Pub Date : 2025-04-21DOI: 10.1109/ACCESS.2025.3561234
Chengwu Liang;Songqi Jiang;Jie Yang;Wei Hu;Yalong Liu;Peiwang Zhu;Guofeng He;Chunlei Shi
Distributed photovoltaic (DPV) power sites in industrial parks are characterized by dispersed layouts, practical fault detection environments, and high safety requirements. Conventional manual DPV O&M systems using handheld sensors are inefficient, expensive, and struggle with fault detection due to sparse industrial data and uni-modal information limitations. To this, this paper proposes an innovative advanced algorithm for DPV fault detection in industrial parks, utilizing a new sparse industrial dataset, “SolarPark,” collected via multi-modal UAVs and annotated through a multi-expert process with uncertainty scoring. By fusing the Convolutional Block Attention Module (CBAM), Bidirectional Feature Pyramid Network (BiFPN), Ghost modules, the algorithm enhances attention to critical photovoltaic fault-related channel information, strengthens multi-scale photovoltaic fault feature fusion, and achieves lightweight efficiency. Combined with multi-modal UAV videos, the proposed industrial DPV fault detection algorithm achieves a precision of 95.4%, effectively ensuring the efficiency of DPV power sites in data-scarce industrial scenarios. Extensive experiments on the developed cloud platform confirm the proposed algorithm’s efficient, cost-effective, and easy to deploy for aerial inspections of DPV O&M systems.
{"title":"A Distributed Photovoltaic Operation and Maintenance Cloud Platform for PV Aerial Inspections With Sparse Industrial Data","authors":"Chengwu Liang;Songqi Jiang;Jie Yang;Wei Hu;Yalong Liu;Peiwang Zhu;Guofeng He;Chunlei Shi","doi":"10.1109/ACCESS.2025.3561234","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3561234","url":null,"abstract":"Distributed photovoltaic (DPV) power sites in industrial parks are characterized by dispersed layouts, practical fault detection environments, and high safety requirements. Conventional manual DPV O&M systems using handheld sensors are inefficient, expensive, and struggle with fault detection due to sparse industrial data and uni-modal information limitations. To this, this paper proposes an innovative advanced algorithm for DPV fault detection in industrial parks, utilizing a new sparse industrial dataset, “SolarPark,” collected via multi-modal UAVs and annotated through a multi-expert process with uncertainty scoring. By fusing the Convolutional Block Attention Module (CBAM), Bidirectional Feature Pyramid Network (BiFPN), Ghost modules, the algorithm enhances attention to critical photovoltaic fault-related channel information, strengthens multi-scale photovoltaic fault feature fusion, and achieves lightweight efficiency. Combined with multi-modal UAV videos, the proposed industrial DPV fault detection algorithm achieves a precision of 95.4%, effectively ensuring the efficiency of DPV power sites in data-scarce industrial scenarios. Extensive experiments on the developed cloud platform confirm the proposed algorithm’s efficient, cost-effective, and easy to deploy for aerial inspections of DPV O&M systems.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"69677-69689"},"PeriodicalIF":3.4,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10972109","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143877679","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-21DOI: 10.1109/ACCESS.2025.3562666
Youngseong Cho;Kyoungil Lim
This study introduces a novel model for predicting control variables in end-to-end autonomous driving by leveraging polynomial and differential networks. Recent advancements in autonomous driving have predominantly focused on methods that incorporate additional supervisory data, such as attention mechanisms and bird’s-eye view images. However, these approaches are often hindered by issues related to computational efficiency and the high costs of data acquisition for real-world applications. In contrast, the proposed method enhances the performance by integrating polynomial and differential networks, facilitating efficient learning while accounting for the physical properties inherent in the data. The results of experiments conducted using the CARLA simulator demonstrate that the proposed model outperforms existing state-of-the-art approaches. The model weights and training code used in these experiments are available at https://github.com/choys0401/polydiff.
{"title":"Polynomial and Differential Networks for End-to-End Autonomous Driving","authors":"Youngseong Cho;Kyoungil Lim","doi":"10.1109/ACCESS.2025.3562666","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3562666","url":null,"abstract":"This study introduces a novel model for predicting control variables in end-to-end autonomous driving by leveraging polynomial and differential networks. Recent advancements in autonomous driving have predominantly focused on methods that incorporate additional supervisory data, such as attention mechanisms and bird’s-eye view images. However, these approaches are often hindered by issues related to computational efficiency and the high costs of data acquisition for real-world applications. In contrast, the proposed method enhances the performance by integrating polynomial and differential networks, facilitating efficient learning while accounting for the physical properties inherent in the data. The results of experiments conducted using the CARLA simulator demonstrate that the proposed model outperforms existing state-of-the-art approaches. The model weights and training code used in these experiments are available at <uri>https://github.com/choys0401/polydiff</uri>.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"69492-69499"},"PeriodicalIF":3.4,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10971419","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143877681","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-21DOI: 10.1109/ACCESS.2025.3562618
Huong Mai do;Tuan Phong Tran;Myungsik Yoo
Augmented reality (AR) in the internet of things requires ultra-low latency, high-resolution video, and fairness in multi-user environments, which pose challenges for traditional cloud and edge computing. To address this shortcoming, we studied AR subtask offloading and resource allocation in a multi-hop, multi-access edge computing federation. Our approach improves the quality of experience (QoE) by optimizing video quality and reducing delay while ensuring fairness, which is modeled as the ratio between provided and required quality. Instead of sequential execution, we adopt parallel AR subtask dependency processing to minimize latency. We propose an improved deep deterministic policy gradient algorithm for efficient solution exploration. Additionally, we implement strict training process monitoring to optimize resource usage and ensure sustainability. Experiments demonstrate that our method improves QoE by nearly 8% compared with TD3 while cutting training time in half.
{"title":"Quality of Experience Optimization for AR Service in an MEC Federation System","authors":"Huong Mai do;Tuan Phong Tran;Myungsik Yoo","doi":"10.1109/ACCESS.2025.3562618","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3562618","url":null,"abstract":"Augmented reality (AR) in the internet of things requires ultra-low latency, high-resolution video, and fairness in multi-user environments, which pose challenges for traditional cloud and edge computing. To address this shortcoming, we studied AR subtask offloading and resource allocation in a multi-hop, multi-access edge computing federation. Our approach improves the quality of experience (QoE) by optimizing video quality and reducing delay while ensuring fairness, which is modeled as the ratio between provided and required quality. Instead of sequential execution, we adopt parallel AR subtask dependency processing to minimize latency. We propose an improved deep deterministic policy gradient algorithm for efficient solution exploration. Additionally, we implement strict training process monitoring to optimize resource usage and ensure sustainability. Experiments demonstrate that our method improves QoE by nearly 8% compared with TD3 while cutting training time in half.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"69821-69839"},"PeriodicalIF":3.4,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10971371","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143877630","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-18DOI: 10.1109/ACCESS.2025.3562391
M. Erdem Isenkul
This paper introduces the Proof-of-Diversity (PoD) protocol, a new consensus mechanism that enhances decentralization, security, and energy efficiency using demographic, geographic, and computational diversity in validator selection. By using a multi-dimensional entropy-based approach, PoD shows high resistance to Sybil attacks, fosters inclusion, and ensures fair participation. Comparative analysis with Tendermint Proof-of-Stake (PoS) and Algorand Proof-of-Stake (Algorand) shows that PoD is more effective in various key metrics, including transaction finality, validator engagement, diversity entropy, energy use, and adaptability. In particular, PoD achieves a shortest average transaction finality time of 72.84 ms over a given period, a notable improvement compared to both Algorand at 215.37 ms and Tendermint PoS at 278.42 ms. In addition, PoD achieves a validator engagement of 85.42%, strengthening its ability to maintain decentralization. PoD also achieves a diversity score of 0.79, better than Tendermint PoS and Algorand, indicating a more fair and inclusive validator selection process. In terms of energy use, PoD achieves a mere 0.0132 kWh per transaction per second (TPS), a considerable improvement compared to its counterparts. In addition, PoD shows better adaptability to changes in step parameters and changes in benefit-cost ratios, further improving validator selection and network optimization. Overall, these results make PoD a scalable and sustainable consensus system that balances diversity, security, and performance in blockchain networks.
{"title":"Proof-of-Diversity (PoD): A Framework for Equitable Blockchain Governance","authors":"M. Erdem Isenkul","doi":"10.1109/ACCESS.2025.3562391","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3562391","url":null,"abstract":"This paper introduces the Proof-of-Diversity (PoD) protocol, a new consensus mechanism that enhances decentralization, security, and energy efficiency using demographic, geographic, and computational diversity in validator selection. By using a multi-dimensional entropy-based approach, PoD shows high resistance to Sybil attacks, fosters inclusion, and ensures fair participation. Comparative analysis with Tendermint Proof-of-Stake (PoS) and Algorand Proof-of-Stake (Algorand) shows that PoD is more effective in various key metrics, including transaction finality, validator engagement, diversity entropy, energy use, and adaptability. In particular, PoD achieves a shortest average transaction finality time of 72.84 ms over a given period, a notable improvement compared to both Algorand at 215.37 ms and Tendermint PoS at 278.42 ms. In addition, PoD achieves a validator engagement of 85.42%, strengthening its ability to maintain decentralization. PoD also achieves a diversity score of 0.79, better than Tendermint PoS and Algorand, indicating a more fair and inclusive validator selection process. In terms of energy use, PoD achieves a mere 0.0132 kWh per transaction per second (TPS), a considerable improvement compared to its counterparts. In addition, PoD shows better adaptability to changes in step parameters and changes in benefit-cost ratios, further improving validator selection and network optimization. Overall, these results make PoD a scalable and sustainable consensus system that balances diversity, security, and performance in blockchain networks.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"69116-69128"},"PeriodicalIF":3.4,"publicationDate":"2025-04-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10969777","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143875085","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-18DOI: 10.1109/ACCESS.2025.3562253
Shintaro Imatani;Kensuke Tobitani;Kyo Akabane
In this study, we propose a novel interaction method, TouchWIM, which combines World in Miniature (WIM) and Hybrid User Interface (HUI) to enhance the efficiency of object manipulation and reduce the workload in spatial design by using Augmented Reality (AR). WIM provides an additional overview perspective in AR by displaying a miniature representation of a room, and HUI enables an accurate and easy input by combining a head-mounted display (HMD) with a tablet. Our system allows the placement and manipulation of objects within a real space by touch interaction with the miniature representation of the room displayed on the tablet. To evaluate TouchWIM, we conducted user studies using a prototype spatial design system, comparing it with existing methods such as Hand-Ray + Direct Touch and WIM alone. The results demonstrated that TouchWIM is the most efficient and reduces the workload for the task of creating a specified spatial layout. This interaction method provides new insights into object manipulation and spatial design in AR.
{"title":"TouchWIM: Object Manipulation in AR Spatial Design With World in Miniature and Hybrid User Interface","authors":"Shintaro Imatani;Kensuke Tobitani;Kyo Akabane","doi":"10.1109/ACCESS.2025.3562253","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3562253","url":null,"abstract":"In this study, we propose a novel interaction method, TouchWIM, which combines World in Miniature (WIM) and Hybrid User Interface (HUI) to enhance the efficiency of object manipulation and reduce the workload in spatial design by using Augmented Reality (AR). WIM provides an additional overview perspective in AR by displaying a miniature representation of a room, and HUI enables an accurate and easy input by combining a head-mounted display (HMD) with a tablet. Our system allows the placement and manipulation of objects within a real space by touch interaction with the miniature representation of the room displayed on the tablet. To evaluate TouchWIM, we conducted user studies using a prototype spatial design system, comparing it with existing methods such as Hand-Ray + Direct Touch and WIM alone. The results demonstrated that TouchWIM is the most efficient and reduces the workload for the task of creating a specified spatial layout. This interaction method provides new insights into object manipulation and spatial design in AR.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"69269-69280"},"PeriodicalIF":3.4,"publicationDate":"2025-04-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10969634","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143875145","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-18DOI: 10.1109/ACCESS.2025.3562343
Swati Soni;Lalita Gupta;Rishav Dubey
The novel deep learning-based time domain single channel speech source separation methods have shown remarkable progress. Recent studies achieve either successful global or local context modeling for monaural speaker separation. Existing CNN-based methods perform local context modeling, and RNN-based or attention-based methods work on the global context of the speech signal. In this paper, we proposed two models which parallelly combine CNN-RNN-based and CNN-attention-based separation modules and perform parallel local and global context modeling. Our models keep maximum global or local context value at a particular time step. These values help our models to separate the speaker signals more accurately. We have conducted the experiments on Libri2mix and Libri3mix datasets. The experimental data demonstrates that our proposed models have outperformed the state-of-the-art methods. Our proposed models remarkably improve SDR and SI-SDR values on Libri2mix and Libri3mix datasets. The proposed parallel CNN-RNN-based and CNN-attention-based separation models achieve average SDR improvement of 2.10 dB and 2.21 dB, respectively, and SI-SDR improvement of 2.74 dB and 2.78 dB, respectively, on the Libri2mix dataset. However, on the Libri3mix dataset, the proposed models achieve 0.57 dB and 0.87 dB average SDR improvement for parallel CNN-RNN-based separation module, and 0.88 dB and 1.4 dB average SI-SDR improvement for CNN-attention-based separation models. Our work indirectly contributes to SDG Goal 10 (Reduced Inequalities) by improving communication tools for diverse linguistic communities. Furthermore, this technology aids SDG Goal 9 (Industry, Innovation, and Infrastructure) by advancing AI-powered assistive technologies, fostering innovation, and building resilient communication systems.
{"title":"Parallel Local and Global Context Modeling of Deep Learning-Based Monaural Speech Source Separation Techniques","authors":"Swati Soni;Lalita Gupta;Rishav Dubey","doi":"10.1109/ACCESS.2025.3562343","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3562343","url":null,"abstract":"The novel deep learning-based time domain single channel speech source separation methods have shown remarkable progress. Recent studies achieve either successful global or local context modeling for monaural speaker separation. Existing CNN-based methods perform local context modeling, and RNN-based or attention-based methods work on the global context of the speech signal. In this paper, we proposed two models which parallelly combine CNN-RNN-based and CNN-attention-based separation modules and perform parallel local and global context modeling. Our models keep maximum global or local context value at a particular time step. These values help our models to separate the speaker signals more accurately. We have conducted the experiments on Libri2mix and Libri3mix datasets. The experimental data demonstrates that our proposed models have outperformed the state-of-the-art methods. Our proposed models remarkably improve SDR and SI-SDR values on Libri2mix and Libri3mix datasets. The proposed parallel CNN-RNN-based and CNN-attention-based separation models achieve average SDR improvement of 2.10 dB and 2.21 dB, respectively, and SI-SDR improvement of 2.74 dB and 2.78 dB, respectively, on the Libri2mix dataset. However, on the Libri3mix dataset, the proposed models achieve 0.57 dB and 0.87 dB average SDR improvement for parallel CNN-RNN-based separation module, and 0.88 dB and 1.4 dB average SI-SDR improvement for CNN-attention-based separation models. Our work indirectly contributes to SDG Goal 10 (Reduced Inequalities) by improving communication tools for diverse linguistic communities. Furthermore, this technology aids SDG Goal 9 (Industry, Innovation, and Infrastructure) by advancing AI-powered assistive technologies, fostering innovation, and building resilient communication systems.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"68607-68621"},"PeriodicalIF":3.4,"publicationDate":"2025-04-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10969763","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143875074","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-18DOI: 10.1109/ACCESS.2025.3562154
Yujia Wang;Wen Long
With the continuous evolution of advanced large language models like GPT, the proliferation of AI-generated fake news presents growing challenges to information dissemination. Traditional text classification methods face difficulties in accurately detecting such content, due to their limited capacity to differentiate between authentic and fabricated news. To address this issue, this paper introduces a novel “Global-Local News Detection Model”, which combines BERT, Bidirectional Long Short-Term Memory (BiLSTM) networks, Text Convolutional Neural Networks (TextCNN), and attention mechanisms to enhance the detection of AI-generated fake news. A new dataset, generated using GPT-4 and covering 42 news categories, was developed to serve as a comprehensive and diverse foundation for training and evaluating the model. Experimental results indicate that the proposed model achieves an accuracy and F1 score of 0.82, surpassing traditional approaches.
{"title":"Global-Local Ensemble Detector for AI-Generated Fake News","authors":"Yujia Wang;Wen Long","doi":"10.1109/ACCESS.2025.3562154","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3562154","url":null,"abstract":"With the continuous evolution of advanced large language models like GPT, the proliferation of AI-generated fake news presents growing challenges to information dissemination. Traditional text classification methods face difficulties in accurately detecting such content, due to their limited capacity to differentiate between authentic and fabricated news. To address this issue, this paper introduces a novel “Global-Local News Detection Model”, which combines BERT, Bidirectional Long Short-Term Memory (BiLSTM) networks, Text Convolutional Neural Networks (TextCNN), and attention mechanisms to enhance the detection of AI-generated fake news. A new dataset, generated using GPT-4 and covering 42 news categories, was developed to serve as a comprehensive and diverse foundation for training and evaluating the model. Experimental results indicate that the proposed model achieves an accuracy and F1 score of 0.82, surpassing traditional approaches.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"69779-69789"},"PeriodicalIF":3.4,"publicationDate":"2025-04-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10969761","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143877678","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-18DOI: 10.1109/ACCESS.2025.3562453
B. B. Shabarinath;Muralidhar Pullakandam
Edge computing allows to do AI processing on devices with limited resources, but the challenge remains high computational costs followed by the energy limitations of such devices making on-device machine learning inefficient, especially for Support Vector Machine (SVM) classifiers. Although SVM classifiers are generally very accurate, they require solving a quadratic optimization problem, making their implementation in real-time embedded devices challenging. While Sequential Minimal Optimization (SMO) has enhanced the efficiency of SVM training, traditional implementations still suffer from high computational cost. In this paper, we propose Parallel SMO, a new algorithm that selects multiple violating pairs in each iteration, allowing batch-wise updates that enhance convergence speed and optimize parallel computation. By buffering kernel values, it minimizes redundant computations, leading to improved memory efficiency and faster SVM training on FPGA architectures. In addition, we present a embedded hardware-efficient FPGA architecture for the integrated SVM learning based on Parallel SMO with SVM inference. It consists of SVM controller that schedules the operations of each clock cycle such that computations and memory access happen concurrently. The dynamic pipeline scheduling employ parameterized modules to schedule linear or nonlinear kernels and produce dimension-based reconfigurable blocks. A configuration signal turns on corresponding sub-blocks and clock-gating unused ones, thus enhancing resource utilization efficiency, energy efficiency, and overall performance. In several benchmarking data sets, the scheme reduces clock cycles per iteration consistently and improves throughput (up to 2427 iterations per second). It achieves up to 98% accuracy in classification with low power consumption, as reflected by training power of $47 mW$ and high energy efficiency (up to $51.64e+3$ iterations per joule). With the assistance of an adaptive kernel datapath, parallel error update execution, and best-pair selection, the scheme facilitates faster convergence, higher throughput, and on-chip inference with resource efficiency maintained.
{"title":"Embedded Hardware-Efficient FPGA Architecture for SVM Learning and Inference","authors":"B. B. Shabarinath;Muralidhar Pullakandam","doi":"10.1109/ACCESS.2025.3562453","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3562453","url":null,"abstract":"Edge computing allows to do AI processing on devices with limited resources, but the challenge remains high computational costs followed by the energy limitations of such devices making on-device machine learning inefficient, especially for Support Vector Machine (SVM) classifiers. Although SVM classifiers are generally very accurate, they require solving a quadratic optimization problem, making their implementation in real-time embedded devices challenging. While Sequential Minimal Optimization (SMO) has enhanced the efficiency of SVM training, traditional implementations still suffer from high computational cost. In this paper, we propose Parallel SMO, a new algorithm that selects multiple violating pairs in each iteration, allowing batch-wise updates that enhance convergence speed and optimize parallel computation. By buffering kernel values, it minimizes redundant computations, leading to improved memory efficiency and faster SVM training on FPGA architectures. In addition, we present a embedded hardware-efficient FPGA architecture for the integrated SVM learning based on Parallel SMO with SVM inference. It consists of SVM controller that schedules the operations of each clock cycle such that computations and memory access happen concurrently. The dynamic pipeline scheduling employ parameterized modules to schedule linear or nonlinear kernels and produce dimension-based reconfigurable blocks. A configuration signal turns on corresponding sub-blocks and clock-gating unused ones, thus enhancing resource utilization efficiency, energy efficiency, and overall performance. In several benchmarking data sets, the scheme reduces clock cycles per iteration consistently and improves throughput (up to 2427 iterations per second). It achieves up to 98% accuracy in classification with low power consumption, as reflected by training power of <inline-formula> <tex-math>$47 mW$ </tex-math></inline-formula> and high energy efficiency (up to <inline-formula> <tex-math>$51.64e+3$ </tex-math></inline-formula> iterations per joule). With the assistance of an adaptive kernel datapath, parallel error update execution, and best-pair selection, the scheme facilitates faster convergence, higher throughput, and on-chip inference with resource efficiency maintained.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"68930-68947"},"PeriodicalIF":3.4,"publicationDate":"2025-04-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10969767","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143875095","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-18DOI: 10.1109/ACCESS.2025.3562495
Ricardo Tirone Fidelis;Ghunter Paulo Viajante;Eric Nery Chaves;Carlos E. Tavares;Augusto W. F. V. Da Silveira;Luciano Coutinho Gomes
This work highlights advances in the torque estimation method for Switched Reluctance Machines (SRMs), focusing on a high-precision torque estimator that integrates classical and modern modeling techniques. The employed method uses cubic splines and Lagrange polynomials to model the inductance surfaces in order to optimize the estimated instantaneous torque. This approach optimizes drive systems, making SRMs more efficient for critical industrial applications, such as electric vehicle propulsion and renewable energy systems. The method, validated through simulations and experiments, presents an accuracy of about 97% in the reconstruction of the inductance surface, which guarantees the high performance of the estimated torque. The presented results indicate that the high detail of the inductance variations in SRMs contributes positively to the real-time and low-computation torque estimation. Thus, this work contributes to the development of more efficient electric drive control systems, which allow advances in sustainable, accurate and effective industrial applications.
{"title":"Torque Estimation in Switched Reluctance Machines: A Comprehensive Approach Involving Inductance Modeling Techniques","authors":"Ricardo Tirone Fidelis;Ghunter Paulo Viajante;Eric Nery Chaves;Carlos E. Tavares;Augusto W. F. V. Da Silveira;Luciano Coutinho Gomes","doi":"10.1109/ACCESS.2025.3562495","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3562495","url":null,"abstract":"This work highlights advances in the torque estimation method for Switched Reluctance Machines (SRMs), focusing on a high-precision torque estimator that integrates classical and modern modeling techniques. The employed method uses cubic splines and Lagrange polynomials to model the inductance surfaces in order to optimize the estimated instantaneous torque. This approach optimizes drive systems, making SRMs more efficient for critical industrial applications, such as electric vehicle propulsion and renewable energy systems. The method, validated through simulations and experiments, presents an accuracy of about 97% in the reconstruction of the inductance surface, which guarantees the high performance of the estimated torque. The presented results indicate that the high detail of the inductance variations in SRMs contributes positively to the real-time and low-computation torque estimation. Thus, this work contributes to the development of more efficient electric drive control systems, which allow advances in sustainable, accurate and effective industrial applications.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"68270-68289"},"PeriodicalIF":3.4,"publicationDate":"2025-04-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10969783","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143875135","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-18DOI: 10.1109/ACCESS.2025.3562432
Guojiang Zheng;Yang Lu;Hui Chen
Alzheimer’s disease (AD), a leading neurodegenerative disorder, progresses from an intermediary stage known as Mild Cognitive Impairment (MCI), characterized by measurable cognitive decline with retained functional independence. Accurate prediction of MCI progression to AD is critical for timely interventions. Existing deep learning-based methods for structural MRI (sMRI) analysis predominantly utilize either Convolutional Neural Networks (CNNs), which effectively capture local features but neglect global context, or Transformer architectures that model global dependencies yet require extensive data and computational resources. Additionally, many methods inadequately leverage longitudinal imaging data, limiting their sensitivity to subtle temporal changes in brain morphology. To overcome these limitations, we introduce EffiSwin-MCI, a novel hybrid deep learning framework integrating EfficientNet and Swin Transformer architectures, specifically designed for longitudinal sMRI analysis. The primary novelty of EffiSwin-MCI lies in its sliding-window attention mechanism, inspired by the Swin Transformer, which effectively integrates localized spatial dependencies within 2D sMRI slices, combined with temporal attention blocks that fuse spatial-temporal features across longitudinal scans at two distinct time points (T1 and T2). EfficientNet-B2 serves as a computationally efficient backbone, extracting hierarchical spatial features crucial for detailed morphological characterization. This alternating spatial and temporal attention strategy uniquely captures progressive local and global structural changes indicative of cognitive decline. Comprehensive experiments conducted on the Alzheimer’s Disease ADNI dataset demonstrate the proposed model’s superior performance compared to state-of-the-art CNN and Transformer-based approaches, achieving an accuracy of 81.69%, recall of 80.27%, precision of 84.35%, and F1-score of 82.27%. EffiSwin-MCI’s interpretability is further validated through Grad-CAM visualizations, highlighting critical neurodegenerative biomarkers such as the hippocampus and amygdala, reinforcing its clinical relevance for early prediction and intervention strategies in MCI management.
{"title":"Deep Learning-Based Framework for Predicting Mild Cognitive Impairment Progression in Neurology Using Longitudinal MRI","authors":"Guojiang Zheng;Yang Lu;Hui Chen","doi":"10.1109/ACCESS.2025.3562432","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3562432","url":null,"abstract":"Alzheimer’s disease (AD), a leading neurodegenerative disorder, progresses from an intermediary stage known as Mild Cognitive Impairment (MCI), characterized by measurable cognitive decline with retained functional independence. Accurate prediction of MCI progression to AD is critical for timely interventions. Existing deep learning-based methods for structural MRI (sMRI) analysis predominantly utilize either Convolutional Neural Networks (CNNs), which effectively capture local features but neglect global context, or Transformer architectures that model global dependencies yet require extensive data and computational resources. Additionally, many methods inadequately leverage longitudinal imaging data, limiting their sensitivity to subtle temporal changes in brain morphology. To overcome these limitations, we introduce EffiSwin-MCI, a novel hybrid deep learning framework integrating EfficientNet and Swin Transformer architectures, specifically designed for longitudinal sMRI analysis. The primary novelty of EffiSwin-MCI lies in its sliding-window attention mechanism, inspired by the Swin Transformer, which effectively integrates localized spatial dependencies within 2D sMRI slices, combined with temporal attention blocks that fuse spatial-temporal features across longitudinal scans at two distinct time points (T1 and T2). EfficientNet-B2 serves as a computationally efficient backbone, extracting hierarchical spatial features crucial for detailed morphological characterization. This alternating spatial and temporal attention strategy uniquely captures progressive local and global structural changes indicative of cognitive decline. Comprehensive experiments conducted on the Alzheimer’s Disease ADNI dataset demonstrate the proposed model’s superior performance compared to state-of-the-art CNN and Transformer-based approaches, achieving an accuracy of 81.69%, recall of 80.27%, precision of 84.35%, and F1-score of 82.27%. EffiSwin-MCI’s interpretability is further validated through Grad-CAM visualizations, highlighting critical neurodegenerative biomarkers such as the hippocampus and amygdala, reinforcing its clinical relevance for early prediction and intervention strategies in MCI management.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"68903-68919"},"PeriodicalIF":3.4,"publicationDate":"2025-04-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10969769","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143875173","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：