检索结果-内蒙古大学图书馆

arXiv 2024年

作者： Zeng, Ling-An Zheng, Wei-Shi School of Artificial Intelligence Sun Yat-sen University Guangdong Zhuhai519082 China School of Computer Science and Engineering Sun Yat-sen University Guangdong Guangzhou510275 China Guangdong Key Laboratory of Information Security Technology Sun Yat-sen University Guangdong Guangzhou510275 China Key Laboratory of Machine Intelligence and Advanced Computing Sun Yat-sen University Ministry of Education Guangdong Guangzhou510275 China

Action quality assessment (AQA) is to assess how well an action is performed. Previous works perform modelling by only the use of visual information, ignoring audio information. We argue that although AQA is highly dependent on visual information, the audio is useful complementary information for improving the score regression accuracy, especially for sports with background music, such as figure skating and rhythmic gymnastics. To leverage multimodal information for AQA, i.e., RGB, optical flow and audio information, we propose a Progressive Adaptive Multimodal Fusion Network (PAMFN) that separately models modality-specific information and mixed-modality information. Our model consists of with three modality-specific branches that independently explore modality-specific information and a mixed-modality branch that progressively aggregates the modality-specific information from the modality-specific branches. To build the bridge between modality-specific branches and the mixed-modality branch, three novel modules are proposed. First, a Modality-specific Feature Decoder module is designed to selectively transfer modality-specific information to the mixed-modality branch. Second, when exploring the interaction between modality-specific information, we argue that using an invariant multimodal fusion policy may lead to suboptimal results, so as to take the potential diversity in different parts of an action into consideration. Therefore, an Adaptive Fusion Module is proposed to learn adaptive multimodal fusion policies in different parts of an action. This module consists of several FusionNets for exploring different multimodal fusion strategies and a PolicyNet for deciding which FusionNets are enabled. Third, a module called Cross-modal Feature Decoder is designed to transfer cross-modal features generated by Adaptive Fusion Module to the mixed-modality branch. Our extensive experiments validate the efficacy of the proposed method, and our method achieves state-of-the-a

关键词： Damage detection

来源：评论

学校读者我要写书评

暂无评论

Benchmarking Deep Models for Salient Object Detection

arXiv

引用

arXiv 2022年

作者： Zhou, Huajun Lin, Yang Yang, Lingxiao Lai, Jianhuang Xie, Xiaohua School of Computer Science and Engineering Sun Yat-sen University Guangzhou510006 China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou510006 China

In recent years, deep network-based methods have continuously refreshed state-of-the-art performance on Salient Object Detection (SOD) task. However, the performance discrepancy caused by different implementation details may conceal the real progress in this task. Making an impartial comparison is required for future researches. To meet this need, we construct a general SALient Object Detection (SALOD) benchmark to conduct a comprehensive comparison among several representative SOD methods. Specifically, we re-implement 14 representative SOD methods by using consistent settings for training. Moreover, two additional protocols are set up in our benchmark to investigate the robustness of existing methods in some limited conditions. In the first protocol, we enlarge the difference between objectness distributions of train and test sets to evaluate the robustness of these SOD methods. In the second protocol, we build multiple train subsets with different scales to validate whether these methods can extract discriminative features from only a few samples. In the above experiments, we find that existing loss functions usually specialized in some metrics but reported inferior results on the others. Therefore, we propose a novel Edge-Aware (EA) loss that promotes deep networks to learn more discriminative features by integrating both pixel- and image-level supervision signals. Experiments prove that our EA loss reports more robust performances compared to existing losses. Code is available at https://***/moothes/SALOD. Copyright © 2022, The Authors. All rights reserved.

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

EDGE ENHANCEMENT NETWORK FOR WEAKLY SUPERVISED SEMANTIC SEGMENTATION

EDGE ENHANCEMENT NETWORK FOR WEAKLY SUPERVISED SEMANTIC SEGM...

引用

2021 IEEE International Conference on Multimedia and Expo, ICME 2021

作者： Yu, Mei Wei, Junbin Wang, Chenhan Jiang, Han Yu, Jian Zhang, Ruixuan Li, Xuewei Yu, Ruiguo College of Intelligence and Computing Tianjin University China Tianjin Key Laboratory of Cognitive Computing and Application China Tianjin Key Laboratory of Advanced Networking China Tianjin International Engineering Institute Tianjin University China OpenBayes Inc Beijing China

ISBN: (纸本)9781665438643

It is difficult and expensive to obtain labels for image semantic segmentation tasks. This has led to more and more researches focusing on weakly supervised semantic segmentation (WSSS) with image-level labels and the class activation maps (CAM) is often used to locate objects. The performance of WSSS methods largely depends on the accuracy of the generated CAMs, but current methods can only perform rough localization, especially cannot fit CAM to the object's edge well. Therefore, we propose the Edge Enhancement Network (EEN), which uses the shallow features of the network to enhance edge information. This allows us to obtain more accurate CAMs that fit the edges only through image-level labels. It greatly improves the accuracy of pseudo ground truth and easily reaches the level of current mainstream methods. After many experiments, it has reached 67.0% mIoU on the Pascal VOC2012 validation set, which exceeds the latest method under the same training settings. © 2021 IEEE

关键词： Semantic Segmentation

来源：评论

学校读者我要写书评

暂无评论

Heterogeneous Dynamic-Aware GNN for RUL Prediction of Aeroengine

引用

IEEE Transactions on Instrumentation and Measurement 2025年 74卷

作者： Jin, Yang Meng, Zhengguo Hu, Qinghua Chen, Dongyue Civil Aviation University of China Tianjin Key Laboratory for Civil Aircraft Airworthiness and Maintenance Tianjin300300 China Civil Aviation University of China College of Aeronautical Engineering Tianjin300300 China Tianjin University College of Intelligence and Computing Tianjin Key Laboratory of Machine Learning Tianjin300350 China

Accurate remaining useful life (RUL) prediction is crucial for achieving predictive maintenance. The performance degradation process of equipment is, however, not only manifested as anomalies in individual variables but also as changes in the constraint relationships among multiple variables. Human operations and changes in operating conditions further complicate the state representation of the performance degradation process, making RUL prediction for complex systems like aeroengines a challenging task. This article proposes a novel graph flow learning model for RUL prediction in such complex systems with dynamic interactive operations, termed the heterogeneous dynamic-aware graph neural network (HDA-GNN). First, various correlation evaluation metrics are used to convert sensor signals into heterogeneous graphs to represent the complex relationships among sensor signals. Then, gated recurrent unit (GRU) and attention mechanisms are employed to dynamically calculate the edge weights of the heterogeneous graphs, and a graph convolutional network (GCN) is used to integrate the features of these dynamic heterogeneous graphs. Finally, the transformer is applied to learn the temporal features within the graph flows. This model is capable of learning dynamic features under human external operation disturbances from multiple perspectives, thereby reducing the impact of the randomness of human operations on dynamic features. Experimental results on the NASA commercial modular aero-propulsion system simulation (C-MAPSS) dataset demonstrate the effectiveness and superiority of this model for RUL prediction tasks in complex systems like aeroengines. © 2025 IEEE.

关键词： Graph neural networks

来源：评论

学校读者我要写书评

暂无评论

Halftone Image Steganography Based on Maximizing Visual Similarity of Block Units 7th

Halftone Image Steganography Based on Maximizing Visual Sim...

引用

7th International Conference on Artificial intelligence and Security, ICAIS 2021

作者： Yin, Xiaolin Yu, Mujian Chen, Lili Wang, Ke Lu, Wei School of Computer Science and Engineering Guangdong Key Laboratory of Information Security Technology Ministry of Education Key Laboratory of Machine Intelligence and Advanced Computing Sun Yat-sen University Guangzhou510006 China Guangdong Science and Technology Innovation Monitoring and Research Center Guangzhou510033 China

ISBN: (纸本)9783030786113

Steganography focuses on imperceptibility of both the human eyes but also to potential analyzers. Based on the human visual system (HVS) of halftone image, most previous visual quality measurements for multi-tone images can not be adopted to design the steganographic schemes. In this paper, we propose a halftone image steganographic scheme based on the maximizing visual similarity of block units aiming at improving the statistical security but also the image visual quality. First, the halftone HVS model is discussed. The similarity of a pixel and its neighboring regions (which are called block units) is an important factor for human eyes. Then the distortion score is designed based on the visual similarity for each pixel in an image. Several visual similarity scores owned by a pixel are obtained in different block units. In general, the maximal visual similarity score means that this pixel is optimal to embed data. Finally, these visual similarity scores are combined with a weight matrix, and the distortion score that measures the embedding distortion is sent into syndrome-trellis code (STC) to minimize the distortion. Experimental results demonstrate the proposed scheme preserves the visual quality and improves the statistical security. © 2021, Springer Nature Switzerland AG.

关键词： Pixels

来源：评论

学校读者我要写书评

暂无评论

Halftone Image Steganography Based on Minimizing Distortion with Pixel Density Transition 7th

Halftone Image Steganography Based on Minimizing Distortion...

引用

7th International Conference on Artificial intelligence and Security, ICAIS 2021

作者： Yu, Mujian Luo, Junwei Xu, Bozhi Chen, Guoliang Lu, Wei School of Computer Science and Engineering Guangdong Key Laboratory of Information Security Technology Ministry of Education Key Laboratory of Machine Intelligence and Advanced Computing Sun Yat-sen University Guangzhou510006 China Guangdong Science and Technology Innovation Monitoring and Research Center Guangzhou510033 China

ISBN: (纸本)9783030786113

Many advanced halftone steganographic schemes focus only on the distortion of human visual perception or the distortion according to statistics. In this paper, a halftone image steganography based on minimizing distortion with density transition is proposed which aims at utilizing the entropy model and pixel density to resist potential steganalysis. First, the entropy model is established on the image database to describe the texture content and transformed to the preliminary distortion score map. Because the form of texture presentation is distinct between halftone images and ordinary binary images, the feature of pixel density is introduced to represent the local intensity in images. Then the pixel density transition adjustment based on the entropy model is presented, which makes the distortion score more reliable. The final additive distortion map is generated by combining the entropy model and the strategy of density transition. To play the advantage of distortion measurement, syndrome-trellis code (STC) is applied with the distortion map to minimize the embedding distortions. Experimental results demonstrate that compared with other halftone steganographic schemes, the proposed method achieves high statistical security and great visual quality with considerable embedding capacity. © 2021, Springer Nature Switzerland AG.

关键词： Pixels

来源：评论

学校读者我要写书评

暂无评论

Approximate k-Nearest Neighbor Query over Spatial Data Federation 28th

Approximate k-Nearest Neighbor Query over Spatial Data Fede...

引用

28th International Conference on Database Systems for advanced Applications, DASFAA 2023

作者： Zhang, Kaining Tong, Yongxin Shi, Yexuan Zeng, Yuxiang Xu, Yi Chen, Lei Zhou, Zimu Xu, Ke Lv, Weifeng Zheng, Zhiming State Key Laboratory of Software Development Environment Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing School of Computer Science and Engineering and Institute of Artificial Intelligence Beihang University Beijing China Department of Computer Science and Engineering HKUST Hong Kong Guangzhou China City University of Hong Kong Hong Kong

ISBN: (纸本)9783031306365

Approximate nearest neighbor query is a fundamental spatial query widely applied in many real-world applications. In the big data era, there is an increasing demand to scale these queries over a spatial data federation, which consists of multiple data owners, each holding a private, disjoint partition of the entire spatial dataset. However, it is non-trivial to enable approximate k-nearest neighbor query over a spatial data federation. This is because stringent security constraints are often imposed to protect the sensitive, privately owned data partitions, whereas naively extending prior secure query processing solutions leads to high inefficiency (e.g., 100 s per query). In this paper, we propose two novel algorithms for efficient and secure approximate k-nearest neighbor query over a spatial data federation. We theoretically analyze their communication cost and time complexity, and further prove their security guarantees and approximation bounds. Extensive experiments show that our algorithms outperform the state-of-the-art solutions with respect to the query efficiency and often yield a higher accuracy. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Nearest neighbor search

来源：评论

学校读者我要写书评

暂无评论

A Novel Gesture Recognition Technique based on sEMG Armband and Leap Motion Controller

A Novel Gesture Recognition Technique based on sEMG Armband ...

引用

2022 IEEE International Conference on Robotics and Biomimetics, ROBIO 2022

作者： Zhang, Huanqing Xie, Jun Xiao, Yi Yang, Yuzhe Ren, Zhiyuan Zhang, Ruiwen Dong, Yuanzhe Tang, Xi Li, Guanglin Fang, Peng School of Mechanical Engineering Xi'An Jiaotong University Xi'an710049 China China Astronauts Research and Training Center National Key Laboratory of Human Factors Engineering Beijing100094 China Shenzhen Institute of Advanced Technology CAS Key Laboratory of Human-Machine Intelligence-Synergy Systems Shenzhen Engineering Laboratory of Neural Rehabilitation Technology Shenzhen518055 China

ISBN: (纸本)9781665481090

With the rapid development of human-computer interaction technology, it has become very common to use various sensors for gesture recognition to control machines or to complete specified tasks. This study proposes a fusion method that uses Leap Motion for gesture recognition and gForcePro+ to evaluate actual force of hands. A three-dimensional (3-D) virtual gesture recognition platform was designed in Unity. Leap Motion was used to control gestures in the virtual environment, and surface electromyography (sEMG) data was used to distinguish different forces of the same gesture. The success of grasping objects was judged by detecting and analyzing the sEMG signals and Leap Motion signals, combined with the size of the grasping objects. Experimental results showed that Leap Motion and gForcePro+ can be used together to control hand motion. In addition, the proposed system can be used to distinguish the same gesture with different force sizes, which can reflect different forces when the hand grasp objects with the same action. It was proved that the fusion technique of gesture recognition system based on Leap Motion and sEMG is effective for online gesture recognition. © 2022 IEEE.

关键词： Gesture recognition

来源：评论

学校读者我要写书评

暂无评论

SpikingVPR: Spiking Neural Network-Based Feature Aggregation for Visual Place Recognition

SpikingVPR: Spiking Neural Network-Based Feature Aggregation...

引用

Sensing, Measurement & Data Analytics in the era of Artificial intelligence (ICSMD), International Conference on

作者： Gulin Wang Ziliang Ren Qieshi Zhang Jun Cheng CAS Key Laboratory of Human-Machine Intelligence-Synergy Systems Shenzhen Institute of Advanced Technology Chinese Academy of Sciences Shenzhen China School of Computer Science and Technology Dongguan University of Technology Dongguan China

ISBN: (数字)9798331529192

ISBN: (纸本)9798331529208

Visual Place Recognition (VPR) is a crucial task in robotics and autonomous driving, where it identifies the geographic location of a query image by matching it with a reference database. VPR faces significant challenges, including appearance variations due to lighting, occlusion, weather, and seasonal changes, alongside the need for real-time processing under low computational cost and latency. To address these challenges, we introduce Spiking Neural Networks (SNN) into the VPR task, specifically focusing on the feature aggregation component. SNN, with their event-driven nature, offer advantages in computational efficiency and energy consumption, making them well-suited for energy-constrained environments such as drones. Our approach transforms the feature aggregation stage of VPR into a spiking, event-driven mechanism, which maintains high recall performance while integrating seamlessly with traditional ANN-based methods. Experimental results demonstrate that our SNN-based method effectively maintains high recall performance, validating its potential in autonomous driving and other related applications.

关键词： Energy consumption Spiking neural networks Transforms Real-time systems Computational efficiency Visual databases Autonomous vehicles Robots Visual place recognition Meteorology

来源：评论

学校读者我要写书评

暂无评论

GAN-SLAM: GAN based Monocular Visual-Inertial Simultaneous Localization and Mapping in Dark Environments 5

GAN-SLAM: GAN based Monocular Visual-Inertial Simultaneous L...

引用

5th International Symposium on Autonomous Systems, ISAS 2022

作者： Zhang, Qieshi Hao, Luoying Li, Hongjian Ren, Ziliang Cheng, Jun Shenzhen Institute of Advanced Technology Chinese Academy of Sciences CAS Key Laboratory of Human-Machine Intelligence-Synergy Systems Shenzhen China Southern University of Science and Technology Shenzhen China School of Science and Technology Dongguan University of Technology Dongguan China

ISBN: (纸本)9781665487085

In recent years, visual-inertial Simultaneous Localization and Mapping (SLAM) techniques have achieved excellent results due to the complementary nature of visual and inertial sensors, the challenges remain in some environments. When encountering a challenging environment, image features become an important factor affecting the success rate and accuracy of localization of SLAM system. Due to the Convolutional Neural Network (CNN) has the advantage of extracting effective features more accurately, this paper makes a new attempt to replace traditional features with those generated by the network. We combine the state-of-the-art CNN of optical flow estimation and iterative optimization model of traditional SLAM framework, which achieves the accuracy and robustness in challenging environments. To show the performance of our method, some state-of-the-art visual-inertial methods are tested and compared with the challenging sequences of public dataset. Experimental results show that the proposed method can achieve better results in accuracy and robustness and works reliably well for challenging environments. © 2022 IEEE.

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：