检索结果-内蒙古大学图书馆

Knowledge Transfer for Body sensor Networks: Characterization and Resistance of Negative Transfer in Overconstrained Environments

引用

IEEE INTERNET OF THINGS JOURNAL 2024年第19期11卷 31959-31973页

作者： Shi, Han Liu, Yang Luo, Hao Zhao, Hai Liaoning Univ Coll Informat Shenyang 110036 Peoples R China Liaoning Univ Sch Phys Shenyang 110036 Peoples R China Northeastern Univ Sch Comp Sci & Engn Shenyang 110169 Peoples R China

Transfer learning (TL) has shown potential for body sensor network (BSN) applications. However, TL performance is influenced by various factors, such as the BSN environment, user-specific physiological signs, or sensor modalities, leading to negative transfer (NT) effects. Limited research has defined NT from perspectives, such as data quality, domain transferability, and transfer components, but the cognition of NT in multimodal BSNs remains insufficient. In this article, we explore the correlation between transfer modes and NT, proposing a knowledge transfer framework composed of common domain adaptation constraints to reveal the NT effects in overconstrained environments. This effect can explain why strongly constrained transfer algorithms are effective but not necessarily optimal in terms of performance. In the three typical BSN scenarios of emotion recognition, fall detection, and daily activity recognition, we demonstrate the NT in overconstrained environments and its potential influencing factors using different constraint combinations and based on different multimodal body sensor fusion architectures. Extensive experiments focus on discussing the impact of category-condition constraints, domain similarity, and fusion architectures on the NT effect in BSNs. This work can provide a new perspective for the design of multimodal BSN knowledge transfer scheme.

关键词： Transfer learning Task analysis Data models Internet of Things Knowledge transfer Accuracy Training Body sensor network (BSN) information fusion negative transfer (NT) transfer learning (TL)

来源：评论

学校读者我要写书评

暂无评论

RGB-LiDAR fusion for accurate 2D and 3D object detection

引用

MACHINE VISION and applications 2023年第5期34卷 1-11页

作者： Mousa-Pasandi, Morteza Liu, Tianran Massoud, Yahya Laganiere, Robert Univeristy Ottawa Ottawa ON Canada

Effective detection of road objects in diverse environmental conditions is a critical requirement for autonomous driving systems. Multi-modal sensor fusion is a promising approach for improving perception, as it enables the combination of information from multiple sensor streams in order to optimize the integration of their respective data. fusion operators are employed within fully convolutional architectures to combine features derived from different modalities. In this research, we present a framework that utilizes early fusion mechanisms to train and evaluate 2D object detection algorithms. Our evaluation shows that sensor fusion outperforms RGB-only detection methods, yielding a boost of +15.07% for car detection, +10.81% for pedestrian detection, and +19.86% for cyclist detection. In our comparative study, we evaluated three arithmetic-based fusion operators and two learnable fusion operators. Furthermore, we conducted a performance comparison between early- and mid-level fusion techniques and investigated the effects of early fusion on state-of-the-art 3D object detectors. Lastly, we provide a comprehensive analysis of the computational complexity of our proposed framework, along with an ablation study.

关键词： Multimodal deep learning sensor fusion fusion operators Object detection Autonomous driving Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

RGB-X Object Detection via Scene-Specific fusion Modules

RGB-X Object Detection via Scene-Specific Fusion Modules

引用

IEEE/CVF Winter conference on applications of Computer Vision (WACV)

作者： Deevi, Sri Aditya Lee, Connor Pandey, Gaurav Gan, Lu Chung, Soon-Jo Nagesh, Sushruth CALTECH Pasadena CA 91125 USA Ford Motor Co Dearborn MI 48121 USA

ISBN: (纸本)9798350318920;9798350318937

Multimodal deep sensor fusion has the potential to enable autonomous vehicles to visually understand their surrounding environments in all weather conditions. However, existing deep sensor fusion methods usually employ convoluted architectures with intermingled multimodal features, requiring large coregistered multimodal datasets for training. In this work, we present an efficient and modular RGB-X fusion network that can leverage and fuse pre-trained single-modal models via scene-specific fusion modules, thereby enabling joint input-adaptive network architectures to be created using small, coregistered multimodal datasets. Our experiments demonstrate the superiority of our method compared to existing works on RGB-thermal and RGB-gated datasets, performing fusion using only a small amount of additional parameters. Our code is available at https://***/dsriaditya999/RGBXfusion.

关键词： applications applications Autonomous Driving Robotics

来源：评论

学校读者我要写书评

暂无评论

Optimizing Long-Term Robot Tracking with Multi-Platform sensor fusion

Optimizing Long-Term Robot Tracking with Multi-Platform Sens...

引用

IEEE/CVF Winter conference on applications of Computer Vision (WACV)

作者： Albanese, Giuliano Mitra, Arka Zaech, Jan-Nico Zhao, Yupeng Chhatkuli, Ajad Van Gool, Luc Swiss Fed Inst Technol Zurich Switzerland Katholieke Univ Leuven Leuven Belgium INSAIT Sofia Bulgaria

ISBN: (纸本)9798350318920;9798350318937

Monitoring a fleet of robots requires stable long-term tracking with re-identification, which is yet an unsolved challenge in many scenarios. One application of this is the analysis of autonomous robotic soccer games at RoboCup. Tracking in these games requires handling of identically looking players, strong occlusions, and non-professional video recordings, but also offers state information estimated by the robots. In order to make effective use of the information coming from the robot sensors, we propose a robust tracking and identification pipeline. It fuses external non-calibrated camera data with the robots' internal states using quadratic optimization for tracklet matching. The approach is validated using game recordings from previous RoboCup World Cup tournaments.

关键词： algorithms algorithms and algorithms applications formulations Machine learning architectures Robotics Video recognition and understanding

来源：评论

学校读者我要写书评

暂无评论

DSP architectures of Covariance Intersection fusion Algorithm for Automotive sensor fusion 14

DSP Architectures of Covariance Intersection Fusion Algorith...

引用

14th International conference on Computing Communication and Networking Technologies, ICCCNT 2023

作者： Praharshita, D.S.L. Achala, G. Srihari, Pathipati Acharya, U. Shripathi Pardhasaradhi, Bethi India National Institute of Technology Karnataka ECE Dept. India Continental Autonomous Mobility India Pvt Ltd Adas Dept. India

ISBN: (纸本)9798350335095

The data fusion from sensors within the automotive vehicle is vital for improved accuracy and safety. The centralized and information matrix fusion (IMF) algorithms are famous for providing an optimal fusion estimate. However, the IMF is not viable in automotive sensor fusion applications due to the limited bandwidth and low hardware resources. Hence, distributed fusion technology is widely adopted in the automotive sensor applications to achieve high-speed and low-area realizations. This paper proposes three digital signal processing (DSP) architectures for covariance intersection (CI) fusion algorithm: Pipelined-traditional CI, adder-ladder CI, and pipelined adder-ladder CI. The proposed DSP architectures are evaluated with hardware resource consumption (multipliers, adders, and delays), maximum achievable frequency, and latency of the architecture. In addition, proposed CI algorithms for Digital Signal Processing (DSP) architectures are compared with IMF DSP architectures. The hardware resources and optimal pipeline stages required for CI with respect to N number of sensors are provided. The traditional pipeline algorithm requires N number of stages where as the proposed pipelined version of adder-ladder CI requires a N-1 pipeline stage with additional 7N-1 and 7N-3 delay elements for even and odd number of sensors to achieve the overall system operating frequency to an operation of multiplier. The proposed DSP architectures are suitable for automotive sensor fusion due to their high operating frequency and low hardware resources. © 2023 IEEE.

关键词： Digital signal processing

来源：评论

学校读者我要写书评

暂无评论

MULTI-sensor INFORMATION fusion TECHNOLOGY: PRINCIPLE, applications AND KEY TECHNOLOGIES EXPLORATIONS 2

MULTI-SENSOR INFORMATION FUSION TECHNOLOGY: PRINCIPLE, APPLI...

引用

2nd International conference on Mechatronic Automation and Electrical Engineering, ICMAEE 2024

作者： Wu, Wenchen Xu, Ruizhe Guangdong Country Garden School Foshan528312 China The Affiliated Middle School of Fujian Normal University Fuzhou350007 China

ISBN: (纸本)9781837242672

Multi-sensor information fusion (MSIF) has emerged as a critical technology in various fields due to the growing complexity of systems and the demand for precise information. This paper explores the fundamental principles, key technologies, and potential applications of MSIF. The paper delves into the design and implementation of MSIF systems, discussing various system architectures such as centralized, distributed, and hybrid models. Key technologies crucial to MSIF, including sensor calibration, data association, filtering methods, and fusion algorithms like Bayesian Criterion, Dempster-Shafer Evidence Theory, Fuzzy Theory, and Neural Networks, are thoroughly examined. The study highlights the importance of data preprocessing and feature extraction techniques in achieving efficient data fusion. Furthermore, the potential of MSIF in diagnosing complex systems is demonstrated through applications in industrial automation, environmental monitoring, and healthcare. MSIF also faces challenges related to data synchronization, algorithm complexity, and the integration of diverse data sources. With ongoing advancements in sensor technology and data processing, MSIF is poised to play an increasingly significant role in intelligent systems, including autonomous vehicles and smart cities. © The Institution of Engineering & Technology 2024.

关键词： Information fusion

来源：评论

学校读者我要写书评

暂无评论

Comparative Analysis and Implementation of Energy-Quality Scalable Multiply-Accumulate architectures for TinyML applications 31

Comparative Analysis and Implementation of Energy-Quality Sc...

引用

31st conference on Electronics Circuits and Systems

作者： Pelayo, Kenn Danielle C. Alvarez, Anastacia B. Chua, Adelson N. Galapon, Fredrick Angelo R. Tan, Allen Jason A. Quizon, Lawrence Roman A. Univ Philippines Diliman Elect & Elect Engn Inst Quezon City Philippines

ISBN: (纸本)9798350377217;9798350377200

Recent trends in machine learning algorithms have shifted from computer vision applications to near-sensor intelligent data analytics, such as wireless sensor nodes, which are considered as ultra-low power devices. This shift presents new challenges in energy consumption, throughput, and area budgets. This work implements and optimizes Bit fusion, a 2D-scalable divide-and-conquer (D&C) MAC architecture with precision-scaling from 8 to 2 bits, through bitwidth optimizations and by trading-off complexity overheads that are not aligned with the target MobileNet-v2 model. The designs are synthesized using 22-nm FD-SOI CMOS technology. It was determined that the throughput trade-offs of 1-level architectures in 2bx2b mode for 17.37% less energy per operation and 26.84% more throughput per area would be preferred. Architecture variations were also explored wherein sum-together (ST) architectures outperformed sum-apart (SA) architectures with 21.03% reduction in energy per operation and 33.32% increase in throughput per area.

关键词： ASIC deep neural networks digital accelerators multiply-accumulate MAC precision-scalable circuits energy-quality scalable circuits

来源：评论

学校读者我要写书评

暂无评论

Beyond fusion: Modality Hallucination-based Multispectral fusion for Pedestrian Detection

Beyond Fusion: Modality Hallucination-based Multispectral Fu...

引用

IEEE/CVF Winter conference on applications of Computer Vision (WACV)

作者： Xie, Qian Cheng, Ta-Ying Zhong, Jia-Xing Zhou, Kaichen Markham, Andrew Trigoni, Niki Univ Oxford Dept Comp Sci Oxford England

ISBN: (纸本)9798350318920;9798350318937

Pedestrian detection is a fundamental task for many downstream applications. Visible and thermal images, as the two most important data types, are usually used to detect pedestrians under various environmental conditions. Many state-of-the-art works have been proposed to use two-stream (i.e., two-branch) architectures to combine visible and thermal information to improve detection performance. However, conventional visible-thermal fusion-based methods have no ability to obtain useful information from the visible branch under poor visibility conditions. The visible branch could even sometimes bring noise into the combined features. In this paper, we present a novel thermal and visible fusion architecture for pedestrian detection. Instead of simply using two branches to separately extract thermal and visible features and then fusing them, we introduce a hallucination branch to learn the mapping from the thermal to the visible domain, forming a novel three-branch feature extraction module. We then adaptively fuse feature maps from all three branches (i.e., thermal, visible, and hallucination). With this new integrated hallucination branch, our network can still get relatively good visible feature maps under challenging low-visibility conditions, thus boosting the overall detection performance. Finally, we experimentally demonstrate the superiority of the proposed architecture over conventional fusion methods.

关键词： algorithms algorithms and algorithms formulations Image recognition and understanding Machine learning architectures

来源：评论

学校读者我要写书评

暂无评论

Multi-view Classification Using Hybrid fusion and Mutual Distillation

Multi-view Classification Using Hybrid Fusion and Mutual Dis...

引用

IEEE/CVF Winter conference on applications of Computer Vision (WACV)

作者： Black, Samuel Souvenir, Richard Temple Univ Dept Comp & Informat Sci Philadelphia PA 19122 USA

ISBN: (纸本)9798350318920;9798350318937

Multi-view classification problems are common in medical image analysis, forensics, and other domains where problem queries involve multi-image input. Existing multi-view classification methods are often tailored to a specific task. In this paper, we repurpose off-the-shelf Hybrid CNN-Transformer networks for multi-view classification with either structured or unstructured views. Our approach incorporates a novel fusion scheme, mutual distillation, and minimal additional parameters. We demonstrate the effectiveness and generalization capability of our approach, MV-HFMD, on multiple multi-view classification tasks and show that it outperforms other multi-view approaches, even task-specific methods. Code is available at https://***/vidarlab/multi-view-hybrid.

关键词： algorithms algorithms and algorithms formulations Image recognition and understanding Machine learning architectures

来源：评论

学校读者我要写书评

暂无评论

Token fusion: Bridging the Gap between Token Pruning and Token Merging.

Token Fusion: Bridging the Gap between Token Pruning and Tok...

引用

IEEE/CVF Winter conference on applications of Computer Vision (WACV)

作者： Kim, Minchul Gao, Shangqian Hsu, Yen-Chang Shen, Yilin Jin, Hongxia Michigan State Univ E Lansing MI 48824 USA Samsung Res Amer Mountain View CA USA

ISBN: (纸本)9798350318920;9798350318937

Vision Transformers (ViTs) have emerged as powerful backbones in computer vision, outperforming many traditional CNNs. However, their computational overhead, largely attributed to the self-attention mechanism, makes deployment on resource-constrained edge devices challenging. Multiple solutions rely on token pruning or token merging. In this paper, we introduce"Token fusion" (ToFu), a method that amalgamates the benefits of both token pruning and token merging. Token pruning proves advantageous when the model exhibits sensitivity to input interpolations, while token merging is effective when the model manifests close to linear responses to inputs. We combine this to propose a new scheme called Token fusion. Moreover, we tackle the limitations of average merging, which doesn't preserve the intrinsic feature norm, resulting in distributional shifts. To mitigate this, we introduce MLERP merging, a variant of the SLERP technique, tailored to merge multiple tokens while maintaining the norm distribution. ToFu is versatile, applicable to ViTs with or without additional training. Our empirical evaluations indicate that ToFu establishes new benchmarks in both classification and image generation tasks concerning computational efficiency and model accuracy.

关键词： 3D algorithms algorithms algorithms and algorithms etc. formulations Generative models for image Image recognition and understanding Machine learning architectures video

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：