检索结果-内蒙古大学图书馆

On the Performance of Active RIS-Assisted Mixed RF-THz Relaying Systems

IEEE Internet of Things Journal 2025年第10期12卷 14938-14951页

作者： Yin, Yiyang Yang, Liang Li, Xingwang Liu, Hongwu Guo, Kefeng Li, Yingsong Hunan University College of Computer Science and Electronic Engineering Changsha410082 China Henan Polytechnic University School of Physics and Electronic Information Engineering Jiaozuo454000 China Shandong Jiaotong University School of Information Science and Electrical Engineering Jinan250357 China Space Engineering University School of Space Information Beijing101407 China Anhui University Key Laboratory of Intelligent Computing and Signal Processing Ministry of Education Hefei230601 China

We investigate the performance of an active reconfigurable intelligent surface (RIS)-assisted mixed radio frequency (RF)-terahertz (THz) relaying system, where the RF signal reaches the relay through the active RIS and is then transmitted to the user via the THz channel. Under this scenario, we analyze the system performance with the relay employing amplify-and-forward (AF) and decode-and-forward (DF) protocols. More specifically, we derive the exact expressions for the cumulative distribution function (CDF) of the end-to-end signal-to-noise ratio (SNR) for both relaying protocols. Based on this, we obtain the exact expressions for the outage probability, average bit error rate (ABER), and average channel capacity (ACC). Furthermore, to gain deeper insights, we derive the asymptotic expressions at high SNRs and obtain diversity order (DO) of the system. Moreover, we extend the analysis to the variable gain relaying scheme. The findings reveal that the DOs for both relaying protocols are determined by the THz channel parameters, and the DO under the AF relaying protocol is twice that of the DF relaying protocol. Finally, we validate that active RIS (A-RIS) can more effectively assist the performance of the mixed RF-THz relaying system compared to passive RIS. © 2014 IEEE.

关键词： Distribution functions

来源：评论

学校读者我要写书评

暂无评论

Toward High-Accuracy and Low-Latency Group Vehicle Trajectory Prediction With Linear UNet-Enhanced Fully Connected Spatial–Temporal Graph Neural Network

引用

IEEE Internet of Things Journal 2025年第9期12卷 11797-11811页

作者： Zhang, Zirui Huang, Xumin Yu, Rong Wu, Maoqiang Kang, Jiawen Xie, Shengli Guangdong University of Technology School of Automation Key Laboratory of Intelligent Information Processing and System Integration of IoT Ministry of Education Guangdong Provincial Key Laboratory of Intelligent Systems and Optimization Integration Guangzhou510006 China South China Normal University School of Electronic Science and Engineering Foshan528225 China Guangdong University of Technology School of Automation 111 Center for Intelligent Batch Manufacturing Based on IoT Technology Guangdong-Hong Kong-Macao Joint Laboratory for Smart Discrete Manufacturing Guangzhou510006 China

Group vehicle trajectory prediction (GVTP) is important for analyzing the traffic states and optimizing the traffic management. However, existing studies have performance bottlenecks in the prediction accuracy and inference latency. To tackle the problems, we propose a linear UNet-enhanced fully connected spatial–temporal GNN (LUFC-STGNN) for GVTP. First, a spatial graph is constructed by integrating prior-based and data-driven methods to capture both explicit and implicit spatial interactions between the vehicles. After that, a comprehensive temporal graph is created to capture the varying strengths of temporal interactions between all vehicles throughout historical timestamps. Furthermore, a fully connected spatial–temporal graph combining the spatial and temporal graphs is introduced to extract the effective spatial–temporal interaction features of the vehicles through the graph convolution operation. Finally, a linear UNet-based temporal dependency encoder (LU-TDE) is designed to further enhance the model’s ability of capturing the potential temporal patterns in the vehicle interactions. The encoder with linear complexity explores the multiscale temporal dependencies from the spatial–temporal interaction features but also reducing the inference latency. Experiments results based on real-world datasets show that compared to state-of-the-art models, our model reduces the average root mean square error over the 5-s prediction horizon by 31% and 10% on the NGSIM and HighD datasets, while reducing the inference latency by at least 1.26 times. © 2014 IEEE.

关键词： Prediction models

来源：评论

学校读者我要写书评

暂无评论

Research on Double Attention Mechanism High-resolution Network for Human Pose Estimation

引用

Engineering Letters 2025年第2期33卷 338-347页

作者： Li, Haisheng Yuan, Rongrong Chen, Jingyin Li, Qiuyi Hu, Cong Guangxi Key Laboratory of Brain-inspired Computing and Intelligent Chips Guangxi Normal University Guangxi Guilin541004 China School of Electronic and Information Engineering Guangxi Normal University Guangxi Guilin541004 China Guangxi Key Laboratory of Automatic Detecting Technology and Instruments Guilin University of Electronic Technology Guangxi Guilin541004 China

Recently, high-resolution networks combining different resolution features have achieved good results in human pose estimation tasks. However, there are still some problems for high-resolution networks, such as insufficient use of important feature information and lost information in the sampling process. This paper proposes a double attention mechanism high-resolution network based on human pose estimation for these problems. The double attention mechanism consists of coordinate and downsampling attention modules. The coordinate attention module can enhance the feature representation of the network by extracting more useful node information. The downsampling attention module reuses the initial feature information to predict the occluded parts better and accurately estimate the corresponding point positions of human joints. Experimental results on MPII and COCO datasets show that the proposed network model effectively improves the accuracy of keypoint estimation in human pose estimation tasks. Especially, our network model achieves 78.1% average precision (AP) on the COCO dataset. © 2025, International Association of Engineers. All rights reserved.

关键词： Importance sampling

来源：评论

学校读者我要写书评

暂无评论

Knowledge Graph-Enhanced Recommendation with Box Embeddings 23rd

Knowledge Graph-Enhanced Recommendation with Box Embeddings

引用

23rd China National Conference on Computational Linguistics, CCL 2024

作者： Liang, Qiuyu Wang, Weihua Lv, Lei Bao, Feilong College of Computer Science Inner Mongolia University Hohhot China National and Local Joint Engineering Research Center of Intelligent Information Processing Technology for Mongolian Hohhot China Inner Mongolia Key Laboratory of Mongolian Information Processing Technology Hohhot China

ISBN: (纸本)9789819783663

Knowledge graphs are used to alleviate the problems of data sparsity and cold starts in recommendation systems. However, most existing approaches ignore the hierarchical structure of the knowledge graph. In this paper, we propose a box embedding method for knowledge graph-enhanced recommendation system. Specifically, the box embedding represents not only the interaction between the user and the item, but also the head entity, the tail entity and the relation between them in the knowledge graph. Then the interaction between the item and the corresponding entity is calculated by the multi-task attention unit. Experimental results show that our method provides a large improvement over previous models in terms of Area Under Curve (AUC) and accuracy in publicly available recommendation datasets with three different domains. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Knowledge graph

来源：评论

学校读者我要写书评

暂无评论

Multiple Weather Images Restoration Using the Task Transformer and Adaptive Mixup Strategy 41st

Multiple Weather Images Restoration Using the Task Transfor...

引用

41st Computer Graphics International Conference, CGI 2024

作者： Wen, Yang Lai, Anyu Qian, Bo Wang, Hao Shi, Wuzhen Cao, Wenming Guangdong Provincial Key Laboratory of Intelligent Information Processing School of Electronic and Information Engineering Shenzhen University Shenzhen China The Department of Comupter Science and Engineering Shanghai Jiao Tong University Shanghai China

ISBN: (纸本)9783031818059

The current state-of-the-art in severe weather removal predominantly focuses on single-task applications, such as rain removal, haze removal, and snow removal. However, real-world weather conditions often consist of a mixture of several weather types, and the degree of weather mixing in autonomous driving scenarios remains unknown. In the presence of complex and diverse weather conditions, a single weather removal model often encounters challenges in producing clear images from severe weather images. Therefore, there is a need for the development of multi-task severe weather removal models that can effectively handle mixed weather conditions and improve image quality in autonomous driving scenarios. In this paper, we introduce a novel multi-task severe weather removal model that can effectively handle complex weather conditions in an adaptive manner. Our model incorporates a weather task sequence generator, enabling the self-attention mechanism to selectively focus on features specific to different weather types. To tackle the challenge of repairing large areas of weather degradation, we introduce Fast Fourier Convolution (FFC) to increase the receptive field. Additionally, we propose an adaptive upsampling technique that effectively processes both the weather task information and underlying image features by selectively retaining relevant information. Our proposed model has achieved state-of-the-art performance on the publicly available dataset. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Restoration

来源：评论

学校读者我要写书评

暂无评论

Complementary Learning Subnetworks Towards Parameter-Efficient Class-Incremental Learning

引用

IEEE Transactions on Knowledge and Data Engineering 2025年第6期37卷 3240-3252页

作者： Li, Depeng Zeng, Zhigang Dai, Wei Suganthan, Ponnuthurai Nagaratnam Huazhong University of Science and Technology School of Artificial Intelligence and Automation Wuhan430074 China Key Laboratory of Image Processing and Intelligent Control of Education Ministry of China Wuhan430074 China China University of Mining and Technology School of Information and Control Engineering Xuzhou221116 China Qatar University Kindi Center for Computing Research College of Engineering Doha2713 Qatar

In the scenario of class-incremental learning (CIL), deep neural networks have to adapt their model parameters to non-stationary data distributions, e.g., the emergence of new classes over time. To mitigate the catastrophic forgetting phenomenon, typical CIL methods either cumulatively store exemplars of old classes for retraining model parameters from scratch or progressively expand model size as new classes arrive, which, however, compromises their practical value due to little attention paid to parameter efficiency. In this paper, we contribute a novel solution, effective control of the parameters of a well-trained model, by the synergy between two complementary learning subnetworks. Specifically, we integrate one plastic feature extractor and one analytical feed-forward classifier into a unified framework amenable to streaming data. In each CIL session, it achieves non-overwritten parameter updates in a cost-effective manner, neither revisiting old task data nor extending previously learned networks;Instead, it accommodates new tasks by attaching a tiny set of declarative parameters to its backbone, in which only one matrix per task or one vector per class is kept for knowledge retention. Experimental results on a variety of task sequences demonstrate that our method achieves competitive results against state-of-the-art CIL approaches, especially in accuracy gain, knowledge transfer, training efficiency, and task-order robustness. Furthermore, a graceful forgetting implementation on previously learned trivial tasks is empirically investigated to make its non-growing backbone (i.e., a model with limited network capacity) suffice to train on more incoming tasks. © 2025 IEEE.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Adaptive Local Neighborhood Window Laplacian Eigenmaps 24

Adaptive Local Neighborhood Window Laplacian Eigenmaps

引用

13th International Conference on computing and Pattern Recognition, ICCPR 2024

作者： Xu, Jiayi Huang, Qirui Zhao, Yang Pei, Jihong Yang, Xuan College of Electronics and Information Engineering Guangdong Key Laboratory of Intelligent Information Processing Shenzhen University Shenzhen Guangdong China College of Computer Science and Software Engineering Shenzhen University Shenzhen Guangdong China

ISBN: (纸本)9798400717482

Manifold learning method is a dimensionality reduction method that treats the data in non-Euclidean space as a Euclidean space in a local scope. However, most existing manifold learning methods cannot obtain the true neighborbood relationship of each sample point well. Therefore, we propose a manifold learning method based on adaptive local neighborhood. This method can adaptively select a neighborhood window. It can be approximated as a local Euclidean space based on the sample characteristics. To implement this method, a specific manifold learning method, named Laplacian Eigenmap of Adaptive Local Neighborhood Window (ALNWLE) is proposed. ALNWLE mainly contains two modules, namely the manifold structure learning module (MSLM) and the manifold structure pattern recognition module (MSPR). MSLM constructs an adaptive local neighborhood adjacency matrix to constrain the learning of the manifold structure. MSPR uses the interpolation method to recognize new sample patterns. The adaptive local neighborhood method proposed in this paper is a general method which is applicable to any manifold learning method. © 2024 Copyright held by the owner/author(s).

关键词： Dimensionality reduction

来源：评论

学校读者我要写书评

暂无评论

AUCPro: AUC-Oriented Provable Robustness Learning

引用

IEEE Transactions on Pattern Analysis and Machine Intelligence 2025年第6期47卷 4579-4596页

作者： Bao, Shilong Xu, Qianqian Yang, Zhiyong He, Yuan Cao, Xiaochun Huang, Qingming Institute of Information Engineering Beijing100093 China University of Chinese Academy of Sciences School of Cyber Security Beijing100049 China Security Department Alibaba Group Hangzhou311121 China Sun Yat-sen University School of Cyber Science and Technology Shenzhen Campus Shenzhen518107 China Beijing101408 China Chinese Academy of Sciences Key Laboratory of Intelligent Information Processing Institute of Computing Technology Beijing100190 China Peng Cheng Laboratory Shenzhen518055 China

The current studies of provable robustness for deep neural networks (DNNs) usually assume that the class distribution is overall balanced. However, in real-world applications especially for safety-sensitive systems, the class distribution often exhibits a long-tailed property. It is well-known that the Area Under the ROC Curve (AUC) is a more proper metric for long-tailed learning problems. Motivated by this fact, an AUC-oriented provable robustness learning framework (named AUCPro) is first proposed in this paper. The key is to construct a proxy model smoothed by the isotropic Gaussian noise and then consider optimizing the proxy model from the AUC-oriented learning point of view. Theoretically, we provide a certified safety region for AUCPro within which the model would be free from the 2 adversarial attacks. Most importantly, we propose a novel standard to theoretically study the robustness generalization toward unseen data for provable robustness learning approaches. To the best of our knowledge, such a problem remains barely considered in the machine learning community. To be specific, under a general principle for performance-robustness trade-off, we prove that the generalization ability of the resulting model could be equivalently expressed as the expected adversarial risk of AUC under 2 perturbation. On top of this, we present two practical settings to explore the excess risk formed by the difference between the empirical risk of AUCPro and the derived generalization performance. Finally, comprehensive experiments speak to the efficacy of our proposed algorithm. © 1979-2012 IEEE.

关键词： Adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

Amplified emission and lasing enabled by high-Q quasi-bound states in asymmetric dielectric metasurfaces via eigenfield perturbation

引用

Optics Express 2025年第10期33卷 21551-21565页

作者： Ke Xu Yuhang Fei Yingying Sun Jian Feng Xuesong Deng Ming Fang Zhixiang Huang The Information Materials and Intelligent Sensing Laboratory of Anhui Province Anhui University Hefei Anhui 230039 China The Key Laboratory of Intelligent Computing and Signal Processing Ministry of Education Anhui University Hefei 230039 China

Symmetry-protected bound states in the continuum represent a distinct class of nonradiative modes in continuous media. By introducing asymmetric parameters into a symmetric metasurface, the inherent symmetry can be broken, transforming BICs—originally decoupled from external radiation channels—into high-Q quasi-BICs with controlled radiation capabilities. However, manufacturing constraints pose a significant challenge in fabricating nanostructures with ultra-small asymmetric parameters, thereby limiting the achievable Q-factor. In this work, we propose an asymmetric perforated metasurface structure that supports BICs. When the perforations are shifted along the x or y axis to a specific position, the linewidth vanishes, and simulations confirm the emergence of a BIC in the system, with a quality factor reaching up to 10 9 . Furthermore, we demonstrate that selective eigenmode perturbation can effectively transform BICs into practical quasi-BICs. Notably, for the same structure, perturbing the weak-field region yields higher Q-factors, offering a flexible approach to optimizing performance. This approach enables precise Q-factor control and enhanced radiation properties. Furthermore, we design a laser cavity structure based on these quasi-BICs, achieving amplified emission and lasing with excellent agreement between simulations and theoretical predictions. These findings expand the BIC library and offer new methods for Q-factor manipulation, paving the way for high-performance optoelectronic devices such as nanolasers, biosensors, and nonlinear harmonic generators.

关键词： Destructive interference Light matter interactions Material properties Nanostructures Optical amplifiers Photonic crystal cavities

来源：评论

学校读者我要写书评

暂无评论

Ingredient-Guided RGB-D Fusion Network for Nutritional Assessment

IEEE Transactions on AgriFood Electronics

引用

IEEE Transactions on AgriFood Electronics 2025年第1期3卷 156-166页

作者： Feng, Zhihui Xiong, Hao Min, Weiqing Hou, Sujuan Duan, Huichuan Liu, Zhonghua Jiang, Shuqiang Shandong Normal University School of Information Science and Enginering Jinan250358 China Macquarie University Centre for Health Informatics Macquarie ParkNSW2109 Australia Chinese Academy of Sciences Key Laboratory of Intelligent Information Processing Institute of Computing Technology Beijing100190 China University of Chinese Academy of Sciences Beijing100049 China Zhejiang Ocean University School of Information Engineering Zhoushan316022 China

The nutritional value of agricultural products is an important indicator for evaluating their quality, which directly affects people's dietary choices and overall well-being. Nutritional assessment studies provide a scientific basis for the production, processing, and marketing of food by analyzing the nutrients they contain. Traditional methods often struggle with suboptimal accuracy and can be time consuming, as well as a shortage of professionals. The progress in artificial intelligence has revolutionized dietary health by offering more accessible methods for food nutritional assessment using vision-based approaches. However, existing vision-based methods using RGB images often face challenges due to varying lighting conditions, impacting the accuracy of nutritional assessment. An alternative is the RGB-D fusion method, which combines RGB images and depth maps. Yet, these methods typically rely on simple fusion techniques that do not ensure precise assessment. Additionally, current vision-based methods struggle to detect small components like oils and sugars on food surfaces, crucial for determining ingredient information and ensuring accurate nutritional assessment. In this pursuit, we propose a novel ingredient-guided RGB-D fusion network that integrates RGB images with depth maps and enables more reliable nutritional assessment guided by ingredient information. Specifically, the multifrequency bimodality fusion module is designed to leverage the correlation between the RGB image and the depth map within the frequency domain. Furthermore, the progressive-fusion module and ingredient-guided module leverage ingredient information to explore the potential correlation between ingredients and nutrients, thereby enhancing the guidance for nutritional assessment learning. We evaluate our approach on a variety of ablation settings on Nutrition5k, where it consistently outperforms state-of-the-art methods. © 2023 IEEE.

关键词： Medicinal chemistry

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：