检索结果-内蒙古大学图书馆

13th IEEE Data Driven Control and Learning Systems Conference, DDCLS 2024

作者： Yang, Dan Peng, Xin Su, Cheng Tan, Dayu Chen, Chaoyang Zhong, Weimin Ministry of Education East China University of Science and Technology Key Laboratory of Smart Manufacturing in Energy Chemical Process Shanghai200237 China Ministry of Education Anhui University Key Laboratory of Intelligent Computing and Signal Processing Anhui Hefei230601 China School of Information and Electrical Engineering Hunan University of Science and Technology Xiangtan China

ISBN: (纸本)9798350361674

Transfer learning aims to leverage the knowledge of the source domain to help the learning of unlabeled target domain models. However, all conventional transfer learning methods assume that samples from source and target domains are always freely accessible. However, this assumption is rarely plausible in real-world due to data privacy. To combat this challenge, we introduce Structure-constrain Source Free Domain Adaptation with regularized Knowledge Distillation of source model (SSFDA-KD). Our method is based on the observations that a deterministic classification result can be obtained when the target data distribution aligns with the source data distribution, and a well-trained model not only yields a deterministic classification for a sample but also has clear category boundaries, which implies samples that are close together belong to the same category. From those, we first generate the source model with labeled source data, which is regarded as a well-trained model. Then, we identify target prototypes with lower self-entropy and assign the pseudo-labels to other samples according to the neighborhood structure. Finally, we refine the target model by minimizing prediction loss, constraining internal structure via triplet loss, and reducing knowledge distillation loss. Experiments of fault diagnosis based on WWTPs are conducted to validate its effectiveness. © 2024 IEEE.

关键词： Domain Knowledge

来源：评论

学校读者我要写书评

暂无评论

Modality Unified Attack for Omni-Modality Person Re-Identification

引用

IEEE Transactions on information Forensics and Security 2025年

作者： Bian, Yuan Liu, Min Yi, Yunqi Wang, Xueping Ma, Yunfeng Wang, Yaonan Hunan University National Engineering Research Center of Robot Visual Perception and Control Technology College of Electrical and Information Engineering Hunan Changsha China Hunan Normal University Hunan Provincial Key Laboratory of Intelligent Computing and Language Information Processing College of Information Science and Engineering Hunan Changsha China

Deep learning based person re-identification (re-id) models have been widely employed in surveillance systems. Recent studies have demonstrated that black-box single-modality and cross-modality re-id models are vulnerable to adversarial examples (AEs), leaving the robustness of multi-modality re-id models unexplored. Due to the lack of knowledge about the specific type of model deployed in the target black-box surveillance system, we aim to generate modality unified AEs for omni-modality (single-, cross- and multi-modality) re-id models. Specifically, we propose a novel Modality Unified Attack method to train modality-specific adversarial generators to generate AEs that effectively attack different omni-modality models. A multi-modality model is adopted as the surrogate model, wherein the features of each modality are perturbed by metric disruption loss before fusion. To collapse the common features of omnimodality models, Cross Modality Simulated Disruption approach is introduced to mimic the cross-modality feature embeddings by intentionally feeding images to non-corresponding modality-specific subnetworks of the surrogate model. Moreover, Multi Modality Collaborative Disruption strategy is devised to facilitate the attacker to comprehensively corrupt the informative content of person images by leveraging a multi modality feature collaborative metric disruption loss. Extensive experiments show that our MUA method can effectively attack the omni-modality re-id models, achieving 55.9%, 24.4%, 49.0% and 62.7% mean mAP Drop Rate, respectively. © 2005-2012 IEEE.

关键词： Modality unified attack Multi Modality Collaborative Disruption Omni-Modality re-id

来源：评论

学校读者我要写书评

暂无评论

Traffic Signal Light Recognition Based on Transformer 12th

Traffic Signal Light Recognition Based on Transformer

引用

12th International Conference on Computer Engineering and Networks, CENet 2022

作者： Ou, Yangze Sun, Yaqi Yu, Xiaozhong Yun, Liuqing College of Computer Science and Technology Hengyang Normal University Hunan Hengyang421008 China Hunan Provincial Key Laboratory of Intelligent Information Processing and Application Hunan Hengyang421008 China

ISBN: (纸本)9789811969003

With more and more developed technology, unmanned driving technology has gradually entered people’s vision. Many cars are now equipped with self-parking technology, which allows the vehicle to enter the garage through a pre-set path without manual operation. At the same time, the traffic light identification mentioned in this paper can also be used for driverless cars, and can also be combined with manual driving to reduce the probability of traffic accidents caused by running red lights. This design is based on Transformer and Resnet convolutional neural network, using feature extraction technology. First of all, the data set used in this algorithm is self-made COCO data set. Then, the corresponding training is carried out to obtain the corresponding model of traffic signal light recognition. Finally, the trained model is used to predict the relevant traffic signal light. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： Convolution

来源：评论

学校读者我要写书评

暂无评论

Deep Learning Image Steganalysis Method Fused with CBAM 12th

Deep Learning Image Steganalysis Method Fused with CBAM

引用

12th International Conference on Computer Engineering and Networks, CENet 2022

作者： Chen, Huasuo Jiao, Ge College of Computer Science and Technology Hengyang Normal University Hunan Hengyang421002 China Hunan Provincial Key Laboratory of Intelligent Information Processing and Application Hunan Hengyang421002 China

ISBN: (纸本)9789811969003

Steganography is a critical technical tool for preventing the disclosure of sensitive information. The detection performance of picture steganography algorithms based on deep learning has to be enhanced in tandem with the ongoing improvement of adaptive steganography algorithm performance. In this paper, a new model SRNet-CBAM based on SRNet fusion channel attention module and spatial attention module was presented to address the challenge of adaptive steganography analysis and the difficulty of model focused analysis for picture favorable regions. In three different embedding rates of WOW, S-UNIWARD, and HUGO algorithms, the experimental results reveal that the SRNet-CBAM model increases the accuracy of the SRNet model by 1.36% on average. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： Steganography

来源：评论

学校读者我要写书评

暂无评论

Traffic Sign Recognition System Based on YOLOv5 12th

Traffic Sign Recognition System Based on YOLOv5

引用

12th International Conference on Computer Engineering and Networks, CENet 2022

作者： Zhang, Lin Sun, Yaqi Chen, Wenli Liang, Xiaoman College of Computer Science and Technology Hengyang Normal University Hunan Hengyang421008 China Hunan Provincial Key Laboratory of Intelligent Information Processing and Application Hunan Hengyang421008 China

ISBN: (纸本)9789811969003

Traffic sign recognition is an important part of intelligent driving and an important content of intelligent transportation system. However, the current traffic sign detection algorithm has some problems, such as high network complexity, low recognition efficiency, poor recognition accuracy, and difficult edge deployment. Aiming at the above problems, this paper proposes to use YOLOv5 single-stage target detection algorithm to identify and detect traffic signs. The input terminal of YOLOv5 enhances and improves adaptive image scaling through Mosaic data, thus reducing information redundancy, reducing computation and improving target detection speed. The CSP structure of Backbone solves the problem of large amount of calculation in reasoning design from the Angle of network structure design. By adopting FPN + CPN structure, the fusion ability of network features is strengthened. Through hardware and software integration design, YOLOv5s model can be deployed to edge equipment, thus forming a set of traffic sign target detection system with fast inference speed, strong anti-interference ability and high identification accuracy. The results show that the overall identification accuracy of traffic signs is up to 98% by adjusting the parameters and improving the algorithm. Through testing 676 images, the recognition accuracy is 96.15% and the average detection speed is 19 frames/s, which meets the requirements of real-time traffic sign detection. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： Traffic signs

来源：评论

学校读者我要写书评

暂无评论

Adaptive Image Steganographic Analysis System Based on Deep Convolutional Neural Network 11th

Adaptive Image Steganographic Analysis System Based on Deep ...

引用

11th International Conference on Computer Engineering and Networks, CENet2021

作者： Jiao, Ge College of Computer Science and Technology Hengyang Normal University HengyangHunan421002 China Hunan Provincial Key Laboratory of Intelligent Information Processing and Application HengyangHunan421002 China

ISBN: (纸本)9789811665530

Adaptive steganography emplaces the message into the hard-to-detect noise area or the complex texture area of the image, so the steganography analysis method based on artificial design features needs to design a very complex feature extraction algorithm to detect the steganography image. In view of the advantages of deep learning steganography in automatic extraction of image features and high detection accuracy, an image steganography analysis algorithm is designed by using deep convolutional neural network, and an image steganography analysis system based on deep learning is developed. The system realizes the functions of neural network structure analysis, image steganography and verification, steganography analysis and so on. The system simplifies the operation process of algorithm comparison, reduces the complexity of algorithm performance evaluation, and verifies the feasibility of the proposed algorithm. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： Feature extraction

来源：评论

学校读者我要写书评

暂无评论

Variational Graph Autoencoders Method Based on Attentional Mechanisms for Overlapping Community Detection 12th

Variational Graph Autoencoders Method Based on Attentional M...

引用

12th International Conference on Computer Engineering and Networks, CENet 2022

作者： Wen, Kunhui Lin, Mugang Zhu, Xuanying Zhao, Huihuang College of Computer Science and Technology Hengyang Normal University Hunan Hengyang421008 China Hunan Provincial Key Laboratory of Intelligent Information Processing and Application Hunan Hengyang421008 China

ISBN: (纸本)9789811969003

Detecting overlapping communities of an attribute network is an important and difficult issue in network science. Traditional methods for overlapping community detection generally considered only its topological structure and neglected node attribute so that the detected communities could be inaccurate. Graph neural networks (GNN) have recently gained attention because of their learning representation ability for node attributes and the topology of a graph and demonstrating superior performance in natural language processing, data mining and pattern recognition, etc. However, GNN methods are still less for overlapping community detection. In this paper, we present an overlapping community detection method based on GNN for attribute networks. It applies a variational graph autoencoder based on attentional mechanisms to learn the representation of nodes in the graph and enhances the representation learning capability of GNN through semi-supervised learning. Various real-world networks of publicly available datasets are used to verify the effectiveness of the proposed approach. The experimental results show that the proposed method outperforms many other methods. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： Graph neural networks

来源：评论

学校读者我要写书评

暂无评论

Attribute Selection Method Based on Artificial Bee Colony Algorithm and Neighborhood Discrimination Matrix Optimization 16th

Attribute Selection Method Based on Artificial Bee Colony Al...

引用

16th International Conference on Bio-Inspired Computing: Theories and Applications, BIC-TA 2021

作者： Ji, Yuxuan Ye, Jun Yang, Zhenyu Ao, Jiaxin Wang, Lei Nanchang Institute of Technology Jiangxi Nanchang330000 China Key Laboratory of Water Information Cooperative Perception and Intelligent Processing of Jiangxi Province Jiangxi Nanchang330000 China

ISBN: (纸本)9789811912559

At present, the existing attribute reduction algorithm combining artificial bee colony and neighborhood rough set basically uses the attribute dependence and the number of attribute subsets as parameters to construct the fitness function, while ignoring the role of heuristic information. As a result, the number of bee colony iterations increases, and the convergence speed is slow. Aiming at this kind of problem, an improved method is proposed. First, a discernibility matrix under the neighborhood rough set is defined;secondly, an attribute importance measurement method of the discernibility matrix under the neighborhood decision system is proposed;The attribute importance of the domain discrimination matrix constructs a new fitness function for the heuristic factor;finally, an attribute selection algorithm for artificial bee colony algorithm and neighborhood discrimination matrix importance optimization is designed. Compared with the original algorithm, the new method reduces the number of generations, accelerates the convergence speed, and retains the minimum attribute reduction collected during each iteration, and multiple minimum attribute reductions can be obtained. The experimental results on the UCI data set prove the feasibility and effectiveness of the algorithm. © 2022, Springer Nature Singapore Pte Ltd.

关键词： Matrix algebra

来源：评论

学校读者我要写书评

暂无评论

Positive-Guided Knowledge Distillation for Document-Level Relation Extraction with Noisy Labeled Data 12th

Positive-Guided Knowledge Distillation for Document-Level R...

引用

12th National CCF Conference on Natural Language processing and Chinese Computing, NLPCC 2023

作者： Zeng, Daojian Zhu, Jianling Jiang, Lincheng Dai, Jianhua Hunan Provincial Key Laboratory of Intelligent Computing and Language Information Processing Hunan Normal University Changsha China Institute of AI and Targeted International Communication Hunan Normal University Changsha China College of Advanced Interdisciplinary Studies National University of Defense Technology Changsha China

ISBN: (纸本)9783031446924

Since one entity may have multiple mentions and relations between entities may stretch across multiple sentences in a document, the annotation of document-level relation extraction datasets becomes a challenging task. Many studies have identified that these datasets contain a large number of noisy labels, hindering performance improvement for the document-level relation extraction task. The previous and most straightforward method is denoising noisy labeled data from a data annotation perspective. However, this time-consuming approach is not suitable for large-scale datasets. In this paper, we propose a novel Positive-Guided Knowledge Distillation (PGKD) model to address the noisy labeled data problem for document-level relation extraction. We design a new teacher-student architecture. The teacher model trained with only positive samples can partially supervise the student model. The positive-guided knowledge distillation algorithm transfers the clean positive-class patterns from the teacher model to the student model. In this way, the student model trained with all samples can efficiently prevent the interference of false negative samples. Extensive experiments on Mix-DocRED demonstrate that PGKD achieves state-of-the-art effectiveness for document-level relation extraction with noisy labeled data. Moreover, PGKD also surpasses other baselines even on the well-annotated Re-DocRED. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Distillation

来源：评论

学校读者我要写书评

暂无评论

Learning to Learn Transferable Generative Attack for Person Re-Identification

arXiv

引用

arXiv 2024年

作者： Bian, Yuan Liu, Min Wang, Xueping Ma, Yunfeng Wang, Yaonan The College of Electrical and Information Engineering Hunan University National Engineering Research Center of Robot Visual Perception and Control Technology Hunan Changsha China The College of Information Science and Engineering Hunan Normal University Hunan Provincial Key Laboratory of Intelligent Computing and Language Information Processing Hunan Changsha China

Deep learning-based person re-identification (re-id) models are widely employed in surveillance systems and inevitably inherit the vulnerability of deep networks to adversarial attacks. Existing attacks merely consider cross-dataset and cross-model transferability, ignoring the cross-test capability to perturb models trained in different domains. To powerfully examine the robustness of real-world re-id models, the Meta Transferable Generative Attack (MTGA) method is proposed, which adopts meta-learning optimization to promote the generative attacker producing highly transferable adversarial examples by learning comprehensively simulated transfer-based cross-model&dataset&test black-box meta attack tasks. Specifically, cross-model&dataset black-box attack tasks are first mimicked by selecting different re-id models and datasets for meta-train and meta-test attack processes. As different models may focus on different feature regions, the Perturbation Random Erasing module is further devised to prevent the attacker from learning to only corrupt model-specific features. To boost the attacker learning to possess cross-test transferability, the Normalization Mix strategy is introduced to imitate diverse feature embedding spaces by mixing multi-domain statistics of target models. Extensive experiments show the superiority of MTGA, especially in cross-model&dataset and cross-model&dataset&test attacks, our MTGA outperforms the SOTA methods by 21.5% and 11.3% on mean mAP drop rate, respectively. The code of MTGA will be released after the paper is accepted. Copyright © 2024, The Authors. All rights reserved.

关键词： Generative adversarial networks

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：