Pub Date : 2026-03-04DOI: 10.1109/ACCESS.2026.3670366
Afnan Shobil;Salah Al-Hagree
With a focus on user reviews from the Google Play Store, this study investigates user perceptions of mobile applications designed for remote monitoring and control of solar energy systems. This study uses sentiment analysis, a powerful natural language processing tool, to extract valuable information from structured and unstructured text data. It classifies opinions and comments as neutral, negative, or positive to understand user experiences, identify significant areas of satisfaction and dissatisfaction, and provide application developers with relevant information by analyzing more than 17,100 user comments across 18 applications that were updated in 2024, collected using the Google-Play-scraper Python script. The reviews were analyzed using the Valence Aware Dictionary and Sentiment Reasoner (VADER) to determine whether they are positive, neutral, or negative. The analysis revealed that a large majority of users, 67.4%, expressed primarily positive sentiments about the ease of use and effectiveness of these apps in monitoring solar systems. Approximately 15.1% of the reviews are negative, and 17.5% are neutral. The analysis leverages classical supervised machine learning (ML) models in combination with lexicon-based sentiment analysis (VADER) to classify the sentiment expressed in comments. Support Vector Machine (SVM) is one of the most effective supervised machine learning methods that can be used for sentiment classification tasks, achieving 91% accuracy, indicating the effectiveness of SVM in sentiment classification when compared to other machine learning methods, such as Decision Tree (DT), Logistic Regression (LR), RidgeClassifier, Random Forest (RF), and Naive Bayes (NB) which achieved accuracies of 90%, 89%, 87%, 68%, and 40%, respectively. However, the study also identified areas for improvement, such as addressing technical bugs, improving response times, and enhancing the user interface. These findings provide valuable insights for developers to improve user experience, enhance app functionality, and ultimately increase user satisfaction with solar monitoring apps. The findings also show that sentiment analysis is a useful technique for classifying and assessing user reviews and feedback.
{"title":"User Sentiment Analysis of Solar Energy Monitoring Apps: Insights From Google Play Reviews","authors":"Afnan Shobil;Salah Al-Hagree","doi":"10.1109/ACCESS.2026.3670366","DOIUrl":"https://doi.org/10.1109/ACCESS.2026.3670366","url":null,"abstract":"With a focus on user reviews from the Google Play Store, this study investigates user perceptions of mobile applications designed for remote monitoring and control of solar energy systems. This study uses sentiment analysis, a powerful natural language processing tool, to extract valuable information from structured and unstructured text data. It classifies opinions and comments as neutral, negative, or positive to understand user experiences, identify significant areas of satisfaction and dissatisfaction, and provide application developers with relevant information by analyzing more than 17,100 user comments across 18 applications that were updated in 2024, collected using the Google-Play-scraper Python script. The reviews were analyzed using the Valence Aware Dictionary and Sentiment Reasoner (VADER) to determine whether they are positive, neutral, or negative. The analysis revealed that a large majority of users, 67.4%, expressed primarily positive sentiments about the ease of use and effectiveness of these apps in monitoring solar systems. Approximately 15.1% of the reviews are negative, and 17.5% are neutral. The analysis leverages classical supervised machine learning (ML) models in combination with lexicon-based sentiment analysis (VADER) to classify the sentiment expressed in comments. Support Vector Machine (SVM) is one of the most effective supervised machine learning methods that can be used for sentiment classification tasks, achieving 91% accuracy, indicating the effectiveness of SVM in sentiment classification when compared to other machine learning methods, such as Decision Tree (DT), Logistic Regression (LR), RidgeClassifier, Random Forest (RF), and Naive Bayes (NB) which achieved accuracies of 90%, 89%, 87%, 68%, and 40%, respectively. However, the study also identified areas for improvement, such as addressing technical bugs, improving response times, and enhancing the user interface. These findings provide valuable insights for developers to improve user experience, enhance app functionality, and ultimately increase user satisfaction with solar monitoring apps. The findings also show that sentiment analysis is a useful technique for classifying and assessing user reviews and feedback.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"14 ","pages":"38421-38433"},"PeriodicalIF":3.6,"publicationDate":"2026-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11421327","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147440528","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}
Fuzzy extractors (FEs) are cryptographic primitives that generate a random key from a sample of a randomness source, together with some helper data that can be used to reproduce the key from a second sample of the source that is “close” to the first one. Reusable FEs can be used securely to derive multiple keys. Robust FEs detect any tempering with the helper data with overwhelming probability. In this paper we make two contributions. First, we construct a computationally secure, robust, and reusable FE (rrFE) that satisfies the strongest notion of reusability security, and its security reduces to the hardness of a new computational assumption that is closely related to LPN problem, for which no efficient quantum algorithm is known. The proof is in the standard model, answering an open question of Canetti et al. (Journal of Cryptology 2021). We implement and evaluate our rrFE using a widely used data set of iris data as the randomness source and compare its performance with a reusable only scheme for the same source. We also introduce and formalize password-protected fuzzy extractors (PPFEs), which use passwords as an additional source of entropy to enhance the security of biometric data against offline attacks (by a constant amount). We further present a PPFE construction with proved security. Second, we motivate and propose a new cryptographic primitive called Password Protected Message Retrieval (PPMR) that enables a user to securely store the helper data on a remote server and later securely recover it on a local device, using one round of communication without requiring a secret key and only relying on a retrieval password. The helper data is used to recover a biometric-based pre-shared key with the server. This removes the need to store sensitive biometric data on the device and allows the AKE protocol to be securely executed from any device that can correctly execute a code (e.g. an uncorrupted terminal).
模糊提取器（FEs）是一种加密原语，它从随机源的样本生成随机密钥，以及一些辅助数据，这些数据可用于从与第一个“接近”的源的第二个样本复制密钥。可重用的fe可以安全地用于派生多个密钥。健壮的FEs以压倒性的概率检测对助手数据的任何篡改。在本文中，我们做了两个贡献。首先，我们构建了一个计算安全、鲁棒和可重用的FE (rrFE)，它满足最强的可重用性安全概念，其安全性降低到与LPN问题密切相关的新计算假设的硬度，而LPN问题目前还没有有效的量子算法。该证明是在标准模型中，回答了Canetti等人的一个开放问题（Journal of Cryptology 2021）。我们使用广泛使用的虹膜数据集作为随机源来实现和评估我们的rrFE，并将其性能与同一来源的可重用方案进行比较。我们还引入并形式化了密码保护的模糊提取器（ppfe），它使用密码作为额外的熵源来增强生物特征数据抵御离线攻击的安全性（通过恒定的数量）。我们进一步提出了一种安全可靠的PPFE结构。其次，我们激发并提出了一种新的加密原语，称为密码保护消息检索（PPMR），它使用户能够安全地将助手数据存储在远程服务器上，然后在本地设备上安全地恢复它，使用一轮通信，而不需要密钥，只依赖于检索密码。helper数据用于与服务器恢复基于生物特征的预共享密钥。这消除了在设备上存储敏感生物识别数据的需要，并允许AKE协议从可以正确执行代码的任何设备安全地执行（例如，未损坏的终端）。
{"title":"Robust and Reusable Fuzzy Extractor for Low-Entropy Distributions and Application to User Authentication","authors":"Somnath Panja;Nikita Tripathi;Shaoquan Jiang;Reihaneh Safavi-Naini","doi":"10.1109/ACCESS.2026.3670084","DOIUrl":"https://doi.org/10.1109/ACCESS.2026.3670084","url":null,"abstract":"Fuzzy extractors (FEs) are cryptographic primitives that generate a random key from a sample of a randomness source, together with some helper data that can be used to reproduce the key from a second sample of the source that is “close” to the first one. Reusable FEs can be used securely to derive multiple keys. Robust FEs detect any tempering with the helper data with overwhelming probability. In this paper we make two contributions. First, we construct a computationally secure, robust, and reusable FE (rrFE) that satisfies the strongest notion of reusability security, and its security reduces to the hardness of a new computational assumption that is closely related to LPN problem, for which no efficient quantum algorithm is known. The proof is in the standard model, answering an open question of Canetti et al. (Journal of Cryptology 2021). We implement and evaluate our rrFE using a widely used data set of iris data as the randomness source and compare its performance with a reusable only scheme for the same source. We also introduce and formalize password-protected fuzzy extractors (PPFEs), which use passwords as an additional source of entropy to enhance the security of biometric data against offline attacks (by a constant amount). We further present a PPFE construction with proved security. Second, we motivate and propose a new cryptographic primitive called Password Protected Message Retrieval (PPMR) that enables a user to securely store the helper data on a remote server and later securely recover it on a local device, using one round of communication without requiring a secret key and only relying on a retrieval password. The helper data is used to recover a biometric-based pre-shared key with the server. This removes the need to store sensitive biometric data on the device and allows the AKE protocol to be securely executed from any device that can correctly execute a code (e.g. an uncorrupted terminal).","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"14 ","pages":"38230-38250"},"PeriodicalIF":3.6,"publicationDate":"2026-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11418906","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147440531","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 : 2026-03-02DOI: 10.1109/ACCESS.2026.3669162
Lei Wang;Heng Zhang;Xiuying Wang;Zhiwei Guan;Mei Xiao;Jian Liu;Mingjiang Wei;Yong Pan
To address the challenge of lane-changing prediction for autonomous vehicles (AVs) within highway weaving sections characterized by dynamic interactions in human-machine mixed traffic flows, this paper proposes a multi-expert collaborative prediction model based on a dynamic conflict field (DCF) and Generative Adversarial Networks, termed G-MoE-WGAN. This model quantifies the dynamic game-theoretic intensity among interacting vehicles via the DCF. The proposed model constructs a Mixture of Experts (MoE) system to facilitate intention decision-making and trajectory generation, employing adversarial training to optimize the physical plausibility and interaction adaptability of the prediction results. Furthermore, a Physics-Informed Neural Network (PINN) is introduced for the reconstruction and smoothing of raw naturalistic driving trajectories to validate the model. Experimental results demonstrate that the G-MoE-WGAN achieves a lane-changing intention classification accuracy of 94.35%, representing an improvement of 3.89% to 16.55% compared to baseline models. Within a 3-second prediction horizon, the Final Displacement Error (FDE) and Average Displacement Error (ADE) of the proposed model are significantly reduced by 8.61%–40.55% and 3.21%–40.27%, respectively. In the 5-second long-term prediction, benefiting from the dynamic expert fusion mechanism, the ADE and FDE metrics exhibit a reduction in error of 3.19%–39.29% relative to comparative methods. The study indicates that the proposed conflict potential representation and the multi-expert adversarial training mechanism effectively capture the spatiotemporal heterogeneity of high-dynamic interactions in weaving sections significantly enhancing the robustness and interpretability of lane-changing predictions in complex scenarios.
{"title":"Multi-Expert Trajectory Prediction for Highway Weaving Sections Using Conflict Potential Energy and GAN","authors":"Lei Wang;Heng Zhang;Xiuying Wang;Zhiwei Guan;Mei Xiao;Jian Liu;Mingjiang Wei;Yong Pan","doi":"10.1109/ACCESS.2026.3669162","DOIUrl":"https://doi.org/10.1109/ACCESS.2026.3669162","url":null,"abstract":"To address the challenge of lane-changing prediction for autonomous vehicles (AVs) within highway weaving sections characterized by dynamic interactions in human-machine mixed traffic flows, this paper proposes a multi-expert collaborative prediction model based on a dynamic conflict field (DCF) and Generative Adversarial Networks, termed G-MoE-WGAN. This model quantifies the dynamic game-theoretic intensity among interacting vehicles via the DCF. The proposed model constructs a Mixture of Experts (MoE) system to facilitate intention decision-making and trajectory generation, employing adversarial training to optimize the physical plausibility and interaction adaptability of the prediction results. Furthermore, a Physics-Informed Neural Network (PINN) is introduced for the reconstruction and smoothing of raw naturalistic driving trajectories to validate the model. Experimental results demonstrate that the G-MoE-WGAN achieves a lane-changing intention classification accuracy of 94.35%, representing an improvement of 3.89% to 16.55% compared to baseline models. Within a 3-second prediction horizon, the Final Displacement Error (FDE) and Average Displacement Error (ADE) of the proposed model are significantly reduced by 8.61%–40.55% and 3.21%–40.27%, respectively. In the 5-second long-term prediction, benefiting from the dynamic expert fusion mechanism, the ADE and FDE metrics exhibit a reduction in error of 3.19%–39.29% relative to comparative methods. The study indicates that the proposed conflict potential representation and the multi-expert adversarial training mechanism effectively capture the spatiotemporal heterogeneity of high-dynamic interactions in weaving sections significantly enhancing the robustness and interpretability of lane-changing predictions in complex scenarios.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"14 ","pages":"34473-34492"},"PeriodicalIF":3.6,"publicationDate":"2026-03-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11417815","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147362377","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 : 2026-03-02DOI: 10.1109/ACCESS.2026.3669451
John S. Fata;Wafa M. Elmannai
This work presents a low-cost FPGA-based accelerator for real-time object detection and classification using a compressed YOLOv3-Tiny model. Existing FPGA-based CNN accelerators excel in one critical performance metric but sacrifice either throughput, accuracy, or power efficiency. This is particularly the case for low-cost devices that are resource-constrained and often heavily rely on off-chip memory which hinders performance. To address these limitations, we introduce three novel contributions: 1) an iterative structured hardware pruning algorithm that removes the least important filters from the YOLO model in small increments; 2) a quantization-aware training (QAT) algorithm that adapts the scaling factor per layer; and 3) a custom RTL memory-mapping controller that prioritizes on-chip BRAM/URAM memory allocation to improve throughput while decreasing power consumption. With this approach, the model size was reduced by $13.3times $ while preserving accuracy. Implemented on a low-cost Kria KV260 FPGA, the approach achieved 93.8% detection accuracy, 24.3 FPS throughput, 41.59 ms latency, and just 2.13W of power consumption. The result was a high-performing, efficient system that directly outperformed comparable low-cost designs. These results demonstrate that balanced, high-performance YOLO inference is attainable on low-cost FPGA hardware without reliance on off-chip memory.
{"title":"Low-Cost FPGA-Enhanced CNN Accelerator for Real-Time YOLO Object Detection and Classification","authors":"John S. Fata;Wafa M. Elmannai","doi":"10.1109/ACCESS.2026.3669451","DOIUrl":"https://doi.org/10.1109/ACCESS.2026.3669451","url":null,"abstract":"This work presents a low-cost FPGA-based accelerator for real-time object detection and classification using a compressed YOLOv3-Tiny model. Existing FPGA-based CNN accelerators excel in one critical performance metric but sacrifice either throughput, accuracy, or power efficiency. This is particularly the case for low-cost devices that are resource-constrained and often heavily rely on off-chip memory which hinders performance. To address these limitations, we introduce three novel contributions: 1) an iterative structured hardware pruning algorithm that removes the least important filters from the YOLO model in small increments; 2) a quantization-aware training (QAT) algorithm that adapts the scaling factor per layer; and 3) a custom RTL memory-mapping controller that prioritizes on-chip BRAM/URAM memory allocation to improve throughput while decreasing power consumption. With this approach, the model size was reduced by <inline-formula> <tex-math>$13.3times $ </tex-math></inline-formula> while preserving accuracy. Implemented on a low-cost Kria KV260 FPGA, the approach achieved 93.8% detection accuracy, 24.3 FPS throughput, 41.59 ms latency, and just 2.13W of power consumption. The result was a high-performing, efficient system that directly outperformed comparable low-cost designs. These results demonstrate that balanced, high-performance YOLO inference is attainable on low-cost FPGA hardware without reliance on off-chip memory.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"14 ","pages":"34614-34642"},"PeriodicalIF":3.6,"publicationDate":"2026-03-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11417804","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147362277","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 : 2026-03-02DOI: 10.1109/ACCESS.2026.3669714
A. Dominguez;B. da Costa Paulo;I. Burguera;I. Tamayo;A. Elosegi;S. Cabrero Barros;M. Zorrilla
Volumetric video represents a transformative advancement in multimedia, offering the ability to capture and render three-dimensional content for immersive and interactive experiences. As the demand for immersive web applications grows, the need for a robust platform to stream live volumetric video on the web becomes more critical. This paper presents a prototype designed to stream volumetric video over 5G networks using web technologies. The platform enables real-time transmission and rendering of volumetric video in point cloud format, compressed with the Draco codec, and streamed via WebSocket and HTTP/DASH protocols. We conducted an empirical study to evaluate the performance of these technologies under different network technologies, transport protocols and scenarios, including the experimental network testbed of a commercial network provider: TELENOR. This new evidence complements our previous empirical study that confirmed the readiness of current devices and browsers for web-based volumetric video streaming. The results highlight significant differences in device performance and offer valuable insights into the limitations and opportunities for the future of volumetric video streaming on the web. Moreover, we are publishing the dataset generated during our empirical evaluation as an additional contribution, as it can be used for further analysis, simulation and model training. Finally, the paper discusses the technical and practical considerations for deploying volumetric video applications, laying the foundation for further advancements in the field.
{"title":"A Web-Ready and 5G-Ready Volumetric Video Streaming Platform: A Platform Prototype and Empirical Study","authors":"A. Dominguez;B. da Costa Paulo;I. Burguera;I. Tamayo;A. Elosegi;S. Cabrero Barros;M. Zorrilla","doi":"10.1109/ACCESS.2026.3669714","DOIUrl":"https://doi.org/10.1109/ACCESS.2026.3669714","url":null,"abstract":"Volumetric video represents a transformative advancement in multimedia, offering the ability to capture and render three-dimensional content for immersive and interactive experiences. As the demand for immersive web applications grows, the need for a robust platform to stream live volumetric video on the web becomes more critical. This paper presents a prototype designed to stream volumetric video over 5G networks using web technologies. The platform enables real-time transmission and rendering of volumetric video in point cloud format, compressed with the Draco codec, and streamed via WebSocket and HTTP/DASH protocols. We conducted an empirical study to evaluate the performance of these technologies under different network technologies, transport protocols and scenarios, including the experimental network testbed of a commercial network provider: TELENOR. This new evidence complements our previous empirical study that confirmed the readiness of current devices and browsers for web-based volumetric video streaming. The results highlight significant differences in device performance and offer valuable insights into the limitations and opportunities for the future of volumetric video streaming on the web. Moreover, we are publishing the dataset generated during our empirical evaluation as an additional contribution, as it can be used for further analysis, simulation and model training. Finally, the paper discusses the technical and practical considerations for deploying volumetric video applications, laying the foundation for further advancements in the field.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"14 ","pages":"34655-34675"},"PeriodicalIF":3.6,"publicationDate":"2026-03-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11418643","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147362279","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}
Stochastic optimal control problems are commonly formulated as optimization problems constrained by stochastic dynamical systems, whose value functions satisfy Hamilton–Jacobi–Bellman (HJB) equations. Owing to their strong nonlinearity and high dimensionality, closed-form solutions of HJB equations are rarely available, thereby motivating the development of robust and highly accurate numerical methods. This research introduces two hybrid spectral–collocation strategies for the numerical solution of stochastic HJB equations, constructed from different combinations of orthogonal polynomial bases. The first strategy utilizes shifted Chebyshev polynomials for time approximation and fractional-order Legendre polynomials for state approximation, while the second utilizes shifted Legendre polynomials in time and fractional-order Chebyshev polynomials in state. A convergence analysis is developed within the Caputo fractional derivative framework to justify the proposed methods and to establish the associated error estimates. The resulting nonlinear algebraic system is then solved using the collocation method. Numerical simulations, including an application to a resource extraction model, confirm that the proposed methods attain a high level of accuracy and exhibit convergence rates in strong agreement with the theoretical predictions. These results demonstrate that the developed hybrid spectral–collocation frameworks constitute reliable and efficient tools for addressing stochastic optimal control problems based on HJB equations.
{"title":"A Hybrid Fractional Chebyshev–Legendre Spectral Collocation Method for Hamilton–Jacobi–Bellman Equations","authors":"Alvian Alif Hidayatullah;Subchan Subchan;Sena Safarina;Tahiyatul Asfihani;R. Mohamad Atok;Irma Fitria;Andriyansah;Kasno Pamungkas","doi":"10.1109/ACCESS.2026.3668899","DOIUrl":"https://doi.org/10.1109/ACCESS.2026.3668899","url":null,"abstract":"Stochastic optimal control problems are commonly formulated as optimization problems constrained by stochastic dynamical systems, whose value functions satisfy Hamilton–Jacobi–Bellman (HJB) equations. Owing to their strong nonlinearity and high dimensionality, closed-form solutions of HJB equations are rarely available, thereby motivating the development of robust and highly accurate numerical methods. This research introduces two hybrid spectral–collocation strategies for the numerical solution of stochastic HJB equations, constructed from different combinations of orthogonal polynomial bases. The first strategy utilizes shifted Chebyshev polynomials for time approximation and fractional-order Legendre polynomials for state approximation, while the second utilizes shifted Legendre polynomials in time and fractional-order Chebyshev polynomials in state. A convergence analysis is developed within the Caputo fractional derivative framework to justify the proposed methods and to establish the associated error estimates. The resulting nonlinear algebraic system is then solved using the collocation method. Numerical simulations, including an application to a resource extraction model, confirm that the proposed methods attain a high level of accuracy and exhibit convergence rates in strong agreement with the theoretical predictions. These results demonstrate that the developed hybrid spectral–collocation frameworks constitute reliable and efficient tools for addressing stochastic optimal control problems based on HJB equations.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"14 ","pages":"34564-34584"},"PeriodicalIF":3.6,"publicationDate":"2026-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11415601","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147362256","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 : 2026-02-27DOI: 10.1109/ACCESS.2026.3668851
Sazeen Taha Abdulrazzaq;Mohammed M. Siddeq;Mohd Asyraf Zulkifley;Mohd Hairi Mohd Zaman;Asraf Mohamed Moubark
Medical imaging is an important contributor to diagnostic accuracy and monitoring of various health conditions, enabling healthcare professionals to gain valuable insights into the internal structures and functions of the body. With the prevalence of telemedicine and big data integration in healthcare, the effective storage and online transmission of these images have grown in importance. The main two challenges in this field are limited transmission bandwidth and storage capacity. Lossless compression techniques are necessary because information must be preserved to guarantee the integrity of medical images. Optimizing and decreasing the duration of compression and decompression processing is a fundamental aspect that warrants attention. In this work, an innovative technique for compressing medical images is developed and evaluated. It shows superior image reconstruction quality and a compression performance of up to 99%. The compression technique involves downscaling medical images through bicubic interpolation. Matrix reduction is subsequently employed to further reduce the size of the interpolated matrix to one-fourth of its dimensions. Finally, efficiency arithmetic coding is added to improve compression. The proposed algorithm emerges as a superior candidate in applications of image compression, outperforming all known advanced compression methods. The effectiveness of this method has been assessed by X-ray, ultrasound, CT, and MRI images sourced from an available database. Various performance metrics, including peak signal-to-noise ratio (PSNR), mean square error, and structural similarity index measure (SSIM), have been utilized to evaluate the quality of the images compressed using the proposed algorithm. The results show that at a high compression ratio of 100:1 (up to 99%), the proposed method achieves a high PSNR and SSIM values.
{"title":"Novel Medical Image Compression/Decompression Technique Based on Bicubic Interpolation With Matrix Reduction Algorithm","authors":"Sazeen Taha Abdulrazzaq;Mohammed M. Siddeq;Mohd Asyraf Zulkifley;Mohd Hairi Mohd Zaman;Asraf Mohamed Moubark","doi":"10.1109/ACCESS.2026.3668851","DOIUrl":"https://doi.org/10.1109/ACCESS.2026.3668851","url":null,"abstract":"Medical imaging is an important contributor to diagnostic accuracy and monitoring of various health conditions, enabling healthcare professionals to gain valuable insights into the internal structures and functions of the body. With the prevalence of telemedicine and big data integration in healthcare, the effective storage and online transmission of these images have grown in importance. The main two challenges in this field are limited transmission bandwidth and storage capacity. Lossless compression techniques are necessary because information must be preserved to guarantee the integrity of medical images. Optimizing and decreasing the duration of compression and decompression processing is a fundamental aspect that warrants attention. In this work, an innovative technique for compressing medical images is developed and evaluated. It shows superior image reconstruction quality and a compression performance of up to 99%. The compression technique involves downscaling medical images through bicubic interpolation. Matrix reduction is subsequently employed to further reduce the size of the interpolated matrix to one-fourth of its dimensions. Finally, efficiency arithmetic coding is added to improve compression. The proposed algorithm emerges as a superior candidate in applications of image compression, outperforming all known advanced compression methods. The effectiveness of this method has been assessed by X-ray, ultrasound, CT, and MRI images sourced from an available database. Various performance metrics, including peak signal-to-noise ratio (PSNR), mean square error, and structural similarity index measure (SSIM), have been utilized to evaluate the quality of the images compressed using the proposed algorithm. The results show that at a high compression ratio of 100:1 (up to 99%), the proposed method achieves a high PSNR and SSIM values.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"14 ","pages":"34600-34613"},"PeriodicalIF":3.6,"publicationDate":"2026-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11415572","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147362273","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 : 2026-02-27DOI: 10.1109/ACCESS.2026.3668903
Jonas Schneider;Carl Pfannschmidt;Peter Nyhuis;Matthias Schmidt
The expansion of product portfolios, the reduction of product life cycles and the volatility of markets pose significant challenges for production systems and their control. Concurrently, these trends present a challenge in achieving an optimal balance between logistical performance and internal company costs. The discordance between the escalating demands of customers on logistics performance, exemplified by metrics such as throughput time and schedule reliability, and the cost-driven corporate objective of minimizing work-in-process and maximizing machine utilization, is becoming increasingly challenging to reconcile. The application of reinforcement learning (RL) is a significant machine learning (ML) approach for overcoming these challenges. In comparison with other ML approaches, RL facilitates direct interaction with production systems and is consequently well suited for controlling them in operational use. Despite the extensive body of research on RL approaches for production control tasks, there is a paucity of literature addressing the influence of these approaches on key logistical targets for logistics performance and costs. The article’s added value derives from the systematic literature review it conducts, which provides researchers and practitioners with an overview of how existing RL approaches influence central logistical target variables. Furthermore, it highlights blind spots in the research landscape. The results indicate the existence of a substantial number of approaches; however, their distribution across control tasks is disproportionate. Furthermore, it is evident that there are distinct discrepancies in the classification system with respect to the impact on logistical target variables.
{"title":"The Role of Reinforcement Learning in Production Control: A Systematic Literature Review","authors":"Jonas Schneider;Carl Pfannschmidt;Peter Nyhuis;Matthias Schmidt","doi":"10.1109/ACCESS.2026.3668903","DOIUrl":"https://doi.org/10.1109/ACCESS.2026.3668903","url":null,"abstract":"The expansion of product portfolios, the reduction of product life cycles and the volatility of markets pose significant challenges for production systems and their control. Concurrently, these trends present a challenge in achieving an optimal balance between logistical performance and internal company costs. The discordance between the escalating demands of customers on logistics performance, exemplified by metrics such as throughput time and schedule reliability, and the cost-driven corporate objective of minimizing work-in-process and maximizing machine utilization, is becoming increasingly challenging to reconcile. The application of reinforcement learning (RL) is a significant machine learning (ML) approach for overcoming these challenges. In comparison with other ML approaches, RL facilitates direct interaction with production systems and is consequently well suited for controlling them in operational use. Despite the extensive body of research on RL approaches for production control tasks, there is a paucity of literature addressing the influence of these approaches on key logistical targets for logistics performance and costs. The article’s added value derives from the systematic literature review it conducts, which provides researchers and practitioners with an overview of how existing RL approaches influence central logistical target variables. Furthermore, it highlights blind spots in the research landscape. The results indicate the existence of a substantial number of approaches; however, their distribution across control tasks is disproportionate. Furthermore, it is evident that there are distinct discrepancies in the classification system with respect to the impact on logistical target variables.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"14 ","pages":"34375-34389"},"PeriodicalIF":3.6,"publicationDate":"2026-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11415616","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147362272","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 : 2026-02-27DOI: 10.1109/ACCESS.2026.3669078
Muhammad Awais Arshad;Haneul Lee;Hosun Lee;Myeongjin Kang;Yeowon Kim;Hyochoong Bang
Unpaired image-to-image translation is fundamental to autonomous driving, robotics, aerospace, remote sensing, and medical imaging, where visual realism must be achieved without compromising scene geometry. Generative Adversarial Network (GAN) based methods remain the most computationally feasible option for real-time deployment; However, they frequently introduce structural distortions, while diffusion and transformer-based models offer stronger controllability at a prohibitive computational cost. We propose CUT-GDC, a compact, structure-aware GAN framework that combines patchwise contrastive learning with gradient-domain constraints to enhance global geometric fidelity. CUT-GDC preserves the efficiency of the GAN-based architecture while enforcing the alignment of edge and gradient information to prevent global layout drift. Extensive experiments on multiple public benchmarks show that CUT-GDC consistently outperforms established GAN-based baselines. Compared with CUT, CUT-GDC reduces the average FID from $210.278~rightarrow ~121.582$ and KID from $0.199~rightarrow ~0.062$ , and improves SSIM from $0.361~rightarrow ~0.501$ . CUT-GDC also yields higher downstream segmentation performance, improving mIoU ($24.7~rightarrow ~28.63$ ), pixel accuracy (68.8% $rightarrow ~70.5$ %), and class accuracy (30.7% $rightarrow ~41.4$ %) relative to CUT on the Cityscapes dataset. Edge-structure evaluation further verifies geometric fidelity, where CUT-GDC consistently improves Canny-IoU and Grad-IoU across validation tasks (e.g., Sim-to-Real IR: 0.659/0.737 vs. 0.256/0.321), confirming superior contour alignment and gradient consistency. Ablation studies on the flower dataset confirm that gradient-domain constraints are a reliable driver of structural gains, reducing FID to 85.035 (vs. 90.647 for CUT and 89.809 for CycleGAN) and raising SSIM to 0.761 (vs. 0.609 and 0.748, respectively). CUT-GDC maintains comparable memory usage and faster training/inference throughput than competing models, enabling practical deployment in real-time and hardware-restricted environments. Qualitative analysis, including t-SNE visualizations, further demonstrates improved domain alignment.
{"title":"Structure-Consistent Contrastive Learning for Unpaired Image Translation With Gradient-Domain Constraints","authors":"Muhammad Awais Arshad;Haneul Lee;Hosun Lee;Myeongjin Kang;Yeowon Kim;Hyochoong Bang","doi":"10.1109/ACCESS.2026.3669078","DOIUrl":"https://doi.org/10.1109/ACCESS.2026.3669078","url":null,"abstract":"Unpaired image-to-image translation is fundamental to autonomous driving, robotics, aerospace, remote sensing, and medical imaging, where visual realism must be achieved without compromising scene geometry. Generative Adversarial Network (GAN) based methods remain the most computationally feasible option for real-time deployment; However, they frequently introduce structural distortions, while diffusion and transformer-based models offer stronger controllability at a prohibitive computational cost. We propose CUT-GDC, a compact, structure-aware GAN framework that combines patchwise contrastive learning with gradient-domain constraints to enhance global geometric fidelity. CUT-GDC preserves the efficiency of the GAN-based architecture while enforcing the alignment of edge and gradient information to prevent global layout drift. Extensive experiments on multiple public benchmarks show that CUT-GDC consistently outperforms established GAN-based baselines. Compared with CUT, CUT-GDC reduces the average FID from <inline-formula> <tex-math>$210.278~rightarrow ~121.582$ </tex-math></inline-formula> and KID from <inline-formula> <tex-math>$0.199~rightarrow ~0.062$ </tex-math></inline-formula>, and improves SSIM from <inline-formula> <tex-math>$0.361~rightarrow ~0.501$ </tex-math></inline-formula>. CUT-GDC also yields higher downstream segmentation performance, improving mIoU (<inline-formula> <tex-math>$24.7~rightarrow ~28.63$ </tex-math></inline-formula>), pixel accuracy (68.8% <inline-formula> <tex-math>$rightarrow ~70.5$ </tex-math></inline-formula>%), and class accuracy (30.7% <inline-formula> <tex-math>$rightarrow ~41.4$ </tex-math></inline-formula>%) relative to CUT on the Cityscapes dataset. Edge-structure evaluation further verifies geometric fidelity, where CUT-GDC consistently improves Canny-IoU and Grad-IoU across validation tasks (e.g., Sim-to-Real IR: 0.659/0.737 vs. 0.256/0.321), confirming superior contour alignment and gradient consistency. Ablation studies on the flower dataset confirm that gradient-domain constraints are a reliable driver of structural gains, reducing FID to 85.035 (vs. 90.647 for CUT and 89.809 for CycleGAN) and raising SSIM to 0.761 (vs. 0.609 and 0.748, respectively). CUT-GDC maintains comparable memory usage and faster training/inference throughput than competing models, enabling practical deployment in real-time and hardware-restricted environments. Qualitative analysis, including t-SNE visualizations, further demonstrates improved domain alignment.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"14 ","pages":"34510-34526"},"PeriodicalIF":3.6,"publicationDate":"2026-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11415593","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147362319","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 : 2026-02-27DOI: 10.1109/ACCESS.2026.3668964
Muhammad Ali;Diana Göhringer
The demand for artificial intelligence applications is rising in every field. And with this demand, machine learning algorithms and techniques are becoming more complicated and compute-intensive. For embedded devices, these compute-intensive and high memory-constrained algorithms are becoming a challenge. Dedicated hardware accelerators are generally the leading solution for embedded systems. This is because of their high performance, low area, and power footprint. Although hardware accelerators are a great solution for machine learning, they lack flexibility and programmability. Another alternative is General-Purpose Processors (GPPs), which provide better flexibility and programmability as compared with hardware accelerators, however, they lack performance, area, and power. Between these two plausible solutions, there is a big gap in terms of different metrics, which can be bridged by Application-Specific Instruction-set Processors (ASIPs). ASIPs are specialized processor systems with an architecture that is tailor-made for a certain application. ASIPs provide better performance as compared with general-purpose processors and have more flexibility and programmability as compared with hardware accelerators. The main goal of an ASIP design is to maximize the performance-to-power ratio for a specific application. This work provides an in-depth survey of different ASIP designs with a focus on Deep Neural Networks (DNNs). For a generic comparison, the proposed related works of ASIP design are classified based on their microarchitecture approach and optimization adopted. The strong and weak points are pointed out in different optimization and microarchitectures, and future trends for ASIP design for fast machine learning are identified.
{"title":"Application-Specific Instruction-Set Processors (ASIPs) for Deep Neural Networks: A Survey","authors":"Muhammad Ali;Diana Göhringer","doi":"10.1109/ACCESS.2026.3668964","DOIUrl":"https://doi.org/10.1109/ACCESS.2026.3668964","url":null,"abstract":"The demand for artificial intelligence applications is rising in every field. And with this demand, machine learning algorithms and techniques are becoming more complicated and compute-intensive. For embedded devices, these compute-intensive and high memory-constrained algorithms are becoming a challenge. Dedicated hardware accelerators are generally the leading solution for embedded systems. This is because of their high performance, low area, and power footprint. Although hardware accelerators are a great solution for machine learning, they lack flexibility and programmability. Another alternative is General-Purpose Processors (GPPs), which provide better flexibility and programmability as compared with hardware accelerators, however, they lack performance, area, and power. Between these two plausible solutions, there is a big gap in terms of different metrics, which can be bridged by Application-Specific Instruction-set Processors (ASIPs). ASIPs are specialized processor systems with an architecture that is tailor-made for a certain application. ASIPs provide better performance as compared with general-purpose processors and have more flexibility and programmability as compared with hardware accelerators. The main goal of an ASIP design is to maximize the performance-to-power ratio for a specific application. This work provides an in-depth survey of different ASIP designs with a focus on Deep Neural Networks (DNNs). For a generic comparison, the proposed related works of ASIP design are classified based on their microarchitecture approach and optimization adopted. The strong and weak points are pointed out in different optimization and microarchitectures, and future trends for ASIP design for fast machine learning are identified.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"14 ","pages":"34545-34563"},"PeriodicalIF":3.6,"publicationDate":"2026-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11415575","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147362270","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：