检索结果-内蒙古大学图书馆

Vision-Based Driving Decision Making Using Multi-Action Deep Q Network

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS 2025年第5期26卷 5816-5831页

作者： Yuan, Sheng Li, Yaochen Zhao, Kai Zhu, Li Guo, Jiaxin Ma, Xinnan Xu, Yuncheng Xi An Jiao Tong Univ Sch Software Engn Xian 710049 Peoples R China

The performance improvement of perception algorithms and integrated validation of driving decision making remains challenging in the fields of computer vision and intelligent transportation systems. In this paper, we propose a novel vision-based framework for driving decision making, which is composed of three stages: object perception, lane line perception and driving decision making. For the object perception stage, an improved object perception model named CenterNet-ARA is developed, composing of a new adversarial training method, a receptive field enhancement module and an adaptive sample allocation equalization strategy to fuse multi-scale feature maps. For the lane line perception stage, a lane line perception method named Lite-MobileTR is proposed, which contains an improved Lite-MobileNetV3 encoder and an improved lite-transformer decoder. Moreover, a noise removal task is incorporated to alleviate the problem of slow convergence speed caused by Hungarian loss function. For the driving decision making stage, a new Multi-Action DQN is proposed utilizing a vehicle curriculum learning strategy and a curiosity exploration strategy to alleviate the problem of random exploration in the learning process. The proposed framework is evaluated on the Tusimple, CULane, TSD-max, and KITTI datasets. Finally, an integration verification is performed in Carla simulator to validate the driving decision making process. The experimental results well demonstrate the effectiveness of the proposed framework.

关键词： Decision making Training Roads computational modeling Data mining Semantics Safety Real-time systems Heuristic algorithms Convergence Object perception lane line perception multi-action DQN driving decision making

来源：评论

学校读者我要写书评

暂无评论

DroneGuard: An Explainable and Efficient Machine Learning Framework for Intrusion Detection in Drone Networks

引用

IEEE INTERNET OF THINGS JOURNAL 2025年第7期12卷 7708-7722页

作者： Ihekoronye, Vivian Ukamaka Ajakwe, Simeon Okechukwu Lee, Jae Min Kim, Dong-Seong Kumoh Natl Inst Technol Dept IT Convergence Engn Gumi 39177 South Korea Kumoh Natl Inst Technol ICT Convergence Res Ctr Gumi 39177 South Korea

Vulnerabilities in drone networks stem from the reliance on GPS and wireless communication technologies, combined with the lack of robust security mechanisms. This study proposes DroneGuard, a comprehensive cybersecurity framework leveraging supervised machine learning (ML) and explainable artificial intelligence (XAI) to detect intrusions and provide insights into the decision-making process of the security model. We explored various feature selection techniques to design a lightweight model suitable for the resource constraints of drones. Additionally, the synthetic minority oversampling technique (SMOTE) is employed to balance target class distribution and mitigate performance degradation, while randomized search cross-validation (RSCV) aids in selecting optimal hyperparameters for model training. Simulation experiments were conducted using a real-time GPS dataset for autonomous vehicles and a cybersecurity dataset containing variants of Denial of Service (DoS) attacks to evaluate the models' performance. Comparison with four ML models using essential evaluation metrics validated the robust performance of the decision tree model, which detected spoofed GPS signals and DoS attacks with high accuracy, low-computational complexity, and minimal false alarm rates. Furthermore, the Shapley additive explanation (SHAP) provides intuitive visual explanations of important features contributing to the detection and classification of both GPS spoofing and DoS attacks. Therefore, DroneGuard offers effective and interpretable security solutions for enhanced drone application and adoption.

关键词： Drones Global Positioning System Security computational modeling Accuracy Wireless sensor networks Machine learning Feature extraction Computer security Support vector machines Cybersecurity Denial of Service (DoS) attacks drone network explainable artificial intelligence (XAI) feature selection (FS) technique GPS spoofing attacks intrusion detection machine learning (ML) models randomized search cross validation UAVs

来源：评论

学校读者我要写书评

暂无评论

Temporally Deformable Convolution for Gait Recognition

引用

IEEE ACCESS 2025年 13卷 6475-6486页

作者： Kim, Juyoung Kim, Beomseong Lee, Heesung Korea Natl Univ Transportat Dept Railrd Elect & Informat Engn Uiwang Si 16106 Gyeonggi Do South Korea Gyeonggi Univ Sci & Technol Dept Artificial Intelligence Siheung Si 15073 Gyeonggi Do South Korea

Gait recognition is a biometric technology that identifies individuals based on the unique characteristics of their gait. With the advancement of deep learning based computer vision technology, gait recognition has significantly improved in performance and gained significant attention due to its non-invasive nature. This paper proposes a vision-based gait recognition model that enhances performance by applying deformable convolution in the temporal dimension. Gait is a periodic behavior, and understanding the body's changing patterns during walking is crucial for improving recognition accuracy. Important features for gait recognition can appear in both short and long moments of the walking process. Therefore, this study proposes a model that performs deformable convolution in the temporal dimension to effectively extract such features. This allows for a more flexible understanding of temporal patterns, eliminating the need for hand-crafted design of filters for extracting temporal features and enabling the model to learn important features for gait recognition that appear in both short and long time intervals in a unified manner. As a result, the proposed model achieved state-of-the-art performance on the CASIA-B dataset.

关键词： Feature extraction Convolution Gait recognition computational modeling Biometrics Deformable models Three-dimensional displays Image recognition Biological system modeling Shape Biometric identification pattern recognition neural networks computer vision machine learning for images

来源：评论

学校读者我要写书评

暂无评论

An Efficient Affine Arithmetic-Based Optimal Dispatch Method for Active Distribution Networks With Uncertainties of Electric Vehicles

引用

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY 2025年第2期16卷 1021-1036页

作者： Dai, Wei Li, Hongzhou Liu, Hui Goh, Hui Hwang Yuan, Xiansong Liu, Yuelin Chen, Baicheng Guangxi Univ Guangxi Key Lab Power Syst Optimizat & Energy Tech Nanning 530004 Peoples R China

Affine Arithmetic (AA) is an effective interval analysis method for addressing uncertainties in power systems. However, previous research on AA-based optimization problems has struggled to accurately capture the uncertainties associated with electric vehicles (EVs) and the cumulative impact of uncertainties on energy storage systems (ESSs). Moreover, the reformulated AA model presents a significant computational challenge due to the high number of variables and constraints. This study proposes an efficient AA-based economic dispatch (AAED) method for active distribution networks incorporating EVs and ESSs while accounting for uncertainties. Specifically, an EV charging load-interval (CLI) model is developed to effectively capture the randomness of plug-in/plug-out times and initial/target energy. A confidence level is defined to prevent excessive conservatism in the CLI model. An ESS model is also formulated within the AA domain to address the cumulative impact of persistent uncertainty, ensuring an accurate state of charge monitoring. To enhance the computational efficiency of the AAED model without sacrificing accuracy, a fast-solving strategy is introduced. This strategy involves eliminating many state variables and constraints and replacing them with derived analytical partial deviation formulations that map the relationship between state and decision variables. Simulation results confirm the effectiveness of the proposed model and method.

关键词： Economics Uncertainty Accuracy computational modeling Simulation Active distribution networks State of charge Reliability Load modeling Arithmetic Affine arithmetic economic dispatch active distribution network electric vehicle multiple uncertainties

来源：评论

学校读者我要写书评

暂无评论

A Parameter Ontology for Simulation Models of Deep Space Habitats

引用

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS 2025年第1期61卷 61-75页

作者： Gratius, Nicolas Berges, Mario Akinci, Burcu Carnegie Mellon Univ Pittsburgh PA 15213 USA

Deep space habitat operations will depend less on ground support than in low Earth orbit because of increasing communication delays. For the crew to be more autonomous, some of the resources dedicated to anomaly response mechanisms must be transferred from the ground to the habitat. Simulation models replicating the behavior of the system are examples of such resources as they can inform decisions to select mitigation strategies. Multiple models are typically developed for a single spacecraft as many domains of expertise and stakeholders are involved. Enabling simulation onboard a habitat will therefore require integration efforts to provide system-wide capabilities. Specifically, model parameters are model components that must be regularly calibrated to accurately represent the system state during operation. Depending on the type of simulation, these can differ greatly in their nature, thereby making difficult the tasks of creating, reading, updating, and deleting parameters. There is no standard specifying how to represent simulation parameters in a multimodel setting. Model ontologies have been shown to support integration, but their representation of model parameters is usually limited, i.e., the required type of parameter attributes and their hierarchical relationships are unknown. For example, quantifying the uncertainties in parameter values is essential in deep space operations but it is unclear how such information should be formally represented. In this article, we propose a parameter ontology extending existing standards with attributes enabling probabilistic parameter identification. This article is validated by comparing related work with the responses of the proposed ontology to competency questions and by evaluating its consistency, conciseness, completeness, and expandability.

关键词： Calibration Task analysis Unified modeling language Space technology computational modeling Buildings Ontologies Autonomous system calibration knowledge base ontology parameter simulation models space habitat vocabulary

来源：评论

学校读者我要写书评

暂无评论

STD2: Swin Transformer-Based Defect Detector for Surface Anomaly Detection

引用

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT 2025年 74卷

作者： Mia, Md Sohag Li, Chunbiao Nanjing Univ Informat Sci & Technol Sch Artificial Intelligence Nanjing 210044 Jiangsu Peoples R China

In the quest for manufacturing excellence, this article presents a pioneering approach to defect detection on steel surfaces. Complex and small defects on steel surfaces manifest in diverse forms, such as irregular shapes and varying sizes, making it challenging to devise a singular detection model capable of effectively detecting these multifarious defect types. Yet, another challenge lies in the close similarity between defects and nondefective components. Additionally, the fluctuating environmental factors exposed to steel structures introduce noise and artifacts into image data, compounding the complexity of precise defect detection. To address these issues, we initially employed the Swin Transformer as the core element of Faster R-CNN. Second, we ingeniously integrated a path aggregation feature pyramid network (PAFPN) into the architecture. This allows us to capture a comprehensive array of defect feature maps and enhance the network's capability to identify defects at various scales. After that, we swapped out the region-of-interest (RoI) pooling for the more sophisticated deformable pooling method to get more precise defect localization information. Finally, the standard intersection over union (IoU) loss is substituted with complete intersection over union (CIoU) to address issues of duplicate proposals and improve overall performance by taking into account the geometry of the bounding boxes in addition to their overlaps. Our proposed Swin Transformer Defect Detector (STD2) has been rigorously evaluated on diverse steel surface defect datasets, including NEU-DET (yielding an impressive mean average precision (mAP) of 81.05%), GC10-DET (with a notable mAP score of 72.38%), and printed circuit board (PCB) (achieving an outstanding mAP of 99.00%), which demonstrate its exceptional performance in surface anomaly detection. The code is available at https://***/Shuvo001/STD2.

关键词： Defect detection Transformers Feature extraction Steel Computer vision Computer architecture Inspection Deep learning computational modeling Adaptation models Complete intersection over union (CIoU) defect detection deformable pooling faster R-CNN path aggregation feature pyramid network (PAFPN) Swin Transformer

来源：评论

学校读者我要写书评

暂无评论

Securing One-Class Federated Learning Classifiers Against Trojan Attacks in Smart Grid

引用

IEEE INTERNET OF THINGS JOURNAL 2025年第4期12卷 4006-4021页

作者： Bondok, Atef H. Badr, Mahmoud Mahmoud, Mohamed Alsabaan, Maazen Fouda, Mostafa M. Abdullah, Mohamed Eastern Connecticut State Univ Dept Comp Sci Willimantic CT 06226 USA SUNY Polytech Inst Coll Engn Dept Network & Comp Secur Utica NY 13502 USA Benha Univ Fac Engn Shoubra Dept Elect Engn Cairo 11672 Egypt Tennessee Technol Univ Dept Elect & Comp Engn Cookeville TN 38505 USA King Saud Univ Coll Comp & Informat Sci Dept Comp Engn Riyadh 11451 Saudi Arabia Idaho State Univ Coll Sci & Engn Dept Elect & Comp Engn Pocatello ID 83209 USA Coll Sci & Engn Ctr Adv Energy Studies Dept Informat & Comp Technol Idaho Falls ID 83401 USA Hamad Bin Khalifa Univ Al Doha Qatar

Existing literature confirms the ability of machine learning to identify fraudulent smart grid power consumers who report false consumption readings to pay less electricity bills. Additionally, federated learning (FL) shows promise as a way to train the detection model without requiring data sharing, thereby safeguarding consumer privacy. However, malicious participants (i.e., clients) in FL training can launch adversarial attacks by training their local models with specially crafted low-consumption data to inject a Trojan into the global model. This Trojan can then be activated during the evaluation phase to evade the detection of false data. To the best of our knowledge, not enough research has been done on this topic in the context of unsupervised learning. The absence of labels in unsupervised learning exacerbates the effectiveness of Trojan attacks and renders it more challenging to design robust defense mechanisms. In this article, we first investigate the vulnerability of one-class classifiers to Trojan attacks. Then, we propose two defense approaches named layerwise close-to-median (LWCM) and Machine Unlearning to counter this attack. In LWCM, by choosing a FL client whose last layer model parameters are near to the median of all clients' last layer parameters to update the global model, we can identify and exclude malicious updates. The idea is that the last layer parameters of honest clients should be similar, whereas those from malicious clients are different. With the majority of clients being honest, the median values are closer to the parameters of these clients, facilitating the detection of malicious clients. In Machine Unlearning, we utilize gradient ascent-based techniques to adapt models by selectively removing attacker-related data points. This is possible because honest clients generate data resembling that of malicious clients and employ a dual-component loss function to maintain model proficiency in recognizing benign power consumption pattern

关键词： Electricity Data models Trojan horses Training Smart grids Detectors computational modeling Data privacy Unsupervised learning Supervised learning And smart grid (SG) automatic metering infrastructure federated learning (FL) one-class classifiers security Trojan attacks

来源：评论

学校读者我要写书评

暂无评论

Simulation of Crowd Evacuation in Asymmetrical Exit Layout Based on Improved Dynamic Parameters Model

引用

IEEE ACCESS 2025年 13卷 7512-7525页

作者： Alidmat, Omar Abu Owida, Hamza Yusof, Umi Kalsom Almaghthawi, Ahmed Altalidi, Askar Alkhawaldeh, Rami S. Abuowaida, Suhaila Alshdaifat, Nawaf Alshaqsi, Jamil Zarqa Univ Fac Informat Technol Dept Comp Sci Zarqa 13110 Jordan Al Ahliyya Amman Univ Fac Engn Dept Med Engn Amman 19328 Jordan Univ Sains Malaysia Sch Comp Sci George Town 11800 Penang Malaysia King Khalid Univ Coll Sci & Art Mahayil Dept Comp Sci Abha 62529 Saudi Arabia King Khalid Univ Coll Sci & Art Mahayil Dept Informat Syst Abha 62529 Saudi Arabia Univ Jordan Dept Comp Informat Syst Aqaba 77110 Jordan Appl Sci Private Univ Fac Informat Technol Amman 11937 Jordan Sultanate Qaboos Univ Informat Syst Dept Muscat Oman

Evacuation systems are crucial for minimizing casualties and property losses during emergencies. Understanding evacuee behavior in overcrowded situations is essential for developing effective evacuation strategies. However, evacuating large crowds from buildings with multiple exits is a challenging problem, especially when the exits are asymmetrical and the crowds are dense. This paper proposes a novel two-dimensional cellular automaton model for multi-exit evacuation, which simplifies evacuee decision-making in an asymmetrical exit layout within dense crowds. The model introduces the dynamic counting area technique, which dynamically adjusts the size and shape of the counting area around each exit based on the evacuee density level. This technique plays a crucial role in preventing the creation of overlapping counting areas between exits, which often leads to overestimated average evacuation time, unit evacuation time, and travel distance. Comparative analysis with previous dynamic parameter models (DPM) reveals notable results: the model achieved an average evacuation time of 201.20 time steps, a unit evacuation time of 0.50 time steps, and a travel distance of 28204 steps. These findings demonstrate the effectiveness of the improved model in addressing evacuation imbalances caused by asymmetrical exit layouts or evacuee distributions. Moreover, the study highlights the pivotal role of evacuee density around exits in determining exit choices during densely crowded emergency situations. The improved model can be applied to various scenarios and settings where multi-exit evacuation is required, such as stadiums, airports, or shopping malls.

关键词： Layout Automata Euclidean distance Decision making computational modeling Analytical models Crowdsourcing Asymmetrical exit layout cellular automaton model crowd simulation dynamic counting area technique multi-exit evacuation

来源：评论

学校读者我要写书评

暂无评论

Mosaic-Mixed Attention-Based Unexpected Traffic Scene Classification

引用

IEEE ACCESS 2025年 13卷 15712-15722页

作者： Lee, Sang-Hyun Lee, Soomok Yun, Ilsoo Ajou Univ Dept Data Network & AI Suwon 16499 South Korea Ajou Univ Dept Mobil Engn Suwon 16499 South Korea Ajou Univ Dept Transportat Syst Engn Suwon 16499 South Korea

With the rapid advancements in artificial intelligence and smart mobility technologies, traffic monitoring systems are evolving quickly. Among these systems, in-vehicle monitoring systems using Object Detection (OD) algorithms are gaining attention for identifying traffic participants in distress. However, current OD algorithms often underperform in complex or unexpected traffic scenarios, such as accidents. In this study, we propose a novel scene classification network that integrates OD with a cross-attention mechanism By leveraging spatial mosaic and mixed attention mechanisms, the network emphasizes spatial relationships and inter-channel correlations, significantly enhancing accuracy in identifying critical traffic events. The detailed evaluation demonstrates improved efficiency and accuracy, underscoring the potential of the system for future traffic incident classification.

关键词： Accidents Feature extraction Roads Classification algorithms Attention mechanisms Accuracy computational modeling Cameras Scene classification Convolutional neural networks Traffic monitoring scene classification attention network image mosaic processing

来源：评论

学校读者我要写书评

暂无评论

DDNet: A Robust, and Reliable Hybrid Machine Learning Model for Effective Detection of Depression Among University Students

引用

IEEE ACCESS 2025年 13卷 49334-49353页

作者： Mumenin, Nasirul Abu Yousuf, Mohammad Alassafi, Madini O. Monowar, Muhammad Mostafa Hamid, Md. Abdul Bangladesh Univ Business & Technol Dept Comp Sci & Engn Dhaka 1216 Bangladesh Jahangirnagar Univ Inst Informat Technol Dhaka 1342 Bangladesh King Abdulaziz Univ Fac Comp & Informat Technol Dept Informat Technol Jeddah 21589 Saudi Arabia

In recent years, the detection of depression among university students has become an increasingly critical issue. This paper presents a depression detection network (DDNet), a novel approach utilizing a three-stage stacked ensemble model to address this challenge. The proposed model incorporates two different Multilayer Perceptron (MLP) and a Stochastic Gradient Descent (SGD) as the 1st stage base classifier, an MLP, and CatBoost (CB) as 2nd stage base classifiers, with a Lasso Regressor (LASSO) serving as the meta-classifier. The hyperparameter of the used models has been optimized using random search. The optimal configuration has been determined through extensive experimentation with various machine learning (ML) models and settings to ensure high performance. Two different datasets (Dataset-1, Dataset-2) of depression detection among university students have been used to evaluate the model. The model has achieved 98.98% and 99.16% accuracy in Dataset-1 and Dataset-2, respectively. Paired t-test has been performed to ensure the statistical significance of the proposed model. To guarantee the model's transparency, SHapley Additive exPlanations (SHAP) were employed, providing interpretability of the predictive factors. Additionally, a variation of Monte Carlo Dropout (MCD) has been used to assess the uncertainty of the model's predictions, ensuring reliability. The results indicate that the proposed model is a promising tool for mental health professionals seeking effective, interpretable, and reliable solutions for early depression detection in educational settings.

关键词： Depression Mental health Predictive models Accuracy Reliability Uncertainty computational modeling Classification tree analysis Support vector machines Random forests Hybrid model depression detection explainable AI uncertainty analysis university students

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：