Search Results (25)

Current Facial expression recognition methods typically extract facial features indiscriminately, incorporating expression-irrelevant information that compromises recognition accuracy. To overcome this, we propose Multi-stage Feature Sparse Constraints (MFSC), a novel model that integrates a Multi-scale Attention-based Sparse Window Selection (MSAWS) mechanism with key region graph learning. Notably, MFSC operates without dependence on pre-extracted facial landmarks, enabling more flexible deployment. The MSAWS mechanism progressively filters redundant features through multi-stage sparse attention, adaptively selecting the most discriminative facial patches. The selected tokens are structured into a dynamic graph to model regional relationships via graph neural networks (GNNs). Critically, our framework further introduces a global-guided fusion module, which effectively integrates fine-grained local features from an IR50 backbone with the global topological features from the GNN through cross-attention. This integration enables complementary strengths, where local details are enhanced by global semantic context. Comprehensive experiments on RAF-DB, FER2013, and AffectNet-7 datasets demonstrate MFSC’s superior performance, achieving state-of-the-art accuracy of 92.31%, 76.21%, and 67.35%, respectively. These results validate the effectiveness of our approach in focusing computational resources on expression-salient regions while maintaining a lightweight and efficient architecture. Full article

Machine translation with large language models has recently attracted growing attention due to its flexibility and strong zero-shot and few-shot capabilities. However, most prompt-based LLM translation methods rely on linear generation or shallow self-refinement, implicitly committing to a single reasoning path. Such designs are brittle when translating long and syntactically complex sources, where reliable translation often requires structured planning and hypothesis exploration. In this paper, we propose TransGoT, a novel machine translation framework inspired by the graph-of-thoughts paradigm, which formulates translation as a structured, multi-stage reasoning process over a graph of intermediate thoughts. TransGoT explicitly decomposes translation into constraint identification, draft generation, and culture- and style-aware refinement, enabling systematic exploration and aggregation of alternative translation hypotheses. To better adapt graph-based reasoning to translation, we design two key mechanisms: (1) Uncertainty-driven thought transformation. Unlike general reasoning tasks, translation uncertainty is often localized and unevenly distributed across tokens, making holistic regeneration inefficient. We therefore design uncertainty-driven thought transformation, which leverages model-internal confidence signals to guide targeted token-level revision; (2) Dispersion-adaptive thought scoring. It emphasizes evaluation criteria with stronger inter-candidate variance to enable robust multi-criteria thought selection. We evaluate TransGoT on the WMT22 benchmarks and experimental results demonstrate that TransGoT consistently outperforms strong LLM-based translation baselines, validating the effectiveness of structured graph-based reasoning for machine translation. Full article

The prevalence of ground fissures in deformation-affected areas has intensified, presenting serious risks to both operational safety and the local natural environment. Fissures in these disturbed terrains are typically characterized by elongated morphologies and large-scale variations, which pose substantial challenges to accurate feature extraction. To address these complexities, this paper proposes a semantic segmentation network termed MGF-UNet. In the shallow layers, we integrate multi-scale feature sensing (MFS) and grouped efficient multi-scale attention (EMA) to sharpen anisotropic textures and boundary details under high-resolution representations. For the deeper layers, a Token-Selective Context Transformer (TSCT) is designed to perform selective global modeling on high-level semantic features, effectively capturing long-range dependencies while preserving the structural integrity of elongated fissures. Meanwhile, we employ feature-wise linear modulation (FiLM) to derive pixel-wise affine parameters from shallow structures, which pre-modulate deep features and strengthen cross-level interactions. In the decoder, a Fourier transform-based adaptive feature fusion (AFF) module suppresses background noise and enhances boundary contrast, followed by cross-scale aggregation for final prediction.Benchmark tests conducted on the mining-area fissure dataset (MFD) and road-based datasets demonstrate that MGF-UNet achieves an accuracy of 78.2%, a Dice score of 81.4%, and an IoU of 68.6%, outperforming existing mainstream networks. The results confirm that MGF-UNet provides an effective solution for automatic fissure extraction in deformation-prone environments, offering significant potential for geohazard monitoring and ecological restoration. Full article

Fine-grained visual classification requires recognizing subtle inter-class differences under substantial intra-class variation. Ancient mural recognition poses additional challenges: severe degradation and complex backgrounds introduce noise that obscures discriminative features, limited annotated data restricts model training, and dynasty-specific artistic styles manifest as periodic brushwork patterns and compositional structures that are difficult to capture. Existing spatial-domain methods fail to model the frequency characteristics of textures and the cross-region semantic relationships inherent in mural imagery. To address these limitations, we propose a Vision Transformer (ViT) framework which integrates frequency-domain enhancement, explicit token-relation modeling, adaptive multi-focus inference, and discriminative metric supervision. A Frequency Channel Attention (FreqCA) module applies 2D FFT-based channel gating to emphasize discriminative periodic patterns and textures. A Cross-Token Relation Attention (CTRA) module employs joint global and local gates to strengthen semantically related token interactions across distant regions. An Adaptive Omni-Focus (AOF) block partitions tokens into importance groups for multi-head classification, while Complementary Tokens Integration (CTI) fuses class tokens from multiple transformer layers. Finally, Adaptive Margin Contrastive Center Loss (AMCCL) improves intra-class compactness and inter-class separability with margins adapted to class-center similarities. Experiments on CUB-200-2011, Stanford Dogs, and a Dunhuang mural dataset show accuracies of 91.15%, 94.57%, and 94.27%, outperforming the ACC-ViT baseline by 1.35%, 1.63%, and 2.20%, respectively. Full article

Financial report question answering (FRQA) presents unique challenges due to the need for precise numerical reasoning, complex table structures, and multi-table associations. Existing approaches often overlook the domain-specific complexities of financial reports and struggle with accurate numerical computation, leading to suboptimal performance in real-world financial intelligence applications. In this study, we propose FinQA-PKD, a framework designed to mitigate these challenges through a novel integration of progressive knowledge distillation and numerical reasoning enhancement. Our method introduces a difficulty-aware curriculum learning strategy that organizes training into two progressive stages, facilitating more effective and stable model learning. To address the limitations of large language models in numerical reasoning, we develop a numerical reasoning enhancement module that automatically decomposes calculation chains, augments numerical tokens, and validates results using a financial formula library. Furthermore, we implement a domain-adaptive selective knowledge distillation strategy, which evaluates teacher model outputs based on numerical accuracy, calculation correctness, and terminology precision, and selectively distills knowledge from high-quality samples. Experimental results in benchmark datasets demonstrate that FinQA-PKD improves numerical and calculation accuracy, achieving competitive performance with reduced computational resources. This framework provides a robust and efficient solution for answering financial report questions in practical financial analysis scenarios. Full article

Accurate aircraft detection in remote sensing imagery is critical for aerospace surveillance, military reconnaissance, and aviation security but remains fundamentally challenged by extreme scale variations, arbitrary orientations, and dense spatial clustering in high-resolution scenes. This paper presents an adaptive attention-guided super-resolution network that integrates multi-scale feature learning with saliency-aware processing to address these challenges. Our architecture introduces three key innovations: (1) A hierarchical coarse-to-fine detection pipeline that first identifies potential regions in downsampled imagery before applying precision refinement, (2) A saliency-aware tile selection module employing learnable attention tokens to dynamically localize aircraft-dense regions without manual thresholds, and (3) A local tile refinement network combining transformer-based super-resolution for target regions with efficient upsampling for background areas. Extensive experiments on DIOR and FAIR1M benchmarks demonstrate state-of-the-art performance, achieving 93.1% AP₅₀ (DIOR) and 83.2% AP₅₀ (FAIR1M), significantly outperforming existing super-resolution-enhanced detectors. The proposed framework offers an adaptive sensing solution for satellite-based aircraft detection, effectively mitigating scale variations and background clutter in real-world operational environments. Full article

Accurately predicting aircraft fuel consumption is vital for aviation safety, operational efficiency, and resource optimization, yet existing models face key limitations. Traditional physical models rely on prior assumptions, while mainstream deep learning models use fixed architectures and time-slice tokens—failing to adapt to distinct flight phases and losing long-range temporal features critical for cross-phase dependency capture. This paper proposes Gate-iInformer, an adaptive framework centered on iInformer with a gating network. It treats flight parameters as independent tokens, integrates Informer to handle long-range dependencies, and uses the gating network to dynamically select pre-trained phase-specific sub-models. Validated on 21,000 Air China 2023 medium-aircraft flights, it reduces MAE and RMSE by up to 53.38% and 44.51%, achieves 0.068 MAE in landing, and outperforms benchmarks. Its prediction latency is under 0.5 s, meeting ADS-B needs. Future work will expand data sources to enhance generalization, boosting aviation intelligent operation. Full article

Visual–language tracking often faces challenges like target deformation and confusion caused by similar objects. These issues can disrupt the alignment between visual inputs and their textual descriptions, leading to cross-modal semantic drift and feature-matching errors. To address these issues, we propose SATrack, a Semantic-Aware Alignment framework for visual–language tracking. Specifically, we first propose the Semantically Aware Contrastive Alignment module, which leverages attention-guided semantic distance modeling to identify hard negative samples that are semantically similar but carry different labels. This helps the model better distinguish confusing instances and capture fine-grained cross-modal differences. Secondly, we design the Cross-Modal Token Filtering strategy, which leverages attention responses guided by both the visual template and the textual description to filter out irrelevant or weakly related tokens in the search region. This helps the model focus more precisely on the target. Finally, we propose a Confidence-Guided Template Memory mechanism, which evaluates the prediction quality of each frame using convolutional operations and confidence thresholding. High-confidence frames are stored to selectively update the template memory, enabling the model to adapt to appearance changes over time. Extensive experiments show that SATrack achieves a 65.8% success rate on the TNL2K benchmark, surpassing the previous state-of-the-art UVLTrack by 3.1% and demonstrating superior robustness and accuracy. Full article

The growing complexity of production systems in the technology sector demands advanced tools to ensure efficiency, flexibility, and cost-effectiveness. This study presents the development of a simulation model for a selective soldering line at a technology manufacturing company in Poland, created during an engineering internship. Using FlexSim 24.2 software, the real production process was replicated, including input/output queues, manual insertion (MI) stations, soldering machines, and quality control points. Special emphasis was placed on implementing dynamic process logic via ProcessFlow, enabling detailed modeling of token flow and system behavior. Through experimentation, various configurations were tested to optimize process time and the number of soldering pallets in circulation. The results revealed that reducing pallets from 12 to 8 maintains process continuity while offering cost savings without impacting performance. An intuitive operator panel was also developed, allowing users to adjust parameters and monitor outcomes in real time. The project demonstrates that simulation not only supports operational decision-making and resource planning but also enhances interdisciplinary communication by visually conveying complex workflows. Ultimately, the study confirms that simulation modeling is a powerful and adaptable approach to production optimization, contributing to long-term strategic improvements and innovation in technologically advanced manufacturing environments. Full article

Recent advances in hyperspectral image (HSI) classification have demonstrated the effectiveness of hybrid architectures that integrate convolutional neural networks (CNNs) and Transformers, leveraging CNNs for local feature extraction and Transformers for global dependency modeling. However, existing fusion approaches face three critical challenges: (1) insufficient synergy between spectral and spatial feature learning due to rigid coupling mechanisms; (2) high computational complexity resulting from redundant attention calculations; and (3) limited adaptability to spectral redundancy and noise in small-sample scenarios. To address these limitations, we propose HyperTransXNet, a novel CNN-Transformer hybrid architecture that incorporates adaptive spectral-spatial fusion. Specifically, the proposed HyperTransXNet comprises three key modules: (1) a Hybrid Spatial-Spectral Module (HSSM) that captures the refined local spectral-spatial features and models global spectral correlations by combining depth-wise dynamic convolution with frequency-domain attention; (2) a Mixture-of-Experts Routing (MoE-R) module that adaptively fuses multi-scale features by dynamically selecting optimal experts via Top-K sparse weights; and (3) a Spatial-Spectral Tokens Enhancer (SSTE) module that ensures causality-preserving interactions between spectral bands and spatial contexts. Extensive experiments on the Indian Pines, Houston 2013, and WHU-Hi-LongKou datasets demonstrate the superiority of HyperTransXNet. Full article

Text-based person search (TPS), a critical technology for security and surveillance, aims to retrieve target individuals from image galleries using textual descriptions. The existing methods face two challenges: (1) ambiguous attribute–noun association (AANA), where syntactic ambiguities lead to incorrect associations between attributes and the intended nouns; and (2) textual noise and relevance imbalance (TNRI), where irrelevant or non-discriminative tokens (e.g., ‘wearing’) reduce the saliency of critical visual attributes in the textual description. To address these aspects, we propose the dependency-aware entity–attribute alignment network (DEAAN), a novel framework that explicitly tackles AANA through dependency-guided attention and TNRI via adaptive token filtering. The DEAAN introduces two modules: (1) dependency-assisted implicit reasoning (DAIR) to resolve AANA through syntactic parsing, and (2) relevance-adaptive token selection (RATS) to suppress TNRI by learning token saliency. Experiments on CUHK-PEDES, ICFG-PEDES, and RSTPReid demonstrate state-of-the-art performance, with the DEAAN achieving a Rank-1 accuracy of 76.71% and an mAP of 69.07% on CUHK-PEDES, surpassing RDE by 0.77% in Rank-1 and 1.51% in mAP. Ablation studies reveal that DAIR and RATS individually improve Rank-1 by 2.54% and 3.42%, while their combination elevates the performance by 6.35%, validating their synergy. This work bridges structured linguistic analysis with adaptive feature selection, demonstrating practical robustness in surveillance-oriented TPS scenarios. Full article

The recent economic turmoil and the increasing volatility of bitcoins have necessitated the need for exploring safe-haven assets for bitcoins. In this quest, the present study aims to investigate the safe haven for bitcoins by examining the dynamic relationship between bitcoins, gold, foreign exchange, and stablecoins. This is achieved by calculating hedge ratios and portfolio weight ratios for various asset classes, by employing adaptive-based techniques such as generalized orthogonal generalized autoregressive conditional heteroscedasticity, corrected dynamic conditional correlation, corrected asymmetric dynamic conditional correlation, and asymmetric dynamic conditional correlation under various market and time-varying conditions. The empirical estimate reveals that all the selected asset classes are effective risk diversifiers for bitcoins. However, among all the asset classes, as per the hedge and portfolio weight ratio, Japanese yen, stablecoin for Japanese yen and Great Britain Pound, and Crypto Holding Frank Token (lowest-cost hedging strategies) are the most effective risk diversifiers when compared with bitcoins. Moreover, while considering external economic shocks, the empirical estimate posits that stablecoins are more stable risk diversifiers compared to the asset class they represent. Furthermore, in terms of the bivariate portfolio analysis formed with bitcoin, this study concludes that the weight of bitcoin is more stable when combined with gold, tether gold, Euro, Great Britain Pound, Swiss franc, and Japanese Yen. Thus, these assets are attractive for long-term investment strategies. This study provides investors and policymakers with significant insight into understanding safe-haven assets for bitcoin’s volatility and constructing a flexible portfolio that is dependent on the investment timeline and the prevailing market conditions. Full article

Various areas of natural language processing (NLP) have greatly benefited from the development of large language models in recent years. This research addresses the challenge of developing efficient tokenizers for transformer-based domain-specific language models. Tokenization efficiency within transformer-based models is directly related to model efficiency, which motivated the research we present in this paper. Our goal in this research was to demonstrate that the appropriate selection of data used for tokenizer training has a significant impact on tokenizer performance. Subsequently, we will demonstrate that efficient tokenizers and models can be developed even if language resources are limited. To do so, we will present a domain-adapted large language model tokenizer developed for masked language modeling of the Serbian legal domain. In this paper, we will present a comparison of the tokenization performance for a domain-adapted tokenizer in version 2 of the SrBERTa language model we developed, against the performances of five other tokenizers belonging to state-of-the-art multilingual, Slavic or Serbian-specific models—XLM-RoBERTa (base-sized), BERTić, Jerteh-81, SrBERTa v1, NER4Legal_SRB. The comparison is performed using a test dataset consisting of 275,660 samples of legal texts written in the Cyrillic alphabet gathered from the Official Gazette of the Republic of Serbia. This dataset contains 197,134 distinct words, while the overall word count is 5,265,352. We will show that our tokenizer, trained upon a domain-adapted dataset, outperforms presented tokenizers by at least 4.5% ranging to 54.62%, regarding the number of tokens generated for the whole test dataset. In terms of tokenizer fertility, we will show that our tokenizer outperforms compared tokenizers by at least 6.39% ranging to 56.8%. Full article

Large-scale match data from esports games like League of Legends are stored in complex JSON files that often exceed the input token limitations of large language models (LLMs), restricting advanced analysis and applications such as automated commentary and strategic insight generation. This paper introduces the League of Legends Match Data Compactor (LoL-MDC), a tool designed to transform extensive match data into a concise and structured format optimized for LLM processing. By systematically summarizing structured match information—including match overviews, player and team statistics, timeline summaries, and algorithmically selected key events—the LoL-MDC significantly reduces the data size from approximately 80,000 tokens to under 2000 tokens while retaining analytical value. This method enables LLMs to generate coherent match summaries, analyze player performances, and identify key momentum shifts more effectively than processing raw JSON files. Additionally, the LoL-MDC integrates a winning probability metric to quantitatively enhance the selection of pivotal game events, ensuring relevance in esports analytics. Experimental evaluations demonstrate that the LoL-MDC improves data processing efficiency while maintaining critical insights. The proposed approach provides a structured and adaptable framework for applying LLMs to esports analytics and can be adapted to other competitive gaming environments, supporting AI-driven applications in match analysis, player performance evaluation, and strategic forecasting. Full article

Traditional methods for requirement identification depend on the manual transformation of unstructured requirement texts into formal documents, a process that is both inefficient and prone to errors. Although requirement knowledge graphs offer structured representations, current named entity recognition and relation extraction techniques continue to face significant challenges in processing the specialized terminology and intricate sentence structures characteristic of the aerospace domain. To overcome these limitations, this study introduces a novel approach for constructing aerospace-specific requirement knowledge graphs using a large language model. The method first employs the GPT model for data augmentation, followed by BERTScore filtering to ensure data quality and consistency. An efficient continual learning based on token index encoding is then implemented, guiding the model to focus on key information and enhancing domain adaptability through fine-tuning of the Qwen2.5 (7B) model. Furthermore, a chain-of-thought reasoning framework is established for improved entity and relation recognition, coupled with a dynamic few-shot learning strategy that selects examples adaptively based on input characteristics. Experimental results validate the effectiveness of the proposed method, achieving F1 scores of 88.75% in NER and 89.48% in relation extraction tasks. Full article