Search Results (316)

Optical remote sensing images are often partially obscured by clouds due to the inability of visible light to penetrate cloud cover, which significantly limits their subsequent applications. Most existing cloud removal methods formulate the problem using low-rank and sparse priors within a discrete representation framework. However, these approaches typically rely on manually designed regularization terms, which fail to accurately capture the complex geostructural patterns in remote sensing imagery. In response to this issue, we develop a continuous blind cloud removal model. Specifically, the cloud-free component is represented using a continuous tensor function that integrates implicit neural representations with low-rank tensor decomposition. This representation enables the model to capture both global correlations and local smoothness. Furthermore, a band-wise sparsity constraint is employed to represent the cloud component. To preserve the information in regions not covered by clouds during reconstruction, a box constraint is incorporated. In this constraint, cloud detection is performed using an adaptive thresholding strategy, and a morphological erosion function is employed to ensure accurate detection of cloud boundaries. To efficiently handle the developed model, we formulate an alternating minimization algorithm that decouples the optimization into three interpretable subproblems: cloud-free reconstruction, cloud component estimation, and cloud detection. Our extensive evaluations on both synthetic and real-world data reveal that the proposed method performs competitively against state-of-the-art cloud removal methods. Full article

This study designed and empirically evaluated a method to enhance information accessibility for museum and art gallery visitors using a small Large Language Model (sLLM) based on the Retrieval-Augmented Generation (RAG) framework. Over 199,000 exhibition descriptions were collected and refined, and a question-answering dataset consisting of 102,000 pairs reflecting user personas was constructed to develop DocentGemma, a domain-optimized language model. This model was fine-tuned through Low-Rank Adaptation (LoRA) based on Google’s Gemma2-9B and integrated with FAISS and OpenSearch-based document retrieval systems within the LangChain framework. Performance evaluation was conducted using a dedicated Q&A benchmark for the docent domain, comparing the model against five commercial and open-source LLMs (including GPT-3.5 Turbo, LLaMA3.3-70B, and Gemma2-9B). DocentGemma achieved an accuracy of 85.55% and a perplexity of 3.78, demonstrating competitive performance in language generation and response accuracy within the domain-specific context. To enhance retrieval relevance, a Spatio-Contextual Retriever (SC-Retriever) was introduced, which combines semantic similarity and spatial proximity based on the user’s query and location. An ablation study confirmed that integrating both modalities improved retrieval quality, with the SC-Retriever achieving a recall@1 of 53.45% and a Mean Reciprocal Rank (MRR) of 68.12, representing a 17.5 20% gain in search accuracy compared to baseline models such as GTE and SpatialNN. System performance was further validated through field deployment at three major exhibition venues in Seoul (the Seoul History Museum, the Hwan-ki Museum, and the Hanseong Baekje Museum). A user test involving 110 participants indicated high response credibility and an average satisfaction score of 4.24. To ensure accessibility, the system supports various output formats, including multilingual speech and subtitles. This work illustrates a practical application of integrating LLM-based conversational capabilities into traditional docent services and suggests potential for further development toward location-aware interactive systems and AI-driven cultural content services. Full article

Individual cow identification is a prerequisite for automatically monitoring behavior patterns, health status, and growth data of each cow, and can provide the assistance in selecting excellent cow individuals for breeding. Despite high recognition accuracy, traditional implantable electronic devices such as RFID (i.e., Radio Frequency Identification) can cause some degree of harm or stress reactions to cows. Image-based methods are widely used due to their non-invasive advantages, but these methods have poor adaptability to different environments and target size, and low detection accuracy in complex scenes. To solve these issues, this study designs a Dy_Conv (i.e., dynamic convolution) module and innovatively constructs a Dynamic_Bottleneck module based on the Dy_Conv and S2Attention (Sparse-shift Attention) mechanism. On this basis, we replaces the first and fourth bottleneck layers of Resnet50 with the Dynamic_Bottleneck to achieve accurate extraction of local features and global information of cows. Furthermore, the QAConv (i.e., query adaptive convolution) module is introduced into the front end of the backbone network, and can adjust the parameters and sizes of convolution kernels to adapt to the scale changes in cow targets and input images. At the same time, NAM (i.e., normalization-based attention module) attention is embedded into the backend of the network to achieve the feature fusion in the channels and spatial dimensions, which contributes to better distinguish visually similar individual cows. The experiments are conducted on the public datasets collected from different cowsheds. The experimental results showed that the Rank-1, Rank-5, and mAP metrics reached 96.8%, 98.9%, and 95.3%, respectively. Therefore, the proposed model can effectively capture and integrate multi-scale features of cow body appearance, enhancing the accuracy of individual cow identification in complex scenes. Full article

Miao embroidery holds significant cultural, economic, and aesthetic value. However, its transmission faces numerous challenges: a high learning threshold, a lack of interest among younger generations, and low production efficiency. These factors have created obstacles to its sustainable development. In the age of artificial intelligence (AI), generative AI is expected to improve the efficiency of pattern innovation and the adaptability of the embroidery industry. Therefore, this study proposes a Miao embroidery pattern generation and application method based on Stable Diffusion and low-rank adaptation (LoRA) fine-tuning. The process includes image preprocessing, data labeling, model training, pattern generation, and embroidery production. Combining objective indicators with subjective expert review, supplemented by feedback from local artisans, we systematically evaluated five representative Miao embroidery styles, focusing on generation quality and their social and business impact. The results demonstrate that the proposed model outperforms the original diffusion model in terms of pattern quality and style consistency, with optimal results obtained under a LoRA scale of 0.8–1.2 and diffusion steps of 20–40. Generated patterns were parameterized and successfully implemented in digital embroidery. This method uses AI technology to lower the skill threshold for embroidery training. Combined with digital embroidery machines, it reduces production costs, significantly improving productivity and increasing the income of embroiderers. This promotes broader participation in embroidery practice and supports the sustainable inheritance of Miao embroidery. It also provides a replicable technical path for the intelligent generation and sustainable design of intangible cultural heritage (ICH). Full article

This study proposes a large language model, Corr-Lora-RAG, designed to address the complexity and uncertainty inherent in corrosion data. A dedicated corrosion knowledge database (CKD) was constructed, and dataset generation code was provided to enhance the model’s reproducibility and adaptability. Based on the Qwen2.5-7B model, the Corr-Lora model was developed by integrating prompt engineering and low-rank adaptation (LoRA) supervised fine-tuning (SFT) techniques, thereby improving the understanding and expression of domain-specific knowledge in the field of corrosion. Furthermore, the Corr-Lora-RAG model was built using retrieval-augmented generation (RAG) technology, enabling dynamic access to external knowledge. Experimental results demonstrate that the proposed model outperforms baseline models in terms of accuracy, completeness, and domain relevance, and exhibits knowledge generation capabilities comparable to those of large-scale language models under limited computational resources. This approach provides an intelligent solution for corrosion risk assessment, standards compliance analysis, and protective strategy formulation, and offers a valuable reference for the development of specialized language models in other engineering fields. Full article

With the growing demand for multilingual financial information, cross-lingual financial sentiment recognition faces significant challenges, including semantic misalignment, ambiguous sentiment expression, and insufficient transferability. To address these issues, a unified multilingual recognition framework is proposed, integrating semantic contrastive learning with a language-adaptive modulation mechanism. This approach is built upon the XLM-R multilingual model and employs a semantic contrastive module to enhance cross-lingual semantic consistency. In addition, a language modulation module based on low-rank parameter injection is introduced to improve the model’s sensitivity to fine-grained emotional features in low-resource languages such as Chinese and French. Experiments were conducted on a constructed trilingual financial sentiment dataset encompassing English, Chinese, and French. The results demonstrate that the proposed model significantly outperforms existing methods in cross-lingual sentiment recognition tasks. Specifically, in the English-to-French transfer setting, the model achieved 73.6% in accuracy, 69.8% in F1-Macro, 72.4% in F1-Weighted, and a cross-lingual generalization score of 0.654. Further improvements were observed under multilingual joint training, reaching 77.3%, 73.6%, 76.1%, and 0.696, respectively. In overall comparisons, the proposed model attained the highest performance across cross-lingual scenarios, with 75.8% in accuracy, 72.3% in F1-Macro, and 74.7% in F1-Weighted, surpassing strong baselines such as XLM-R+SimCSE and LaBSE. These results highlight the model’s superior capability in semantic alignment and generalization across languages. The proposed framework demonstrates strong applicability and promising potential in multilingual financial sentiment analysis, public opinion monitoring, and multilingual risk modeling. Full article

Hyperspectral image (HSI) denoising is an important preprocessing step for downstream applications. Fully characterizing the spatial-spectral priors of HSI is crucial for denoising tasks. In recent years, denoising methods based on low-rank subspaces have garnered widespread attention. In the low-rank matrix factorization framework, the restoration of HSI can be formulated as a task of recovering two subspace factors. However, hyperspectral images are inherently three-dimensional tensors, and transforming the tensor into a matrix for operations inevitably disrupts the spatial structure of the data. To address this issue and better capture the spatial-spectral priors of HSI, this paper proposes a modeling approach named low-rank Tucker decomposition with subspace implicit neural representation (LRTSINR). This data-driven and model-driven joint modeling mechanism has the following two advantages: (1) Tucker decomposition allows for the characterization of the low-rank properties across multiple dimensions of the HSI, leading to a more accurate representation of spectral priors; (2) Implicit neural representation enables the adaptive and precise characterization of the subspace factor continuity under Tucker decomposition. Extensive experiments demonstrate that our method outperforms a series of competing methods. Full article

Generative model-assisted design has become a trend, providing a new paradigm for knitted fabric image generation. The FLUX diffusion model was chosen to generate images in this study and was compared to other generative models. In order to effectively apply the large model to specialized verticals, an efficient fine-tuning method, low-rank adaptation, was used. Experiments showed that the method allows a pre-trained model to stably generate knitted fabric images in batches through easily understandable text prompts. The generated images have clear textures and correct structures, and can display the surface characteristics of knitted fabrics generated by using different yarn specifications and yarn bristles. Moreover, the unit tissue structural similarity index measure (SSIM) is 0.6528, which is very similar to real fabrics. This research expands the application of fabric generation in the field of deep learning. This method is highly efficient, low-cost, and capable of stably simulating knitted fabrics, which can be used to rapidly expand the image design materials of knitted fabrics. Full article

Smart contracts have become integral to decentralized applications, yet their programmability introduces critical security risks, exemplified by high-profile exploits such as the DAO and Parity Wallet incidents. Existing vulnerability detection methods, including static and dynamic analysis, as well as machine learning-based approaches, often struggle with emerging threats and rely heavily on large, labeled datasets. This study investigates the effectiveness of open-source, lightweight large language models (LLMs) fine-tuned using parameter-efficient techniques, including Quantized Low-Rank Adaptation (QLoRA), for smart contract vulnerability detection. We introduce the EVuLLM dataset to address the scarcity of diverse evaluation resources and demonstrate that our fine-tuned models achieve up to 94.78% accuracy, surpassing the performance of larger proprietary models, while significantly reducing computational requirements. Moreover, we emphasize the advantages of lightweight models deployable on local hardware, such as enhanced data privacy, reduced reliance on internet connectivity, lower infrastructure costs, and improved control over model behavior, factors that are especially critical in security-sensitive blockchain applications. We also explore Retrieval-Augmented Generation (RAG) as a complementary strategy, achieving competitive results with minimal training. Our findings highlight the practicality of using locally hosted LLMs for secure, efficient, and reproducible smart contract analysis, paving the way for broader adoption of AI-driven security in blockchain ecosystems. Full article

With the increasing demand for high integration, low cost, and large capacities in satellite systems, integrating the antenna and microwave component into the same system has become appealing to the satellite engineer. The dual C-type gold ribbon, performing as the key electromagnetic signal bridge between the microwave component and the antenna, has a significant impact on the electrical performance of satellite antennas. However, during its manufacturing and operating, the interconnection geometry undergoes deformation due to mounting errors and environmental loads. Consequently, these parasitic geometry parameters can significantly increase energy loss during the signal transmission. To address this issue, this paper has proposed a method for determining the design range of the geometrical parameters of the dual C-type gold ribbon, and applied it to the performance prediction of the microstrip antennas and the parameter optimization of the gold ribbon. In this study, a mechanical response analysis of the antennas in the operating environment has been carried out and the manufacturing disturbance has been considered to calculate the geometry fluctuation range. Then, the significance ranking of the geometry parameters has been determined and the key parameters have been selected. Finally, the chaos feedback adaptive whale optimization algorithm–back propagation neural network has been used as a surrogate model to establish the relationship between the geometry parameters and the antenna electromagnetic performance, and the multi-objective red-billed blue magpie optimization algorithm has been combined with the surrogate model to optimize the configuration parameters. This paper provides theoretical guidance for the interconnection geometry design and the optimization of the integration module of the antennas and microwave components. Full article

Multimodal knowledge graphs have recently been successfully applied to tasks such as those relating to information retrieval, question and answer, and recommender systems. In this study, we propose a dual-path fine-tuning mechanism technique with a low-rank adapter and an embedded cueing layer, aiming to improve the generalization and accuracy of the model in analogical reasoning tasks. The low-rank fine-tuning (LoRA) technique with rank-stable scaling factor is used to fine-tune the MKGformer model, and a cue-embedding layer is innovatively added to the input layer, which enables the model to better grasp the scale of the relationship between entities according to the dynamic cue vectors during the fine-tuning process and ensures that the model achieves the best results during training. The experimental results show that the R-MKG model improves several evaluation indexes by more than 20%, which is significantly better than the traditional DoRA and FA-LoRA methods. This research provides technical support for multimodal knowledge graph analogical reasoning. We hope that our work will bring benefits and inspire future research. Full article

In an increasingly complex global context, higher education faces the challenge of preparing professionals who are innovative, committed, and socially responsible. Entrepreneurial competence is particularly prominent among the key skills required to meet this goal, given its significant personal and social impact. This study examines how final-year undergraduate students at the University of Deusto (Spain) perceive the importance of entrepreneurial competence—defined as a set of transversal skills, knowledge, and attitudes enabling initiative and opportunity recognition across various contexts—rather than entrepreneurial competence strictly understood as business creation. The sample included 267 students from different faculties. Descriptive, comparative, and ordinal logistic regression analyses (SPSS) were used. The results show that, while entrepreneurial competence was given significant importance, it was ranked comparatively low relative to other competencies. Significant differences by gender were observed, with women rating entrepreneurial competence more highly than men. The faculty variable showed slight disparities, and there were no relevant differences between campuses. These findings highlight the need to reinforce the integration of entrepreneurial competence into educational curricula on a transversal basis, adapting the teaching of this competence to the sociocultural context of students, as well as the need to increase students’ awareness of the importance of entrepreneurial competence. It is proposed that further research should focus on the relationships between intrapreneurship, gender, and academic disciplines, in order to enrich entrepreneurial competence education and its impact on the employability and social commitment of students. Full article

In clinical decision support systems (CDSSs), where accurate classification of drug–drug interactions (DDIs) can directly affect treatment safety and outcomes, identifying drug interactions is a major challenge, introducing a scalable approach for classifying DDIs utilizing a finely-tuned biomedical language model. The method shown here uses BiomedBERT, a domain-specific version of bidirectional encoder representations from transformers (BERT) that was pre-trained on biomedical literature, to reduce the number of resources needed during fine-tuning. Low-rank adaptation (LoRA) was used to fine-tune the model on the DrugBank dataset. The objective was to classify DDIs into two clinically distinct categories, that is, synergistic and antagonistic interactions. A pseudo-labeling strategy was created to deal with the problem of not having enough labeled data. A curated ground-truth dataset was constructed using polarity-labeled interaction entries from DrugComb and verified DrugBank antagonism pairs. The fine-tuned model is used to figure out what kinds of interactions there are in the rest of the unlabeled data. A checkpointing system saves predictions and confidence scores in small pieces, which means that the process can be continued and is not affected by system crashes. The framework is designed to log every prediction it makes, allowing results to be refined later, either manually or through automated updates, without discarding low-confidence cases, as traditional threshold-based methods often do. The method keeps a record of every output it generates, making it easier to revisit earlier predictions, either by experts or with improved tools, without depending on preset confidence cutoffs. It was built with efficiency in mind, so it can handle large amounts of biomedical text without heavy computational demands. Rather than focusing on model novelty, this research demonstrates how existing biomedical transformers can be adapted to polarity-aware DDI classification with minimal computational overhead, emphasizing deployment feasibility and clinical relevance. Full article

Recent advancements in generative artificial intelligence have significantly progressed the field of text-to-music generation, enabling users to create music from natural language descriptions. Despite the success of various models, such as MusicLM, MusicGen, and AudioLDM, the current approaches struggle to capture fine-grained genre-specific characteristics, precisely control musical attributes, and handle underrepresented cultural data. This paper introduces a novel, lightweight fine-tuning method for the AudioLDM framework using low-rank adaptation (LoRA). By updating only selected attention and projection layers, the proposed method enables efficient adaptation to musical genres with limited data and computational cost. The proposed method enhances controllability over key musical parameters such as rhythm, emotion, and timbre. At the same time, it maintains the overall quality of music generation. This paper represents the first application of LoRA in AudioLDM, offering a scalable solution for fine-grained, genre-aware music generation and customization. The experimental results demonstrate that the proposed method improves the semantic alignment and statistical similarity compared with the baseline. The contrastive language–audio pretraining score increased by 0.0498, indicating enhanced text-music consistency. The kernel audio distance score decreased by 0.8349, reflecting improved similarity to real music distributions. The mean opinion score ranged from 3.5 to 3.8, confirming the perceptual quality of the generated music. Full article

Landslide displacement prediction is crucial for disaster mitigation, yet traditional methods often fail to capture the complex, non-stationary spatiotemporal dynamics of slope evolution. This study introduces an enhanced prediction framework that integrates multi-scale signal processing with dynamic, geology-aware graph modeling. The proposed methodology first employs the Maximum Overlap Discrete Wavelet Transform (MODWT) to denoise raw Global Navigation Satellite System (GNSS)-monitored displacement time series data, enhancing the underlying deformation features. Subsequently, a geology-aware graph is constructed, using the temporal correlation of displacement series as a practical proxy for physical relatedness between monitoring nodes. The framework’s core innovation lies in a dynamic graph optimization model with low-rank constraints, which adaptively refines the graph topology to reflect time-varying inter-sensor dependencies driven by factors like mining activities. Experiments conducted on a real-world dataset from an active open-pit mine demonstrate the framework’s superior performance. The DCRNN-proposed model achieved the highest accuracy among eight competing models, recording a Root Mean Square Error (RMSE) of 2.773 mm in the Vertical direction, a 39.1% reduction compared to its baseline. This study validates that the proposed dynamic graph optimization approach provides a robust and significantly more accurate solution for landslide prediction in complex, real-world engineering environments. Full article