Computers & Electrical Engineering最新文献

筛选
英文 中文
Optimized fractional order Takagi-Sugeno Fuzzy-PID power system stabilizer: An enhanced dung beetle optimization approach 优化分数阶Takagi-Sugeno模糊pid电力系统稳定器:一种增强的屎壳虫优化方法
IF 4.9 3区 计算机科学
Computers & Electrical Engineering Pub Date : 2025-10-15 DOI: 10.1016/j.compeleceng.2025.110767
Intissar Hattabi , Aissa Kheldoun , Rafik Bradai , Hamza Belmadani
{"title":"Optimized fractional order Takagi-Sugeno Fuzzy-PID power system stabilizer: An enhanced dung beetle optimization approach","authors":"Intissar Hattabi ,&nbsp;Aissa Kheldoun ,&nbsp;Rafik Bradai ,&nbsp;Hamza Belmadani","doi":"10.1016/j.compeleceng.2025.110767","DOIUrl":"10.1016/j.compeleceng.2025.110767","url":null,"abstract":"<div><div>This paper introduces a novel Fractional Order Takagi-Sugeno Fuzzy-PID (FO-TSF-PID) controller, optimized using an enhanced Dung Beetle Optimization (EDBO) algorithm, to improve the damping of low-frequency oscillations in power systems. The controller's design involves simultaneous optimization of membership functions (MFs) and gains, enhancing performance, particularly under three-phase fault conditions. The optimization process, executed through the EDBO algorithm, is both flexible and straightforward to implement. The FO-TSF-PID controller was tested on a two-area power system subjected to three symmetrical faults. Performance evaluations demonstrated the controller's superiority over the standard Fractional Order PID (FOFPID) controller, achieving significant improvements in inter-area and local-area eigenvalues. Specifically, inter-area improvements were 87.08 % with PSO, 83.86 % with EO, 81.29 % with DBO, and 78.89 % with EDBO, while local-area improvements were 71.01 % with PSO, 70.52 % with EO, 65.73 % with DBO, and 64.32 % with EDBO. Comparative analysis against traditional controllers such as Lead-Lag Power System Stabilizer (PSS), Proportional-Integral-Derivative (PID), and Fractional Order PID (FOPID) consistently showed the FO-TSF-PID controller's enhanced stability and robustness. Further comparisons revealed that the EDBO-optimized FO-TSF-PID controller achieved 99.94 %, 99.93 %, and 99.95 % enhancements compared to those optimized using PSO, EO, and DBO, respectively.</div><div>The results indicate that the EDBO-optimized FO-TSF-PID controller excels in reducing settling time, minimizing overshoot, and improving steady-state error, thus proving its efficacy in stabilizing power systems.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"128 ","pages":"Article 110767"},"PeriodicalIF":4.9,"publicationDate":"2025-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145320473","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
SASTNet: Self-Attention-Enabled Spatio-Temporal Network for single scene video anomaly detection SASTNet:用于单场景视频异常检测的自关注时空网络
IF 4.9 3区 计算机科学
Computers & Electrical Engineering Pub Date : 2025-10-15 DOI: 10.1016/j.compeleceng.2025.110749
Rashmiranjan Nayak, Umesh Chandra Pati, Santos Kumar Das
{"title":"SASTNet: Self-Attention-Enabled Spatio-Temporal Network for single scene video anomaly detection","authors":"Rashmiranjan Nayak,&nbsp;Umesh Chandra Pati,&nbsp;Santos Kumar Das","doi":"10.1016/j.compeleceng.2025.110749","DOIUrl":"10.1016/j.compeleceng.2025.110749","url":null,"abstract":"<div><div>Video anomaly detection is the process of localizing abnormal video patterns spatiotemporally. The extraction of spatiotemporal features incorporating both local and global spatiotemporal dependency from the video is one of the essential and challenging tasks for video anomaly detection. In order to solve this issue, this article proposes a novel deep reconstruction model-based <strong>S</strong>elf-<strong>A</strong>ttention-enabled <strong>S</strong>patio-<strong>T</strong>emporal <strong>Net</strong>work (SASTNet) model that uses Self-Attention-enabled Convolutional Bidirectional Long-Short-Term-Memory-based Auto-Encoder (SAConvBiLSTMAE) architecture. A novel Self-Attention-enabled Convolutional Bidirectional Long-Short-Term-Memory (SAConvBiLSTM) block is designed to learn the global spatiotemporal dependencies during the frame reconstruction in an end-to-end autoencoder framework for detecting video anomalies. The foundation of the proposed SASTNet model is supported by a comprehensive mathematical problem formulation using a deep reconstruction approach for detecting video anomalies in a single-scene scenario. The selection of an appropriate threshold is one of the crucial factors in video anomaly detection problems. Hence, an efficient strategy has been implemented to select the optimal threshold by utilizing the Receiver Operating Characteristics (ROC) curve for detecting video anomalies. Finally, extensive experimental analysis and comparison with state-of-the-art approaches using three bench-marked single-scene video anomaly datasets, such as UCSD Ped1, UCSD Ped2, and CUHK Avenue, indicate the superiority of the proposed framework in video anomaly detection.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"128 ","pages":"Article 110749"},"PeriodicalIF":4.9,"publicationDate":"2025-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145319523","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 fitness-assignment method for evolutionary constrained multi-objective optimization 演化约束多目标优化的适应度分配方法
IF 4.9 3区 计算机科学
Computers & Electrical Engineering Pub Date : 2025-10-15 DOI: 10.1016/j.compeleceng.2025.110769
Oladayo S. Ajani , Sri Srinivasa Raju M. , Anand Paul , Rammohan Mallipeddi
{"title":"A fitness-assignment method for evolutionary constrained multi-objective optimization","authors":"Oladayo S. Ajani ,&nbsp;Sri Srinivasa Raju M. ,&nbsp;Anand Paul ,&nbsp;Rammohan Mallipeddi","doi":"10.1016/j.compeleceng.2025.110769","DOIUrl":"10.1016/j.compeleceng.2025.110769","url":null,"abstract":"<div><div>The effectiveness of Constrained Multi-Objective Evolutionary Algorithms (CMOEAs) depends on their ability to explore diverse feasible regions within the problem search space by utilizing information from both feasible and infeasible solutions. While many high-performing CMOEAs have been proposed, they are often too complex due to their underlying multi-stage or multi-population design. To simplify the process, fitness-assignment-based CMOEAs have been proposed that integrate feasibility information into traditional methods from unconstrained multi-objective optimization. However, these approaches are not scalable in terms of performance because it is difficult to design a fitness assignment method that can simultaneously account for constraint violation, convergence, and diversity. Hence, in this paper, we propose an effective single-population fitness assignment-based CMOEA referred to as <span><math><msubsup><mrow><mi>I</mi></mrow><mrow><mi>S</mi><mi>D</mi><msup><mrow><mi>E</mi></mrow><mrow><mo>+</mo></mrow></msup></mrow><mrow><mi>c</mi></mrow></msubsup></math></span> that can explore different feasible regions in the search space. <span><math><msubsup><mrow><mi>I</mi></mrow><mrow><mi>S</mi><mi>D</mi><msup><mrow><mi>E</mi></mrow><mrow><mo>+</mo></mrow></msup></mrow><mrow><mi>c</mi></mrow></msubsup></math></span> is a fitness assignment-based algorithm, that is an efficient fusion of constraint violation (c), Shift-based Density Estimation (SDE), and sum of objectives (+ ). This fusion facilitates the efficient use of information from infeasible solutions and ensures the algorithm can effectively span diverse feasible regions in the search space. The performance of <span><math><msubsup><mrow><mi>I</mi></mrow><mrow><mi>S</mi><mi>D</mi><msup><mrow><mi>E</mi></mrow><mrow><mo>+</mo></mrow></msup></mrow><mrow><mi>c</mi></mrow></msubsup></math></span> evaluated in terms of Hypervolume and runtime complexity is favorably compared against 9 baseline CMOEAs on 6 different benchmark suites with diverse characteristics. The code of the proposed <span><math><msubsup><mrow><mi>I</mi></mrow><mrow><mi>S</mi><mi>D</mi><msup><mrow><mi>E</mi></mrow><mrow><mo>+</mo></mrow></msup></mrow><mrow><mi>c</mi></mrow></msubsup></math></span> is publicly available at <span><span>https://github.com/RammohanMallipeddi/Matlab-Codes-for-cISDE-</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"128 ","pages":"Article 110769"},"PeriodicalIF":4.9,"publicationDate":"2025-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145319530","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
Quantum-resilient trust in motion: Smart-contract-driven authentication for next-gen vehicular networks 运动中的量子弹性信任:下一代车辆网络的智能合约驱动认证
IF 4.9 3区 计算机科学
Computers & Electrical Engineering Pub Date : 2025-10-14 DOI: 10.1016/j.compeleceng.2025.110754
Ahmed S. Zaghloul , Mohamed AbdElAzim Mohamed , Hanan M. Amer , Mahmoud A. Shawky
{"title":"Quantum-resilient trust in motion: Smart-contract-driven authentication for next-gen vehicular networks","authors":"Ahmed S. Zaghloul ,&nbsp;Mohamed AbdElAzim Mohamed ,&nbsp;Hanan M. Amer ,&nbsp;Mahmoud A. Shawky","doi":"10.1016/j.compeleceng.2025.110754","DOIUrl":"10.1016/j.compeleceng.2025.110754","url":null,"abstract":"<div><div>Vehicular ad-hoc networks (VANETs) enable the exchange of critical traffic-related information among vehicles, thereby enhancing road safety and improving traffic management. However, the open and dynamic nature of VANETs exposes them to a range of security threats, including message tampering, impersonation, and privacy breaches. Traditional cryptographic methods, such as elliptic-curve Diffie–Hellman, are increasingly vulnerable to emerging quantum computing capabilities, raising significant concerns about the future resilience of VANET security. To address these challenges, this paper proposes a lattice-based conditional privacy-preserving authentication scheme (SC-LB-CPPA) that leverages the immutable characteristics of blockchain technology. The proposed scheme employs smart contract-based blockchain for dynamic public parameter updates across regions, integrating lattice-based cryptography for quantum-resistant security. Comprehensive security analyses, including informal evaluations and formal proofs using the random oracle model, demonstrate the scheme’s robustness against impersonation and message modification attacks. Furthermore, the scheme’s effectiveness is validated through a practical implementation on the <span><math><mrow><mi>R</mi><mi>e</mi><mi>m</mi><mi>i</mi><mi>x</mi><mi>V</mi><mi>M</mi></mrow></math></span>. Finally, a comprehensive performance evaluation is conducted to assess the computational and communication overheads, along with the gas costs of the proposed smart contract functions. The comparative analysis confirms the superiority of the proposed approach in terms of signature generation and verification latency, as well as signature size.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"128 ","pages":"Article 110754"},"PeriodicalIF":4.9,"publicationDate":"2025-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145319524","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
FedTinyMed: Federated learning enabled tiny multi task machine learning model for smart healthcare monitoring for IoMT FedTinyMed:支持联邦学习的微型多任务机器学习模型,用于IoMT的智能医疗监控
IF 4.9 3区 计算机科学
Computers & Electrical Engineering Pub Date : 2025-10-14 DOI: 10.1016/j.compeleceng.2025.110761
Shakir Khan , Kumar Perumal , Hadeel Alsolai , Abeer Aljohani
{"title":"FedTinyMed: Federated learning enabled tiny multi task machine learning model for smart healthcare monitoring for IoMT","authors":"Shakir Khan ,&nbsp;Kumar Perumal ,&nbsp;Hadeel Alsolai ,&nbsp;Abeer Aljohani","doi":"10.1016/j.compeleceng.2025.110761","DOIUrl":"10.1016/j.compeleceng.2025.110761","url":null,"abstract":"<div><div>This research develops a FedTinyMed, an end-to-end cognitive healthcare model suitable for Internet of Medical Things (IoMT) by combining Tiny Machine Learning (TinyML) and Federated Learning (FL). The model composed of various real time physiological sensors on the wearable devices which includes Inertial Sensors (IS), Pulse Oximeter Sensor (POS), EEG sensor, and Blood Pressure Sensor (BPS) for continuously monitoring the crucial healthcare indicators. A Unified Multi Task Learning Transformer (UMLFormer) is developed by this research which ensures parallel inference among varied healthcare tasks with shared model parameters that effectively diminishes the computational overhead. With the proper training of four benchmark datasets such as Wisconsin, Breast Cancer, PIMA, and Parkinson’s heart disease the proposed research gains end-to-end latency of 100 ms, specificity of 91.5 %, sensitivity of 97.4 %, and F1-score of 95.8 % those superiors the conventional models on the resource constrained devices. By employing Joint Reinforcement Learning enabled Quantization and Pruning (JRL-QP) our research guarantees lesser memory footprint without minimizing the accuracy on real time TinyML deployment. When compared to the state-of-the-art works, the proposed FedTinyMed enables three major advantages which includes (i) higher scalability for decentralized IoMT deployments, (ii) privacy preservation from secure FedAvg aggregation, and cloud independence for ejecting over reliance on higher latency network connections. With those processes the proposed FedTinyMed enables a robust and practical solution for real time smart healthcare monitoring.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"128 ","pages":"Article 110761"},"PeriodicalIF":4.9,"publicationDate":"2025-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145319531","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
In-memory computing platform based on a novel single-MTJ non-volatile SRAM design for intermittent AI computation 基于新型单mtj非易失SRAM设计的内存计算平台用于间歇人工智能计算
IF 4.9 3区 计算机科学
Computers & Electrical Engineering Pub Date : 2025-10-14 DOI: 10.1016/j.compeleceng.2025.110755
Nima Eslami, Seyed Hassan Hadi Nemati, Mohammad Hossein Moaiyeri
{"title":"In-memory computing platform based on a novel single-MTJ non-volatile SRAM design for intermittent AI computation","authors":"Nima Eslami,&nbsp;Seyed Hassan Hadi Nemati,&nbsp;Mohammad Hossein Moaiyeri","doi":"10.1016/j.compeleceng.2025.110755","DOIUrl":"10.1016/j.compeleceng.2025.110755","url":null,"abstract":"<div><div>This paper presents a low-power, non-volatile SRAM cell based on a single in-plane magnetic tunnel junction (I-MTJ), specifically designed for edge artificial intelligence (AI) applications where the power supply is limited and intermittent. The proposed design supports in-situ data backup and restore, while integrating in-memory computing (IMC) to eliminate the gap between computation and memory. The single-ended SRAM enables in-place write operations, fundamental logic functions, Binary Content-Addressable Memory (BCAM), and storage of intermediate computation results within the same clock cycle as they are generated. Post-layout simulations in 7 nm FinFET technology demonstrate 55 % higher read stability, 30 % higher write stability, and 99 % lower storage power compared to conventional SRAM cells. Evaluation with MobileNet-V1 under power-failure scenarios shows a 67 % reduction in delay and 82 % lower power consumption compared to prior IMC solutions.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"128 ","pages":"Article 110755"},"PeriodicalIF":4.9,"publicationDate":"2025-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145319525","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
Secure and anonymous model training using pixel-level encryption for privacy-preserving AI with obfuscation-based representation learning 使用像素级加密的安全匿名模型训练,用于基于混淆的表示学习的隐私保护AI
IF 4.9 3区 计算机科学
Computers & Electrical Engineering Pub Date : 2025-10-13 DOI: 10.1016/j.compeleceng.2025.110757
Qaiser Razi , Amit Chougule , G.S.S. Chalapathi , Vikas Hassija
{"title":"Secure and anonymous model training using pixel-level encryption for privacy-preserving AI with obfuscation-based representation learning","authors":"Qaiser Razi ,&nbsp;Amit Chougule ,&nbsp;G.S.S. Chalapathi ,&nbsp;Vikas Hassija","doi":"10.1016/j.compeleceng.2025.110757","DOIUrl":"10.1016/j.compeleceng.2025.110757","url":null,"abstract":"<div><div>We present a novel framework for privacy-preserving model training that embeds pixel-level AES-ECB encryption directly into the deep learning pipeline, enabling computation on encrypted medical images without exposing raw data. To address the challenge of learning from visually obfuscated inputs, we introduce an obfuscation-based representation learning mechanism that extracts task-relevant features while preserving utility. Extensive experiments on COVID-19, Malaria, and Pneumonia classification tasks demonstrate that our method achieves competitive performance, with only (<span><math><mo>∼</mo></math></span>1.7%) accuracy degradation compared to training on unencrypted data. Comprehensive quantitative evaluations, including Structural Similarity Index (SSIM), Peak Signal-to-Noise Ratio (PSNR), Entropy, Histogram Chi-Square Distance, and Number of Pixel Change Rates (NPCR), confirm strong visual obfuscation, ensuring data confidentiality. Our framework integrates seamlessly with existing medical imaging workflows, supporting deployment in hospitals, telemedicine systems, and cross-institutional collaborations, making it a practical solution for privacy-preserving AI in regulated healthcare environments.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"128 ","pages":"Article 110757"},"PeriodicalIF":4.9,"publicationDate":"2025-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145319528","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
Robust image segmentation through a coupled fractional deformable model: Addressing speckle noise and severe intensity inhomogeneity 通过耦合分数阶变形模型的鲁棒图像分割:解决散斑噪声和严重的强度不均匀性
IF 4.9 3区 计算机科学
Computers & Electrical Engineering Pub Date : 2025-10-13 DOI: 10.1016/j.compeleceng.2025.110768
Ankit Kumar , Subit K. Jain
{"title":"Robust image segmentation through a coupled fractional deformable model: Addressing speckle noise and severe intensity inhomogeneity","authors":"Ankit Kumar ,&nbsp;Subit K. Jain","doi":"10.1016/j.compeleceng.2025.110768","DOIUrl":"10.1016/j.compeleceng.2025.110768","url":null,"abstract":"<div><div>Intensity inhomogeneity and multiplicative noise present a significant challenge in image segmentation, particularly when they occur together. In contrast to additive noise, multiplicative noise has a much more detrimental impact on image integrity. Consequently, a significant number of presently available deformable techniques deliver unsatisfactory outcomes in such situations. This article aims to tackle these challenges by incorporating an inhomogeneity correction equation combining region and edge information and introducing a despeckling equation. The proposed model simultaneously performs image despeckling and segmentation, robustly handling intensity inhomogeneity and Rayleigh speckle noise. Furthermore, this framework eliminates the need for intricate re-initialization and mitigates numerical instability issues throughout the evolution process. Moreover, the well-posedness of the proposed coupled fractional deformable model (CFDM) is also established. We conducted extensive experiments on a wide variety of synthetic, natural, and ultrasound images that were affected by multiplicative noise and severe intensity inhomogeneity. Our thorough quantitative and qualitative analysis indicates that the CFDM performs competitively in comparison to six traditional models, seven state-of-the-art deformable models, and four fractional deformable models, demonstrating consistent improvements across various evaluation metrics. Furthermore, the CFDM is insensitive to the initial contour positioning and resilient against varying noise levels and severe intensity inhomogeneity.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"128 ","pages":"Article 110768"},"PeriodicalIF":4.9,"publicationDate":"2025-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145319526","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
Diversity-regularized multi-objective optimized Random Forest for fall detection using cost-sensitive and key point feature-based learning 基于代价敏感和关键点特征学习的多样性正则化多目标优化随机森林跌倒检测
IF 4.9 3区 计算机科学
Computers & Electrical Engineering Pub Date : 2025-10-13 DOI: 10.1016/j.compeleceng.2025.110684
Aayushi Bansal, Rewa Sharma, Mamta Kathuria
{"title":"Diversity-regularized multi-objective optimized Random Forest for fall detection using cost-sensitive and key point feature-based learning","authors":"Aayushi Bansal,&nbsp;Rewa Sharma,&nbsp;Mamta Kathuria","doi":"10.1016/j.compeleceng.2025.110684","DOIUrl":"10.1016/j.compeleceng.2025.110684","url":null,"abstract":"<div><div>In healthcare, accurate fall detection is essential for elder care, where timely intervention can prevent severe injuries and fatalities. This study proposes a Diversity-Regularized Multi-Objective Optimized Random Forest model, integrating spatial and temporal feature learning for improved fall detection. To optimize the classifier, Differential Evolution (DE) is employed for hyperparameter tuning, balancing Gini impurity, recall, and diversity regularization. This regularization component, seldom explored in traditional Random Forest architectures, enables enhanced generalization on imbalanced data by reducing overfitting. To address class imbalance, cost-sensitive learning is applied that emphasizes on fall detection by assigning higher penalties to misclassified falls, thereby enhancing recall. Sequential data augmentation further improves the model’s ability to generalize to diverse fall scenarios. Key spatial features are extracted from human poses using Detectron 2 for key point detection, allowing the system to capture intricate posture dynamics. These features, including joint angles and distances, uniquely support fall and non-fall differentiation. The model incorporates sliding window-based temporal sequence modeling, enabling enhanced motion pattern recognition. Additionally, confidence-based sample weighting dynamically adjusts learning based on keypoint reliability, improving robustness against uncertain detections. Post-fall analysis further refines predictions by smoothing false negatives in sequential frames. Experiments on CAUCA and UR Fall dataset demonstrate that the proposed model outperforms traditional methods, achieving 97.50% accuracy on UR Fall Dataset. Ablation studies confirm that integrating temporal sliding windows and post-fall detection significantly enhances model performance. The results establish proposed model as a scalable and robust solution for fall detection in healthcare and assistive environments.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"128 ","pages":"Article 110684"},"PeriodicalIF":4.9,"publicationDate":"2025-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145319527","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
An explainable lightweight convolutional neural network with global–local context-enhanced channel attention for grape leaf disease detection 一种具有全局-局部上下文增强通道关注的可解释轻量级卷积神经网络用于葡萄叶片病害检测
IF 4.9 3区 计算机科学
Computers & Electrical Engineering Pub Date : 2025-10-11 DOI: 10.1016/j.compeleceng.2025.110745
G. Loganathan, M. Palanivelan
{"title":"An explainable lightweight convolutional neural network with global–local context-enhanced channel attention for grape leaf disease detection","authors":"G. Loganathan,&nbsp;M. Palanivelan","doi":"10.1016/j.compeleceng.2025.110745","DOIUrl":"10.1016/j.compeleceng.2025.110745","url":null,"abstract":"<div><div>Grape cultivation plays a vital role worldwide by supporting the agricultural economy and supplying a variety of grape-based commodities. The susceptibility of grapes to various leaf diseases presents a substantial risk to both quality and yield. Conventional methods for identifying leaf diseases require expert knowledge, which restricts efficiency and scalability. To overcome these challenges, we introduce ELCAM-Net, an automated lightweight channel attention-based convolutional neural network with explainable artificial intelligence. The ELCAM-Net model classifies grape leaf images into various categories, including disease types such as black measles, black rot, and leaf blight, along with a healthy category. To ensure lightweight architecture and computational efficiency, SqueezeNet is selected as the backbone model. The proposed framework incorporates a global–local context-enhanced channel attention module into the SqueezeNet architecture. This integration refines the channel weights of the feature maps, enabling the backbone model to ignore irrelevant features and focus on the key characteristics of the disease. Consequently, it improves the classification accuracy. The effectiveness of the ELCAM-Net model has been evaluated using a publicly accessible grape leaf image dataset, achieving an accuracy of 99.41% and surpassing the backbone model by 2.64%. Furthermore, the decision-making process of ELCAM-Net is visualized using explainable artificial intelligence methods, such as gradient-weighted class activation mapping heatmaps. By integrating model predictions with corresponding heatmaps, the proposed method provides an explainable framework for grape leaf disease detection.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":"128 ","pages":"Article 110745"},"PeriodicalIF":4.9,"publicationDate":"2025-10-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145319529","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
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
相关产品
×
本文献相关产品
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信