Search Results (859)

High-resolution remote-sensing semantic segmentation requires models to simultaneously capture global scene semantics and preserve fine-grained local structures. Although satellite-pretrained vision foundation models provide strong transferable representations, the features extracted by a frozen backbone remain insufficiently adapted to dense prediction, particularly for representing high-frequency details and multiscale local patterns. In addition, correcting residual prediction errors with dense full-map refinement introduces substantial computational redundancy, since hard errors are typically concentrated in only a small subset of locations. To address these challenges, we propose ADVMSeg, an efficient remote-sensing semantic segmentation framework built upon a frozen satellite-pretrained DINOv3 backbone. Specifically, we introduce a Spatial-Frequency Adapter (SF-Adapter) to improve backbone-level dense feature adaptation by jointly modeling global frequency responses and multiscale local spatial details in a lightweight bottleneck space. We further design an Adaptive Sparse Refinement (ASR) module after the pixel decoder, which identifies hard regions from coarse predictions via uncertainty and boundary cues, and performs targeted local cross-attention refinement only on selected critical locations. Extensive experiments on GID-15, LoveDA, and ISPRS Potsdam validate the effectiveness of the proposed framework. Under the unified setting, ADVMSeg achieves 63.1% mIoU on GID-15, 63.5% mIoU on LoveDA, and 81.4% mIoU on ISPRS Potsdam. These results validate the effectiveness of jointly improving backbone-level feature adaptation and prediction-stage computation allocation under the evaluated setting of frozen DINOv3, and three representative remote-sensing semantic-segmentation datasets. Full article

Impulsive alpha (α)-stable noise, characterized by heavy tails and intense outliers, is a key ingredient in simulating financial, medical, seismic, and digital communication technologies. It poses versatile challenges to conventional machine learning (ML) algorithms in predicting noise parameters for multidisciplinary artificial intelligence (AI)-embedded devices. In this study, we adopted a two-phase methodology to investigate the complexity and performance of supervised ML algorithms while classifying impulsive noise parameters. We generated synthetic datasets of α-stable noise distributions for experimentation in a controlled environment. It was followed by experimental evaluation to derive the complexity and performance of ML classifiers—k-nearest neighbors (KNN), Support Vector Machine (SVM), Naïve Bayes (NB), Decision Tree (DT), and Random Forest (RF). Moreover, we employed a very high channel noise level of −15 dB in the test datasets to ensure that the derived analysis applies to real-world devices. The results demonstrate the high performance of DT and RF in structured binary classification of the α regime and the sign of skewness, while incurring satisfactory computational costs. However, SVM and kNN are comparatively more robust for multi-class classification, albeit with higher memory and training costs. On the contrary, NB fails to address the skewed and impulsive behavior of α-stable noise. We observed that even the most effective classifiers struggle to achieve perfect accuracy in multi-class classification. Overall, the experimental results reveal significant trade-off relationships between the complexity and performance of ML classifiers. Conclusively, simple models are well-suited for coarse-grained tasks, such as α-approximation and sign-of-skewness classification. In contrast, sophisticated models can be deployed to predict noise parameters to some extent. Our study provides a clear set of trade-offs for future applied AI devices that address adversarial and impulsive noise. Full article

3D multi-object tracking (MOT) for autonomous driving remains challenging due to frequent identity switches in crowded scenes, trajectory fragmentation during occlusions, and the difficulty of adapting association strategies to varying scene complexities. While existing methods rely on fixed geometric or appearance-based associations, they struggle to handle ambiguous cases and detection failures. We present an adaptive multi-level 3D MOT framework that achieves robust tracking through three key innovations: (1) multi-granularity temporal modeling that captures both fine-grained short-term motion and coarse long-term trends via dual-scale spatio-temporal attention, enabling accurate motion prediction across different object dynamics; (2) Transformer-based Appearance Association that employs cross-attention to model global inter-object relationships, resolving ambiguous associations in crowded scenarios where geometric cues alone fail; and (3) scene-adaptive learned thresholds that automatically adjust association strictness based on object density, motion complexity, and occlusion levels, avoiding the one-size-fits-all limitations of fixed thresholds. Our hierarchical four-level tracking strategy progressively handles cases from easy geometric matching (Level 1) to complex interval-frame recovery (Level 4), with SOT-based virtual detection generation bridging detector failures. Extensive experiments on the nuScenes benchmark demonstrate state-of-the-art performance. Full article

Personality assessment serves as a key building block in intelligent information systems that enable human-centered modeling. Unlike cognitive tests, personality assessments rely primarily on self-reports and are therefore susceptible to faking. Forced-choice (FC) formats partially mitigate this problem, yet socially desirable responding remains a systematic source of bias. Traditional approaches rely on expert-annotated social desirability (SD) ratings to construct FC item blocks and infer respondents’ personality traits from block-level rankings. This rating procedure is labor-intensive and coarse-grained. Furthermore, existing methods neglect the non-linear SD interactions between respondents and items, which act as structured adversarial noise that hinders the recovery of true latent traits. To address these challenges, we propose the Social Desirability-aware Forced-Choice Diagnosis (SDFCD) approach. Our approach adopts a knowledge-guided learning paradigm by leveraging large language models (LLMs) to distill fine-grained, continuous SD ratings, thereby replacing sparse expert ratings. We then introduce a decoupled neural interaction module that jointly represents latent personality traits and SD tendencies, enabling the modeling of respondent–item SD interactions. Experiments on real assessment data demonstrate that our method significantly outperforms baseline FC models in personality trait diagnostic performance and model interpretability. This study highlights the potential of LLMs for automated, fine-grained SD quantification and offers a scalable path toward more trustworthy personality assessment. Full article

Creep is one of the main failure mechanisms of materials at elevated temperatures, and the creep rate curve is a key descriptor of creep deformation and damage evolution. However, existing creep models are mainly phenomenological or stage-wise, and the physical origin of the bathtub-shaped creep rate curve over the full creep process has not been systematically clarified. In this study, creep damage is treated as an aging failure process of a material system, and a physically interpretable hierarchical model is established based on statistical physics for disordered complex systems. By linking the evolution and interaction of microscopic material units with macroscopic creep behavior, the proposed model provides a unified description of the primary, secondary, and tertiary creep stages and offers a theoretical explanation for the bathtub-shaped creep rate curve. Validation using representative metallic and composite material cases shows that the model can reasonably reproduce the overall three-stage creep rate evolution, with residual sums of squares of 1.3088 and 0.5369, respectively. These results demonstrate the ability of the model to capture full-process creep behavior in different material systems. The main advantage of the proposed approach is its physical interpretability within a unified framework, while its current limitation is that the validation remains limited in scale and broader benchmark comparisons with conventional methods are still needed. This work provides a statistical perspective for creep behavior modeling and for understanding the microscopic mechanisms and interactions underlying creep degradation in structural materials. Full article

This paper presents a multimodal sensing approach for fine-grained soccer action recognition using synchronized mm-wave FMCW radar and multiview RGB cameras. A TI IWR1443BOOST FMCW radar and three Sony IMX296 global-shutter cameras were used to record seven soccer-related actions in different movement directions in an outdoor environment. Range–Doppler radar processing is applied to extract global mel features and CFAR-localized block representations of mel and radar spectrogram features to capture both coarse and fine micro-Doppler characteristics. Camera features are derived from bounding box, HOG, optical flow, and pose estimations. Classification is performed using logistic regression as the classical model and various deep models. Performance is evaluated using cross-validation. Radar alone achieved moderate performance (0.897 F1_macro using TCN), successfully identifying coarse motion but showing limited separability for dribbling-based actions. Camera-only models achieve near-perfect accuracy (≥0.997 F1_macro using 1D-CNN), with the confusion matrices being nearly perfectly diagonal already. The best performance is obtained from a cross-modal transformer with multiple cameras (0.998 F1_macro). These results demonstrate that a camera by itself performs strongly for the action recognition task but also that radar–camera fusion can improve robustness and enhance the discrimination of finer soccer player movements for outdoor analytics and player monitoring applications. Full article

Coastal reclaimed areas are characterized by complex strata and high groundwater levels, and pile foundations in such areas often suffer from insufficient uplift resistance. Compared with conventional cast-in-place piles, squeezed branch piles exhibit superior uplift performance; however, studies on squeezed branch piles in reclaimed areas remain limited. To investigate the uplift bearing performance of squeezed branch piles in the complex strata of coastal reclaimed areas, in situ full-scale uplift tests were conducted in the Shenzhen Binhai Avenue (Headquarters Base Section) traffic reconstruction project. Based on the actual physical and mechanical properties of the soil strata, a three-dimensional numerical model was established and validated against the load–displacement curves obtained from the in situ full-scale uplift tests. On this basis, the uplift bearing performance of squeezed branch piles, the differences in uplift bearing performance between branch and plate structures, and their applicable strata were analyzed. The plate structure and different branch configurations of squeezed branch piles exhibit distinct symmetric configuration characteristics, and these configuration differences influence the overall uplift bearing performance. The results show that the load–displacement curves of the uplift piles are generally smooth, without obvious abrupt rises or drops, exhibiting a gradual variation pattern, and the maximum pile-head displacements are all less than 100 mm. The mobilization of the bearing capacity of the branch and plate structures exhibits a distinct temporal and sequential pattern, with the plate structures at shallower embedment depths mobilized earlier than those at greater depths. Compared with conventional cast-in-place pile foundations, the presence of branches and plates endows squeezed branch piles with better elastic mechanical behavior and higher rebound ratios during unloading. Under identical stratum and loading conditions, the uplift bearing performance of the plate is 133% higher than that of the six-radial-branch configuration, while that of the six-radial-branch configuration is 34% higher than that of the four-radial-branch configuration. It is recommended to adopt the six-radial-branch configuration in clayey sandy gravel strata and the plate configuration in gravelly clayey soil and completely weathered coarse-grained granite strata, whereas neither branches nor plates are recommended in soil-like strongly weathered coarse-grained granite strata. Full article

Molecular engineering has traditionally followed a structure–function paradigm based on well-defined, folded architectures. However, intrinsically disordered proteins and regions (IDPs/IDRs) reveal that nature also exploits disorder as a functional design strategy. Here, we argue that intrinsic disorder can be understood as a biomimetic design principle for molecular and materials engineering. From a soft matter perspective, IDRs function through statistical ensembles, weak multivalent interactions, and collective behavior rather than fixed structure, with sequence features encoding a molecular grammar that governs phase behavior, viscoelasticity, and responsiveness. These principles closely parallel those found in associative polymers and colloidal systems. Recent advances in coarse-grained modeling, machine learning, and inverse design further enable disorder to be treated as a controllable engineering variable. By reframing intrinsic disorder as a programmable and bioinspired design strategy, this Perspective highlights its potential for the development of adaptive and responsive biomimetic materials. Full article

The critical brain hypothesis proposes that neural systems operate near a phase transition to optimize information processing. A key method for investigating this hypothesis is the phenomenological renormalization group (pRG), which looks for scale-invariant features across levels of coarse-graining. One such feature is the power-law scaling of eigenvalues of covariance matrices of coarse-grained variables. However, the estimation of this scaling exponent, $\mu$, often relies on linear regression over arbitrarily selected ranges of the plot of eigenvalues versus rank. This heuristic “eyeballing” introduces uncontrolled bias and complicates the interpretation of observed scaling relationships. In order to obtain a more robust estimation of $\mu$, we do not fit the standard eigenvalue-vs-rank relationship, but rather the density of eigenvalues, for which standard protocols exist for fitting power laws to empirical data distributions. We demonstrate this approach using a synthetic model that replicates the scaling signatures of neural data while providing control over the system’s exponents as well as neural data obtained from publicly available Neuropixels recordings. We also establish standards for the minimal data required to quantify power-law behavior in a pRG eigenvalue analysis. Our approach contributes a tool for understanding the fundamental limitations imposed by spatial and temporal constraints of experimental datasets, which is required to rigorously assess the neural criticality hypothesis. Full article

The dynamic modulus of asphalt mixtures is a key design parameter in pavement design, which significantly impacts the mechanical properties of asphalt pavements. This study simulated dynamic modulus tests of asphalt mixtures using the three-dimensional (3D) discrete element method (DEM) to investigate mechanical behaviors such as the loading-bearing ratio of individual aggregates. Fine-grained AC-13 and medium-grained AC-20 asphalt mixture models were randomly constructed in the DEM program using user-defined methods. The dynamic modulus and phase angle values of the asphalt mixtures were predicted. By comparing laboratory experiments with DEM simulation results, the model was validated, and the effects of temperature and loading frequency on the dynamic modulus were explored. Further exploration was conducted on the loading-bearing ratio and mechanical interactions among aggregates of different sizes within the mixtures. The results show that the 3D DEM model can accurately predict the dynamic modulus and phase angle of asphalt mixtures. Temperature and frequency have an impact on these parameters, and the increase in gradation has an impact on the loading-bearing ratio, due to the proportion of coarse aggregates. Full article

Conventional casting of 5N5 high-purity aluminum often results in coarse grains, microstructural inhomogeneity, and a low equiaxed grain area fraction. Vacuum casting in a graphite mold was integrated with multidirectional mechanical vibration to refine and homogenize the solidification microstructure. A three-factor, three-level Box–Behnken design combined with response surface methodology was employed to optimize pouring temperature (A), mold temperature (B), and vibration frequency (C), with the average grain size (Y₁) minimized and the average shape factor (Y₂) and equiaxed grain area fraction (Y₃) maximized. Analysis of variance indicated statistically significant quadratic models with a non-significant lack of fit. The predicted optimum (A ≈ 714 °C, B ≈ 363 °C, C ≈ 37 Hz) was validated experimentally, producing a refined and highly equiaxed structure (Y₁ ≈ 0.85 ± 0.02 mm, Y₂ ≈ 0.84 ± 0.04, Y₃ ≈ 88.6 ± 2.11%), consistent with model predictions. Multidirectional vibration strengthens melt convection and interfacial shear, which is considered to promote grain multiplication and increase the number of effective nuclei, thereby accelerating the columnar-to-equiaxed transition and improving microstructural uniformity. Full article

Encrypted traffic classification is essential for network management and security, yet payload inspection is ineffective under modern protocols such as Transport Layer Security (TLS) and Quick UDP Internet Connections (QUIC). Existing metadata-based methods perform well for coarse-grained tasks but often fail to distinguish structurally similar applications because they model temporal behavior only implicitly or coarsely. We propose the Bi-Directional Directed Temporal Graph (BiDT), a framework based on a Directed Temporal Interaction Graph (DTIG) and a Bi-Directional GraphSAGE (BiGraphSAGE). The DTIG represents packets as nodes and explicitly encodes inter-arrival times (IATs) as directed edge attributes, preserving both causal structure and communication rhythm. The BiGraphSAGE then aggregates temporal interaction features from forward and backward perspectives. We evaluated the BiDT on the VNAT benchmark and validated it on ISCX-VPN. On the challenging 10-class VNAT dataset, the BiDT achieves 98.57% accuracy and outperforms strong baselines, including complete separation of easily confused protocols such as SCP and SFTP. The results on ISCX-VPN further confirm the effectiveness of the proposed design. These findings show that explicit temporal edge modeling is effective for fine-grained encrypted traffic classification. Full article

A comprehensive analytical framework has been developed for the spatio-temporal forecasting of spring wheat yield in risk-prone rainfed agricultural zones. The study is grounded in 25-year time series integrating remote sensing data, meteorological reanalysis products, and soil parameters. The implementation of the XGBoost algorithm enabled the modeling of complex nonlinear biophysical relationships. To account for spatial autocorrelation and Tobler’s First Law of Geography, a two-level validation strategy was employed. The interpolation performance achieved an accuracy of R² = 0.69 (RMSE = 0.33 t/ha), while extrapolation to unseen regions yielded R² = 0.65 (RMSE = 0.35 t/ha), demonstrating the robustness and transferability of the proposed architecture. Application of the TreeSHAP interpretability framework revealed the dominant influence of agroclimatic drivers, highlighting the critical role of April soil moisture recharge and the significance of winter insolation as a proxy for snow cover persistence and surface albedo dynamics. The superiority of NDWI over NDVI for detecting latent water stress during the grain-filling stage was empirically confirmed. Unlike prior frameworks that rely predominantly on growing-season vegetation indices, the present study demonstrates that pre-seasonal agroclimatic drivers—particularly winter solar radiation and April moisture recharge—exert a stronger influence on yield than mid-season NDVI in arid rainfed systems. Geospatial analysis identified a pronounced domain shift in foothill and irrigated clusters, attributed to the coarse spatial resolution of climate grids and the irrigation-induced decoupling of crop phenology from precipitation regimes. Full article

With the continuous emergence of exam question types, accurate classification of knowledge points is crucial for intelligent exam analysis. Existing methods focus on text or text–image fusion but largely ignore spatial layout. To address this limitation, we propose a heterogeneous layout-aware cross-modal framework for knowledge point classification. The architecture begins with an encoding module where independent text and layout encoders extract semantic content and spatial configurations, respectively. We then design a layout-aware enhancing module consisting of two parallel cross-modal blocks, namely a Layout-Aware Text-Enhancing block and a Context-Aware Layout-Enhancing block. This module supports the bidirectional fusion of text and layout features and generates a comprehensive representation that integrates both semantic and spatial information. Furthermore, a dynamic router with top-k expert selection is introduced to dynamically adapt to question-specific knowledge distributions and focus on core knowledge points for precise classification. Experimental results demonstrate that our method effectively integrates text and layout information, significantly enhancing performance on the proposed QType-EDU dataset. The approach achieves 91.56% accuracy for coarse-grained classification and 80.58% for fine-grained classification, with an overall F1-score of 91.39%, surpassing all baseline models. Full article

The geometric signatures of macroscopic interfaces in the two-dimensional critical Ising model strictly adhere to Schramm–Loewner Evolution (SLE) theory. In this study, we propose a physics-driven generative approach using Super-Resolution Generative Adversarial Networks (SRGANs) to approximate the inverse coarse-graining operation to generate larger configurations. From the perspective of Geometric Deep Learning (GDL), we leverage the geometric priors of Convolutional Neural Networks (CNNs)—specifically their translational and rotational symmetries—to effectively encode the universal physical laws of the Ising Hamiltonian. This inductive bias allows the model to be trained on small scales yet be generalized to large-scale systems ($2048 \times 2048$) while preserving physical conservation. To accommodate spin discreteness, we employ an L1-based loss function to maintain domain wall sharpness. SLE analysis and long-range correlation functions confirm that the model reproduces critical dynamics and conformal invariance, successfully serving as a physics-preserving inverse coarse-graining transformation framework. Full article