In the world of engineering equipment, customized parts possess enormous size spans and intricate designs. Current research is insufficient in terms of adsorption score and time consumption. This study introduces a novel approach leveraging neural network acceleration and the Grey Wolf Optimization (GWO) algorithm to tackle these challenges. First, the network model and the synthesized part image are combined to classify the part’s shape. Then, the improved GWO algorithm will search for the adsorption pose for parts that are categorized as irregular kinds. Finally, in order to accelerate the search process, the 1024 vectors of part images and the parallel optimal poses are saved for subsequent retrieval. Numerous experiments demonstrate that the network model demonstrates high accuracy, approaching 100%, and its judgment time is 285 times lower than the traditional method. In addition, compared with the current SOTA research, our adsorption score improves 4.68%, which fully demonstrates the advantage of our method in industrial production. This is the first report that a deep neural network is used to accelerate the adsorption of industrial parts well.

Translational development of somatic cell genome editing requires monitoring the extent of editing in the body at a given time, the specificity of editing, durability, and the potential for adverse events. A noninvasive approach that can identify edited cells invivo is beneficial for addressing these and related questions. These studies used total-body positron emission tomography (PET) to identify somatic cell gene editing invivo in fetal rhesus macaques. Rhesus dams were screened to confirm they were seronegative for AAV serotype 9 and SaCas9 antibodies, then selected for the study. Fetuses were administered a dual imaging vector (AAV9/SaCas9 and AAV9/PCSK9 gRNA/HSV-sr39TK) in utero using an ultrasound-guided fetal intrahepatic approach in the second trimester (1012 vector genomes/fetus). After maternal intravenous administration of 9-(4-(18)F-Fluoro-3-[hydroxymethyl]butyl)guanine ((18)F-FHBG) (∼3 mCi/kg), PET imaging was performed in the second and third trimesters. PET imaging provided evidence of editing in the fetal liver, which was sustained. Appropriate insertion of the promoterless HSV-sr39TK reporter in frame with the PCSK9 gene was confirmed in the fetal liver near term using RNA sequencing, and correctly targeted insertions were observed. These studies have shown that total-body PET can provide insights into gene-edited somatic cells in utero and without evidence of adverse effects.

<sec><title>Objective</title> As one of the basic well-beings and primary needs for the survival and development of mankind, health is closely influenced by a variety of multidimensional and complex influencing factors. Leisure-time physical activity (LTPA) is a physical activity of large flexibility in such factors as activity strength, duration, location selection and others, and together with the outstanding regulation effect on the physiological and psychological aspects, has seen a continuous and gradually expanded role in daily life. It is therefore regarded as the most promising type of changeable physical activities. Urban parks as a spatial store of comprehensive health resources have been rooted in green space system. They are combined with various service resources, providing a physical environment for LTPA. However, the majority of current research on LTPA predominantly concentrates on the direct relationship between the environment and behavior. Visual and auditory cues are typically presented in a parallel manner, lacking joint modeling at the same spatiotemporal location. Emotional perception, as a crucial psychological pivot through which environmental cues influence behavior, has not undergone systematic scrutiny. This has led to difficulties in the precise implementation of relevant design strategies. In the context of park settings, visual and auditory cues frequently co-occur at the same spatiotemporal point. They may influence an individual's subjective evaluations of safety, pleasure, and vitality through emotional regulation, thereby affecting their activity intensity and inclination to remain in the area. Consequently, there is an urgent need to conduct empirical research using the framework of "audio-visual environmental elements−emotional perception−leisure physical activities". </sec><sec><title>Methods</title> In this paper, 144 landscape nodes in Taizi River Park, Liaoyang City, Liaoning Province were classified meticulously. In terms of visual object, the image acquisition method is conducted on a sunny day, and the circumferential three pictures are shot for each landscape node. ArcGIS 10.8 software was applied to match the image with GPS spatial data. The mask2Former semantic segmentation model was applied to the image data. According to Mapillary Vistas dataset, leisure-time physical activity related image vectors were embedded to build customized model according to the research purpose. The acquired image data were analyzed by this method. Color features (color saturation, color richness, and color harmony) of obtained image were analyzed using K-means algorithm. A “human-machine duel” score sheet was used for the emotional evaluation of the environment. For the acoustics, the environmental sound cues and the source category for the audio part were noted according to the ISO/TS <styled-content style-type="number">12913</styled-content>-2 standard and the soundscape screening of the Swedish soundscape quality scale protocol. Combining above-mentioned audiovisual data with other methods including questionnaire surveys and behavioral observation data, multiple stepwise regression model and mediation effect analysis were carried out to investigating the direct and mediating relationships among environmental elements of urban park, visitors’ emotional states and leisure physical activities. </sec><sec><title>Results</title> Through the analysis of the audio-visual environment, emotional evaluation, and behavioral observation data collected in the field, this study has yielded a series of crucial findings. Visual and auditory elements within urban parks, including the blue view ratio, sky openness, natural sounds, and light guidance, can significantly enhance visitors' positive emotional experiences (such as a sense of security, vitality, and fulfillment). This, in turn, further increases the likelihood of light-intensity and moderate-intensity LTPA. Emotional perception exerted a partial mediating effect along multiple pathways, providing quantitative evidence for the "environment−emotion−behavior" framework. Specifically, sky openness and natural sound sources influenced light LTPA through emotional perception, while spatial enclosure affected moderate LTPA. These results suggest that an open skyline and a favorable water-related environment are conducive to eliciting positive emotions and promoting gentle physical activities. Moreover, moderately enclosed green spaces significantly facilitate moderate activities by enhancing the sense of security. In contrast, no significant emotional mediation pathway was identified for high-intensity physical activities. This might be attributed to the fact that high-intensity activities are more goal-driven and performance-oriented. Environmental factors primarily act on such activities by directly influencing aspects such as safety and convenience, rather than indirectly through emotions. </sec><sec><title>Conclusion</title> This research delved into the interrelationship of "environment−emotion−behavior" within the context of urban parks, quantitatively validating the pivotal mediating role of emotional perception in the process where the audiovisual environment influences LTPA. This has advanced the comprehension of the mechanisms through which environmental elements facilitate behaviors. On a practical level, the research findings offer significant implications for enhancing the health-promoting benefits of urban parks. Looking ahead, the analytical framework and optimization strategies established in this study can serve as a theoretical foundation and practical reference for related disciplines. They can also drive the research agenda of integrating emotional perception into environmental interventions to promote behavioral change, thereby providing a scientific underpinning for the creation of healthy cities and active spaces. </sec>

Multimodal imaging using hybrid imaging agents is a promising strategy for diagnosing and evaluating pathologies after image-guided surgical interventions. Combining optical and radioactive imaging techniques provides a comprehensive approach to monitoring and diagnosing infections, which would be more effective than routine nuclear clinical tracers for SPECT or PET imaging, thereby enabling more effective treatment as in image-guided surgery. This review summarizes the latest developments in hybrid imaging agents and vectors for radioactive and optical imaging of bacterial, fungal, and viral infections. We pinpoint the pitfalls in the current preclinical landscape for developing infection imaging tracers. Besides diagnosing and tracking pathogens, the role of optical imaging in diagnosing and aiding antimicrobial interventions, including image-guided surgery, is discussed. Finally, practical considerations are addressed for multimodal workflow challenges in preclinical infection imaging with hybrid tracers.

Abstract Uterine fibroids are commonly benign but significantly impact women’s health due to undiagnosed asymptomatic progression and lack of awareness. While hysterectomy and myomectomy remain the primary treatments for uterine fibroids, patients need non-surgical alternatives in overall fibroid management. Modern scientific methods such as Artificial Intelligence (AI)-driven diagnostics combined with ancient healing techniques can provide alternative curing methods for affected patients. In this review, we have highlighted three key non-surgical approaches along with molecular basis of uterine fibroid development for futuristic personalized therapies. The non-surgical methods include (i) AI driven diagnostics using 3D Super-Resolution Diffusion-Weighted Imaging (SR-DWI) and support vector machines (SVMs), (ii) Phytochemical-based therapeutics such as curcumin ( Curcuma longa ), resveratrol ( Vitis vinifera ), and epigallocatechin gallate (EGCG, Camellia sinensis ), (iii) Physical therapies like Yoga and acupuncture. Advanced imaging techniques improve diagnostic accuracy which allows timely intervention. Phytochemicals regulate fibrotic pathways, while yoga and acupuncture alleviate stress and support uterine health, providing a comprehensive non-surgical treatment approach. Additionally, we also describe the molecular basis for uterine fibroid development in terms of germline mutations, epigenetics and signalling pathways which are well correlated with our proposed phytochemical and physical therapies. Overall, we emphasize that the AI driven diagnostics combined with ancient therapeutic approaches offer an adjunctive strategy for effective fibroid management thereby potentially reducing the need for hysterectomy.

_ The landscape of reserves management continues to evolve rapidly, shaped by advances in digital technologies, changing regulatory frameworks, and strategic financial priorities across the upstream and midstream sectors. The papers featured here highlight how artificial intelligence (AI), carbon management, and risk analytics are redefining the way companies assess and manage their assets in increasingly complex environments. Three key themes emerge from the selected works. Expanding AI Applications. Machine learning is being embedded across workflows—from optimizing field operations in real time to identifying anomalies in financial statements. This broader application of AI and data science strengthens both reserves assurance and corporate governance. AI-driven asset optimization, in particular, is proving transformative in balancing capital efficiency with operational reliability. Shift Toward Probabilistic Project Management. A notable transition is underway from deterministic planning to probabilistic modeling. By integrating cost and schedule risk, operators can better quantify potential delays and their financial implications, an essential capability for resilient reserves planning amid market volatility. Subsurface Storage and Long-Term Strategy Integration. The emergence of carbon capture and storage (CCS) introduces new technical and regulatory dimensions to reserves management. The papers discuss evolving frameworks that link CCS deployment to corporate sustainability targets and financing strategies, reflecting the growing importance of aligning reserves practices with environmental, social, and governance commitments and investor expectations. Collectively, these trends signal a more mature and interconnected discipline, one that is multidisciplinary, data-driven, and increasingly responsive to both technical realities and societal imperatives shaping the future of energy. Summarized Papers in This December 2025 Issue SPE 221836 - Integrated Cost and Schedule Risk Analysis Eases Consequences of Project Delays by Christopher Britton, SPE, OMV. SPE 222600 - Image Hashing and Vector Databases Improve Detection of Fraud in Financial Statements by Dalia Albuqaytah, Sarafudheen Tharayil, and Muhammad Azmi Idris, Saudi Aramco. SPE 225882 - Independent Study Addresses Causes of Challenges in Asia Pacific Flagship CCS Project by Ali Sabzabadi, SPE, Ian Gladman, and Bill Billingsley, RISC Advisory. Recommended Additional Reading at OnePetro: www.onepetro.org. SPE 225849 - CCS in the State of Alaska—Regulatory Framework and Commercial Selection Criteria for Transoceanic CO2 Imports by N. Fulford, Gaffney, Cline, and Associates, et al. SPE 225371 - AI-Enabled Integrated Asset Optimization for CAPEX and OPEX by Fernando Gutierrez, Tachyus, et al. SPE 225633 - Financing Strategies for Oil and Gas Companies Pursuing Long-Term Energy Transition Under ESG Regulations by Y. Akin, Turkish Petroleum

Abstract The accelerating demands of Artificial Intelligence (AI) Dynamic Random‐Access Memory and the Internet of Things (IoT) applications call for next‐generation non‐volatile memory devices with high speed and nanoscale integration. BiFeO 3 (BFO) ferroelectric thin films are promising candidates for beyond‐von Neumann architectures but suffer from intrinsic defects, particularly oxygen vacancies, which degrade polarization stability and increase the leakage current. Herein, we report an atomic‐scale lattice engineering approach based on precisely controlled Fe 3+ ‐ion implantation to modulate the [FeO 6 ] octahedral framework in BFO. Atomic‐resolution scanning transmission electron microscopy imaging and polarization vector mapping reveal that optimized implantation induces controlled in‐plane compressive strain, which systematically modifies octahedral tilt patterns. This strain‐mediated distortion simultaneously suppresses oxygen vacancies, enhances Fe 3+ –O 2− orbital hybridization, and facilitates the formation of ordered defect–dipole arrays. Consequently, the implanted films demonstrate a 50% enhancement in remanent polarization along with a 10‐fold reduction in leakage current. By establishing a quantitative relationship between implantation dose, octahedral distortion, and ferroelectric response, this work provides a predictive model for strain–defect–dipole engineering in perovskite oxides, offering a generalizable strategy for optimizing performance in oxide nanoelectronics.

Vector-based image search, which represents images and search queries as vectors in a high-dimensional space, performs image search by identifying images whose vector representations closely match the vector representation of the query. Search engine optimization has been traditionally achieved by using relevant keywords to increase organic search engine discoverability. For images to be surfaced as a result of a search query, these relevant keywords have been inserted into the page that hosts the image or embedded directly into an image as metadata that can be decoded and indexed by a text-based search engine. However, when a search engine uses vectors to operate, the static nature of an image's vector representation prevents adaptation of content to increase discoverability. In this paper, we present a novel approach to alter an image in a way that enhances its discoverability with a known vector-based search engine while minimizing the visual impact of these alterations for the average user. The alteration is guided by the maximization of intended search queries and the minimization of unwanted search queries.

A Formalization of Image Vectorization by Region Merging

This study assesses and contrasts the efficacy of raw picture pixels and image vectors as features in face expression classification. The CKPLUS dataset is utilized, and the issue of class imbalance is tackled by data augmentation. The dataset is partitioned into a 70% training set and 30% validation set. The training set consists of 175 images for each class, while the validation set consists of 75 images. The features are displayed using Matplotlib for raw pixels and t-SNE for vector features, then categorized using Random Forest and CNN classifiers. The performance is evaluated by utilizing confusion matrices, accuracy, precision, recall, and F1-score. The findings indicate that the Random Forest algorithm, when combined with vector features, obtains the maximum level of accuracy (99.6190%). Additionally, CNNs using raw pixel features also demonstrate strong performance. The precision, recall, and F1-scores exhibit similarity among the different approaches, with Random Forest (vector feature) and 2D CNN (raw pixels) showing somewhat better performance compared to other methods. These findings suggest that vector features have superior performance when used in conjunction with Random Forest, whereas raw pixel features are more successful when utilized with CNN.

Sunrise iii is a balloon-borne solar observatory dedicated to investigating the physics governing the magnetism and dynamics in the lower solar atmosphere. The observatory is designed to operate in the stratosphere, at heights around 36 km (above 99% of Earth’s atmosphere), to avoid image degradation due to turbulence in the Earth’s lower atmosphere, to gain access to the NUV wavelengths down to 309 nm, and to enable (when flown during summer solstice) observing the Sun uninterruptedly 24 hours/day. It is composed of a balloon gondola (equivalent to a spacecraft bus) carrying a 1-m aperture telescope (the largest solar telescope to-date to fly in the stratosphere on a balloon) feeding an imaging vector magnetograph and two spectropolarimeters aiming at acquiring high spatial resolution high cadence time series maps of the solar vector magnetic fields, plasma flows, and temperature in the photosphere and chromosphere. In July 2024 Sunrise iii successfully completed a six and a half days long stratospheric flight from Kiruna (Sweden) to Northern Canada at an average altitude of 36 km. This was the third successful flight of the Sunrise observatory, which had previously flown in 2009 and 2013. For this flight it was upgraded substantially with a new and improved suite of three instruments carried by a completely new gondola with upgraded pointing control system. This article focuses on describing the design and flight performance of the Sunrise iii gondola and all its subsystems. It describes the gondola mechanical structure, its power system, its command and control system, and in particular its pointing control system which was key for achieving high spatial and spectral resolution images of the solar photosphere and chromosphere by the three instruments.

Sediment accumulation in irrigation channels poses a significant challenge to water resource management, impacting hydraulic efficiency and agricultural sustainability. This study introduces an innovative multidisciplinary framework that integrates advanced image analysis (FIJI/ImageJ 1.54p), statistical validation (RStudio), and vector field modeling with a novel Sinusoidal Morphodynamic Bedload Transport Equation (SMBTE) to predict sediment deposition patterns with high precision. Conducted along the Malacatos River in La Tebaida Linear Park, Loja, Ecuador, the research captured a natural sediment transport event under controlled flow conditions, transitioning from pressurized pipe flow to free-surface flow. Observed sediment deposition reduced the hydraulic cross-section by approximately 5 cm, notably altering flow dynamics and water distribution. The final SMBTE model (Model 8) demonstrated exceptional predictive accuracy, achieving RMSE: 0.0108, R2: 0.8689, NSE: 0.8689, MAE: 0.0093, and a correlation coefficient exceeding 0.93. Complementary analyses, including heatmaps, histograms, and vector fields, revealed spatial heterogeneity, local gradients, and oscillatory trends in sediment distribution. These tools identified high-concentration sediment zones and quantified variability, providing actionable insights for optimizing canal design, maintenance schedules, and sediment control strategies. By leveraging open-source software and real-world validation, this methodology offers a scalable, replicable framework applicable to diverse water conveyance systems. The study advances understanding of sediment dynamics under subcritical (Fr ≈ 0.07) and turbulent flow conditions (Re ≈ 41,000), contributing to improved irrigation efficiency, system resilience, and sustainable water management. This research establishes a robust foundation for future advancements in sediment transport modeling and hydrological engineering, addressing critical challenges in agricultural water systems.

ABSTRACT Medical images play a crucial role in diagnosis, treatment procedures and overall healthcare. Nevertheless, they also pose substantial risks to patient confidentiality and safety. Safeguarding the confidentiality of patients' data has become an urgent and practical concern. We present a novel approach for reversible data hiding for colour medical images. In a hybrid domain, we employ AlexNet, tuned with watershed transform (WST) and L‐shaped fractal Tromino encryption. Our approach commences by constructing the host image's feature vector using a pre‐trained AlexNet model. Next, we use the watershed transform to convert the extracted feature vector into a vector for a topographic map, which we then encrypt using an L‐shaped fractal Tromino cryptosystem. We embed the secret image in the transformed image vector using a histogram‐based embedding strategy to enhance payload and visual fidelity. When there are no attacks, the RDHNet exhibits robust performance, can be reversed to the original image and maintains a visually appealing stego image, with an average PSNR of 73.14 dB, an SSIM of 0.9999 and perfect values of NC = 1 and BER = 0 under normal conditions. The proposed RDHNet demonstrates a robust ability to withstand detrimental geometric and noise‐adding attacks as well as various steganalysis methods. Furthermore, our RDHNet method initiative demonstrates efficacy in tackling contemporary confidentiality issues.

Watermarking is the process of embedding and extracting a watermark design on a digital cover to prove the image’s copyright or ownership, thereby securing the image’s authenticity. The proposed method in this paper uses a combination of honey encryption and reversible cellular automata for image watermarking. This method has two main phases: In the first phase, first, the initial matrix of the image is converted to a vector form. Then, the image vector is initially diffused using the XOR operator. After that, the initial key space is created in the context of honey encryption. Subsequently, the diffusion matrix transformation function is applied according to honey encryption and the key. Finally, the reversible cellular automata transformation is performed on the encrypted matrix. In the second phase, the matrix resulting from the previous step is stored in the cover image. For this purpose, the discrete wavelet transform is used to perform watermarking without changing the visual information of the image. This method has been able to minimize the changes in the cover image information and maximize the level of confidentiality of the information. The results demonstrate that the proposed method significantly outperforms the compared methods in terms of imperceptibility, achieving a significantly lower mean squared error of 13.55 and mean absolute error of 3.05, and a higher peak signal-to-noise ratio of 36.89. Furthermore, normalized correlation analysis of the proposed method exhibits its higher robustness against various attacks, including noise and JPEG compression than the compared approaches.

Recently, text-guided scalable vector graphics (SVG) synthesis has shown great promise in domains like iconography and sketching. However, existing Text-to-SVG methods often face challenges in editability, visual quality, and diversity. To address these issues, we propose a novel framework for text-guided SVG synthesis that significantly enhances editability, quality, and diversity. To enhance the editability of output SVGs, we introduce a Hierarchical Image VEctorization (HIVE) framework that operates at the semantic object level and supervises the optimization of components within the vector object. This approach facilitates the decoupling of vector graphics into distinct objects and component levels. Our proposed HIVE algorithm, informed by image segmentation priors, not only ensures a more precise representation of vector graphics but also enables fine-grained editing capabilities within vector objects. To improve the diversity of output SVGs, we present a Vectorized Particle-based Score Distillation (VPSD) approach. VPSD addresses over-saturation issues in existing methods and enhances sample diversity. A pre-trained reward model is incorporated to re-weight vector particles, improving aesthetic appeal and enabling faster convergence. Additionally, we design a novel adaptive vector primitives control strategy, which allows for the dynamic adjustment of the number of primitives, thereby enhancing the presentation of graphic details. Extensive experiments validate the effectiveness of the proposed method, demonstrating its superiority over baseline methods in terms of editability, visual quality, and diversity. We also show that our new method supports up to six distinct vector styles, capable of generating high-quality vector assets suitable for stylized vector design and poster design.

This paper takes the innovative design of traditional patterns as the research object, exploring the application of digital technology in the cultural diversity symbiosis model and its implementation path. First, traditional patterns are collected and digitized; pattern samples are extracted from folk art, historical documents, and intangible cultural heritage works, and structured storage is performed through image recognition and vectorization technology. Secondly, combined with the Generative Adversarial Network (GAN), the traditional patterns are transferred and reconstructed to generate innovative design solutions to ensure the continuity of their cultural elements. Then, the generated patterns are applied to a variety of design scenarios, including clothing, architectural decoration, and digital media design, through parametric design technology, and feedback data is collected using a user participatory design platform. Finally, based on the theory of multicultural symbiosis, the correlation between traditional patterns and modern design styles is established, and their adaptability and symbiosis in different cultural contexts are analyzed. The results show that the patterns generated based on GAN have a high innovation score. Digital technology can effectively promote the innovative design of traditional patterns. At the same time, it can realize the integration of tradition and modernity through the cultural diversity symbiosis model, providing a new path for the inheritance and development of intangible cultural heritage.

A ocultação de informações tem sido abordada em vários estudos. Quando a informação é ocultada em imagens digitais, pode ser realizada em três domínios: domínio espacial (espaço original dos pixels da imagem), domínio da frequência (ou domínio transformado, como o domínio da Transformada Cosseno Discreta) e domínio da imagem comprimida. No último caso, pode-se citar a inserção de dados em imagens comprimidas por quantização vetorial (QV). Este artigo aborda o problema da partição do dicionário de código no cenário de inserção de marca d'água invisível em imagens digitais comprimidas por QV. Neste artigo, duas técnicas são investigadas para fins de partição: o algoritmo PSO (Particle Swarm Optimization) e o EFA (Enhanced Fireworks Algorithm) como alternativas ao Algoritmo Genético. O desempenho das técnicas é avaliado com relação ao tempo de execução dos algoritmos. A robustez da marca d'água contra uma variedade de ataques é avaliada para dicionários de código particionados com os algoritmos mencionados.

With the rapid development of the Internet, the existence of fake news and its rapid spread has brought many negative effects to the society. Consequently, the fake news detection task has become increasingly important over the past few years. Existing methods are predominantly unimodal methods or the multimodal representation of unimodal fusion for fake news detection. However, the large number of model parameters and the interference of noisy data increase the risk of overfitting. Thus, we construct an information enhancement and contrast learning framework by introducing Improved Low-rank Multimodal Fusion approach for Fake News Detection (ILMF-FND), which aims to reduce the noise interference and achieve efficient fusion of multimodal feature vectors with fewer parameters. In detail, an encoder extracts the feature vectors of text and images, which are subsequently refined using the Multi-gate Mixture-of-Experts. The refined features are mapped into the same space for semanteme sharing. Then, a cross-modal fusion is performed, resulting in that an efficient and highly precision fusion of text and image features is done with fewer parameters. Besides, we design an adaptive mechanism that can adjust the weights of the final components according to the modal fitness before inputting them into the classifier to achieve the best detection results in the current state. We evaluate the performance of ILMF-FND and the competitive baselines on two public datasets, i.e., Twitter and Weibo. The results indicate that our ILMF-FND greatly minimizes the number of parameters while outperforming the best baseline in terms of accuracy by 0.2% and 1.1% on the Weibo and Twitter datasets, respectively.

This paper presents an artificial intelligence-based model for the classification of maritime vessel images obtained by cameras operating in the visible part of the electromagnetic spectrum. It incorporates both the deep learning techniques for initial image representation and traditional image processing and machine learning methods for subsequent image classification. The presented model is therefore a hybrid approach that uses the Inception v3 deep learning model for the purpose of image vectorization and a combination of SVM, kNN, logistic regression, Naïve Bayes, neural network, and decision tree algorithms for final image classification. The model is trained and tested on a custom dataset consisting of a total of 2915 images of maritime vessels. These images were split into three subsections: training (2444 images), validation (271 images), and testing (200 images). The images themselves encompassed 11 distinctive classes: cargo, container, cruise, fishing, military, passenger, pleasure, sailing, special, tanker, and non-class (objects that can be encountered at sea but do not represent maritime vessels). The presented model accurately classified 86.5% of the images used for training purposes and therefore demonstrated how a relatively straightforward model can still achieve high accuracy and potentially be useful in real-world operational environments aimed at sea surveillance and automatic situational awareness at sea.

Abstract We present a fully automated technique that segments raster images into smooth shaded regions and reconstructs them using an optimal mix of solid fills, linear gradients, and radial gradients. Our method leverages a novel discontinuity‐aware segmentation strategy and gradient reconstruction algorithm to accurately capture intricate shading details and produce compact Bézier curve representations. Extensive evaluations on both designer‐created art and generative images demonstrate that our approach achieves high visual fidelity with minimal geometric complexity and fast processing times. This work offers a robust and versatile solution for converting detailed raster images into scalable vector graphics, addressing the evolving needs of modern design workflows.