Search Results (20,229)

Growing deployment of Convolutional Neural Networks in safety-critical domains has intensified the need for transparent, discriminative explanations, yet gradient-based methods such as Grad-CAM compute explanations for a single class score, producing saliency maps that activate regions shared with competing alternatives. This limitation is particularly consequential in multi-class and fine-grained recognition settings, where non-discriminative explanations reduce the practical utility of the method and impede reliable model auditing. This research study contributes to the body of knowledge by proposing Contrastive Gradient-Flow Interpretation (CGFI), an operator on gradient-weighted attribution maps that applies the contrastive principle that discrimination is sharpened by modelling what a class is not, to generate class-discriminative explanatory maps. CGFI explicitly models the top-K competing classes, computes their aggregated gradient attribution, and subtracts their influence from the target class attribution at the feature-map level, emphasising regions uniquely characteristic of the predicted class. It was evaluated across four backbones (Grad-CAM, Grad-CAM++, XGrad-CAM, Score-CAM), with and without contrast, on ResNet-50, VGG-16 and VGG-19. CGFI reduced the target-rival rank correlation across all 1162 image-backbone combinations evaluated, and improved the target probability share in 11 of 12, while preserving attribution fidelity. These findings suggest that contrastive attribution is a principled mechanism for improving explanation specificity. Full article

Historic buildings are important carriers of traditional Chinese culture; however, they currently face the dual challenges of long-term deterioration and inappropriate interventions. In Zhejiang Province, more than 75,000 immovable cultural heritage sites have been incorporated into the protection system, while a large number of historic buildings remain in urgent need of conservation. In response to the complexity of conventional digital reconstruction processes and the low efficiency of authenticity-oriented restoration, this study proposes an AI-generated content (AIGC)-driven Methodological Framework for efficient and authenticity-oriented digital reconstruction of historic buildings. Taking Xiangxian Hall in Changle Village, Lanxi, Zhejiang, as a case study, this research follows authenticity-oriented criteria, including accurate morphological characteristics, realistic materials and textures, realistic color and light–shadow effects representation, precise proportion and scale, and comprehensive historical and cultural representation. Based on these criteria, a systematic evaluation framework for authenticity-oriented digital reconstruction is established. Furthermore, an AIGC-driven full-process Methodological Framework is proposed that integrates intelligent knowledge retrieval, cross-modal generation models, AI-assisted image-based 3D reconstruction, and inverse rendering. The results demonstrate that the proposed approach, supported by AIGC technologies, can deeply integrate multi-source information, including field investigation data, the existing conditions of historic buildings, and historical and cultural context, thereby achieving efficient and authenticity-oriented digital reconstruction of historic buildings. Furthermore, it provides an integrated technical pathway and practical solution for the conservation and digital inheritance of historic buildings. Full article

Emerging and novel ovine viruses pose increasing threats to animal health, livestock productivity, trade, food security, and public health preparedness. Their emergence is driven by complex interactions among animal movement, mixed-species production systems, wildlife–livestock interfaces, arthropod vectors, environmental changes, and viral evolution. This review critically compares current strategies for identifying major, emerging, re-emerging, zoonotic, and newly recognized ovine viruses, with emphasis on their analytical sensitivity, turnaround time, throughput, operational cost, accessibility, field applicability, validation status, and capacity for novel-virus detection. Conventional diagnostic approaches, including virus isolation, serology, antigen detection, histopathology, and immunohistochemistry, remain essential for confirmation and flock-level surveillance but may be limited by slow turnaround, dependence on specialized facilities, reduced sensitivity at low viral loads, and an inability to identify highly divergent or unknown viruses. Targeted molecular assays, including PCR, RT-PCR, qPCR, multiplex assays, digital PCR, isothermal amplification, and CRISPR-based diagnostics, have improved detection speed and sensitivity but generally require prior knowledge of viral genomic targets. Genomic and metagenomic approaches, including whole-genome sequencing, next-generation sequencing, nanopore sequencing, viral metagenomics, bioinformatics, and phylogenetic analysis, provide broader detection capabilities by enabling characterization of viral diversity, outbreak tracing, co-infection identification, and discovery of previously unrecognized viruses. However, their interpretation remains challenging due to low viral abundance, poor sample quality, host nucleic acid background, contamination, incomplete reference databases, limited computational capacity, and the inability of sequence detection alone to confirm disease causality. Therefore, future ovine virus surveillance requires integration of molecular diagnostics with active, passive, outbreak-based, risk-based, vector, wildlife, and animal-movement surveillance within a One Health framework. Linking genomic information with ecological, epidemiological, and environmental data will be essential for transforming ovine virus surveillance from reactive diagnosis toward proactive preparedness. Advances in standardized sampling, validated field diagnostics, affordable sequencing, curated databases, bioinformatics capacity, and cross-sector data sharing will strengthen early recognition, risk assessment, and preparedness against emerging viral threats in sheep. Full article

Several prominent theories have asserted atypical arousal as a biological feature of autism. However, there is mixed evidence regarding atypical arousal patterns, particularly resting-state arousal, in autistic children, with research increasingly using psychophysiological measures, such as electrodermal activity (EDA), to help elucidate arousal levels in individuals who might not otherwise be able to identify such states. This study analyzed extant EDA data from 135 autistic children (6–12 years) collected during a baseline rest period in a dental environment to determine the relationships between tonic (mean skin conductance level; SCL) and phasic (frequency of non-specific skin conductance responses; NS-SCRs) arousal and participant characteristics, including sensory over-responsiveness, autism traits, expressive language, and general anxiety. EDA data were collected using pre-gelled electrodes placed on the distal phalanges of the participant’s non-dominant hand using the BIOPAC MP150 system. Data preprocessing and epoch analyses were conducted using BIOPAC AcqKnowledge software. Conditional linear and, for the outcome variable anxiety, quadratic growth models were used to assess the temporal (within-participant across observations) relationship between EDA and behavioral variables. The results revealed a significant time-by-autism trait interaction for SCL, as well as linear and quadratic relationships between anxiety and NS-SCRs. No significant relationships were found between EDA arousal parameters and sensory over-responsiveness or expressive language abilities. A significant covariance relationship between SCL and NS-SCR frequency was also observed. Collectively, these results support previous characterizations of unique, within-autism-population arousal profiles related to autism traits, anxiety, and tonic-to-phasic covariance patterns. Full article

Fine-grained maturity assessment of Sichuan pepper (Zanthoxylum bungeanum cv. Hanyuan) is challenging: CNNs are data-hungry and opaque, and vision-language models (VLMs) achieve only chance-level accuracy ($16.67\%$) on this six-way task when used end-to-end. We present KG-RAG, a retrieval-augmented classification framework that combines CNN visual features with a 25,881-triplet knowledge graph built from VLM-extracted structured attributes. KG-RAG introduces three key components: (i) a hierarchical attribute consistency score (HACS) that generalizes Jaccard re-ranking via mutual-information weighting and family-level regularization; (ii) a confidence-guided retrieval gate for adaptive parametric/non-parametric fusion, with calibration as a secondary benefit; and (iii) a two-stage VLM curriculum that repurposes a VLM—inaccurate as an end-to-end classifier but reliable as an attribute extractor—into a structured knowledge provider. On 2114 expert-annotated images (inter-annotator Cohen’s $\kappa =0.89$), $10\times 5$-fold cross-validation shows consistent gains across four backbones (+1.25 to +4.65 percentage points), with the largest gain in the low-data regime (+14.89 pp at 10% training data). Statistical significance is assessed via paired t-tests with repeated-CV caveats. Full article

Rare-earth elements (REEs) constrain the deployment of wind generation and electric mobility, on which EU climate targets depend. The EU accordingly developed five criticality assessments between 2011 and 2023, led by the European Commission’s DG GROW and supported by JRC methodological work. This article asks whether assessment capacity has been converted into industrial capacity. Combining EU, USGS and IEA indicators with documented capacity evidence, we classify each assessment cycle on two ordinal scales: policy response and structural outcome, interpreted separately. Exposure remained high: China’s share of global REE processing stayed within an 87–95% band, the end-of-life recycling input rate remained below 1%, and EU separation, magnet production and recycling capacity showed no sustained net expansion during the paired period. Between 2011 and 2020, policy responses became more formalised, while an early recycling-capacity gain was reversed, and no sustained structural improvement remained. The 2023 cycle is reported on the response side only, while 2024–2025 developments are treated as a post-period update. Japan, an illustrative contrast, absorbed the same 2010 shock under a different policy configuration. The EU case indicates an implementation constraint rather than a knowledge deficit, bearing directly on the feasibility of EU climate and circular-economy targets. Full article

Mineral prospectivity mapping (MPM) increasingly integrates geological, geochemical, geophysical, remote-sensing, and structural evidence through machine-learning workflows. Predictive accuracy alone, however, does not establish that a prospectivity map is geologically credible or useful for exploration decisions. This critical narrative review examines explainable artificial intelligence (XAI) for MPM through the connections among model behavior, mineral-system knowledge, sampling, spatial validation, uncertainty, and field evidence. We distinguish methods demonstrated in representative MPM studies from general explanation tools and proposed applications. Study-level comparisons show that SHAP and permutation-based attribution can support evidence-layer auditing and target interpretation, while their meaning depends on correlated predictors, label construction, and evaluation design. Spatially separated evaluation tests a different generalization problem from random splitting; neither replaces newly acquired field evidence. Geological plausibility, model faithfulness, explanation stability, and decision utility therefore require separate assessment. We synthesize practical pathways for geological knowledge embedding and three-dimensional modeling, identify limits in current graph explanations and uncertainty reporting, and propose a minimum reporting checklist. Future priorities include geospatial foundation models, source-traceable language tools, three-dimensional prospectivity and four-dimensional extensions incorporating geological time, knowledge-guided hypothesis generation, integrated exploration systems, and field-based evaluation of explanations. Full article

Carbon-based nanomaterials have emerged as promising tools for addressing major challenges in sustainable agriculture, including declining soil fertility, climate change, resource inefficiency, and increasing environmental pollution. Among these materials, carbon nanotubes (CNTs) and carbon quantum dots (CQDs) have attracted considerable attention because of their unique physicochemical properties and diverse interactions with plants and soil systems. This review provides a comprehensive comparative analysis of CNTs and CQDs, emphasizing how their structural characteristics govern environmental fate, plant uptake, physiological responses, and stress mitigation mechanisms. CNTs primarily function as one-dimensional nanostructures that improve soil properties, facilitate nutrient delivery, and immobilize environmental contaminants, whereas CQDs, owing to their ultrasmall size, excellent water dispersibility, and intrinsic fluorescence, actively regulate plant metabolism, photosynthesis, nutrient acquisition, and antioxidant defense. Their distinct transport pathways, rhizosphere interactions, and subcellular localization are critically evaluated alongside recent advances in synthesis, surface functionalization, and physicochemical modification. The review further summarizes current evidence regarding their roles in enhancing tolerance to heavy metal toxicity, salinity, and drought stress through modulation of reactive oxygen species scavenging, osmotic regulation, ion homeostasis, and stress-responsive signaling pathways. Potential phytotoxicity, environmental persistence, ecological risks, and green synthesis strategies are also discussed to provide a balanced assessment of their agricultural applications. Emerging opportunities for synergistic CNT–CQD composite systems are discussed, together with major knowledge gaps and future research priorities, including machine learning-assisted nanomaterial design, multi-omics characterization, long-term field validation, and life cycle-based risk assessment. This review establishes a conditional, context-dependent structure–behavior–function conceptual framework that provides theoretical guidance for the rational design and safe implementation of carbon nanomaterials in next-generation sustainable agriculture. Full article

Web security audits of SQL injection (SQLi) vulnerabilities still depend primarily on the analyst’s skill and knowledge to select the right tools, interpret results, and determine whether there is any vulnerability. This work presents SQLiAgent, an open-source multi-agent framework that automates the complete cycle of an SQLi audit—from initial recognition to post-exploitation assisted by artificial intelligence (AI)—in order to facilitate the detection and subsequent patching of vulnerabilities. SQLiAgent integrates well-established tools from the pentesting ecosystem within a decoupled and traceable workflow, and it adds a mode assisted by large language models (LLMs) whose contribution is especially relevant in post-exploitation: automated interpretation of database schema, identification of credential tables and autonomous generation of privilege-escalation SQL statements from the inferred structure of the database. The tool has been validated on the OWASP Broken Web Applications (BWA) environment, used as a reproducible testbed that makes it possible to measure its effectiveness and to establish a baseline for comparison with other systems. The results obtained show that both modes reach 100% coverage per application and that the AI mode improves detection, locating a larger number of vulnerable pages than the non-AI mode. The incorporation of AI demonstrates a clear advantage in post-exploitation, in tasks such as the automatic generation of privilege escalation or generation of technical security reports. Full article

To address the limited automation of 3D reconstruction caused by the inability of point clouds to represent the hidden structures and construction logic of Ming–Qing large timber buildings, this study proposes a hidden structural parameter inference method that integrates point clouds with traditional construction rules. First, a unified parameter space is established to provide a structured representation of building components and their associated parameters. Second, traditional construction knowledge is formalized into computable proportional, relational, and spatial constraints. Finally, hidden structural parameters are inferred through hierarchical constraint propagation, and the inferred results are used to generate HBIM components. The proposed method was validated using the sub-eave columns and their associated components of the Dabei Hall of Chongshan Temple in Taiyuan. Among 1848 hidden structural parameters, 1800 were successfully inferred, corresponding to a solvability rate of 97.4%. The results demonstrate that the proposed method can effectively infer hidden structural parameters under the available observations and construction-rule constraints and use the rule-consistent inference results to generate parametric HBIM components. This study extends HBIM beyond geometric representation toward knowledge-driven model representation, providing a knowledge-enhanced modeling approach for the digital documentation and structural understanding of traditional timber architecture. Full article

Reliable semantic interaction between physical entities and their virtual counterparts is fundamental to digital-twin operation. In challenging wireless environments, however, channel noise and fading corrupt continuous semantic representations, causing semantic drift, token substitutions, repetitive generation, and premature termination at the receiver. This article proposes LM-DeepSC, a language model-aided text semantic communication framework for digital twins. The framework combines end-to-end joint source–channel semantic transmission with a trainable continuous semantic feature adapter at the receiver. The adapter projects channel-corrupted features produced by the semantic decoder into continuous conditioning representations that a frozen pretrained language model can directly exploit. Consequently, the receiver jointly exploits residual communication evidence, contextual dependencies, and pretrained linguistic knowledge to reconstruct the source text without first committing to an error-prone intermediate token sequence. Experiments over additive white Gaussian noise and Rayleigh fading channels show that LM-DeepSC consistently outperforms DeepSC in multi-order BLEU scores and sentence similarity under the same transmitted channel-symbol budget. Evaluations over ten independent AWGN channel realizations further demonstrate a relative reduction of 46.3–71.0% in token substitution rate. On MASSIVE control instructions unseen during communication-model training, LM-DeepSC improves downstream intent-classification accuracy by 6.3–15.5 percentage points over DeepSC. These results demonstrate the effectiveness of the proposed LM-assisted receiver for robust text interaction under noisy wireless channels, with potential application to digital-twin systems. Full article

To address the adverse effects of complex acquisition conditions on continuous sign language recognition (CSLR) performance and the increasing computational cost of standard self-attention with increasing video sequence length, this paper proposes a continuous sign language recognition method based on temporal–frequency enhancement and DCT linear attention. First, a multi-scene continuous sign language dataset was constructed from the recordings of nine signers. The dataset contains 3502 video clips, 1208 lexical items, and nine acquisition scenarios, covering indoor and outdoor environments, strong and weak illumination, and static and dynamic backgrounds. These acquisition scenarios were designed to introduce diverse visual conditions during data collection. Because the available evaluation uses a random video-level split without scene-disjoint grouping, the results describe performance under the recorded mixed conditions and do not establish generalization to unseen signers or unseen acquisition scenarios. Second, using a Video Swin Transformer and spatial global average pooling as the feature-extraction front end, a temporal–frequency-enhanced teacher model was developed through a temporal branch, a short-time Fourier transform (STFT) frequency branch, and gated fusion, thereby jointly exploiting temporal and local frequency information. On this basis, a lightweight student configuration was developed using DCT-kernelized linear attention in the temporal-modeling pathway. Response-level knowledge distillation was further introduced to mitigate the recognition-performance loss associated with lightweight linearized modeling. Experimental results show that the teacher model achieves word error rates (WERs) of 19.8%, 23.1%, and 37.2% on PHOENIX14, CSL-Daily, and the self-constructed multi-scene dataset, respectively. After knowledge distillation, the student model achieves WERs of 21.2%, 24.1%, and 38.8% on the three datasets, with gaps of 1.4, 1.0, and 1.6 percentage points relative to the teacher model, respectively. At the complete-configuration level, the reported FLOPs decrease from 12.5 G for the teacher model to 4.2 G for the student model. On an NVIDIA RTX 3090 Ti GPU with a batch size of 1 and a 200-frame input, the single-sample inference latency decreases from 48.5 ms to 36.8 ms. These results indicate that, under the specified experimental conditions, the lightweight student configuration achieves lower computational cost and inference latency at the expense of only a limited loss in recognition performance. Full article

Endometriosis has historically been characterized by silence, stigma, and limited public awareness, yet in recent years, the condition has increasingly appeared in contemporary fiction. This paper examines how literary representations of endometriosis contribute to the production of knowledge about the illness, focusing on Sally Rooney’s Conversations with Friends and Madeline Docherty’s Gender Theory. Both novels belong to the popular sad girl literary tradition and are therefore particularly significant sites for examining how endometriosis is represented in widely read contemporary fiction. Adopting a medical humanities perspective, the paper reads these novels in dialogue with scholarship on endometriosis, illuminating experiential dimensions of the condition that are often difficult to capture through biomedical discourse alone. The analysis engages with the concept of menstrual etiquette and largely focuses on the social and epistemic injustices experienced by protagonists living with a stigmatized, gendered illness. Rather than treating literature as a reflection of medical knowledge, the article argues that fiction functions as a mode of inquiry, generating valuable insights into the embodied, emotional, and social realities of living with endometriosis. By bringing these experiences into mainstream fiction, the novels not only deepen understandings of endometriosis within the medical humanities but also contribute to broader public awareness of one of the world’s most common yet persistently overlooked diseases. Full article

There is growing interest in the role of the gut microbiome in immune development and homeostasis during childhood. Increasing evidence suggests that early alterations in microbial composition might influence susceptibility to both respiratory and rheumatic diseases through interconnected pathways. While the gut–lung and gut–joint axes have been investigated separately, their integration into a biologically specific and testable model remains largely unexplored. In this narrative review, the literature was searched in PubMed/MEDLINE and Scopus for evidence syntheses published from 2020 to 2026; 37 eligible evidence syntheses were included, comprising 26 addressing the gut–lung/respiratory domain and 11 addressing the gut–joint/rheumatic domain. We examine current knowledge of microbiome-mediated interactions linking the gut, lungs, and joints in childhood. We discuss the development of the paediatric microbiome and the influence of early-life factors, including mode of delivery, breastfeeding, infections, antibiotic exposure, and environmental determinants, on immune programming. We then summarise evidence supporting the gut–lung axis in paediatric respiratory diseases and the gut–joint axis in juvenile idiopathic arthritis (JIA), highlighting recurrent but non-specific findings such as dysbiosis, impaired epithelial barrier integrity, altered microbial metabolite production, and immune dysregulation. Recent paediatric epidemiological evidence also showed that JIA was more prevalent among children with asthma than among those without asthma (0.81% vs. 0.23%), with asthma associated with approximately twofold higher odds of JIA after propensity-score weighting (OR 2.10, 95% CI 1.56–2.81), although this association does not establish microbiome-mediated causality. We finally critically examine whether more specific mechanisms, including gut-primed immune-cell trafficking, antigen-specific amplification, and metabolite–receptor convergence, could provide testable links between pulmonary and articular inflammation. On this basis, we propose the gut–lung–joint axis as a conceptual, hypothesis-generating model. Future research integrating microbiome profiling, immune-cell clonality, immunophenotyping, and metabolomics will be required to test its predictions and clarify its clinical relevance. Full article

Macedonian is a low-resource language for automatic speech recognition: annotated speech data are scarce, dialectal variation is substantial, and existing evaluations focus almost entirely on read speech in Standard Macedonian. This paper presents a unified pipeline that converts Macedonian speech, including regional dialects, into accurate transcripts and structured, queryable knowledge, a capability required for applications such as searchable parliamentary archives, broadcast transcription and subtitling, and dialectological documentation. Parameter-efficient adaptation of Whisper large-v3-turbo via low-rank adaptation is evaluated against strong zero-shot and language-specific baselines on four newly curated dialect corpora (Ohrid, Veles, Tikvesh, and Gostivar) and three Standard Macedonian corpora, two of which were collected for this work. The adapted model reduces word error rate by 57–70% relative to the strongest zero-shot baseline and by 33–66% relative to the language-specific BUKI Whisper 2.0 model on dialectal speech, with comparable improvements over zero-shot baselines on standard Macedonian speech, while updating only about 0.7% of parameters. Beyond transcription, the pipeline adds speaker diarization with cross-recording speaker linking and a graph-based retrieval-augmented generation component that enables speaker-, topic-, and time-aware querying of diarized transcripts, evaluated on long-form Macedonian parliamentary recordings. Together, these results establish parameter-efficient adaptation, speaker-aware processing, and graph-based retrieval as a practical and transferable framework for transforming under-resourced speech into accessible, structured knowledge. Full article