This paper presents a quantitative and qualitative comparison of three concurrency frameworks for embedded real-time systems: the C-based FreeRTOS kernel and the rust-based real-time interrupt-driven concurrency (RTIC) and Embassy frameworks. All three frameworks are implemented on an STM32F3 ARM Cortex-M4 platform using three representative applications: a single-task Blinky benchmark, a UART-to-LED event-signaling workload, and a Crane low profile vessel (LPV) subsystem integrating UART, CAN, and PWM tasks. Timing metrics, including task latency, periodic jitter, interrupt latency, and frame-to-register latency, are measured using the ARM data watchpoint and trace (DWT) cycle counter, while flash and RAM utilization are obtained through binary analysis. Across the Crane application, FreeRTOS achieves the lowest mean task-switch latency at 1110 cycles, compared with 2308 cycles for RTIC and 18,333 cycles for Embassy. RTIC provides the smallest memory footprint, using 17,224 bytes of flash and 1308 bytes of RAM, while Embassy and FreeRTOS use 20,784 and 20,444 bytes of flash and 5692 and 21,944 bytes of RAM, respectively. Embassy also shows highly repeatable timing behavior in selected application-level measurements, including a constant 442-cycle frame-to-register latency. Qualitatively, the study evaluates memory safety, concurrency reliability, predictability, scalability, and developer ergonomics. The results suggest that FreeRTOS is preferable for applications requiring minimal latency, RTIC is well suited for safety-oriented and RAM-constrained systems, and Embassy is attractive for scalable asynchronous I/O workloads.
Embodied intelligent robots demand comprehensive 3D scene understanding via multimodal sensor fusion for complex physical environments, yet existing methods struggle with insufficient heterogeneous data complementarity and inflexible fusion weight optimization. This study presents Artificial Hummingbird Algorithm (AHA)-fusion, a vision–LiDAR–tactile fusion framework optimized by the AHA for contact-aware embodied 3D scene perception. The framework integrates a multimodal feature extraction pipeline using ResNet-50, PointNet++, and lightweight CNNs; an improved Artificial Hummingbird Algorithm (I-AHA) with chaotic traversal and elite learning to optimize fusion weights globally; and a contact-aware dynamic weight modulation mechanism for real-time weight adjustment based on tactile contact states. A multilevel temporal synchronization strategy achieves 3.2 ± 1.1 ms alignment accuracy (Acc). Experiments on nuScenes, EmbodiedScan, and a self-built tactile manipulation dataset show AHA-Fusion attains 70.2% mean average precision (mAP) and 73.8% nuScenes detection score (NDS) for 3D object detection and 67.4% mean intersection over union (mIoU) for semantic segmentation, outperforming state-of-the-art methods. Ablation studies confirm I-AHA surpasses learnable attention weights by 2.3 percentage points with enhanced stability, while the contact-aware mechanism boosts contact region detection by 5.2 percentage points. A small-scale real-robot pilot experiment further suggests that the proposed framework can support online contact-aware manipulation, although broader real-world validation remains necessary. These results demonstrate the potential of optimization-guided trimodal fusion for robust embodied 3D perception.
Due to the rapid growth of online and technology development, every people have an own mobile to search their doubts and easily find it out within a second. Hypertext induced topic search (HITS) is a link analyzed algorithm utilized to find reliable websites that satisfy broad-topic queries. To identify hub page that link to different authority pages and enhance better topic search recommendations, various conventional approaches have been proposed in recent times. Yet, these methods consume more computational duration and complexity ranges in the training and testing performance, enhancing poor performance. Moreover, it fails to manage with high-dimensional large quantity of input noisy data, resulting inaccurate outcomes. Therefore, a novel deep learning-based HITS model is developed to capture complex patterns in sparse, sequential and contextual data, thereby enabling personalized, dynamic topic search recommendations that customize the individual preferences in social media platforms. The applications of the adaptive HITS (AHITS) algorithm precisely focus on effective hyperlink selection. While analyzing the specific topics, the relevance of content in HITS algorithm is specifically improved using the random search-based sharpbelly fish optimization (RS-SBFO) strategy is designed. Then, the link contents are processed in the preprocessing phase. Here, the provided contents are preprocessed to eliminate noise and other related artifacts. Then, the acquired preprocessed contents are provided to the developed multihead attention-based bidirectional encoder representations from transformer with residual generative long short-term memory (MHA-BRGLSTM) model for topic modeling and refinement. Here, the developed MHA-BRGLSTM provides an effective decision regarding the specific topics and related analysis by utilizing deep context analysis. Thus, the acquired outcomes are topic modeled and refined. Experimental validations are conducted to compare the effectiveness of the developed framework.
The continuous demand for reliable and privacy-sensitive face recognition and detection has led to improvements in machine learning and federated learning (FL) frameworks. While the centralized framework of existing solutions is beneficial for data collection, it presents significant privacy concerns and challenges due to the diversity of data types. To overcome these challenges, this study presents a new, scientifically grounded hybrid approach that combines ensemble deep learning with FL architecture. The authors achieve this by utilizing various deep learning models, such as AlexNet and vision transformers (ViTs), in conjunction with residual networks (ResNets), enhancing the overall efficiency of data prediction and enhancing the versatility of non-independent and identically distributed (Non-IID) datasets. The last component, FL, is the need to store personal user data in centralized repositories. Major developments include the application of secure aggregation techniques and differential privacy to avoid leakage of data, minimizing overhead communication, and designing strategic learning for different clients’ data distribution. Real-world testing on a distributed dataset with 50,000 facial images shows that the proposed framework can achieve a recognition accuracy of up to 95.6%, which is better than the best single-model FL systems and other similar benchmark methods. Due to the framework’s ability to process large volumes of data from noisy and imbalanced data sets, surveillance, healthcare, and the internet of things are viable application areas. This work highlights the possibility of developing such cascaded solutions using ensemble deep learning embedded in a FL architecture.
Fusion of multimodal medical images has become an essential reference for clinical diagnosis and a prominent topic in information fusion research. However, balancing the fusion performance with computational efficiency remains a challenging issue. In this paper, we propose an effective and efficient hybrid fusion method for multimodal medical images that integrates the merits of both spatial and transform domains, rather than relying on a single approach. First, a highly efficient multi-resolution tool, the framelet transform (FT), is employed to decompose medical images from different modalities into a series of low- and high-frequency sub-images. Next, an improved version of the structure tensor is designed to fuse the low-frequency sub-images, while a modified side-window filter (SWF) model is applied to fuse the high-frequency sub-images. Finally, the inverse FT is used to reconstruct the final fused image. Extensive simulation experiments on more than 120 pairs of multimodal medical images validate the performance of the proposed method, with results demonstrating superior visual quality and objective evaluation metrics compared to several recently published state-of-the-art approaches.
Image dehazing enhances visibility and image quality, enabling better decision-making in critical applications such as surveillance, remote sensing, and autonomous driving. However, single-image dehazing remains challenging due to the need for smoothness in homogeneous regions, accurate edge preservation, texture fidelity, and artifact-free reconstruction. In this paper, we propose RABNet, a novel deep learning framework for single-image dehazing. Traditional approaches often suffer from lossy reconstruction and poor performance under uneven haze distribution. RABNet addresses these limitations by integrating three key modules: the Hazy Feature Extraction Module (HFEM), Visibility Restoration Module (VRM), and Dehazed Image Reconstruction Module (DIRM). The HFEM enhances the network’s ability to perceive haze characteristics across varying lighting, contrast, and structural scales, providing a strong foundation for restoration. The VRM promotes feature diversity by employing multiple activation functions and an attention mechanism to refine visibility cues. Finally, the DIRM reconstructs the clean image by estimating the residual haze component and combining it with the input. By modeling haze-related features effectively, RABNet restores the underlying clean image while preserving color, texture, and structural details. Extensive experiments on both synthetic and real-world datasets, including FRIDA, O-HAZE, I-HAZE, and NH-HAZE, demonstrate the robustness and superior performance of RABNet, achieving state-of-the-art results in terms of PSNR and SSIM.
Insider threats are a widespread and severe issue of critical infrastructure and cyber-physical systems (CPSs), in which traditional security measures are frequently not sufficient. Despite these machine learning (ML) methods being utilized in detecting insider threats, most of the data mining techniques utilized are based either on misuse or anomaly methods, which limits their flexibility and resistance in dynamic operations. In order to overcome these shortcomings, this study presents a hybrid deep learning framework that is built on ensemble learning, and it can detect the presence of insider threat by integrating the spatial and temporal behavioral features. The architecture suggested is based on the convolutional neural networks (CNNs), long short-term memory (LSTM), gated recurrent unit (GRU), and bidirectional LSTM (Bi-LSTM) models. The conventional layers can be extracted to obtain spatial features of input sequences that are based on logs, whereas recurrent layers are used to forecast temporal dynamics and changing user behavior. The heterogeneous models are combined via a soft-averaging ensemble approach, which boosts the generalizability and robustness. Recursive feature elimination (RFE) is used to reduce dimensionality to enhance interpretability using a random forest-based ranking of feature importance. Moreover, synthetic minority over-sampling technique (SMOTE) is applied to address the issue of the existing class imbalance, making the model more sensitive to infrequent insider threats. Evaluation of the cyber emergency response team (CERT) insider threat dataset using experimental methods confirms that the proposed ensemble model is much more effective than traditional baselines, with a detection accuracy of 90% and a ROC area under the curve (AUC) score of 0.96, a true positive rate (TPR) of 0.96 and a fairly low false positive rate (FPR) of 0.128. This article helps to fill the gap between approaches based on misuse and others on anomaly, and a scalable, real-time-capable detection solution, as well as provides a solid architecture of human-centric security in digitized and complex CPS environments.
Soft errors pose a significant threat to the reliability of digital circuits, especially in harsh environments. While traditional error detection and correction techniques, such as Hamming codes and dual modular redundancy (DMR), have been widely employed, they often incur significant hardware overhead. This article presents a novel approach to enhancing the fault tolerance of finite impulse response (FIR) filters using multivalued logic (MVL). By leveraging the inherent redundancy of MVL, we propose several fault detection schemes based on current-mode MVL (CMMVL). These proposed architectures offer a significant reduction in hardware overhead compared to state-of-the-art techniques, often achieving at least a 60% decrease in area overhead. However, due to the power consumption characteristics of CMMVL, the proposed methods may introduce a higher power overhead. Among the proposed schemes, the time-division multiplexed checker (TDMC), technique demonstrates on average a 30% reduction in power overhead compared to the best previous method. Our proposed architectures achieve an area overhead of less than 40%, a significant reduction compared to the 110% overhead of conventional DMR. Among these, the TDMC architecture also reduces power overhead to 77%, outperforming DMR’s 108% power overhead. However, due to the high delay of the checker circuit, our methods do not show a significant improvement in the power-delay product (PDP). In the best-case prelayout scenario, the TDMC overhead is similar to DMR. Unfortunately, in the post-layout simulation, this overhead increases by as much as 110%.
Artistic image style transfer enables the synthesis of visually creative images by blending the structure of a content image with the textures of an artistic reference. However, existing models struggle to simultaneously preserve structural content and achieve coherent multi-style fusion, often producing distorted outputs with limited controllability. To address these limitations, we propose Fusion GAN, a dual-branch generative framework that integrates a content-style encoder, a multi-style adaptive fusion block (MSAFB), and perceptual constraints derived from VGG16. The architecture enables controlled blending of multiple style statistics while maintaining spatial consistency. Unlike standard GAN-based style transfer approaches, the proposed model incorporates a hybrid loss formulation combining perceptual content loss, Gram-matrix-based style loss, and adversarial realism regularization. The model is trained using MS-COCO (content) and WikiArt (style) datasets under a fully reproducible experimental setup with fixed hyperparameters and repeated trials. An innovative image recognition approach for real-time video surveillance is presented using a fusion generative adversarial network (fusion GAN). The method improves recognition robustness by separating structural features from appearance variations and adaptively fusing visual information under dynamic surveillance conditions. Quantitative results indicate partial improvements in structural retention but also highlight significant challenges, such as high Fr & eacute;chet inception distance (FID) and low peak signal-to-noise ratio (PSNR)/structural similarity index (SSIM) values, demonstrating that the model remains an exploratory prototype rather than a high-performance system. Qualitative outputs show successful stylistic blending, while the analysis identifies key failure points and areas for optimization. Overall, fusion GAN provides a structured and extensible foundation for multi-style transfer, but further refinement is needed to achieve competitive realism and perceptual quality. The performance of the proposed fusion GAN model is evaluated using the MS-COCO dataset for content images and the WikiArt dataset for artistic styles. Evaluation metrics such as FID, SSIM, and PSNR are employed to assess the model's effectiveness in content preservation and style fidelity.
To solve the problem of lack of market insight and insufficient innovation creativity in entrepreneurship of college students, this paper aims to explore the application of data mining and predictive analysis technology in this field. Quantitative research methods are employed. First, large-scale data is collected, including relevant information, such as market trends, user needs, and competitor intelligence. Second, cluster analysis techniques are used to analyze and interpret these data. Third, historical data and features are leveraged to build predictive analytics models to identify potential market opportunities and innovations to evaluate the potential success of startup projects through predictive analytics. Finally, statistical analysis methods are used to quantify and compare data to obtain objective results. The results show that the number of users in specific fields is growing at a rate of 10% per year and is expected to reach 200 million in the next 3 years. Based on the predictive analysis model, the entrepreneurial success probability of the entrepreneurial project is evaluated, and the predictive model shows that the probability of project success is 80%. This data support helps entrepreneurs make informed plans, optimize project management and operational strategies, and provide more opportunities for college students to start their businesses. ig data-driven predictive analytics and business intelligence (BI) for enhancing college student entrepreneurship success. Therefore, the integration of big data and BI technology into college students' entrepreneurship plays an important role, which can provide entrepreneurs with accurate market information and data support and improve the success rate of entrepreneurship.
As the demand for low-power electronic products grows, asynchronous circuits are considered a good alternative for addressing power consumption issues. Applying dynamic voltage scaling (DVS) in asynchronous circuits can further improve their power efficiency. However, asynchronous circuits face challenges, such as performance analysis considering voltage, temperature, and process variations. This paper proposes a new statistical performance analysis model for asynchronous pipelines. This model can be applied to two different styles of asynchronous circuits. The results show that this model has reasonable accuracy on estimated mean delay (2% error on average) compared to detailed analysis carried out with low-level Monte Carlo (MC) circuit simulations.
The traditional method of identifying abnormal behavior in athletes during long-distance track and field events, particularly through video surveillance, has limitations such as misjudgments and oversight. To address these challenges, this study explores the use of artificial intelligence (AI) and intelligent surveillance technology for event safety management. However, existing video-based abnormal behavior recognition methods face significant difficulties in adapting to marathon scenarios due to the complexities of outdoor surveillance footage. In response, this paper proposes innovative methods for recognizing abnormal behavior and identifying athletes using bib numbers, thereby enhancing the accuracy of behavior detection in marathon environments. Additionally, an AI-powered system for athlete identification and abnormal behavior detection is introduced to improve event monitoring and safety. The contributions of this study include the development of an attention-residual-based abnormal behavior recognition algorithm, a rotation-based object detection model for accurate athlete identification, and a deep learning-based intelligent recognition system for effective event management.
With the rapid growth of multimedia communication, protecting image data has become increasingly critical. This article proposes a novel 3-stage hyperchaos-based augmented image encryption technique (3SHAIET) that utilizes a three-stage process with chaotic systems of increasing dimensionality (e.g., six-dimensional [6D], 8D, and 9D) to enhance security against cryptographic attacks. At each stage, the image is converted into a 1D bit-stream array, encrypted using dynamically generated S-boxes, and XORed with keys derived from the chaotic systems, ensuring high sensitivity to initial conditions and strong diffusion and confusion properties. To achieve real-time performance, the algorithm is realized on a field-programmable gate array (FPGA), leveraging hardware acceleration for the lowest possible latency encryption. The FPGA realization is shown to provide a 165 times improvement in encryption rate over its corresponding software implementation. Experimental results demonstrate the algorithm’s robustness, with an extremely large key space of 25475 and resistance to statistical (PSNR of 8.11 dB, MAE of 81.74, and entropy of 7.999) and differential attacks (NPCR of 99.6216% and UACI of 32.0534%). Additionally, the resulting encrypted images’ bit-streams successfully pass all National Institute of Standards and Technology (NIST) randomness tests, further validating the algorithm’s security. These results make the proposed 3SHAIET algorithm a highly efficient and secure solution for modern multimedia systems. The objective of 3SHAIET is to jointly maximize cryptographic strength (via multistage hyperchaotic keys and dynamic S-boxes) and real-time efficiency (via FPGA realization) for secure multimedia transmission and storage.
The posit number system represents a significant advancement aimed at replacing the current IEEE floating-point standard in a seamless manner. With its notable dynamic range and gradually tapering precision, a smaller posit can closely match the performance of a larger floating-point number in representing decimal values. Multiplication is a fundamental arithmetic operation that is essential in a wide range of applications, particularly in fields such as image processing, signal processing, neural networks (NNs), and machine learning. Given the considerable power consumption, area requirements, and latency associated with multiplication, it is imperative to explore optimization strategies in these areas. This article provides a comprehensive review of both exact and inexact (approximate) posit multiplier designs. It includes a detailed comparative evaluation of their error rates and circuit characteristics, aimed at fostering a deeper understanding of the distinctive features of various designs. This study examines Booth-based posit multipliers and logarithmic posit multipliers, categorizing Booth multipliers into exact and inexact types. The posit multipliers are implemented and synthesized using the Cadence RTL Compiler in Verilog HDL, while error characterization is conducted using the soft posit library in Python. In this article, power, area, and delay are compared in relation to the mean relative error distance (MRED). The comparative results indicate that the logarithmic-based posit multiplier is hardware-efficient but has low accuracy. In contrast, the Booth posit multiplier offers superior accuracy, despite having higher performance metrics. Notably, the logarithmic multiplier, referred to as posit logarithmic-approximate multiplier (PLAM), provides a substantial decrease in power, area, and delay by at least 92%, 82%, and 78%, respectively, compared to all the Booth multipliers. The approximation error of PLAM is analyzed, including metrics such as MRED, to assess performance relative to exact posit multipliers. The posit logarithmic multiplier was validated using various NN architectures, including LeNet-5, VGG11, and ResNet-18. The results indicate that posit logarithmic multiplier achieves inference accuracy comparable to traditional floating-point multipliers while also enhancing hardware efficiency.
Facial emotion recognition has poor robustness and low recognition accuracy in complex lighting, posture changes, and occlusion scenes. This study aims to design a high-performance convolutional neural network (CNN) model to improve the recognition accuracy and generalization ability of seven basic emotions in complex environments. FER2013, CK+ and Japanese female cultural specific expression (JAFFE) datasets are selected, and data preprocessing is performed through grayscale, histogram equalization and size normalization; secondly, random rotation, horizontal flipping and brightness perturbation are used for data enhancement to improve the generalization of the model; then, a 12-layer CNN model is constructed, including four convolutional blocks, two fully connected layers and an output layer, and Dropout (0.5) is used to prevent overfitting; the Adam optimizer is used to iterate 100 epochs on the training data, with cross entropy as the loss function, and the early stopping mechanism is used to optimize the hyperparameters on the validation set. The highest accuracy rate reaches 99.2% on the FER2013 test set, and the average accuracy rates of 97.3% and 88.3% are obtained in the cross-dataset tests of CK+ and JAFFE, respectively. Key performance indicators show that the average recall rate is 90.7%; the precision rate is 90.4%; the F1-score is 90.5%; the accuracy rate is still 85.2% in the standard mask occlusion test scenario. The proposed CNN model significantly improves the accuracy and robustness of emotion recognition under complex conditions through end-to-end feature learning and data enhancement strategies, providing an effective technical solution for real-time emotion analysis systems.
With the rapid development of computer science and information technology, augmented reality (AR) technology has been widely used in the field of visual communication design. AR is needed in visual communication design because it allows blending the real world with the virtual one, which cannot be done with the help of traditional 2D design methods. This paper aims to enhance the application of reality technology in visual communication design and evaluate its effect and advantages and disadvantages through experiments. In this paper, AR technology is used to virtually superimpose images, videos, and other elements in the real scene to achieve colorful visual communication effects. Meanwhile, the artificial intelligence (AI) algorithm is used to optimize the AR content to improve its visual quality and realism. Finally, the effect and user experience of traditional graphic design and AR design are compared through experiments. The research results show that AR technology can enhance the visualization effect of information, increase the user participation by nearly 20% year-on-year, and enhance the memory durability by 4.37% compared with before. AR technology can also create a unique experience different from traditional design media. It shows that AR technology can effectively improve the effect and user experience of visual communication design and provide more abundant and diversified design tools and means for visual communication designers.
With the widespread application of local affine (LA) motion models in various video coding standards, this study explores the implementation methods and performance changes of introducing a global registration model in an encoder that already includes a LA motion model. First, a coding scheme combining global and local registration is achieved by incorporating global registration computation, optimizing reference frame selection strategies, and macroblock mode selection strategies. Second, through experiments, the impact of introducing a global warp motion model and a global translational (GT) registration model on performance is further compared. The results indicate that the introduction of a global warp motion model leads to functional redundancy and mutual interference, with higher computational complexity and limited overall benefits. On the other hand, introducing a GT registration model can complement and enhance the coding performance for translation scenarios, working in synergy with the LA model, while maintaining lower computational complexity and greater practicality.
Generative adversarial networks (GANs), a subset of deep learning, have demonstrated breakthrough performance in domains such as computer vision (CV) and natural language processing (NLP), particularly in surveillance, autonomous driving, and automated programing assistance. Based on game theory principles, GANs utilize a generator–discriminator architecture to produce high-quality synthetic data. This study conducts a systematic literature review (SLR) to comprehensively assess the development, applications, limitations, and security-related advancements of GANs. It examines foundational models and key architectural variants, providing a critical evaluation of their roles in NLP and CV. This research explores the integration of GANs into the domain of security, highlighting their applications in information security, cybersecurity, and artificial intelligence (AI)-driven defense mechanisms. The study also discusses prominent evaluation metrics such as inception score (IS), Fréchet inception distance (FID), structural similarity index measure (SSIM), and peak signal-to-noise ratio (PSNR) to assess GAN performance. Key strengths of GANs, including their ability to generate high-resolution data and support domain adaptation, are emphasized as driving factors for their continued evolution and adoption.
Securing reusable hardware intellectual property (IP) cores used in system-on-chip (SoC) designs is crucial, due to global design supply chain that may introduce different points of security vulnerability. One of the major threats includes an untrustworthy entity in the SoC design house attempting piracy or falsely claiming ownership of the IP design. Further, owing to the importance of handling transient fault in hardware IP designs, design of fault-detectable IP designs has become a standard practice in the community. However, these fault-detectable IP designs are also similarly prone to hardware threats such as IP piracy and false claim of IP ownership. Therefore, robust sturdy countermeasure for fault-detectable IP designs against such threats is essential. This paper presents a detective countermeasure using proposed novel hardware watermarking methodology for transient fault-detectable IP designs. The proposed IP watermarking methodology introduces a novel multivariate encoded high-level synthesis (HLS) scheduling based multimodal security framework. The proposed approach is capable of embedding a robust, unique, and nonreplicable watermark in the HLS register allocation phase of fault-detectable IP design. The proposed watermarking technique is more robust than the prior watermarking approaches in terms of reduced probability of coincidence (PC; upto ~10−8), stronger tamper tolerance (TT; upto ~10130), and lower watermark decoding probability at 0% design cost overhead.
This study presents an innovative temperature-induced random noise correction method for complementary metal oxide semiconductor (CMOS) spatial cameras using an attention mechanism-enhanced long short-term memory (LSTM) model. The model, specifically designed to address pixel drift and random noise issues in CMOS space cameras due to temperature variations, incorporates a multilayer LSTM network with an attention mechanism. This study comprehensively examines the temperature-induced variations in noise characteristics of CMOS cameras across diverse thermal conditions, encompassing in-depth analyses of both dark-field and light-field scenarios. Through detailed pixel-level analysis, the study quantifies the influence of temperature on pixel values and critical performance parameters such as internal nonuniformity within the camera. The experimental results show that under the dark field condition, the fitting variance between the predicted value and the measured value ranges from 0.29585 to 5.798307. After correction in light field conditions, the average variance of images decreases to 0.29, the mean signal-to-noise ratio (SNR) increases to 80, and the photo response nonuniformity (PRNU) mean drops to 0.0161%. Compared to precorrection levels, these key metrics show significant improvements, with an average 83.57-fold reduction, 1.89-fold increase, and 84.98-fold decrease, respectively. These results confirm the effectiveness of the deep learning method in correcting temperature-induced noise, highlighting the potential for practical engineering applications.
