Pub Date : 2025-01-28DOI: 10.1109/ACCESS.2025.3535844
Abdelmgeid A. Ali;Mohamed T. Hammad;Hassan S. Hassan
Brain tumors are among the deadliest diseases, leading researchers to focus on improving the accuracy of tumor classification—a critical task for prompt diagnosis and effective treatment. Recent advancements in brain tumor diagnosis have significantly increased the use of deep learning techniques, particularly pre-trained models, for classification tasks. These models serve as feature extractors or can be fine-tuned for specific tasks, reducing both training time and data requirements. However, achieving high accuracy in multi-class brain tumor classification remains a major challenge, driving continued research in this area. Key obstacles include the need for expert interpretation of deep learning model outputs and the difficulty of developing highly accurate categorization systems. Optimizing the hyperparameters of Convolutional Neural Network (CNN) architectures, especially those based on pre-trained models, plays a crucial role in improving training efficiency. Manual hyperparameter adjustment is time-consuming and often results in suboptimal outcomes. To address these challenges, we propose an advanced approach that combines transfer learning with enhanced coevolutionary algorithms. Specifically, we utilize EfficientNetB3 and DenseNet121 pre-trained models in conjunction with the Co-Evolutionary Genetic Algorithm (CEGA) to classify brain tumors into four categories: gliomas, meningiomas, pituitary adenomas, and no tumors. CEGA optimizes the hyperparameters, improving both convergence speed and accuracy. Experiments conducted on a Kaggle dataset demonstrate that CEGA-EfficientNetB3 achieved the highest accuracy of 99.39%, while CEGA-DenseNet121 attained 99.01%, both without data augmentation. These results outperform cutting-edge methods, offering a rapid and reliable method for brain tumor classification. This approach has great potential to support radiologists and physicians in making timely and accurate diagnoses.
{"title":"A Co-Evolutionary Genetic Algorithm Approach to Optimizing Deep Learning for Brain Tumor Classification","authors":"Abdelmgeid A. Ali;Mohamed T. Hammad;Hassan S. Hassan","doi":"10.1109/ACCESS.2025.3535844","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3535844","url":null,"abstract":"Brain tumors are among the deadliest diseases, leading researchers to focus on improving the accuracy of tumor classification—a critical task for prompt diagnosis and effective treatment. Recent advancements in brain tumor diagnosis have significantly increased the use of deep learning techniques, particularly pre-trained models, for classification tasks. These models serve as feature extractors or can be fine-tuned for specific tasks, reducing both training time and data requirements. However, achieving high accuracy in multi-class brain tumor classification remains a major challenge, driving continued research in this area. Key obstacles include the need for expert interpretation of deep learning model outputs and the difficulty of developing highly accurate categorization systems. Optimizing the hyperparameters of Convolutional Neural Network (CNN) architectures, especially those based on pre-trained models, plays a crucial role in improving training efficiency. Manual hyperparameter adjustment is time-consuming and often results in suboptimal outcomes. To address these challenges, we propose an advanced approach that combines transfer learning with enhanced coevolutionary algorithms. Specifically, we utilize EfficientNetB3 and DenseNet121 pre-trained models in conjunction with the Co-Evolutionary Genetic Algorithm (CEGA) to classify brain tumors into four categories: gliomas, meningiomas, pituitary adenomas, and no tumors. CEGA optimizes the hyperparameters, improving both convergence speed and accuracy. Experiments conducted on a Kaggle dataset demonstrate that CEGA-EfficientNetB3 achieved the highest accuracy of 99.39%, while CEGA-DenseNet121 attained 99.01%, both without data augmentation. These results outperform cutting-edge methods, offering a rapid and reliable method for brain tumor classification. This approach has great potential to support radiologists and physicians in making timely and accurate diagnoses.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"21229-21248"},"PeriodicalIF":3.4,"publicationDate":"2025-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10856009","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143105642","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 : 2025-01-28DOI: 10.1109/ACCESS.2025.3535796
Ahmad Qurthobi;Robertas Damaševičius;Vytautas Barzdaitis;Rytis Maskeliūnas
As a complex ecosystem composed of flora and fauna, the forest has always been vulnerable to threats. Previous researchers utilized environmental audio collections, such as the ESC-50 and UrbanSound8k datasets, as proximate representatives of sounds potentially present in forests. This study focuses on the application of deep learning models for forest sound classification as an effort to establish an early threats detection system. The research evaluates the performance of several pre-trained deep learning models, including MobileNet, GoogleNet, and ResNet, on the limited FSC22 dataset, which consists of 2,025 forest sound recordings classified into 27 categories. To improve classification capabilities, the study introduces a hybrid model that combines neural network (CNN) with a Bidirectional Long-Short-Term Memory (BiLSTM) layer, designed to capture both spatial and temporal features of the sound data. The research also employs Pareto-Mordukhovich-optimized Mel Frequency Cepstral Coefficients (MFCC) for feature extraction, improving the representation of audio signals. Data augmentation and dimensionality reduction techniques were also explored to assess their impact on model performance. The results indicate that the proposed hybrid CNN-BiLSTM model significantly improved classification loss and accuracy scores compared to the standalone pre-trained models. GoogleNet, with an added BiLSTM layer and augmented data, achieved an average reduced loss score of 0.7209 and average accuracy of 0.7852, demonstrating its potential to classify forest sounds. Improvements in loss score and classification performance highlight the potential of hybrid models in environmental sound analysis, particularly in scenarios with limited data availability.
{"title":"Robust Forest Sound Classification Using Pareto-Mordukhovich Optimized MFCC in Environmental Monitoring","authors":"Ahmad Qurthobi;Robertas Damaševičius;Vytautas Barzdaitis;Rytis Maskeliūnas","doi":"10.1109/ACCESS.2025.3535796","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3535796","url":null,"abstract":"As a complex ecosystem composed of flora and fauna, the forest has always been vulnerable to threats. Previous researchers utilized environmental audio collections, such as the ESC-50 and UrbanSound8k datasets, as proximate representatives of sounds potentially present in forests. This study focuses on the application of deep learning models for forest sound classification as an effort to establish an early threats detection system. The research evaluates the performance of several pre-trained deep learning models, including MobileNet, GoogleNet, and ResNet, on the limited FSC22 dataset, which consists of 2,025 forest sound recordings classified into 27 categories. To improve classification capabilities, the study introduces a hybrid model that combines neural network (CNN) with a Bidirectional Long-Short-Term Memory (BiLSTM) layer, designed to capture both spatial and temporal features of the sound data. The research also employs Pareto-Mordukhovich-optimized Mel Frequency Cepstral Coefficients (MFCC) for feature extraction, improving the representation of audio signals. Data augmentation and dimensionality reduction techniques were also explored to assess their impact on model performance. The results indicate that the proposed hybrid CNN-BiLSTM model significantly improved classification loss and accuracy scores compared to the standalone pre-trained models. GoogleNet, with an added BiLSTM layer and augmented data, achieved an average reduced loss score of 0.7209 and average accuracy of 0.7852, demonstrating its potential to classify forest sounds. Improvements in loss score and classification performance highlight the potential of hybrid models in environmental sound analysis, particularly in scenarios with limited data availability.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"20923-20944"},"PeriodicalIF":3.4,"publicationDate":"2025-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10856116","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143105849","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 : 2025-01-28DOI: 10.1109/ACCESS.2025.3535804
Zachary M. Choffin;Lingyan Kong;Yu Gan;Nathan Jeong
The integration of a non-invasive microwave imaging system with a machine learning algorithm could improve food quality and food safety. In this paper, a S- and C-band microwave imaging system that utilizes DAS (Delay and Sum) beamforming with an automated high-frequency switching network is built to scan watermelons and determine their ripeness. A total of 288 images were collected from eight different watermelons varying the height and angle of capture. A convolutional neural network (CNN) was employed to assess the ripeness level, which was determined by analyzing the Brix sugar content. The results show 86% accuracy for ripeness classification in three fold cross validation. This novel approach demonstrates the potential of combining microwave imaging with machine learning for non-destructive food quality assessment, offering a scalable and reliable tool for real-time evaluation of fruit ripeness and quality.
{"title":"A CNN-Based Microwave Imaging System for Detecting Watermelon Ripeness","authors":"Zachary M. Choffin;Lingyan Kong;Yu Gan;Nathan Jeong","doi":"10.1109/ACCESS.2025.3535804","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3535804","url":null,"abstract":"The integration of a non-invasive microwave imaging system with a machine learning algorithm could improve food quality and food safety. In this paper, a S- and C-band microwave imaging system that utilizes DAS (Delay and Sum) beamforming with an automated high-frequency switching network is built to scan watermelons and determine their ripeness. A total of 288 images were collected from eight different watermelons varying the height and angle of capture. A convolutional neural network (CNN) was employed to assess the ripeness level, which was determined by analyzing the Brix sugar content. The results show 86% accuracy for ripeness classification in three fold cross validation. This novel approach demonstrates the potential of combining microwave imaging with machine learning for non-destructive food quality assessment, offering a scalable and reliable tool for real-time evaluation of fruit ripeness and quality.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"21413-21421"},"PeriodicalIF":3.4,"publicationDate":"2025-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10856151","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143105632","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 : 2025-01-28DOI: 10.1109/ACCESS.2025.3535535
Md. Atiqur Rahman;Israt Jahan;Maheen Islam;Taskeed Jabid;Md Sawkat Ali;Mohammad Rifat Ahmmad Rashid;Mohammad Manzurul Islam;Md. Hasanul Ferdaus;Md Mostofa Kamal Rasel;Mahmuda Rawnak Jahan;Shayla Sharmin;Tanzina Afroz Rimi;Atia Sanjida Talukder;Md. Mafiul Hasan Matin;M. Ameer Ali
Classifying sleep disorders, such as obstructive sleep apnea and insomnia, is crucial for improving human quality of life due to their significant impact on health. The traditional expert-based classification of sleep stages, particularly through visual inspection, is challenging and prone to errors. This fact highlights the need for accurate machine learning algorithms (MLAs) for analyzing, monitoring, and diagnosing sleep disorders. This paper compares the MLAs for sleep disorder classification, specifically targeting None, Sleep Apnea, and Insomnia, using the Sleep Health and Lifestyle Dataset. We conducted two experiments. In the first one, we selected five key features from the feature spaces using the Gradient Boosting Regressor based on the Mean Decrease Impurity (MDI) technique. We chose two key features using the same methodology in the second experiment. We utilized 15 machine learning classifiers, and Gradient Boosting, Voting, Catboost, and Stacking Classifiers achieved an identical classification accuracy of 97.33%, with Precision, Recall, F1-score of 0.9733, and Specificity of 0.9569 in the original feature space. Among these, Gradient Boosting had the highest AUC of 0.9953 and was 3.36, 5.86, and 20.16 times faster than Voting, Catboost, and Stacking Classifiers, respectively. In the second experiment, the Decision Tree achieved the highest accuracy of 96% in the original and engineered feature spaces and was 149.33 times faster in the engineered feature space. Thus, this research proposes Gradient Boosting as the most effective method, outperforming all state-of-the-art techniques by achieving the highest accuracy, precision, recall, F1-score, and AUC, highlighting its superior classification performance and computational efficiency.
{"title":"Improving Sleep Disorder Diagnosis Through Optimized Machine Learning Approaches","authors":"Md. Atiqur Rahman;Israt Jahan;Maheen Islam;Taskeed Jabid;Md Sawkat Ali;Mohammad Rifat Ahmmad Rashid;Mohammad Manzurul Islam;Md. Hasanul Ferdaus;Md Mostofa Kamal Rasel;Mahmuda Rawnak Jahan;Shayla Sharmin;Tanzina Afroz Rimi;Atia Sanjida Talukder;Md. Mafiul Hasan Matin;M. Ameer Ali","doi":"10.1109/ACCESS.2025.3535535","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3535535","url":null,"abstract":"Classifying sleep disorders, such as obstructive sleep apnea and insomnia, is crucial for improving human quality of life due to their significant impact on health. The traditional expert-based classification of sleep stages, particularly through visual inspection, is challenging and prone to errors. This fact highlights the need for accurate machine learning algorithms (MLAs) for analyzing, monitoring, and diagnosing sleep disorders. This paper compares the MLAs for sleep disorder classification, specifically targeting None, Sleep Apnea, and Insomnia, using the Sleep Health and Lifestyle Dataset. We conducted two experiments. In the first one, we selected five key features from the feature spaces using the Gradient Boosting Regressor based on the Mean Decrease Impurity (MDI) technique. We chose two key features using the same methodology in the second experiment. We utilized 15 machine learning classifiers, and Gradient Boosting, Voting, Catboost, and Stacking Classifiers achieved an identical classification accuracy of 97.33%, with Precision, Recall, F1-score of 0.9733, and Specificity of 0.9569 in the original feature space. Among these, Gradient Boosting had the highest AUC of 0.9953 and was 3.36, 5.86, and 20.16 times faster than Voting, Catboost, and Stacking Classifiers, respectively. In the second experiment, the Decision Tree achieved the highest accuracy of 96% in the original and engineered feature spaces and was 149.33 times faster in the engineered feature space. Thus, this research proposes Gradient Boosting as the most effective method, outperforming all state-of-the-art techniques by achieving the highest accuracy, precision, recall, F1-score, and AUC, highlighting its superior classification performance and computational efficiency.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"20989-21004"},"PeriodicalIF":3.4,"publicationDate":"2025-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10856004","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143105912","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 : 2025-01-27DOI: 10.1109/ACCESS.2025.3535093
Florian Quatresooz;Claude Oestges
Atmospheric turbulence influence on optical wave propagation, referred to as optical turbulence, has long been studied for astronomical applications and is now being addressed for free-space optical communication links between ground and satellites. While challenges overlap, models developed for astronomical applications are not fully transferable to optical communications. This paper provides a literature review of optical turbulence models, i.e., models giving vertical profiles of the refractive index structure parameter $C_{n}^{2}$ , highlighting differences between astronomical and optical communication sites. It presents different classifications of available $C_{n}^{2}$ models, based on the atmospheric layer they target and their necessary input parameters. Boundary layer $C_{n}^{2}$ models are also addressed, and recent machine learning approaches for $C_{n}^{2}$ modeling are discussed. Additionally, commonly used metrics for comparing $C_{n}^{2}$ profiles are introduced. Therefore, this work provides important insights into optical turbulence model selection, enabling accurate site characterization and informed optical terminal design.
{"title":"Cₙ² Modeling for Free-Space Optical Communications: A Review","authors":"Florian Quatresooz;Claude Oestges","doi":"10.1109/ACCESS.2025.3535093","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3535093","url":null,"abstract":"Atmospheric turbulence influence on optical wave propagation, referred to as optical turbulence, has long been studied for astronomical applications and is now being addressed for free-space optical communication links between ground and satellites. While challenges overlap, models developed for astronomical applications are not fully transferable to optical communications. This paper provides a literature review of optical turbulence models, i.e., models giving vertical profiles of the refractive index structure parameter <inline-formula> <tex-math>$C_{n}^{2}$ </tex-math></inline-formula>, highlighting differences between astronomical and optical communication sites. It presents different classifications of available <inline-formula> <tex-math>$C_{n}^{2}$ </tex-math></inline-formula> models, based on the atmospheric layer they target and their necessary input parameters. Boundary layer <inline-formula> <tex-math>$C_{n}^{2}$ </tex-math></inline-formula> models are also addressed, and recent machine learning approaches for <inline-formula> <tex-math>$C_{n}^{2}$ </tex-math></inline-formula> modeling are discussed. Additionally, commonly used metrics for comparing <inline-formula> <tex-math>$C_{n}^{2}$ </tex-math></inline-formula> profiles are introduced. Therefore, this work provides important insights into optical turbulence model selection, enabling accurate site characterization and informed optical terminal design.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"21279-21305"},"PeriodicalIF":3.4,"publicationDate":"2025-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10855431","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143105543","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}
Audio-language models (ALMs) generate linguistic descriptions of sound-producing events and scenes. Advances in dataset creation and computational power have led to significant progress in this domain. This paper surveys 69 datasets used to train ALMs, covering research up to September 2024 (https://github.com/GLJS/audio-datasets). The survey provides a comprehensive analysis of dataset origins, audio and linguistic characteristics, and use cases. Key sources include YouTube-based datasets such as AudioSet, with over two million samples, and community platforms such as Freesound, with over one million samples. The survey evaluates acoustic and linguistic variability across datasets through principal component analysis of audio and text embeddings. The survey also analyzes data leakage through CLAP embeddings, and examines sound category distributions to identify imbalances. Finally, the survey identifies key challenges in developing large, diverse datasets to enhance ALM performance, including dataset overlap, biases, accessibility barriers, and the predominance of English-language content, while highlighting specific areas requiring attention: multilingual dataset development, specialized domain coverage and improved dataset accessibility.
{"title":"Audio-Language Datasets of Scenes and Events: A Survey","authors":"Gijs Wijngaard;Elia Formisano;Michele Esposito;Michel Dumontier","doi":"10.1109/ACCESS.2025.3534621","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3534621","url":null,"abstract":"Audio-language models (ALMs) generate linguistic descriptions of sound-producing events and scenes. Advances in dataset creation and computational power have led to significant progress in this domain. This paper surveys 69 datasets used to train ALMs, covering research up to September 2024 (<uri>https://github.com/GLJS/audio-datasets</uri>). The survey provides a comprehensive analysis of dataset origins, audio and linguistic characteristics, and use cases. Key sources include YouTube-based datasets such as AudioSet, with over two million samples, and community platforms such as Freesound, with over one million samples. The survey evaluates acoustic and linguistic variability across datasets through principal component analysis of audio and text embeddings. The survey also analyzes data leakage through CLAP embeddings, and examines sound category distributions to identify imbalances. Finally, the survey identifies key challenges in developing large, diverse datasets to enhance ALM performance, including dataset overlap, biases, accessibility barriers, and the predominance of English-language content, while highlighting specific areas requiring attention: multilingual dataset development, specialized domain coverage and improved dataset accessibility.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"20328-20360"},"PeriodicalIF":3.4,"publicationDate":"2025-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10854210","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143106001","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 : 2025-01-27DOI: 10.1109/ACCESS.2025.3535233
Qusay Alghazali;Husam Al-Amaireh;Tibor Cinkler
This work addresses the critical challenge of energy consumption in Mobile Edge Computing (MEC), a burgeoning field that extends cloud computing capabilities to the edge of cellular networks. Given the exponential growth of mobile devices and the resultant surge in energy demands, there is an urgent need for efficient energy management strategies to ensure sustainable development and operation of MEC infrastructures. This paper introduces a comprehensive framework for reducing energy consumption in MEC environments by leveraging advanced optimization techniques and energy-efficient resource allocation algorithms. We propose a novel approach that dynamically adjusts the computational resources based on the current network load and the type of services requested, thus minimizing unnecessary energy consumption. We derive and propose an optimized energy consumption for local processing. Then, we study the two network scenarios: Non-Orthogonal Multiple Access (NOMA) and Massive Multiple-Input Multiple-Output (mMIMO). We propose an optimized energy consumption algorithm in NOMA based on the derived processing resource requirements. Then, in mMIMO, we derive optimized power allocation algorithms. Simulations validate the effectiveness of our proposed framework, demonstrating significant energy savings.
{"title":"Energy-Efficient Resource Allocation in Mobile Edge Computing Using NOMA and Massive MIMO","authors":"Qusay Alghazali;Husam Al-Amaireh;Tibor Cinkler","doi":"10.1109/ACCESS.2025.3535233","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3535233","url":null,"abstract":"This work addresses the critical challenge of energy consumption in Mobile Edge Computing (MEC), a burgeoning field that extends cloud computing capabilities to the edge of cellular networks. Given the exponential growth of mobile devices and the resultant surge in energy demands, there is an urgent need for efficient energy management strategies to ensure sustainable development and operation of MEC infrastructures. This paper introduces a comprehensive framework for reducing energy consumption in MEC environments by leveraging advanced optimization techniques and energy-efficient resource allocation algorithms. We propose a novel approach that dynamically adjusts the computational resources based on the current network load and the type of services requested, thus minimizing unnecessary energy consumption. We derive and propose an optimized energy consumption for local processing. Then, we study the two network scenarios: Non-Orthogonal Multiple Access (NOMA) and Massive Multiple-Input Multiple-Output (mMIMO). We propose an optimized energy consumption algorithm in NOMA based on the derived processing resource requirements. Then, in mMIMO, we derive optimized power allocation algorithms. Simulations validate the effectiveness of our proposed framework, demonstrating significant energy savings.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"21456-21470"},"PeriodicalIF":3.4,"publicationDate":"2025-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10855410","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143105711","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 : 2025-01-27DOI: 10.1109/ACCESS.2025.3535158
Edgar F. Ladeira;Bruno M. C. Silva
The emissions of pollutants and radioactive gases are the main causes of several environmental disasters that may cause premature deaths. The significant impact of these gases on public health is a major concern, especially in remote and rural regions. In these areas, access to security and health services is scarce, and real-time monitoring of citizens and the conditions in which they live is very difficult. Without means to monitor or predict, healthcare and government stakeholders typically act too late when indoor incidents occur. Hence, this paper presents a digital decision support system that uses Machine Learning (ML) for monitoring and prediction of incidents related with indoor hazardous gases. This system is implemented on top of an Internet of Things (IoT) ecosystem named RuraLTHINGS. This project, developed by the University of Beira Interior, Portugal, monitors the quality of air in remote and rural regions in real-time. The platform aims to predict and notify residents and other stakeholders about environmental conditions and prevent the risk of exposure to dangerous gases. The system uses ML techniques to analyze the collected data and provide future predictions using unidirectional Long Short-Term Memory (LSTM) layers overlaid on bidirectional LSTM layers, meaning layers stacked together, which was the model architecture that delivered the best results in this context. This paper presents the validation of the digital platform and the ML model using a real test bed environment. The model successfully predicted future data trends related to indoor monitoring of hazardous gases.
{"title":"A Machine Learning-Based Platform for Monitoring and Prediction of Hazardous Gases in Rural and Remote Areas","authors":"Edgar F. Ladeira;Bruno M. C. Silva","doi":"10.1109/ACCESS.2025.3535158","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3535158","url":null,"abstract":"The emissions of pollutants and radioactive gases are the main causes of several environmental disasters that may cause premature deaths. The significant impact of these gases on public health is a major concern, especially in remote and rural regions. In these areas, access to security and health services is scarce, and real-time monitoring of citizens and the conditions in which they live is very difficult. Without means to monitor or predict, healthcare and government stakeholders typically act too late when indoor incidents occur. Hence, this paper presents a digital decision support system that uses Machine Learning (ML) for monitoring and prediction of incidents related with indoor hazardous gases. This system is implemented on top of an Internet of Things (IoT) ecosystem named RuraLTHINGS. This project, developed by the University of Beira Interior, Portugal, monitors the quality of air in remote and rural regions in real-time. The platform aims to predict and notify residents and other stakeholders about environmental conditions and prevent the risk of exposure to dangerous gases. The system uses ML techniques to analyze the collected data and provide future predictions using unidirectional Long Short-Term Memory (LSTM) layers overlaid on bidirectional LSTM layers, meaning layers stacked together, which was the model architecture that delivered the best results in this context. This paper presents the validation of the digital platform and the ML model using a real test bed environment. The model successfully predicted future data trends related to indoor monitoring of hazardous gases.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"20297-20315"},"PeriodicalIF":3.4,"publicationDate":"2025-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10855430","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143105968","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}
People nowadays can easily synthesize high fidelity fake images with different types of image content due to the rapid advances of deep learning technologies. Detecting such images and attributing them to their generative models (GMs) is crucial. Existing deep learning methods attempt to identify and classify GM-specific artifacts but often struggle with content-independence and generalizability. In this paper, we observe that while GMs leave unique artifacts in the frequency domain, they are coupled with the image content. Based on this observation, we propose a novel deep learning-based solution that learns input-adaptive masks to highlight GMs’ artifacts and achieve high accuracy on the synthesized image attribution task. In addition, we observed that GMs’ artifacts in the frequency domain remain intact in sub-images of the original image, and they are even retained when the images are distorted. To further improve the accuracy of the proposed solution, we leverage the characteristics of GMs artifacts in sub-images and distorted images to make our network perform more effectively. Our evaluation results show that our proposed solution outperforms other state-of-the-art methods on unseen image types, showing great generalizability.
{"title":"An Efficient Frequency Domain Based Attribution and Detection Network","authors":"Junbin Zhang;Yixiao Wang;Hamid Reza Tohidypour;Panos Nasiopoulos","doi":"10.1109/ACCESS.2025.3534829","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3534829","url":null,"abstract":"People nowadays can easily synthesize high fidelity fake images with different types of image content due to the rapid advances of deep learning technologies. Detecting such images and attributing them to their generative models (GMs) is crucial. Existing deep learning methods attempt to identify and classify GM-specific artifacts but often struggle with content-independence and generalizability. In this paper, we observe that while GMs leave unique artifacts in the frequency domain, they are coupled with the image content. Based on this observation, we propose a novel deep learning-based solution that learns input-adaptive masks to highlight GMs’ artifacts and achieve high accuracy on the synthesized image attribution task. In addition, we observed that GMs’ artifacts in the frequency domain remain intact in sub-images of the original image, and they are even retained when the images are distorted. To further improve the accuracy of the proposed solution, we leverage the characteristics of GMs artifacts in sub-images and distorted images to make our network perform more effectively. Our evaluation results show that our proposed solution outperforms other state-of-the-art methods on unseen image types, showing great generalizability.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"19909-19921"},"PeriodicalIF":3.4,"publicationDate":"2025-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10855423","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143105732","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 : 2025-01-27DOI: 10.1109/ACCESS.2025.3534678
Kevin S. Anderson;Joshua S. Stein;Marios Theristis
The performance of photovoltaic (PV) modules is determined by the interplay between their inherent characteristics and the prevailing weather conditions. Although the impacts of different characteristics (e.g., low-light behavior, spectral mismatch, temperature coefficient, etc) are known, they have not been quantified over large geographic regions. This study uses the Climate Specific Energy Rating (CSER) and specific yield metrics as criteria to determine how different PV modules perform across climates in the contiguous United States (CONUS) and identifies the underlying drivers behind the observed variations. The annual CSER and specific yield of various PV technologies vary by more than 10% and 30%, respectively, across the CONUS. As expected, temperature has the most significant impact on CSER, affecting CSER by up to 13.1%, while spectral effects account for up to 4.9% variation in the case of cadmium telluride modules. Additionally, minor differences in parameter estimation procedures are shown to result in CSER differences of up to 1.5% in some climates. Furthermore, the IEC 61853-4 reference climatic datasets are found to overestimate CSER by 2–4% relative to climatic data for locations of actual PV systems in the United States. A new set of reference locations that accurately represents CSER across CONUS is proposed as an alternative to the IEC 61853-4 reference datasets.
{"title":"Variation in Photovoltaic Energy Rating and Underlying Drivers Across Modules and Climates","authors":"Kevin S. Anderson;Joshua S. Stein;Marios Theristis","doi":"10.1109/ACCESS.2025.3534678","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3534678","url":null,"abstract":"The performance of photovoltaic (PV) modules is determined by the interplay between their inherent characteristics and the prevailing weather conditions. Although the impacts of different characteristics (e.g., low-light behavior, spectral mismatch, temperature coefficient, etc) are known, they have not been quantified over large geographic regions. This study uses the Climate Specific Energy Rating (CSER) and specific yield metrics as criteria to determine how different PV modules perform across climates in the contiguous United States (CONUS) and identifies the underlying drivers behind the observed variations. The annual CSER and specific yield of various PV technologies vary by more than 10% and 30%, respectively, across the CONUS. As expected, temperature has the most significant impact on CSER, affecting CSER by up to 13.1%, while spectral effects account for up to 4.9% variation in the case of cadmium telluride modules. Additionally, minor differences in parameter estimation procedures are shown to result in CSER differences of up to 1.5% in some climates. Furthermore, the IEC 61853-4 reference climatic datasets are found to overestimate CSER by 2–4% relative to climatic data for locations of actual PV systems in the United States. A new set of reference locations that accurately represents CSER across CONUS is proposed as an alternative to the IEC 61853-4 reference datasets.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"21064-21073"},"PeriodicalIF":3.4,"publicationDate":"2025-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10854476","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143105739","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：