Search Results (102)

Tea bud detection is a fundamental prerequisite for accurate tea yield estimation and intelligent mechanical harvesting. However, existing detection methods face several critical challenges, including ineffective extraction of multi-scale features, weak feature saliency for small tea bud targets, and the prevalent imaging issue in which the central regions of tea images are in focus while peripheral areas suffer from defocus blur. These factors collectively result in a high rate of missed detections, severely limiting detection accuracy and subsequent application performance. To overcome these technical bottlenecks, this paper proposes a novel tea bud detection framework, termed REB-Tea, which integrates image clarity optimization with multi-scale feature enhancement. First, the Restormer image restoration network is employed to improve overall image clarity and enhance the discriminative representation of tea bud features. Subsequently, a bidirectional feature pyramid network (BiFPN) structure and an efficient multi-scale attention (EMA) mechanism are incorporated into the neck of the YOLOv5 model to strengthen multi-scale feature fusion and guide the network to focus on fine-grained tea bud features across different scales, thereby improving detection performance for small and densely distributed targets. Experimental results based on 10-fold cross-validation demonstrate that the proposed REB-Tea model achieves an average mAP₅₀ of 95.5% on the Longjing 43 tea test set, representing a 9.9 percentage point improvement over the baseline YOLOv5 model, and Welch’s independent two-sample t-test verifies that this accuracy increment is highly statistically significant. Moreover, the model exhibits reliable detection performance across different tea varieties, including Cuifeng and Fuding White Tea. Specifically, the mAP₅₀ reaches 88.3% on Cuifeng, which shares similar appearance characteristics with Longjing, and 78.1% on Fuding White Tea, which has noticeably different appearance characteristics from Longjing. These results confirm the effectiveness of the REB-Tea framework in addressing challenges such as out-of-focus blurring, weak feature saliency, and multi-scale feature extraction. Overall, the proposed approach significantly enhances tea bud detection accuracy in natural environments and provides robust technical support for intelligent tea harvesting applications. Full article

We provide an intrinsic construction of the central conics in the real hyperbolic plane ${\mathbb{H}}^2$, whereby each conic C is the composition of a unique pair of Steiner conics (those generated by collineations). The composition is achieved by elliptic curve addition on intersection points of the two components with their orthogonal trajectories, which have a natural representation as genus 1 curves in any inversive model of ${\mathbb{H}}^2$. The central Steiner conics that have a focal axis L are identified with the subgroup $G\left(L\right)$ of collineations generated by reflections in the lines perpendicular to L. We obtain a $G\left(L\right)$-equivariant partition of the central conics by defining the fiber over $g\in G\left(L\right)$ to be the set of compositions C such that $\pi \left(C\right)=g$. Here, $\pi \left(C\right)$ is the unique Steiner conic tangent to C at the points on L, and is the product of the two elements in $G\left(L\right)$ that represent the components of C. We use the terminology of fibers strictly in an incidence-geometric sense. Full article

Group equivariant convolutional neural networks (G-CNNs) achieve superior sample efficiency by encoding symmetry into network architecture, yet their computational overhead (up to 3.78× slower inference and 4.63× more multiply–accumulate operations) hinders deployment on resource-constrained edge devices. Existing pruning methods cannot be applied directly: arbitrarily removing weights breaks the group representation structure and degrades equivariance. We characterize the complete design space of equivariance-preserving compression, proving that exactly two axes leave a convolutional layer equivariant: irrep-bundle pruning, which reduces irreducible-representation multiplicities, and orbit-wise pruning, which removes complete spatial orbits from kernel supports; via Schur’s lemma, no third structure-preserving axis exists. This completeness result, rather than the use of representation theory itself, is our central contribution. We turn it into practice through direct sub-filter extraction, which yields real convolutional parameter reduction (up to 83%) and 1.4–2.9× measured inference speedup, unlike masking, which gives no real speedup. Across three datasets (MNIST, CIFAR-10, EuroSAT) and three symmetry groups (C₄, D₄, SO(2)), compression is nearly lossless on strongly symmetric data: the 4-layer EuroSAT model drops only 1.07% at 83% reduction. On weakly symmetric data (CIFAR-10), the pruned model can even gain 2.6 points, but our analysis attributes this to relaxing a mismatched equivariance constraint rather than to pruning itself; the value of pruning over from-scratch training scales with the data’s symmetry strength. Full article

This article reconsiders the methodological primacy of representation in early modern art history by shifting attention from image to material. Taking Rembrandt’s Polish Nobleman (1637) as its point of departure, it argues that narrative interpretation—long central to the discipline—has obscured the material conditions that make images possible. Rather than assembling meaning from pictorial elements, the essay follows the painting’s support: a Baltic oak panel sourced from the woodlands of the Polish–Lithuanian Commonwealth. From this perspective, the artwork emerges not simply as an autonomous image but as the endpoint of an extractive chain linking forestry, peasant labour, river transport, and long-distance trade. Drawing on agronomic manuals, estate records, and economic histories, the article reconstructs these dispersed threads as “story matter”: fragments that, brought into relation, begin to cohere into an alternative mode of narration. In doing so, it advances “material literacy” as a methodological reorientation—an attunement to substances, processes, and infrastructures that precede and exceed representation. Recovering these histories does not replace interpretation but expands its scope, opening art history to ecological and infrastructural forms of storytelling. Full article

Social media platforms have become central sites of identity construction, where visibility and legitimacy are shaped through algorithmic systems, aesthetic conventions, and platform economies. This paper approaches online personas through the lens of illusionary selves, understood here as online personas experienced as authentic while being shaped by sociotechnical processes, examining how they are produced through sociotechnical processes entangling design practices, generative artificial intelligence(AI), and cultural expectations. We present an AI-mediated critical design inquiry into how generative systems translate and normalize visual patterns of online self-imaging. Using a pix2pix-based model trained on 630 internet celebrity selfies, facial images are abstracted into dot-based representations and aggregated across selfie angles, foregrounding repetition and normalization. An interactive design installation links bodily orientation and numerical parameters to generative output in real time, introducing perceptual friction in self-imaging. A total of 30 participants engaged with the system in situated contexts, and their experiences were documented through observation, video recording, and a 5-point Likert questionnaire across three dimensions: perceptual friction, awareness of algorithmic mediation, and reflective responses to self-presentation. Results indicate high levels of perceptual friction (mean [M] = 4.21), strong awareness of algorithmic mediation (M = 4.29), and consistent reflective unease (M = 4.07). Through situated use, the system renders algorithmic mediation tangible and positions AI as an implicated actor in identity construction. This work contributes a conceptual framing of AI-mediated critical design, showing how generative and interactive systems operate as epistemic devices interrogating online persona construction. Full article

Background/Objectives: Egocentric and allocentric spatial transformations are central to spatial cognition, yet it is unknown whether they rely on the same neural mechanisms. The goal of this study was to examine whether egocentric transformations engage the neural processes associated with mental rotation in visual–spatial working memory. Methods: High-density EEG was recorded while participants performed two matched pointing-direction tasks, in which they indicated the direction toward a target location, while instructed to use either allocentric array rotation or egocentric perspective-taking. Response times and accuracy were recorded, and event-related potential (ERP) responses were analyzed as a function of rotation angle (100° vs. 160°) and differences between front and back pointing directions. Results: Response times increased with rotation angle in both tasks, whereas a front–back asymmetry in accuracy was observed only in perspective-taking. Both tasks showed rotation-related ERP modulation, but the timing and spatial distribution of these effects differed across tasks. In the array-rotation task, rotation-related ERP effects were observed over right-parieto–occipital regions at 460–510 ms. In the perspective-taking task, the ERP effects were observed over left-central regions at 400–470 ms and 520–610 ms. ERP differences between front and back directions were robust and widespread in the egocentric condition but limited in the allocentric condition. Conclusions: Perspective-taking does not show the posterior rotation-related ERP effect associated with mental rotation of object representations in visual–spatial working memory. Instead, it appears to reflect updating of the observer-centered reference frame, consistent with simulated self-motion processes involving vestibular and proprioceptive signals. Full article

This paper presents a computationally feasible/time-effective Scan-to-BrIM workflow for generating a highly detailed digital model of a complex steel pedestrian bridge. The proposed methodology integrates rapid and accurate point cloud acquisition with advanced parametric modelling and structural information management. First, a high-resolution point cloud is produced using a fast survey strategy that ensures the geometric precision required for a faithful representation of the existing structure. Second, the point cloud is processed in Rhinoceros/Grasshopper, where a custom Python (version 3.13) algorithm automatically detects and generates reference planes containing the structural components, enabling the creation of a consistent and fully parametric BrIM model. The latter approach includes metric normalization, voxel-based downsampling, reliable under tested conditions ground and outlier removal, and PCA (Principal Component Analysis)-based reorientation, followed by guided slicing of the point cloud and projection of each slice onto its section plane. The proposed workflow achieved a geometric RMSE of 2.5 mm with a total processing time of 7.3 h. The resulting parametric model achieves geometric consistency with the source point cloud within an operational tolerance range of approximately 5–10 mm, in line with the requirements of structural applications. Finally, the model is organised and managed within the BrIM environment and then transferred to a downstream FEM environment for preliminary structural application. The workflow is tested on a case study of a 40-m steel pedestrian bridge located in central Italy. Results demonstrate that the integrated approach provides a reproducible and semi-automated solution that reduces manual intervention in Scan-to-BrIM processes for producing accurate parametric models of steel pedestrian bridges, supporting structural assessment, asset management, and future maintenance strategies. Full article

Urban soundscape management is a central challenge to the livability and sustainability of cities and requires approaches that complement level indicators with frameworks capable of integrating context, use and experience. In this framework, the present work applies a multilayer methodology to the Old Town of Bilbao, understood as a useful case study to explore the applicability of soundscape reading in historic centers with intense coexistence of commercial, hospitality and catering uses, pedestrian, logistical and cultural uses. The methodology is organized into two phases. The first focuses on the recording and documentation of control points and routes through sound fieldwork, perceptual descriptions and homogeneous systematization of information. From this corpus, a qualified sound map and a first visual characterization of the sound identity are elaborated. The second phase presented in this article, consists of the interpretative synthesis of the corpus through five analytical dimensions and the preparation of fragments and sound sequences conceived for future application through reactivated listening. The results are presented at three levels: (1) a traceable documentary corpus of records, files and synthetic representations; (2) a comparative reading by dimensions that identifies spatial contrasts between interior, exterior and perimeter, as well as relationships between urban form, uses, persistence, masking and salience; and (3) a set of operational audio materials prepared for subsequent comparison with inhabitants and users. In a transversal way, type–token reading distinguishes between the diversity of sounds and dominance by repetition. The article does not yet carry out participatory validation of these materials; its contribution consists of proposing and applying a traceable analytical protocol as a basis for future phases of social contrast and applied discussion. Full article

This paper presents a detailed re-examination of the conformalcompactification $\overline{\mathbb{M}}$ of Minkowski space $\mathbb{M}$, constructed as the projective null cone of the six-dimensional space ${\mathbb{R}}^{4,2}$. We provide an explicit and basis-independent formulation, emphasizing geometric clarity. A central result is the explicit identification of $\overline{\mathbb{M}}$ with the unitary group $U\left(2\right)$ via a diffeomorphism, offering a clear matrix representation for points in the compactified space. We then systematically construct and analyze the action of the full conformal group $\mathrm{O}(4,2)$ and its connected component ${\mathrm{SO}}_0(4,2)$ on this manifold. A key contribution is the detailed study of the double cover, $\tilde{\mathbb{M}}$, which is shown to be diffeomorphic to $S^3\times S^1$. This construction resolves the non-effectiveness of the $\mathrm{SO}(4,2)$ action on $\overline{\mathbb{M}}$, yielding an effective group action on the covering space. A significant portion of our analysis is devoted to a precise and novel geometric characterization of the conformal infinity. Moving beyond the often-misrepresented “double cone” description, we demonstrate that the infinity of the double cover, ${\tilde{\mathbb{M}}}_{\infty }$, is a squeezed torus (specifically, a horn cyclide), while the simple infinity, ${\overline{\mathbb{M}}}_{\infty }$, is a needle cyclide. We provide explicit parametrizations and graphical representations of these structures. Finally, we explore the embedding of five-dimensional constant-curvature spaces, whose boundary is the compactified Minkowski space. The paper aims to clarify long-standing misconceptions in the literature and provides a robust, coordinate-free geometric foundation for conformal compactification, with potential implications for cosmology and conformal field theory. Full article

The increasing deployment of Internet of Things (IoT) devices introduces significant security challenges, while privacy concerns limit centralized data aggregation for intrusion detection. Federated learning (FL) offers a decentralized alternative, yet the interaction between feature representation, model architecture, and data heterogeneity remains insufficiently understood in IoT malware detection. This study provides a controlled comparative analysis of centralized and federated learning, optionally using amino acid encoding, under IID and Non-IID conditions using a 10,000-sample subset of the CTU–IoT–Malware–Capture dataset. First, we evaluate raw tabular features versus amino acid-based feature encoding, followed by a lightweight multi-layer perceptron (2882 parameters) versus a deeper residual network (70,532 parameters), across binary and multi-class classification tasks. In the binary setting, centralized training achieved up to 98.6% accuracy, while federated IID training reached 98.6%, with differences within statistical variance. Under Non-IID conditions, performance decreased modestly (0.1–0.5 percentage points), and accuracy was consistently lower when using encoded features compared with raw features. The degradation is smaller in deeper architectures and may offer improved stability under highly skewed federated conditions. In the four-class setting, the complex network achieved up to 97.8% accuracy with raw features, while amino acid encoding achieves up to 93.3%. The results show that federated learning can achieve performance comparable to centralized training under moderate heterogeneity, that lightweight architectures are sufficient for low-dimensional IoT traffic features, and that feature compression via amino acid encoding does not inherently mitigate Non-IID effects. These findings clarify the relative impact of representation, heterogeneity, and architectural capacity in practical FL-based IoT intrusion detection systems. Full article

Fluid-driven fracture processes are central to the development of subsurface energy systems such as geothermal and hydrocarbon reservoirs. Although phase-field formulations have become a widely used tool for describing fracture initiation and growth, the diffuse representation of cracks makes it difficult to resolve flow behavior accurately inside discrete fracture networks (DFNs) and to represent hydro-mechanical coupling in a sharp-interface sense. This study develops a hybrid-dimensional iterative framework for lubrication-flow simulation in deformable fractured geomaterials. By leveraging phase-field point clouds together with non-conforming discretization schemes for both the solid matrix and fracture domains, the proposed framework enables the dynamic reconstruction of evolving fracture networks. The theoretical formulation and numerical implementation of the coupling strategy are presented in detail. Hydraulic benchmark examples verify the performance of the fluid flow solver under various physical conditions. The classical Sneddon problem and Khristianovic–Geertsma–de Klerk (KGD) model are employed to validate the solid deformation solver, confirming accurate predictions of crack opening displacement and mesh independence in fracture width calculation. Additional simulations with complex pre-existing fracture patterns further demonstrate the applicability of the framework to coupled hydro-mechanical analysis in fractured media. Full article

Under high renewable energy penetration, wind power forecasting faces pronounced challenges due to strong randomness and uncertainty, making conventional point-forecast-centric paradigms insufficient for risk-aware and reliable power system scheduling. An uncertainty-aware temporal transformer framework for wind power forecasting is presented, integrating probabilistic modeling with deep temporal representation learning to jointly optimize prediction accuracy and uncertainty characterization. Crucially, rather than treating uncertainty quantification merely as a post-processing step, the central conceptual contribution lies in modularizing uncertainty directly within the attention mechanism. A probability-driven temporal attention mechanism is incorporated at the encoding stage to emphasize high-variability and high-risk time slices during feature aggregation, while a multi-quantile output and interval modeling strategy is adopted at the prediction stage to directly learn the conditional distribution of wind power, enabling simultaneous point and interval forecasts with statistical confidence. Extensive experiments on multiple public wind power datasets demonstrate that the proposed method consistently outperforms traditional statistical models, deep temporal models, and deterministic transformers, as validated by formal statistical significance testing. Specifically, the method achieves an MAE of $0.089$, an RMSE of $0.132$, and a MAPE of $10.84\%$ on the test set, corresponding to reductions of approximately $8\%$–$10\%$ relative to the deterministic transformer. In uncertainty evaluation, a PICP of $0.91$ is attained while compressing the MPIW to $0.221$ and reducing the CWC to $0.241$, indicating a favorable balance between coverage reliability and interval compactness. Compared with mainstream probabilistic forecasting methods, the model further reduces RMSE while maintaining coverage levels close to the 90% target, effectively mitigating excessive interval conservatism. Moreover, by adaptively generating heteroscedastic intervals that widen during high-volatility events and narrow under stable conditions, the model achieves a highly focused and effective capture of critical uncertainty information. Full article

This paper argues that digital identity in AI-mediated environments has become a central mechanism through which contemporary societies organise recognition, participation, and belonging. Digital identity is no longer simply a technical representation of the individual. It is produced through infrastructural processes of classification, ranking, and credibility signalling that determine who becomes visible, who is treated as legitimate, and who is able to participate meaningfully in social and civic life. The paper develops a conceptual framework that treats AI-driven platforms as social infrastructures rather than neutral intermediaries. It shows how identity is inferred through data-driven systems rather than negotiated through social interaction, how recognition is operationalised through visibility and credibility metrics rather than ethical judgement, and how participation becomes conditional on algorithmic allocation of attention rather than guaranteed by access alone. Visibility is identified as the key conversion point through which inferred identity becomes social consequence. Drawing on interdisciplinary literature, the analysis demonstrates that misrecognition, exclusion, and inequality in platform environments are not primarily the result of isolated error or intentional bias. They are patterned outcomes of ordinary optimisation processes that distribute legitimacy and opportunity unevenly across social groups. These dynamics reshape group formation, harden social boundaries, and concentrate risk among populations that are already more vulnerable to misrecognition and reduced contestability. The paper concludes that governing digital identity is a societal challenge rather than a purely technical one. As platforms increasingly perform institutional functions without equivalent accountability, digital identity governance becomes a critical site of social ordering. Addressing this challenge requires public standards for how visibility, recognition, and participation are allocated, meaningful avenues for contestation, and protections against the normalisation of stratified belonging in AI-mediated societies. Full article

Artificial neural networks (ANNs) are widely used to approximate nonlinear mappings, yet their ability to capture thermodynamic behaviour in dynamic physical systems remains insufficiently characterised. This study investigates how representational capacity influences surrogate modelling accuracy for a crank-angle-resolved internal combustion engine (ICE) simulation with a maximum dynamic state dimension of six. Two feedforward ANN configurations are evaluated: a low-capacity 5–5 architecture containing 84 trainable parameters and a high-capacity 25–25–25 architecture containing 1554 parameters (18.5× larger). Both networks approximate the nonlinear mapping from five embedded operating parameters to four peak thermodynamic outputs (maximum pressure, pressure phasing, maximum temperature, and temperature phasing). Evaluation across 53,178 operating points demonstrates that the high-capacity configuration reduces root mean squared error by factors of 30–50× relative to the low-capacity network, decreasing peak temperature error from 17.68 K to 0.36 K and peak pressure error from 0.116 MPa to 0.0025 MPa. Although both models achieve coefficients of determination exceeding 0.99, the low-capacity network exhibits heavy-tailed residual distributions and regime-dependent error amplification, whereas the high-capacity model reduces both central dispersion and extreme-case error. These results demonstrate that high correlation alone does not guarantee engineering reliability in nonlinear thermodynamic systems. Distribution-level analysis, including percentile and extreme-case characterisation, is required to evaluate engineering robustness. The findings provide a quantitative framework linking ANN capacity, nonlinear dynamic system representation, and predictive robustness. Full article

Place recognition is a fundamental challenge for robotics and autonomous vehicles. While visual place recognition has achieved high precision, cross-modal place recognition—specifically, visual localization within large-scale point cloud maps—remains a formidable problem. Existing methods often struggle with the significant domain gap between modalities and can be computationally prohibitive, especially those processing raw 3D point clouds. Furthermore, they frequently fail to learn features invariant to viewpoint and scale variations, limiting generalization to unseen environments. In this paper, we formulate cross-modal recognition as a problem of learning a scale-invariant, unified embedding space. Our framework employs a hierarchical Swin Transformer to extract multi-scale features from unified 2D representations of both modalities. The central principle of our method is a multi-scale self-distillation paradigm, which recasts feature learning as an intra-modal knowledge transfer task. Specifically, the coarse-scale “teacher” features provide supervision for the fine-scale “student” features. The final inter-modal alignment is then achieved via a global contrastive loss, exclusively leveraging the semantically rich “teacher” embeddings to ensure a reliable and discriminative matching. Extensive experiments on the KITTI and KITTI-360 datasets demonstrate that our method achieves state-of-the-art performance. Notably, using only the KITTI-trained model without fine-tuning, Recall@1 exceeds 60% on all evaluable sequences of KITTI-360 at a 10 m threshold. Code and pre-trained models will be made publicly available upon acceptance. Full article