Monitoring and analyzing sea-ice changes is critically important for navigation through polar regions, ecosystem monitoring, and environmental risk assessment, and Synthetic Aperture Radar(SAR) imagery is one of the most reliable and useful data sources for this goal. SAR images exhibit complex backscatter patterns that are generally not interpretable without technical expertise. Consequently, accurate annotation of SAR images requires domain experts to identify the corresponding sea ice classes, resulting in labelled datasets that are both limited and often imbalanced across classes.A further challenge is that SAR image channels differ from those of conventional images, with HH, HV, and incidence angle each containing important physical information. Although current image generation methodologies have strong capabilities in RGB images and have solved several limitations, including training instability and high computational cost, some challenges in SAR image generation remain. These issues, such as limited preservation of channel-specific physical information and mode collapse, motivate us to work in generative modelling and seek to overcome them. In this study, we propose a Flow Matching framework with a class-informed prior. Instead of using a standard Gaussian distribution as the initial point for learning, the Class-Conditioned Flow Matching (CFM) model estimates the mean and variance of each image class and uses them to better approximate the data distribution from a more informative starting point. Another contribution of this paper is the proposed Swin Transformer Convolutional Hybrid (SCM-UNet) model, which generates images using a modified U-Net architecture. Experiments on the chosen dataset demonstrate that the proposed model improves visual metrics in generated images (FSD=19.23, PSNR=28.0, and SSIM=0.83) and effectively enhances the downstream task. Results show that, when trained exclusively on generated data and evaluated on real SAR images, the classification model achieved an average accuracy of 91.61% and an F1-score exceeding 91.5%.
The observed increase in cyber threat intelligence data, the diversity of sources, and relational complexity make it difficult for analysts to directly assess and interpret threat profiles from raw Structured Threat Information Expression (STIX) packages. This study utilises an object-relational model to transform all object and relationship types into a single unified representation and reconstruct the same graph model across three different visualisation software packages (plotly, matplotlib, pyvis). The proposed analytical framework generates extended threat models by combining multiple STIX datasets through entity extraction, creating observable entities from each dataset, and completing relationships among them; thus mapping relationship types to the operational meanings of attacks and enabling more comprehensive risk prioritisation. The case study is based on a public OASIS STIX example package. The enriched graph is dominated by relationship types such as uses, indicates, pattern-refers-to, and attributed-to, and the highest risk scores are assigned to the nodes labelled “Privilege Escalation” and “Ugly Gorilla”. Scalability tests performed on 1000-node synthetic directed graphs revealed a processing time of 8.08 seconds, memory consumption of 124.3 MB, and a frame rate of 215.05 fps during interactive preview. These findings demonstrate that the proposed framework offers a unified and viable foundation for visual analysis, risk prioritisation, and scalable analytical reporting of STIX-based cyber threat intelligence.
This paper investigates the problem of blind detection of orthogonal space–time block codes (OSTBCs) with quadrature amplitude modulation (QAM) in multiple-input multiple-output (MIMO) systems over quasistatic flat Rayleigh fading channels. To resolve the inherent rotational ambiguity in blind OSTBC detection, we propose a structurally constrained QAM constellation that enables unique symbol recovery without the use of pilot signals. Building on this design, we develop a low-complexity iterative detector, referred to as the iterative maximum-likelihood with averaged initial channel estimate (IML-AICE) detector, which jointly estimates the channel and transmitted symbols. The proposed detector incorporates a novel initialization strategy and an iterative refinement mechanism inspired by clairvoyant maximum-likelihood detection, leading to improved convergence and detection accuracy. The proposed framework enables reliable blind recovery of OSTBC symbols, thereby improving spectral efficiency by eliminating pilot overhead. Simulation results demonstrate that the proposed IML-AICE detector consistently outperforms existing trained and blind detection schemes across a range of signal-to-noise ratios and system configurations at low computational complexity.
Sea ice is considered one of the most valuable sources of information for maintaining the balance of Earth’s climate system and preventing excessive warming. This study introduces a lightweight yet accurate model designed for real-time sea ice classification to provide an accessible and practical tool for operational use. We propose the SCM-Net, a deep learning model for sea ice classification applications, and compare its performance against other state-of-the-art models, including MobileNet, Residual Network (ResNet), Visual Geometry Group Network (VGGNet), Vision Transformers (ViT), and Shifted Window Transformers (SwinT). The Swin Transformer Convolutional Hybrid model (SCM-Net) is a lightweight model with around 45 times less parameters, enabling usage in real time applications. The results demonstrate that the proposed SCMNet model achieves a comparable and even better accuracy in comparison to other models. Moreover, the proposed model significantly reduces the number of parameters while improving inference efficiency. These results show that the proposed model is well suited for real time sea ice classification applications.
Ensuring continuous operation and minimizing unexpected failures are critical priorities in industrial manufacturing. Due to this requirement, smart manufacturing has transformed maintenance strategies, moving from traditional scheduled approaches to predictive maintenance (PdM), which leverages actual machine health conditions. A powerful technique that enables accurate PdM is Remaining Useful Life (RUL) estimation. Accurate RUL prediction allows the assessment of machine health, thereby facilitating timely and appropriate maintenance decisions to sustain continuous operation and reduce repair costs. While several existing models have been developed for RUL estimation, they often struggle to capture long-term dependencies in time series data, limiting their predictive accuracy. In this study, we propose a novel architecture that combines a one-dimensional Convolutional Neural Network (1D-CNN) with two consecutive transformer encoders enhanced by Fourier transforms to address these challenges. The performance of the proposed model was evaluated based on two well-known benchmark datasets, C-MAPSS and its improved version, N-CMAPSS. The experiment results show that our approach outperforms current state-of-the-art methods in both prediction accuracy and computational efficiency, demonstrating its potential for practical applications.
