal-deformation mechanism, yielding a positive, even, normalized, and localized density kernel satisfying the partition of unity. The resulting construction provides normalized compact-interval and whole-line quasi-interpolation operators for Banach space-valued functions. Quantitative pointwise and uniform convergence estimates are established through the first modulus of continuity and are extended to higher-order and Caputo–Bochner fractional approximation. Numerical diagnostics support the theoretical kernel properties, and fractional approximation experiments compare the SNN and classical NN operators under common computational conditions. A controlled blind-prediction experiment on a synthetic monthly temperature-like series uses a strict fit–validation–test protocol and a parameter-matched operator comparison, with seasonal ARIMA and MLP models as external baselines. Across five independent realizations, the SNN attains the best mean predictive performance, with $R^2=0.9500$, NMAE $=0.0452$, and NRMSE $=0.0570$. A vector-valued experiment in $Y={\mathbb{R}}^2$ further illustrates the non-scalar applicability of the Banach space framework. In addition, the normalized SNN kernel weights provide an intrinsic node-level interpretation mechanism without requiring an external post hoc explainability method. Full article

The widespread adoption of Location-Based Services (LBSs) has significantly enhanced daily convenience, yet the upload and storage of user trajectory data pose substantial privacy leakage risks. Addressing the limitations of existing privacy protection methods in achieving personalized adaptation and balancing privacy with utility, this paper proposes a personalized privacy protection strategy for location trajectories based on weighted Kullback–Leibler (KL) divergence. The approach first employs a Markov transition matrix to model user movement patterns, utilizes quadtree-based dynamic grid partitioning for adaptive encoding of the state space, and introduces sensitivity scores weighted by dwell duration and visit frequency to identify critical privacy-sensitive points. It then develops an exponential decay perturbation mechanism combining regularization parameters and distortion thresholds to preserve trajectory spatial usability while protecting sensitive transitions. By quantifying privacy leakage through weighted KL divergence and measuring data utility via distortion metrics, a linearly weighted composite index is constructed, enabling personalized parameter optimization via grid search. Experimental results on the real-world Geolife dataset demonstrate that compared to three differential privacy baselines, this method reduces privacy leakage (measured by weighted KL divergence), improves POI Recall rates, and decreases average geographic errors. Paired t-tests confirm that all improvements are statistically significant (p < 0.001) with large effect sizes, validating its effectiveness and superiority in balancing privacy protection and data usability. Full article

Fake-speech detection is commonly studied through acoustic artifacts, speaker-level spoofing cues, or visual inconsistencies, while less attention has been given to meaning-level manipulation, where the spoken content is altered while the speech remains natural and speaker-consistent. This study investigates the relative and complementary contributions of interpretable speech-centered features for detecting meaning-level fake speech. Specifically, it examines transcript-level linguistic and psycholinguistic-style features, text–audio emotion-alignment features, and rhythm-based audio descriptors. A feature-aware gated-fusion framework is used to analyze and combine these three feature groups, with separate branches encoding each feature type and learned branch-level weights adaptively controlling their contributions to binary classification. The framework was evaluated on FakeSpeech+, an audio-only dataset designed for meaning-level manipulation, using a strict leakage-controlled repeated-seed protocol that prevents source-pair, filepath, exact-transcript, and combined group overlap across training, validation, and test partitions. The gated-fusion model achieved 0.842 accuracy, a 0.840 F1-score, and 0.919 AUC. Analysis of the learned fusion weights indicated that transcript-level features contributed most strongly, followed by text–audio emotion-alignment features, while rhythm features received the lowest contribution. These findings provide evidence that meaning-level fake-speech detection can benefit from jointly examining linguistic content, emotional alignment, and rhythmic characteristics, while also highlighting differences in the relative contributions of these feature groups. Full article

To address the broadband noise distribution, complex excitation sources, and insufficient low-frequency attenuation of conventional porous acoustic packages in commercial vehicle cabs, this study proposes a PU–six-cavity Helmholtz metamaterial composite acoustic package for broadband noise reduction. The proposed structure consists of a 30 mm PU porous layer for mid-to-high-frequency dissipation and a 30 mm six-cavity Helmholtz metamaterial layer for low-frequency absorption, forming a 60 mm composite acoustic package. A full-vehicle acoustic model of a commercial vehicle cab was established in VA One to identify the A-weighted sound pressure level (SPL) spectrum at the driver position. The results show that the PU porous acoustic package improves the mid- and high-frequency noise response, whereas pronounced peaks remain in the low- and low-to-mid-frequency ranges. To enhance these bands, the six sub-cavities of the Helmholtz metamaterial were tuned to 200, 250, 315, 400, 500, and 630 Hz through spatial partitioning and cavity grouping. COMSOL (version 6.1) simulations and particle velocity distributions confirmed the multi-peak absorption mechanism and the selective excitation of the corresponding sub-cavities. The composite structure was further validated through impedance-tube measurements, full-vehicle acoustic simulations, and in-vehicle tests. The VA One simulation shows that the total A-weighted SPL at the driver position decreases from 68.25 dB for the PU porous package to 66.01 dB after introducing the six-cavity Helmholtz metamaterial, corresponding to an additional reduction of 2.24 dB. A preliminary in-vehicle test under a stationary idling condition shows that the total A-weighted SPL near the driver’s ear decreases from 55.83 dB(A) to 54.56 dB(A). These results demonstrate that the proposed PU–six-cavity Helmholtz metamaterial composite acoustic package combines broadband porous dissipation with low-frequency resonant absorption, providing a feasible solution for broadband noise control in commercial vehicle cabs. Full article

Background: Accurate identification and classification of white blood cell (WBC) subtypes in peripheral blood smears (PBS) is essential for the diagnosis and monitoring of hematological malignancies, including leukemia. Conventional manual microscopy, although clinically established, is labor-intensive and subject to inter- and intra-observer variability. Deep learning-based object detection offers a route to automation, yet most prior studies are limited by small datasets, restricted cell taxonomies, or single-microscope acquisition. This study evaluates a YOLOv11-large (YOLOv11L) detector for simultaneous localization and classification of 13 leukemia-relevant WBC subtypes plus an artifact class (14 classes total), trained on the large-scale, multi-domain, open-source LeukemiaAttri dataset. Methods: From the LeukemiaAttri dataset, 18,664 annotated images (67,347 objects) acquired at 40× and 100× magnification were partitioned by stratified sampling into training (70%), validation (15%), and test (15%) sets. The training set was expanded to 65,785 images through extensive geometric, photometric, and AugMix augmentation. A YOLOv11L model pretrained on MS COCO was fine-tuned for 250 epochs (640 × 640 input) on a single NVIDIA H200 SXM GPU, using an auto-selected optimizer (momentum 0.9; weight decay 5 × 10⁻⁴), automatic mixed precision (AMP), and mosaic augmentation for the first 240 epochs. Results: On an internal held-out test set, the model achieved an mAP50 of 93.9%, mAP50-95 of 77.9%, precision of 94.1%, recall of 88.8%, and an F1 score of 0.913, with similar performance in the validation and test sets. Class-wise average precision (AP) ranged from 89.3% (monocyte) to 98.2% (monoblast), confirming consistent detection across morphologically diverse subtypes. Conclusions: The YOLOv11L detector achieved high performance across all 14 categories on the internal test set, with metrics exceeding those previously reported for subset-specific baselines. These findings support further evaluation of the model as a decision-support tool for peripheral blood smear analysis. External validation is required to determine its clinical utility and generalizability. Full article

This paper develops a unified asymptotic theory for inverse-probability-weighted conditional U-statistics of arbitrary fixed order in the presence of missing-at-random responses and infinite-dimensional functional covariates. The target is a conditional higher-order functional generated by a measurable response kernel and evaluated locally on a separable Banach space. Localization is formulated through delta sequences, providing a common framework for kernel, partition, regressogram, orthogonal series, and related smoothing procedures without recourse to finite-dimensional density arguments. For bounded kernels, we establish uniform almost-complete convergence over pseudo-compact functional domains and obtain a sharp decomposition into deterministic localization bias and stochastic fluctuation. The latter is governed by the localized-kernel variance, the envelope of the delta sequence, the metric complexity of the indexing domain, and the small-ball concentration of the functional covariate. Unbounded kernels are treated under explicit weighted moment, truncation, and summability conditions. The feasible theory quantifies the additional perturbation induced by estimating the propensity score and identifies conditions under which this first-stage uncertainty is asymptotically negligible. Pointwise distributional theory is derived through a denominator linearization combined with the Hoeffding decomposition of the centered localized kernel. The Gaussian limit is driven by the first projection, while the higher-order canonical components are shown to be negligible under explicit local-mass, moment, and noncancellation assumptions. This yields oracle-equivalent feasible inference, a consistent first-projection variance estimator, and asymptotically valid studentized confidence intervals. A finite-grid adaptive comparison principle is also developed for data-driven resolution selection. The scope of the theory is illustrated through conditional rank functionals, discrimination with incomplete labels, metric-learning criteria, and functional prediction. Synthetic and semi-synthetic studies based on functional classification, phoneme log-periodograms, and growth trajectories document the finite-sample interaction between covariate-dependent label observation, local information loss, propensity estimation, and inverse-weighting variance. Full article

Stochastic metaheuristic feature selection (FS) may generate unstable feature subsets across repeated runs, undermining reproducibility in Internet of Things (IoT) intrusion detection systems (IDSs). This paper presents Head–Tail Population Initialization (HTPI), a feature-importance-guided initialization approach for enhancing the stability of Adaptive Hybrid Genetic Algorithm–Simulated Annealing (AHGA-SA)-based FS. HTPI uses feature-importance scores to split candidate features into Head and Tail groups and initializes candidate subsets by prioritizing Head features and sampling Tail features with importance-based weights. HTPI is integrated into AHGA-SA as an incremental extension, termed HTPI-AHGA-SA, and modifies only the initialization and reinitialization steps. Experiments on eight IoT-oriented IDS datasets using 50 runs per configuration, with seeds paired across methods, showed significantly higher Nogueira stability under HTPI-AHGA-SA on all datasets after Holm correction, with non-overlapping 95% leave-one-run-out jackknife confidence intervals in every case. These results characterize algorithmic cross-run stability under a fixed data partition. All absolute differences in dataset-level mean F1 Macro remained below $0.003$; formal equivalence at this margin was supported for six datasets, while dataset-specific security-metric trade-offs remained. On three representative datasets, post hoc explainable artificial intelligence (XAI) analyses indicated generally higher permutation importance (PI)-based cross-run consistency and measurable predictive utility in the selected Head and Tail portions under retraining. Full article

Hyperspectral unmixing (HU) requires effective modeling of spectral–spatial information and local–global feature interactions to achieve accurate abundance estimation and endmember extraction. Although Transformer-based HU methods are effective in capturing long-range dependencies, they often neglect the intrinsic spectral consistency of hyperspectral data and do not fully exploit the global spectral–spatial correlations in hyperspectral image cubes. To address these issues, this paper proposes a Spectral Consistency–guided Transformer with Volumetric Linear Self-Attention (SCVolFormer) for hyperspectral unmixing. A Spectral Consistency Block (SCB) is introduced to preserve consistency across adjacent spectral bands and produce physically meaningful feature representations. A spectral grouping strategy is further adopted to partition the high-dimensional spectrum into locally continuous subspaces, reducing computational cost. In addition, a shared-weight Transformer encoder with Volumetric Linear Self-Attention (VolLSA) is designed to model interactions between the spectral and spatial dimensions and capture long-range dependencies within hyperspectral image cubes. A decoder is then used to estimate abundance maps and reconstruct hyperspectral images. Experiments on one synthetic dataset and three real hyperspectral datasets demonstrate that SCVolFormer outperforms state-of-the-art methods in abundance estimation and endmember extraction, confirming the effectiveness of spectral consistency guidance and volumetric attention modeling for hyperspectral unmixing. Full article

Ovarian cancer remains a leading cause of gynaecological cancer mortality, and ultrasound-based deep learning systems for its diagnosis are typically built as separate post hoc processes for classification, segmentation, and interpretability, which introduces workflow inefficiencies and may produce inconsistent predictions. This work addresses that limitation. We propose UM-TOTA (Unified Multi-Task Ovarian Tumour Architecture), a Vision Transformer (ViT)-based architecture that performs eight-class tumour classification, three-class malignancy detection, tumour segmentation, and clinical concept interpretability within a single unified framework. We integrate a concept bottleneck guided by the IOTA and O-RADS clinical guidelines to enable transparent decision-making through medical concepts that clinicians can understand, and we employ combined adaptive t-vMF Dice and boundary-enhanced segmentation losses with progressive task weighting to stabilise multi-task optimisation. We evaluated the model on the Multi-Modality Ovarian Tumor Ultrasound (MMOTU) 2D dataset under two protocols: image-level 5-fold stratified cross-validation, and the patient-disjoint partition released with the dataset. Under cross-validation, UM-TOTA achieved 80.26% ± 1.10% accuracy (97.06% one-vs-rest macro specificity) for eight-class classification, 90.88% ± 1.14% accuracy (90.41% specificity) for malignancy detection, and 77.29% ± 1.29% Dice for segmentation. Under the patient-disjoint partition, which excludes any overlap of patients between training and testing, the corresponding values were 78.46%, 89.13%, and 75.41%, a reduction of under 2.2 percentage points on every metric. The UM-TOTA reduced the computational parameter load by approximately 65.1% relative to sequential single-task pipelines. The learned concepts aligned with established malignancy criteria, identifying vascularisation, solid components, and papillary projections as key predictors. This unified approach offers an efficient and interpretable framework for clinical ovarian ultrasound workflows. Full article

Urban green-space provision depends not only on green quantity but also on how greenery is positioned relative to buildings as urban form evolves. However, most existing assessments aggregate greenery within predefined spatial units and rely on static snapshots, thereby obscuring both its distance-sensitive distribution around buildings and its variation across urban-development trajectories. This study develops a Supply–Form framework to examine this relationship across heterogeneous development trajectories. Using multi-temporal very-high-resolution imagery from Beijing and Tianjin, we introduce Quantified Green-Space Supply (QGS), a distance-weighted indicator that accumulates greenery around buildings over nested ranges of 5–2000 m. Building-expansion trajectories are then linked to evolution-stratified XGBoost models interpreted with consensus-grouped Partition SHAP, allowing recurrent morphological groups and their within-group contributions to be compared across contexts. QGS was more strongly associated with the selected urban-form variables than conventional green coverage ($R^2=0.717$ versus $0.633$). Both cities showed higher mean QGS in 2022 than in 2012, but their internal patterns diverged: Beijing developed more continuous central low-QGS clusters, whereas Tianjin’s central cold spots contracted and fragmented. Across 15 trajectory-specific models, building density was the leading morphological group in 12, while the importance of patch size, edge structure, aggregation–adjacency, and spatial configuration varied by trajectory. Nonlinear responses showed a stable positive association for building density, diminishing returns for edge density and dispersion, adverse effects at very high adjacency, and intermediate-range benefits for patch dominance and compactness. The framework advances green-space assessment from aggregate greenness mapping to a relational, trajectory-conditioned diagnosis of potential surrounding green-space supply, providing a basis for context-sensitive urban renewal. Full article