Computers & Electrical Engineering最新文献_第7页

Privacy-preserving medical image retrieval with traceability and verifiability 具有可追溯性和可验证性的保护隐私的医学图像检索

IF 4.9 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Computers & Electrical Engineering

Pub Date : 2025-12-10 DOI: 10.1016/j.compeleceng.2025.110896

Ruihua Xu , Qiuyu Zhang , Zhaoheng Chen

The development of cloud computing technology reduces the storage burden of medical image retrieval systems; however, it also introduces significant risks of privacy leakage for medical data stored in the cloud. Conventional linear chaotic encryption approaches are mainly restricted to pixel-level operations, lack sufficient sensitivity to small variations in the plaintext, and therefore do not satisfy the strict security requirements of medical images. Furthermore, existing mechanisms for dynamic data verification frequently depend on complex hash structures, which result in high computational costs and provide inadequate protection against the illegal distribution and misuse of sensitive medical images. To overcome these limitations, we present a privacy-preserving medical image retrieval framework with traceability and verifiability (PPTV). In PPTV embedding space first reserved by employing a nonlinear chaotic encryption approach, followed by bit-level three-channel sequential encryption applied to color medical images, and user information is embedded through ciphertext watermarking based on chaotic mapping. A strongly balanced dynamic verification tree with unique indexing is constructed to support efficient verification of dynamic image data. Traceability of malicious access is realized by extracting the embedded ciphertext watermark that carries user information. Experimental evaluation indicates that PPTV achieves retrieval accuracies of 92.8 % and 97.7 % on the IDRiD and COVID datasets, respectively, confirming its ability to provide both high retrieval precision and robust privacy preservation.

云计算技术的发展减轻了医学图像检索系统的存储负担；然而，它也为存储在云中的医疗数据带来了隐私泄露的重大风险。传统的线性混沌加密方法主要局限于像素级操作，对明文的微小变化缺乏足够的灵敏度，因此不能满足医学图像严格的安全要求。此外，现有的动态数据验证机制经常依赖于复杂的哈希结构，这导致了高计算成本，并且对敏感医学图像的非法分发和滥用提供了不足的保护。为了克服这些限制，我们提出了一个具有可追溯性和可验证性（PPTV）的隐私保护医学图像检索框架。在PPTV嵌入空间中，首先采用非线性混沌加密方法保留空间，然后将位级三通道顺序加密应用于彩色医学图像，并通过基于混沌映射的密文水印嵌入用户信息。构造了具有唯一索引的强平衡动态验证树，以支持对动态图像数据的有效验证。通过提取嵌入的携带用户信息的密文水印，实现对恶意访问的跟踪。实验评估表明，PPTV在IDRiD和COVID数据集上的检索准确率分别达到92.8%和97.7%，证明了其具有较高的检索精度和鲁棒性的隐私保护能力。

{"title":"Privacy-preserving medical image retrieval with traceability and verifiability","authors":"Ruihua Xu , Qiuyu Zhang , Zhaoheng Chen","doi":"10.1016/j.compeleceng.2025.110896","DOIUrl":"10.1016/j.compeleceng.2025.110896","url":null,"abstract":"<div><div>The development of cloud computing technology reduces the storage burden of medical image retrieval systems; however, it also introduces significant risks of privacy leakage for medical data stored in the cloud. Conventional linear chaotic encryption approaches are mainly restricted to pixel-level operations, lack sufficient sensitivity to small variations in the plaintext, and therefore do not satisfy the strict security requirements of medical images. Furthermore, existing mechanisms for dynamic data verification frequently depend on complex hash structures, which result in high computational costs and provide inadequate protection against the illegal distribution and misuse of sensitive medical images. To overcome these limitations, we present a privacy-preserving medical image retrieval framework with traceability and verifiability (PPTV). In PPTV embedding space first reserved by employing a nonlinear chaotic encryption approach, followed by bit-level three-channel sequential encryption applied to color medical images, and user information is embedded through ciphertext watermarking based on chaotic mapping. A strongly balanced dynamic verification tree with unique indexing is constructed to support efficient verification of dynamic image data. Traceability of malicious access is realized by extracting the embedded ciphertext watermark that carries user information. Experimental evaluation indicates that PPTV achieves retrieval accuracies of 92.8 % and 97.7 % on the IDRiD and COVID datasets, respectively, confirming its ability to provide both high retrieval precision and robust privacy preservation.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"130 ","pages":"Article 110896"},"PeriodicalIF":4.9,"publicationDate":"2025-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145736917","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Joint estimation of the state-of-charge and state-of-energy of lithium-ion batteries under diverse temperatures and discharging conditions using a data-driven model 使用数据驱动模型对不同温度和放电条件下锂离子电池的充电状态和能量状态进行联合估计

IF 4.9 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Computers & Electrical Engineering

Pub Date : 2025-12-10 DOI: 10.1016/j.compeleceng.2025.110907

Baoliang Chen, Jujin Pan, Yonggui Liu

The state-of-charge (SOC) and state-of-energy (SOE) of lithium-ion batteries (LIBs) play a pivotal role in BMS. However, accurately estimating these states is challenging due to the varying operating conditions of batteries including unpredictable discharging behaviors and fluctuating ambient temperatures, which significantly impact battery performance. To overcome this challenge, this study proposes a novel method that can achieve accurate co-estimation of SOC and SOE under complex conditions without relying on any prior knowledge or specialized parameterization designs. Specifically, a data-driven model called Graphformer is introduced to meet this end, which has the following advantages. (a) It utilizes the ScaleGraph block to comprehensively excavate the spatio-temporal dependencies within battery sensor data, applying the FFT to analyze multi-scale temporal information while enhancing spatial representation through an adaptive graph convolution module. (b) It effectively discovers long-term dependencies in sensor data using the multi-head ProbSparse self-attention module. Extensive experiments are conducted to verify the efficacy of the proposed method over diverse driving cycles at ambient temperatures ranging from 0°C to 40°C. The results demonstrate its strong robustness and superior predictive ability, with an average root mean square error of 0.1691% on test samples from the LG battery dataset.

锂离子电池（lib）的荷电状态（SOC）和能量状态（SOE）在BMS中起着至关重要的作用。然而，由于电池的工作条件变化，包括不可预测的放电行为和波动的环境温度，这些都会对电池的性能产生重大影响，因此准确估计这些状态是具有挑战性的。为了克服这一挑战，本研究提出了一种新的方法，可以在不依赖任何先验知识或专门参数化设计的情况下实现复杂条件下SOC和SOE的精确联合估计。具体地说，引入了一种称为Graphformer的数据驱动模型来满足这一目的，它具有以下优点。(a)利用ScaleGraph块全面挖掘电池传感器数据中的时空依赖关系，应用FFT分析多尺度时间信息，同时通过自适应图卷积模块增强空间表征。(b)使用多头ProbSparse自关注模块有效地发现传感器数据中的长期依赖关系。进行了大量的实验来验证所提出的方法在环境温度范围从0°C到40°C的不同驾驶循环中的有效性。结果表明，该方法具有较强的鲁棒性和较好的预测能力，对LG电池数据集的测试样本的平均均方根误差为0.1691%。

{"title":"Joint estimation of the state-of-charge and state-of-energy of lithium-ion batteries under diverse temperatures and discharging conditions using a data-driven model","authors":"Baoliang Chen, Jujin Pan, Yonggui Liu","doi":"10.1016/j.compeleceng.2025.110907","DOIUrl":"10.1016/j.compeleceng.2025.110907","url":null,"abstract":"<div><div>The state-of-charge (SOC) and state-of-energy (SOE) of lithium-ion batteries (LIBs) play a pivotal role in BMS. However, accurately estimating these states is challenging due to the varying operating conditions of batteries including unpredictable discharging behaviors and fluctuating ambient temperatures, which significantly impact battery performance. To overcome this challenge, this study proposes a novel method that can achieve accurate co-estimation of SOC and SOE under complex conditions without relying on any prior knowledge or specialized parameterization designs. Specifically, a data-driven model called Graphformer is introduced to meet this end, which has the following advantages. (a) It utilizes the ScaleGraph block to comprehensively excavate the spatio-temporal dependencies within battery sensor data, applying the FFT to analyze multi-scale temporal information while enhancing spatial representation through an adaptive graph convolution module. (b) It effectively discovers long-term dependencies in sensor data using the multi-head <em>ProbSparse</em> self-attention module. Extensive experiments are conducted to verify the efficacy of the proposed method over diverse driving cycles at ambient temperatures ranging from 0°C to 40°C. The results demonstrate its strong robustness and superior predictive ability, with an average root mean square error of 0.1691% on test samples from the LG battery dataset.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"130 ","pages":"Article 110907"},"PeriodicalIF":4.9,"publicationDate":"2025-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145737002","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

GA-CNN-BiGRU-IDS: A robust framework for Intrusion Detection System based on GA for data augmentation and hybrid CNN-BiGRU model for spatiotemporal feature extraction GA-CNN-BiGRU- ids：基于GA的数据增强和CNN-BiGRU混合模型的时空特征提取的鲁棒入侵检测系统框架

IF 4.9 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Computers & Electrical Engineering

Pub Date : 2025-12-09 DOI: 10.1016/j.compeleceng.2025.110900

Rahma Jablaoui, Noureddine Liouane

The vital importance of Intrusion Detection Systems (IDS) in securing modern IoT and cloud infrastructures is challenged by the evolving sophistication of network threats. Deep learning (DL)-based IDS, while promising, are often constrained by acute class imbalance and the difficulty of capturing discriminative spatiotemporal patterns within complex network traffic. To overcome these limitations, this paper proposes GA-CNN-BiGRU-IDS, a novel IDS framework integrating an advanced data augmentation technique with a hybrid DL architecture. The proposed genetic algorithm (GA)-inspired augmentation scheme incorporates a one-point crossover operation to maintain feature coherence and a cosine similarity-optimized mutation operator to preserve the original data distribution, effectively mitigating class imbalance through high-quality synthetic minority instances. The subsequent hybrid CNN-BiGRU model autonomously learns robust feature representations; the convolutional layers extract spatial hierarchies from the data, while the bidirectional gated recurrent units capture long-range temporal dependencies. The framework is rigorously evaluated on the well-established Netflow-based NF-UNSW-NB15 dataset under binary and multiclass scenarios. Experiments show that our methodology reaches leading results, particularly in detecting rare and sophisticated attack categories. A comprehensive comparative analysis confirms the framework’s superiority over both ablation and existing benchmark approaches, highlighting its potential to greatly improve the reliability of IDS in complex and imbalanced network environments.

入侵检测系统（IDS）在保护现代物联网和云基础设施方面的重要性受到网络威胁日益复杂的挑战。基于深度学习（DL）的入侵检测虽然很有前途，但往往受到严重的类不平衡和难以在复杂的网络流量中捕获歧视性时空模式的限制。为了克服这些限制，本文提出了GA-CNN-BiGRU-IDS，这是一种集成了先进数据增强技术和混合深度学习架构的新型IDS框架。基于遗传算法的增强方案采用单点交叉操作保持特征一致性，采用余弦相似度优化突变算子保持原始数据分布，通过高质量的合成少数派实例有效缓解类失衡。随后的CNN-BiGRU混合模型自主学习鲁棒特征表示；卷积层从数据中提取空间层次，而双向门控循环单元捕获长期时间依赖性。该框架在基于netflow的NF-UNSW-NB15数据集上进行了严格的二元和多类场景评估。实验表明，我们的方法达到了领先的结果，特别是在检测罕见和复杂的攻击类别。综合比较分析证实了该框架优于消融和现有基准方法，突出了其在复杂和不平衡网络环境中大大提高IDS可靠性的潜力。

{"title":"GA-CNN-BiGRU-IDS: A robust framework for Intrusion Detection System based on GA for data augmentation and hybrid CNN-BiGRU model for spatiotemporal feature extraction","authors":"Rahma Jablaoui, Noureddine Liouane","doi":"10.1016/j.compeleceng.2025.110900","DOIUrl":"10.1016/j.compeleceng.2025.110900","url":null,"abstract":"<div><div>The vital importance of Intrusion Detection Systems (IDS) in securing modern IoT and cloud infrastructures is challenged by the evolving sophistication of network threats. Deep learning (DL)-based IDS, while promising, are often constrained by acute class imbalance and the difficulty of capturing discriminative spatiotemporal patterns within complex network traffic. To overcome these limitations, this paper proposes GA-CNN-BiGRU-IDS, a novel IDS framework integrating an advanced data augmentation technique with a hybrid DL architecture. The proposed genetic algorithm (GA)-inspired augmentation scheme incorporates a one-point crossover operation to maintain feature coherence and a cosine similarity-optimized mutation operator to preserve the original data distribution, effectively mitigating class imbalance through high-quality synthetic minority instances. The subsequent hybrid CNN-BiGRU model autonomously learns robust feature representations; the convolutional layers extract spatial hierarchies from the data, while the bidirectional gated recurrent units capture long-range temporal dependencies. The framework is rigorously evaluated on the well-established Netflow-based NF-UNSW-NB15 dataset under binary and multiclass scenarios. Experiments show that our methodology reaches leading results, particularly in detecting rare and sophisticated attack categories. A comprehensive comparative analysis confirms the framework’s superiority over both ablation and existing benchmark approaches, highlighting its potential to greatly improve the reliability of IDS in complex and imbalanced network environments.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"130 ","pages":"Article 110900"},"PeriodicalIF":4.9,"publicationDate":"2025-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145736915","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Learning based vertex prediction for high capacity reversible data hiding in 3D meshes 基于学习的三维网格高容量可逆数据隐藏顶点预测

IF 4.9 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Computers & Electrical Engineering

Pub Date : 2025-12-08 DOI: 10.1016/j.compeleceng.2025.110898

Mohsin Shah , Muhammad Nawaz Khan , Sokjoon Lee , Byoung Koo Kim , Inam Ullah

Reversible data hiding (RDH) is a prominent information hiding method that enables the lossless embedding of additional data within digital multimedia cover files. RDH guarantees perfect reversibility, enabling the receiver to reconstruct the original cover media following data extraction. RDH in 3D meshes has gained increasing attention due to its widespread applications. A critical challenge in RDH for 3D meshes is to accurately predict vertex positions to minimize distortion during data embedding using prediction error expansion (PEE). In this paper, we propose a novel multilayer perceptron based predictor (MLPP) by dividing the vertices of a 3D mesh model into two sets and using one set (reference set) to predict the other set (embedding set) for data embedding. The 1-ring neighbor vertices of the reference set are arranged into a fixed dimension feature vector to train a lightweight and computationally efficient multilayer perceptron network. The proposed network learns from the local geometric structure of 3D meshes to predict embedding vertices and produces sharp prediction errors histogram centered at zero. Furthermore, the small prediction errors are expanded for data embedding, leading to higher capacity and lower distortion. Experimental results demonstrate that the proposed MLPP attains better performance in terms of prediction accuracy, embedding capacity and embedding distortion.

可逆数据隐藏（RDH）是一种重要的信息隐藏方法，它可以在数字多媒体封面文件中无损嵌入附加数据。RDH保证了完美的可逆性，使接收器能够在数据提取后重建原始覆盖介质。三维网格中的RDH由于其广泛的应用而受到越来越多的关注。在三维网格的RDH中，一个关键的挑战是使用预测误差扩展（PEE）来准确预测顶点位置，以减少数据嵌入过程中的失真。本文提出了一种基于多层感知器的预测器（MLPP），将三维网格模型的顶点划分为两组，并使用一组（参考集）预测另一组（嵌入集）进行数据嵌入。将参考集的1环相邻顶点排列成固定维数的特征向量，训练出轻量级且计算效率高的多层感知器网络。该网络从三维网格的局部几何结构中学习预测嵌入点，并产生以零为中心的明显预测误差直方图。此外，将较小的预测误差扩展到数据嵌入，从而提高了容量和降低了失真。实验结果表明，该方法在预测精度、嵌入容量和嵌入失真等方面都取得了较好的效果。

{"title":"Learning based vertex prediction for high capacity reversible data hiding in 3D meshes","authors":"Mohsin Shah , Muhammad Nawaz Khan , Sokjoon Lee , Byoung Koo Kim , Inam Ullah","doi":"10.1016/j.compeleceng.2025.110898","DOIUrl":"10.1016/j.compeleceng.2025.110898","url":null,"abstract":"<div><div>Reversible data hiding (RDH) is a prominent information hiding method that enables the lossless embedding of additional data within digital multimedia cover files. RDH guarantees perfect reversibility, enabling the receiver to reconstruct the original cover media following data extraction. RDH in 3D meshes has gained increasing attention due to its widespread applications. A critical challenge in RDH for 3D meshes is to accurately predict vertex positions to minimize distortion during data embedding using prediction error expansion (PEE). In this paper, we propose a novel multilayer perceptron based predictor (MLPP) by dividing the vertices of a 3D mesh model into two sets and using one set (reference set) to predict the other set (embedding set) for data embedding. The 1-ring neighbor vertices of the reference set are arranged into a fixed dimension feature vector to train a lightweight and computationally efficient multilayer perceptron network. The proposed network learns from the local geometric structure of 3D meshes to predict embedding vertices and produces sharp prediction errors histogram centered at zero. Furthermore, the small prediction errors are expanded for data embedding, leading to higher capacity and lower distortion. Experimental results demonstrate that the proposed MLPP attains better performance in terms of prediction accuracy, embedding capacity and embedding distortion.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"130 ","pages":"Article 110898"},"PeriodicalIF":4.9,"publicationDate":"2025-12-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145736922","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Noise-robust anomaly detection framework using temporal features for industrial applications 基于时序特征的工业应用噪声鲁棒异常检测框架

IF 4.9 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Computers & Electrical Engineering

Pub Date : 2025-12-08 DOI: 10.1016/j.compeleceng.2025.110885

Hyeonah Jang , Taehan Lee , Bum Jun Kim , Hyeyeon Choi , Sang Woo Kim

In the steel industry, the hot rolling process is the final step where pressure is applied to flatten the steel. Accidents such as steel plate tearing can occur during this process, leading to financial losses and safety hazards. Therefore, an automated system for immediate accident detection is necessary. In this paper, we propose an unsupervised learning method for accident detection that is robust to noise in the analysis of surveillance videos from the hot rolling process. Real industrial datasets are often noisy, and in the hot rolling process, steam frequently obscures steel plates, degrading detection accuracy. Additionally, the number of actual accident videos is extremely limited, making it impossible to use them in the training process. To address this problem, we leverage the fact that video data consists of consecutive frames and that the differences between them are small in normal videos. We propose a framework that examines feature differences between consecutive video frames, where small differences indicate normal states, while large differences suggest accidents. Moreover, to overcome the challenge of distinguishing normal and abnormal data in noisy environments, we propose a method for extracting features that are less affected by noise, allowing the model to concentrate on the target object and effectively detect anomalies. Our proposed method achieved an F₁-score of 99.6% and an accuracy of 99.6% at the clip level on a limited and imbalanced real-world dataset.

在钢铁工业中，热轧过程是施加压力使钢材变平的最后一步。在此过程中会发生钢板撕裂等事故，造成经济损失和安全隐患。因此，一个自动的事故检测系统是必要的。在本文中，我们提出了一种无监督学习的事故检测方法，该方法在热轧过程监控视频分析中对噪声具有鲁棒性。真实的工业数据集通常是有噪声的，并且在热轧过程中，蒸汽经常使钢板模糊，降低了检测精度。此外，实际事故视频的数量非常有限，因此无法在培训过程中使用它们。为了解决这个问题，我们利用视频数据由连续帧组成的事实，并且在正常视频中它们之间的差异很小。我们提出了一个框架来检查连续视频帧之间的特征差异，其中小的差异表明正常状态，而大的差异表明事故。此外，为了克服在噪声环境中区分正常和异常数据的挑战，我们提出了一种提取受噪声影响较小的特征的方法，使模型能够专注于目标物体并有效地检测异常。我们提出的方法在有限和不平衡的真实数据集上实现了99.6%的f1得分和99.6%的准确率。

{"title":"Noise-robust anomaly detection framework using temporal features for industrial applications","authors":"Hyeonah Jang , Taehan Lee , Bum Jun Kim , Hyeyeon Choi , Sang Woo Kim","doi":"10.1016/j.compeleceng.2025.110885","DOIUrl":"10.1016/j.compeleceng.2025.110885","url":null,"abstract":"<div><div>In the steel industry, the hot rolling process is the final step where pressure is applied to flatten the steel. Accidents such as steel plate tearing can occur during this process, leading to financial losses and safety hazards. Therefore, an automated system for immediate accident detection is necessary. In this paper, we propose an unsupervised learning method for accident detection that is robust to noise in the analysis of surveillance videos from the hot rolling process. Real industrial datasets are often noisy, and in the hot rolling process, steam frequently obscures steel plates, degrading detection accuracy. Additionally, the number of actual accident videos is extremely limited, making it impossible to use them in the training process. To address this problem, we leverage the fact that video data consists of consecutive frames and that the differences between them are small in normal videos. We propose a framework that examines feature differences between consecutive video frames, where small differences indicate normal states, while large differences suggest accidents. Moreover, to overcome the challenge of distinguishing normal and abnormal data in noisy environments, we propose a method for extracting features that are less affected by noise, allowing the model to concentrate on the target object and effectively detect anomalies. Our proposed method achieved an F<sub>1</sub>-score of 99.6% and an accuracy of 99.6% at the clip level on a limited and imbalanced real-world dataset.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"130 ","pages":"Article 110885"},"PeriodicalIF":4.9,"publicationDate":"2025-12-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145736923","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

YOLOv8 with innovative dilated residual and attention modules for mammographic tumor detection YOLOv8具有创新的扩展残余和关注模块，用于乳房x线摄影肿瘤检测

IF 4.9 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Computers & Electrical Engineering

Pub Date : 2025-12-07 DOI: 10.1016/j.compeleceng.2025.110903

Zahra Raeisi , Amirsadegh Roshanzamir , Fatemeh Abedi Lomer , Reza Ahmadi Lashaki

Early and accurate detection of breast cancer remains a critical challenge in medical imaging. In this study, we propose a novel deep learning-based framework built upon the YOLOv8 architecture to enable real-time and precise breast cancer detection. To enhance its capability in handling mammographic variability, we introduce a Multi-Scale Dilated Residual (MSDR) block that replaces the original C2f modules in YOLOv8-L’s backbone. This module leverages dilated convolutions and residual pathways to capture multi-scale contextual features while preserving spatial detail, which is vital for localizing tumors of varying sizes. Additionally, a Feature-Focused Attention (FFA) block is proposed to improve the model’s discrimination power by emphasizing tumor-relevant regions and suppressing background noise, resulting in more accurate classification and segmentation performance. We conduct extensive experiments across three benchmark breast imaging datasets CBIS-DDSM, INbreast, and BUSI, to evaluate the model’s detection accuracy and generalization capability across diverse clinical scenarios. Results demonstrate that the proposed framework consistently outperforms existing state-of-the-art methods in both performance and computational efficiency, highlighting its potential for real-time clinical deployment.

早期和准确检测乳腺癌仍然是医学影像学的一个关键挑战。在这项研究中，我们提出了一种基于YOLOv8架构的新型深度学习框架，以实现实时和精确的乳腺癌检测。为了增强其处理乳房x线摄影变异性的能力，我们引入了一个多尺度膨胀残差（MSDR）块，取代了YOLOv8-L骨干中的原始C2f模块。该模块利用扩张卷积和残差通路来捕获多尺度上下文特征，同时保留空间细节，这对于定位不同大小的肿瘤至关重要。此外，我们还提出了Feature-Focused Attention (FFA) block，通过强调肿瘤相关区域和抑制背景噪声来提高模型的识别能力，从而获得更准确的分类和分割性能。我们在三个基准乳腺成像数据集（CBIS-DDSM、INbreast和BUSI）上进行了广泛的实验，以评估该模型在不同临床场景下的检测准确性和泛化能力。结果表明，所提出的框架在性能和计算效率方面始终优于现有的最先进的方法，突出了其实时临床部署的潜力。

{"title":"YOLOv8 with innovative dilated residual and attention modules for mammographic tumor detection","authors":"Zahra Raeisi , Amirsadegh Roshanzamir , Fatemeh Abedi Lomer , Reza Ahmadi Lashaki","doi":"10.1016/j.compeleceng.2025.110903","DOIUrl":"10.1016/j.compeleceng.2025.110903","url":null,"abstract":"<div><div>Early and accurate detection of breast cancer remains a critical challenge in medical imaging. In this study, we propose a novel deep learning-based framework built upon the YOLOv8 architecture to enable real-time and precise breast cancer detection. To enhance its capability in handling mammographic variability, we introduce a Multi-Scale Dilated Residual (MSDR) block that replaces the original C2f modules in YOLOv8-L’s backbone. This module leverages dilated convolutions and residual pathways to capture multi-scale contextual features while preserving spatial detail, which is vital for localizing tumors of varying sizes. Additionally, a Feature-Focused Attention (FFA) block is proposed to improve the model’s discrimination power by emphasizing tumor-relevant regions and suppressing background noise, resulting in more accurate classification and segmentation performance. We conduct extensive experiments across three benchmark breast imaging datasets CBIS-DDSM, INbreast, and BUSI, to evaluate the model’s detection accuracy and generalization capability across diverse clinical scenarios. Results demonstrate that the proposed framework consistently outperforms existing state-of-the-art methods in both performance and computational efficiency, highlighting its potential for real-time clinical deployment.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"130 ","pages":"Article 110903"},"PeriodicalIF":4.9,"publicationDate":"2025-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145736924","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

H-Dense FHNet: Hierarchical dense forward harmonic network for depression detection H-Dense FHNet：用于降噪检测的分层密集正向谐波网络

IF 4.9 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Computers & Electrical Engineering

Pub Date : 2025-12-06 DOI: 10.1016/j.compeleceng.2025.110865

Amol Vishwanath Dhumane , Nihar M. Ranjan , Mubin Tamboli , Jayashree Rajesh Prasad , Rajesh Shardanand Prasad

Objective

Depression is a pervasive mental health state that impacts millions worldwide, exemplified by the constant loss of interest, sadness, and several physical and emotional indicators. Even with its widespread occurrence, many individuals fail to receive a timely diagnosis or adequate treatment. Nevertheless, developing an automated scheme capable of detecting depression signs accurately from the text remains a complex task. This article proposes a new approach termed Hierarchical Dense Forward Harmonic Network (H-Dense FHNet) for depression detection in text sentences.

Methodology

The input text sentence is fetched using the selected dataset. Consequently, tokenization is implemented to split the sentence into tokens with the help of Bidirectional Encoder Representations from Transformers (BERT). Further, feature extraction is accomplished, and features like verb vectors, capitalized words, elongated units, count vectors of categories, adjective vectors, length of text, degree adverbs vectors, and punctuation vectors are mined. Ultimately, depression detection is accomplished by the presented H-Dense FHNet, a unified framework of Hierarchical Attention Network (HAN), DenseNet, and harmonic analysis.

Result

The evaluation of the presented H-Dense FHNet shows that it obtained maximal F1-score, recall, and precision of 92.996 %, 93.655 %, and 92.766 % correspondingly.

抑郁症是一种普遍存在的精神健康状态，影响着全球数百万人，其典型表现为不断丧失兴趣、悲伤以及一些身体和情感指标。即使它广泛发生，许多人未能得到及时的诊断或适当的治疗。然而，开发一种能够从文本中准确检测抑郁迹象的自动化方案仍然是一项复杂的任务。本文提出了一种新的文本句子降噪检测方法——层次密集前向谐波网络（H-Dense FHNet）。使用选定的数据集获取输入文本句子。因此，在双向编码器表示（BERT）的帮助下，实现了标记化，将句子拆分为多个标记。进一步，完成特征提取，挖掘动词向量、大写词、拉长单位、类别计数向量、形容词向量、文本长度、程度副词向量、标点符号向量等特征。最终，消沉检测由提出的H-Dense FHNet、分层注意网络（HAN）、DenseNet和谐波分析的统一框架完成。结果对所提出的H-Dense FHNet的评价结果表明，其最高的f1分、召回率和准确率分别为92.996%、93.655 %和92.766%。

{"title":"H-Dense FHNet: Hierarchical dense forward harmonic network for depression detection","authors":"Amol Vishwanath Dhumane , Nihar M. Ranjan , Mubin Tamboli , Jayashree Rajesh Prasad , Rajesh Shardanand Prasad","doi":"10.1016/j.compeleceng.2025.110865","DOIUrl":"10.1016/j.compeleceng.2025.110865","url":null,"abstract":"<div><h3>Objective</h3><div>Depression is a pervasive mental health state that impacts millions worldwide, exemplified by the constant loss of interest, sadness, and several physical and emotional indicators. Even with its widespread occurrence, many individuals fail to receive a timely diagnosis or adequate treatment. Nevertheless, developing an automated scheme capable of detecting depression signs accurately from the text remains a complex task. This article proposes a new approach termed Hierarchical Dense Forward Harmonic Network (H-Dense FH<img>Net) for depression detection in text sentences.</div></div><div><h3>Methodology</h3><div>The input text sentence is fetched using the selected dataset. Consequently, tokenization is implemented to split the sentence into tokens with the help of Bidirectional Encoder Representations from Transformers (BERT). Further, feature extraction is accomplished, and features like verb vectors, capitalized words, elongated units, count vectors of categories, adjective vectors, length of text, degree adverbs vectors, and punctuation vectors are mined. Ultimately, depression detection is accomplished by the presented H-Dense FH<img>Net, a unified framework of Hierarchical Attention Network (HAN), DenseNet, and harmonic analysis.</div></div><div><h3>Result</h3><div>The evaluation of the presented H-Dense FH<img>Net shows that it obtained maximal F1-score, recall, and precision of 92.996 %, 93.655 %, and 92.766 % correspondingly.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"130 ","pages":"Article 110865"},"PeriodicalIF":4.9,"publicationDate":"2025-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145685506","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Transformer-based motion-visual integrated fusion for isolated sign language recognition 基于变换的运动视觉融合孤立手语识别

IF 4.9 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Computers & Electrical Engineering

Pub Date : 2025-12-06 DOI: 10.1016/j.compeleceng.2025.110902

S. Renjith , Aneesh Varghese , Manazhy Rashmi , Poorna S.S.

Sign Language Recognition (SLR) is a critical technology that facilitates seamless interaction between hearing-impaired individuals and the broader community. Among the various tasks in SLR, isolated sign word recognition remains particularly challenging due to high visual similarity between signs and variability in hand motion dynamics. To address these challenges, we propose Transformer-based Motion and Visual Integrated Fusion for Isolated Sign Language Recognition (TransMoVIF), a novel dual-stream transformer-based framework that effectively combines skeletal motion trajectories and visual appearance cues for enhanced recognition. The proposed model extracts fine-grained motion signatures by encoding trajectory-level features—such as velocity, displacement, and curvature from hand keypoints. Simultaneously, a visual stream processes RGB keyframes identified using a crucial frame selection algorithm based on pose variations and structural similarity. These two complementary modalities are fused using a cross-attention mechanism, allowing the model to learn intricate relationships between motion dynamics and visual features. Unlike conventional late-fusion methods, TransMoVIF enables deep integration of semantic and temporal patterns across streams, enhancing its ability to differentiate between visually similar signs with distinct motions. Extensive experiments on two publicly available isolated SLR datasets demonstrate that TransMoVIF outperforms state-of-the-art unimodal and fusion-based models in terms of both accuracy and robustness, establishing a new benchmark for isolated sign language recognition.

手语识别（SLR）是一项关键技术，它促进了听障人士与更广泛的社区之间的无缝互动。在单反的各种任务中，由于手势之间的高度视觉相似性和手部运动动力学的可变性，孤立的手势单词识别仍然是特别具有挑战性的。为了解决这些挑战，我们提出了基于变压器的运动和视觉集成融合的孤立手语识别（TransMoVIF），这是一种新的基于双流变压器的框架，有效地结合了骨骼运动轨迹和视觉外观线索，以增强识别。该模型通过对手部关键点的速度、位移和曲率等轨迹级特征进行编码，提取出细粒度的运动特征。同时，视觉流处理使用基于姿态变化和结构相似性的关键帧选择算法识别的RGB关键帧。这两种互补的模式使用交叉注意机制融合，允许模型学习运动动力学和视觉特征之间的复杂关系。与传统的后期融合方法不同，TransMoVIF可以跨流深度整合语义和时间模式，增强其区分具有不同动作的视觉相似符号的能力。在两个公开可用的孤立单反数据集上进行的大量实验表明，TransMoVIF在准确性和鲁棒性方面都优于最先进的单峰模型和基于融合的模型，为孤立手语识别建立了新的基准。

{"title":"Transformer-based motion-visual integrated fusion for isolated sign language recognition","authors":"S. Renjith , Aneesh Varghese , Manazhy Rashmi , Poorna S.S.","doi":"10.1016/j.compeleceng.2025.110902","DOIUrl":"10.1016/j.compeleceng.2025.110902","url":null,"abstract":"<div><div>Sign Language Recognition (SLR) is a critical technology that facilitates seamless interaction between hearing-impaired individuals and the broader community. Among the various tasks in SLR, isolated sign word recognition remains particularly challenging due to high visual similarity between signs and variability in hand motion dynamics. To address these challenges, we propose Transformer-based Motion and Visual Integrated Fusion for Isolated Sign Language Recognition (TransMoVIF), a novel dual-stream transformer-based framework that effectively combines skeletal motion trajectories and visual appearance cues for enhanced recognition. The proposed model extracts fine-grained motion signatures by encoding trajectory-level features—such as velocity, displacement, and curvature from hand keypoints. Simultaneously, a visual stream processes RGB keyframes identified using a crucial frame selection algorithm based on pose variations and structural similarity. These two complementary modalities are fused using a cross-attention mechanism, allowing the model to learn intricate relationships between motion dynamics and visual features. Unlike conventional late-fusion methods, TransMoVIF enables deep integration of semantic and temporal patterns across streams, enhancing its ability to differentiate between visually similar signs with distinct motions. Extensive experiments on two publicly available isolated SLR datasets demonstrate that TransMoVIF outperforms state-of-the-art unimodal and fusion-based models in terms of both accuracy and robustness, establishing a new benchmark for isolated sign language recognition.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"130 ","pages":"Article 110902"},"PeriodicalIF":4.9,"publicationDate":"2025-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145685507","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A learning-based outage prediction method for resilient electricity distribution systems in response to extreme weather events 基于学习的极端天气下弹性配电系统停电预测方法

IF 4.9 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Computers & Electrical Engineering

Pub Date : 2025-12-06 DOI: 10.1016/j.compeleceng.2025.110881

Mohammad Hassan Amirioun, S. Sepehr Tabatabaei, Amin Asgari

Electricity distribution systems are vulnerable to damage from extreme weather events like hurricanes and floods. Although predicting outage locations in these systems is a significant challenge, it provides operators with critical data for implementing proactive measures. This paper presents a decision tree-based learning method to predict potential outages in distribution branches during hurricanes. The challenges of input data, component diversity, and numerous affecting factors are highlighted and effectively addressed. Our model considers all potentially effective static and dynamic features to estimate the damage risk for each branch. The data for training and testing the classifier were acquired from historical records and synthesized samples based on expert knowledge, with a separate set of real data used for validation. Beyond outage prediction, the classifier also serves as a feature selection tool by identifying the most discriminative features. Numerical simulations confirmed a high level of accuracy with a negligible error rate. The method was successfully implemented on a modified IEEE 33-bus distribution system.

配电系统很容易受到飓风和洪水等极端天气事件的破坏。尽管预测这些系统的停机位置是一项重大挑战，但它为作业者提供了实施主动措施的关键数据。本文提出了一种基于决策树的学习方法来预测飓风期间配电分支的潜在停电。输入数据、组件多样性和众多影响因素的挑战得到了突出和有效解决。我们的模型考虑了所有潜在有效的静态和动态特征来估计每个分支的损害风险。训练和测试分类器的数据来自历史记录和基于专家知识的合成样本，并使用一组单独的真实数据进行验证。除了停机预测之外，分类器还可以通过识别最具判别性的特征来作为特征选择工具。数值模拟证实了高水平的精度和可忽略不计的错误率。该方法已在改进的IEEE 33总线配电系统上成功实现。

{"title":"A learning-based outage prediction method for resilient electricity distribution systems in response to extreme weather events","authors":"Mohammad Hassan Amirioun, S. Sepehr Tabatabaei, Amin Asgari","doi":"10.1016/j.compeleceng.2025.110881","DOIUrl":"10.1016/j.compeleceng.2025.110881","url":null,"abstract":"<div><div>Electricity distribution systems are vulnerable to damage from extreme weather events like hurricanes and floods. Although predicting outage locations in these systems is a significant challenge, it provides operators with critical data for implementing proactive measures. This paper presents a decision tree-based learning method to predict potential outages in distribution branches during hurricanes. The challenges of input data, component diversity, and numerous affecting factors are highlighted and effectively addressed. Our model considers all potentially effective static and dynamic features to estimate the damage risk for each branch. The data for training and testing the classifier were acquired from historical records and synthesized samples based on expert knowledge, with a separate set of real data used for validation. Beyond outage prediction, the classifier also serves as a feature selection tool by identifying the most discriminative features. Numerical simulations confirmed a high level of accuracy with a negligible error rate. The method was successfully implemented on a modified IEEE 33-bus distribution system.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"130 ","pages":"Article 110881"},"PeriodicalIF":4.9,"publicationDate":"2025-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145685503","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Enhancing Alzheimer’s disease risk prediction using brain age gap: A randomized regression model for brain age prediction with magnetic resonance imaging 利用脑年龄差距增强阿尔茨海默病风险预测：磁共振成像预测脑年龄的随机回归模型

IF 4.9 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Computers & Electrical Engineering

Pub Date : 2025-12-06 DOI: 10.1016/j.compeleceng.2025.110905

Raveendra Pilli , Tripti Goel , R. Murugan

Brain age prediction from magnetic resonance imaging (MRI) using machine learning and deep learning models is gaining importance as a key biomarker for assessing brain health and development. The brain age gap (the difference between chronological and predicted age) can be a biomarker to analyze neurocognitive abnormalities and disease-related brain changes. Random vector functional link (RVFL) network, a type of single-hidden-layer feedforward network, offers strong generalization capabilities. To enhance age prediction, this study incorporates Universum learning principles, leveraging additional informative data points from the same domain but with a distinct distribution. We propose a novel Universum-RVFL model to refine regression-based age estimation. Experimental validation on neuroimaging datasets demonstrates superior predictive performance compared to baseline methods, achieving a mean absolute error of 2.84 years and a root mean square error of 3.69 years in healthy individuals. The significance of the brain age gap is further examined in individuals with mild cognitive impairment (MCI), early MCI (EMCI), late MCI (LMCI), and Alzheimer’s disease (AD). Results indicate a significantly larger brain age gap in AD-affected individuals, reflecting accelerated brain aging and atrophy. Additionally, we employ the Cox proportional hazards model to predict the risk of MCI-to-AD conversion, revealing that the brain age gap is a strong predictor of AD progression compared to other covariates.

利用机器学习和深度学习模型从磁共振成像（MRI）中预测大脑年龄，作为评估大脑健康和发育的关键生物标志物，正变得越来越重要。大脑年龄差距（实际年龄和预测年龄之间的差异）可以作为分析神经认知异常和与疾病相关的大脑变化的生物标志物。随机向量功能链接（RVFL）网络是一种单隐层前馈网络，具有很强的泛化能力。为了增强年龄预测，本研究结合了Universum的学习原理，利用来自同一领域但具有不同分布的额外信息数据点。我们提出了一种新的Universum-RVFL模型来改进基于回归的年龄估计。神经影像学数据集的实验验证表明，与基线方法相比，该方法的预测性能优越，健康个体的平均绝对误差为2.84年，均方根误差为3.69年。在轻度认知障碍（MCI）、早期MCI （EMCI）、晚期MCI （LMCI）和阿尔茨海默病（AD）患者中，进一步研究了脑年龄差距的意义。结果表明，ad患者的脑年龄差距明显更大，反映了大脑衰老和萎缩的加速。此外，我们采用Cox比例风险模型来预测mci到AD转换的风险，表明与其他协变量相比，大脑年龄差距是AD进展的一个强有力的预测因子。

{"title":"Enhancing Alzheimer’s disease risk prediction using brain age gap: A randomized regression model for brain age prediction with magnetic resonance imaging","authors":"Raveendra Pilli , Tripti Goel , R. Murugan","doi":"10.1016/j.compeleceng.2025.110905","DOIUrl":"10.1016/j.compeleceng.2025.110905","url":null,"abstract":"<div><div>Brain age prediction from magnetic resonance imaging (MRI) using machine learning and deep learning models is gaining importance as a key biomarker for assessing brain health and development. The brain age gap (the difference between chronological and predicted age) can be a biomarker to analyze neurocognitive abnormalities and disease-related brain changes. Random vector functional link (RVFL) network, a type of single-hidden-layer feedforward network, offers strong generalization capabilities. To enhance age prediction, this study incorporates Universum learning principles, leveraging additional informative data points from the same domain but with a distinct distribution. We propose a novel Universum-RVFL model to refine regression-based age estimation. Experimental validation on neuroimaging datasets demonstrates superior predictive performance compared to baseline methods, achieving a mean absolute error of 2.84 years and a root mean square error of 3.69 years in healthy individuals. The significance of the brain age gap is further examined in individuals with mild cognitive impairment (MCI), early MCI (EMCI), late MCI (LMCI), and Alzheimer’s disease (AD). Results indicate a significantly larger brain age gap in AD-affected individuals, reflecting accelerated brain aging and atrophy. Additionally, we employ the Cox proportional hazards model to predict the risk of MCI-to-AD conversion, revealing that the brain age gap is a strong predictor of AD progression compared to other covariates.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"130 ","pages":"Article 110905"},"PeriodicalIF":4.9,"publicationDate":"2025-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145685502","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0