Search Results (6,003)

With the widespread adoption of electric vehicles, the power converter, as a key component, plays a crucial role. Traditional fault detection methods often face challenges in real-time performance and computational efficiency, making it difficult to meet the demands of electric vehicle power converters for efficient and accurate fault diagnosis. To address this challenge, this paper proposes a novel fault detection model—SpikingShuffleNet. This paper first designs an efficient SpikingShuffle Unit that integrates grouped convolutions and channel shuffle techniques, effectively reducing the model’s computational complexity by optimizing feature extraction and channel interaction. Next, by appropriately stacking SpikingShuffle Units and refining the network architecture, a complete lightweight diagnostic network is constructed for real-time fault detection in electric vehicle power converters. Finally, the Mixed Local Channel Attention mechanism is introduced to address the potential limitations in feature representation caused by grouped convolutions, further enhancing fault detection accuracy and robustness by balancing local detail preservation and global feature integration. Experimental results show that SpikingShuffleNet exhibits excellent accuracy and robustness in the fault detection task for power converters, fulfilling the real-time fault diagnosis requirements for low-power embedded devices. Full article

Transformers were initially introduced for natural language processing, leveraging the self-attention mechanism. They require minimal inductive biases in their design and can function effectively as set-based architectures. Additionally, transformers excel at capturing long-range dependencies and enabling parallel processing, which allows them to outperform traditional models, such as long short-term memory (LSTM) networks, on sequence-based tasks. In recent years, transformers have been widely adopted in computer vision, driving remarkable advancements in the field. Previous surveys have provided overviews of transformer applications across various computer vision tasks, such as object detection, activity recognition, and image enhancement. In this survey, we focus specifically on image classification. We begin with an introduction to the fundamental concepts of transformers and highlight the first successful Vision Transformer (ViT). Building on the ViT, we review subsequent improvements and optimizations introduced for image classification tasks. We then compare the strengths and limitations of these transformer-based models against classic convolutional neural networks (CNNs) through experiments. Finally, we explore key challenges and potential future directions for image classification transformers. Full article

This paper introduces RWA-BFT, a reputation-weighted asynchronous Byzantine Fault Tolerance (BFT) consensus algorithm designed to address the scalability and performance challenges of blockchain systems in large-scale IoT scenarios. Traditional centralized IoT architectures often face issues such as single points of failure and insufficient reliability, while blockchain, with its decentralized and tamper-resistant properties, offers a promising solution. However, existing blockchain consensus mechanisms struggle to meet the high throughput, low latency, and scalability demands of IoT applications. To address these limitations, RWA-BFT adopts a two-layer blockchain architecture; the first layer leverages reputation-based filtering to reduce computational complexity by excluding low-reputation nodes, while the second layer employs an asynchronous consensus mechanism to ensure efficient and secure communication among high-reputation nodes, even under network delays. This dual-layer design significantly improves performance, achieving higher throughput, lower latency, and enhanced scalability, while maintaining strong fault tolerance even in the presence of a substantial proportion of malicious nodes. Experimental results demonstrate that RWA-BFT outperforms HB-BFT and PBFT algorithms, making it a scalable and secure blockchain solution for decentralized IoT applications. Full article

Aiming at the optimization of the big data infrastructure in China’s healthcare system, this study proposes a lightweight time series physician demand prediction model, which is especially suitable for the field of telemedicine. The model incorporates multi-head attention mechanisms and generates statistical information, which significantly improves the ability to process nonlinear data, adapt to different data sources, improve the computational efficiency, and process high-dimensional features. By combining variational autoencoders and LSTM units, the model can effectively capture complex nonlinear relationships and long-term dependencies, and the multi-head attention mechanism overcomes the limitations of traditional algorithms. This lightweight architecture design not only improves the computational efficiency but also enhances the stability in high-dimensional data processing and reduces feature redundancy by combining the normalization process with statistics. The experimental results show that the model has wide applicability and excellent performance in a telemedicine consulting service system. Full article

This paper introduces a novel joint model architecture for Speech Emotion Recognition (SER) that integrates both discrete and dimensional emotional representations, allowing for the simultaneous training of classification and regression tasks to improve the comprehensiveness and interpretability of emotion recognition. By employing a joint loss function that combines categorical and regression losses, the model ensures balanced optimization across tasks, with experiments exploring various weighting schemes using a tunable parameter to adjust task importance. Two adaptive weight balancing schemes, Dynamic Weighting and Joint Weighting, further enhance performance by dynamically adjusting task weights based on optimization progress and ensuring balanced emotion representation during backpropagation. The architecture employs parallel feature extraction through independent encoders, designed to capture unique features from multiple modalities, including Mel-frequency Cepstral Coefficients (MFCC), Short-term Features (STF), Mel-spectrograms, and raw audio signals. Additionally, pre-trained models such as Wav2Vec 2.0 and HuBERT are integrated to leverage their robust latent features. The inclusion of self-attention and co-attention mechanisms allows the model to capture relationships between input modalities and interdependencies among features, further improving its interpretability and integration capabilities. Experiments conducted on the IEMOCAP dataset using a leave-one-subject-out approach demonstrate the model’s effectiveness, with results showing a 1–2% accuracy improvement over classification-only models. The optimal configuration, incorporating the joint architecture, dynamic weighting, and parallel processing of multimodal features, achieves a weighted accuracy of 72.66%, an unweighted accuracy of 73.22%, and a mean Concordance Correlation Coefficient (CCC) of 0.3717. These results validate the effectiveness of the proposed joint model architecture and adaptive balancing weight schemes in improving SER performance. Full article

Large language models (LLMs) have emerged as powerful tools for natural language processing tasks, revolutionizing the field with their ability to understand and generate human-like text. As the demand for more sophisticated LLMs continues to grow, there is a pressing need to address the computational challenges associated with their scale and complexity. This paper presents a comprehensive survey of hardware accelerators designed to enhance the performance and energy efficiency of large language models. By examining a diverse range of accelerators, including GPUs, FPGAs, and custom-designed architectures, we explore the landscape of hardware solutions tailored to meet the unique computational demands of LLMs. The survey encompasses an in-depth analysis of architecture, performance metrics, and energy efficiency considerations, providing valuable insights for researchers, engineers, and decision-makers aiming to optimize the deployment of LLMs in real-world applications. Full article

Anomalous Sound Detection (ASD) is crucial for ensuring industrial equipment safety and enhancing production efficiency. However, existing methods, while pursuing high detection accuracy, are often associated with high computational complexity, making them unsuitable for resource-constrained environments. This study proposes an efficient self-supervised ASD framework that integrates spectral features, lightweight neural networks, and various anomaly scoring methods. Unlike traditional Log-Mel features, spectral features retain richer frequency domain details, providing high-quality inputs that enhance detection accuracy. The framework includes two network architectures: the lightweight ASDNet, optimized for resource-limited scenarios, and SpecMFN, which combines SpecNet and MobileFaceNet for advanced feature extraction and classification. These architectures employ various anomaly scoring methods, enabling complex decision boundaries to effectively detect diverse anomalous patterns. Experimental results demonstrate that ASDNet achieves an average AUC of 94.42% and a pAUC of 87.18%, outperforming existing methods by 6.75% and 9.34%, respectively, while significantly reducing FLOPs (85.4 M, a 93.81% reduction) and parameters (0.51 M, a 41.38% reduction). SpecMFN achieves AUC and pAUC values of 94.36% and 88.60%, respectively, with FLOPs reduced by 86.6%. These results highlight the framework’s ability to balance performance and computational efficiency, making it a robust and practical solution for ASD tasks in industrial and resource-constrained environments. Full article

Convolutional neural networks (CNNs) are increasingly recognized as an important and potent artificial intelligence approach, widely employed in many computer vision applications, such as facial recognition. Their importance resides in their capacity to acquire hierarchical features, which is essential for recognizing complex patterns. Nevertheless, the intricate architectural design of CNNs leads to significant computing requirements. To tackle these issues, it is essential to construct a system based on field-programmable gate arrays (FPGAs) to speed up CNNs. FPGAs provide fast development capabilities, energy efficiency, decreased latency, and advanced reconfigurability. A facial recognition solution by leveraging deep learning and subsequently deploying it on an FPGA platform is suggested. The system detects whether a person has the necessary authorization to enter/access a place. The FPGA is responsible for processing this system with utmost security and without any internet connectivity. Various facial recognition networks are accomplished, including AlexNet, ResNet, and VGG-16 networks. The findings of the proposed method prove that the GoogLeNet network is the best fit due to its lower computational resource requirements, speed, and accuracy. The system was deployed on three hardware kits to appraise the performance of different programming approaches in terms of accuracy, latency, cost, and power consumption. The software programming on the Raspberry Pi-3B kit had a recognition accuracy of around 70–75% and relied on a stable internet connection for processing. This dependency on internet connectivity increases bandwidth consumption and fails to meet the required security criteria, contrary to ZYBO-Z7 board hardware programming. Nevertheless, the hardware/software co-design on the PYNQ-Z2 board achieved an accuracy rate of 85% to 87%. It operates independently of an internet connection, making it a standalone system and saving costs. Full article

The coal mining industry in Northern Shaanxi is robust, with a prevalent use of the local dialect, known as “Shapu”, characterized by a distinct Northern Shaanxi accent. This study addresses the practical need for speech recognition in this dialect. We propose an end-to-end speech recognition model for the North Shaanxi dialect, leveraging the Conformer architecture. To tailor the model to the coal mining context, we developed a specialized corpus reflecting the phonetic characteristics of the dialect and its usage in the industry. We investigated feature extraction techniques suitable for the North Shaanxi dialect, focusing on the unique pronunciation of initial consonants and vowels. A preprocessing module was designed to accommodate the dialect’s rapid speech tempo and polyphonic nature, enhancing recognition performance. To enhance the decoder’s text generation capability, we replaced the Conformer decoder with a Transformer architecture. Additionally, to mitigate the computational demands of the model, we incorporated Connectionist Temporal Classification (CTC) joint training for optimization. The experimental results on our self-established voice dataset for the Northern Shaanxi coal mining industry demonstrate that the proposed Conformer–Transformer–CTC model achieves a 9.2% and 10.3% reduction in the word error rate compared to the standalone Conformer and Transformer models, respectively, confirming the advancement of our method. The next step will involve researching how to improve the performance of dialect speech recognition by integrating external language models and extracting pronunciation features of different dialects, thereby achieving better recognition results. Full article

Background: This study focuses on the critical task of blood vessel segmentation in medical image analysis, essential for diagnosing cardiovascular diseases and enabling effective treatment planning. Although deep learning architectures often produce very high segmentation results in medical images, coronary computed tomography angiography (CTA) images are more challenging than invasive coronary angiography (ICA) images due to noise and the complexity of vessel structures. Methods: Classical architectures for medical images, such as U-Net, achieve only moderate accuracy, with an average Dice score of 0.722. Results: This study introduces Morpho-U-Net, an enhanced U-Net architecture that integrates advanced morphological operations, including Gaussian blurring, thresholding, and morphological opening/closing, to improve vascular integrity, reduce noise, and achieve a higher Dice score of 0.9108, a precision of 0.9341, and a recall of 0.8872. These enhancements demonstrate superior robustness to noise and intricate vessel geometries. Conclusions: This pre-processing filter effectively reduces noise by grouping neighboring pixels with similar intensity values, allowing the model to focus on relevant anatomical structures, thus outperforming traditional methods in handling the challenges posed by CTA images. Full article

This study investigates the integration of reinforcement learning (RL) with optimal control to enhance precision and energy efficiency in industrial robotic manipulation. A novel framework is proposed, combining Deep Deterministic Policy Gradient (DDPG) with a Linear Quadratic Regulator (LQR) controller, specifically applied to the ABB IRB120, a six-degree-of-freedom (6-DOF) industrial manipulator, for pick-and-place tasks in warehouse automation. The methodology employs an actor–critic RL architecture with a 27-dimensional state input and a 6-dimensional joint action output. The RL agent was trained using MATLAB’s Reinforcement Learning Toolbox and integrated with ABB’s RobotStudio simulation environment via TCP/IP communication. LQR controllers were incorporated to optimize joint-space trajectory tracking, minimizing energy consumption while ensuring precise control. The novelty of this research lies in its synergistic combination of RL and LQR control, addressing energy efficiency and precision simultaneously—an area that has seen limited exploration in industrial robotics. Experimental validation across 100 diverse scenarios confirmed the framework’s effectiveness, achieving a mean positioning accuracy of 2.14 mm (a 28% improvement over traditional methods), a 92.5% success rate in pick-and-place tasks, and a 22.7% reduction in energy consumption. The system demonstrated stable convergence after 458 episodes and maintained a mean joint angle error of 4.30°, validating its robustness and efficiency. These findings highlight the potential of RL for broader industrial applications. The demonstrated accuracy and success rate suggest its applicability to complex tasks such as electronic component assembly, multi-step manufacturing, delicate material handling, precision coordination, and quality inspection tasks like automated visual inspection, surface defect detection, and dimensional verification. Successful implementation in such contexts requires addressing challenges including task complexity, computational efficiency, and adaptability to process variability, alongside ensuring safety, reliability, and seamless system integration. This research builds upon existing advancements in warehouse automation, inverse kinematics, and energy-efficient robotics, contributing to the development of adaptive and sustainable control strategies for industrial manipulators in automated environments. Full article

The retreat of Arctic sea ice has opened new maritime routes, offering faster shipping opportunities; however, these routes present significant navigational challenges due to the harsh ice conditions. To address these challenges, this paper proposes a deep learning-based Arctic ice risk management architecture with multiple modules, including ice classification, risk assessment, ice floe tracking, and ice load calculations. A comprehensive dataset of 15,000 ice images was created using public sources and contributions from the Canadian Coast Guard, and it was used to support the development and evaluation of the system. The performance of the YOLOv8n-cls model was assessed for the ice classification modules due to its fast inference speed, making it suitable for resource-constrained onboard systems. The training and evaluation were conducted across multiple platforms, including Roboflow, Google Colab, and Compute Canada, allowing for a detailed comparison of their capabilities in image preprocessing, model training, and real-time inference generation. The results demonstrate that Image Classification Module I achieved a validation accuracy of 99.4%, while Module II attained 98.6%. Inference times were found to be less than 1 s in Colab and under 3 s on a stand-alone system, confirming the architecture’s efficiency in real-time ice condition monitoring. Full article

This study is focused on developing a machine learning (ML) meta-model to predict the progression of a multiple steam generator tube rupture (MSGTR) accident in the APR1400 reactor. The accident was simulated using the thermal–hydraulic code RELAP5/SCDAPSIM/MOD3.4. The model incorporates a mitigation strategy executed through operator interventions. Following this, uncertainty quantification employing the Best Estimate Plus Uncertainty (BEPU) methodology was undertaken by coupling RELAP5/SCDAPSIM/MOD3.4 with the statistical software, DAKOTA 6.14.0. The analysis concentrated on critical safety parameters, including Reactor Coolant System (RCS) pressure and temperature, as well as reactor vessel upper head (RVUH) void fraction. These simulations generated a comprehensive dataset, which served as the foundation for training three ML architectures: Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), and Convolutional LSTM (CNN+LSTM). Among these models, the CNN+LSTM hybrid configuration demonstrated superior performance, excelling in both predictive accuracy and computational efficiency. To bolster the model’s transparency and interpretability, Integrated Gradients (IGs)—an advanced Explainable AI (XAI) technique—was applied, elucidating the contribution of input features to the model’s predictions and enhancing its trustworthiness. Full article

Conventional weather forecasting relies on numerical weather prediction (NWP), which solves atmospheric equations using numerical methods. The Korea Meteorological Administration (KMA) adopted the Met Office Global Seasonal Forecasting System version 6 (GloSea6) NWP model from the UK and runs it on a supercomputer. However, due to high task demands, the limited resources of the supercomputer have caused job queue delays. To address this, the KMA developed a low-resolution version, Low GloSea6, for smaller-scale servers at universities and research institutions. Despite its ability to run on less powerful servers, Low GloSea6 still requires significant computational resources like those of high-performance computing (HPC) clusters. We integrated deep learning with Low GloSea6 to reduce execution time and improve meteorological research efficiency. Through profiling, we confirmed that deep learning models can be integrated without altering the original configuration of Low GloSea6 or complicating physical interpretation. The profiling identified “tri_sor.F90” as the main CPU time hotspot. By combining the biconjugate gradient stabilized (BiCGStab) method, used for solving the Helmholtz problem, with a deep learning model, we reduced unnecessary hotspot calls, shortening execution time. We also propose a convolutional block attention module-based Half-UNet (CH-UNet), a lightweight 3D-based U-Net architecture, for faster deep-learning computations. In experiments, CH-UNet showed 10.24% lower RMSE than Half-UNet, which has fewer FLOPs. Integrating CH-UNet into Low GloSea6 reduced execution time by up to 71 s per timestep, averaging a 2.6% reduction compared to the original Low GloSea6, and 6.8% compared to using Half-UNet. This demonstrates that CH-UNet, with balanced FLOPs and high predictive accuracy, offers more significant execution time reductions than models with fewer FLOPs. Full article

The widespread use of the internet, coupled with the increasing production of digital content, has caused significant challenges in information security and manipulation. Deepfake detection has become a critical research topic in both academic and practical domains, as it involves identifying forged elements in artificially generated videos using various deep learning and artificial intelligence techniques. In this dissertation, an innovative model was developed for detecting deepfake videos by combining the Quantum Transfer Learning (QTL) and Class-Attention Vision Transformer (CaiT) architectures. The Deepfake Detection Challenge (DFDC) dataset was used for training, and a system capable of detecting spatiotemporal inconsistencies was constructed by integrating QTL and CaiT technologies. In addition to existing preprocessing methods in the literature, a novel preprocessing function tailored to the requirements of deep learning models was developed for the dataset. The advantages of quantum computing offered by QTL were merged with the global feature extraction capabilities of the CaiT. The results demonstrated that the proposed method achieved a remarkable performance in detecting deepfake videos, with an accuracy of 90% and ROC AUC score of 0.94 achieved. The model’s performance was compared with other methods evaluated on the DFDC dataset, highlighting its efficiency in resource utilization and overall effectiveness. The findings reveal that the proposed QTL-CaiT-based system provides a strong foundation for deepfake detection and contributes significantly to the academic literature. Future research should focus on testing the model on real quantum devices and applying it to larger datasets to further enhance its applicability. Full article