检索结果-内蒙古大学图书馆

arXiv 2024年

作者： Chen, Jiayi Deng, Chunhua School of Computer Science and Technology Wuhan University of Science and Technology Wuhan430065 China Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System Wuhan430065 China

With the advancement of video analysis technology, the multi-object tracking (MOT) problem in complex scenes involving pedestrians is gaining increasing importance. This challenge primarily involves two key tasks: pedestrian detection and re-identification. While significant progress has been achieved in pedestrian detection tasks in recent years, enhancing the effectiveness of re-identification tasks remains a persistent challenge. This difficulty arises from the large total number of pedestrian samples in multi-object tracking datasets and the scarcity of individual instance samples. Motivated by recent rapid advancements in meta-learning techniques, we introduce MAML MOT, a meta-learning-based training approach for multi-object tracking. This approach leverages the rapid learning capability of meta-learning to tackle the issue of sample scarcity in pedestrian re-identification tasks, aiming to improve the model’s generalization performance and robustness. Experimental results demonstrate that the proposed method achieves high accuracy on mainstream datasets in the MOT Challenge. This offers new perspectives and solutions for research in the field of pedestrian multi-object tracking. Copyright © 2024, The Authors. All rights reserved.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Radar HRRP Open Set Recognition Using Hierarchical Prototype Learning

Radar HRRP Open Set Recognition Using Hierarchical Prototype...

引用

IEEE International Conference on Radar

作者： Yichen Liu Yanhua Wang Liang Zhang Binhua Yu Xiaojing Yang Yu-ang Zhang Le Zheng Difan Zou School of Information and Electronics Beijing Institute of Technology Beijing China Beijing Key Laboratory of Embedded Real-Time Information Processing Technology Beijing China Department of Computer Science University of Hong Kong Hongkong China HKU Musketeers Foundation Institute of Data Science University of Hong Kong Hongkong China

ISBN: (数字)9798350362381

ISBN: (纸本)9798350362398

High-resolution range profiles (HRRP) are increasingly employed in radar automatic target recognition (RATR). In practical applications of RATR, the environment is open and dynamic, and the recognition model is likely to encounter targets of unseen categories. However, traditional methods focus on a close-set scenario, where all categories in the test set are known during training. To bridge this gap, this paper proposes an open-set recognition (OSR) method based on hierarchical classification, developed from prototype learning. This method improves the generalization performance for recognizing known categories and identifies unknown categories while providing information about how the unknown categories relate to the known ones. Specifically, first, a four-layer category hierarchy is built based on prior knowledge to guide the training and testing stages. Next, we propose a hierarchical prototype loss (HPL) to constrain the feature space extracted by the prototype network so that the feature distribution of objects is consistent with the hierarchical category structure. Lastly, the trained prototype network makes predictions following the hierarchical structure. This operation could provide relationship information between the unknown and known categories. Extensive experiments on measured HRRP data validate the effectiveness of our proposed method for open-set recognition.

关键词： Training Bridges Target recognition Prototypes Radar Feature extraction Testing

来源：评论

学校读者我要写书评

暂无评论

Generative Multi-View HRRP Recognition Based on Cascade Generation and Fusion Network

Generative Multi-View HRRP Recognition Based on Cascade Gene...

引用

IEEE International Conference on Radar

作者： Qiang Zhou Bingqian Yu Yanhua Wang Liang Zhang Le Zheng Difan Zou Xin Zhang School of Information and Electronics Beijing Institute of Technology Beijing China Beijing Key Laboratory of Embedded Real-Time Information Processing Technology Beijing China Department of Computer Science University of Hong Kong Hongkong China HKU Musketeers Foundation Institute of Data Science University of Hong Kong Hongkong China

ISBN: (数字)9798350362381

ISBN: (纸本)9798350362398

High-resolution range profile (HRRP) is critical for radar target recognition. However, HRRP data for non-cooperative targets are often sparsely collected, leading to limited performance of HRRP target recognition methods. To address the issue of sparse view target recognition, we propose a generative multi-view HRRP recognition (GMVR) model. Firstly, by leveraging the pattern learning and generation ability of deep generative models (DGMs), we propose a viewpoint feature expansion module (VFEM) to generate multi-view features adjacent to sparse viewpoints. Subsequently, based on feature-level fusion, we proposed a multi-view feature fusion module (MVFM). Aided by attention mechanisms, the generated features can be adaptively integrated, thereby reducing the decision bias caused by single-viewpoint HRRP samples. Additionally, we employ transfer learning to mitigate the small sample problem caused by sparse viewpoints. Experimental results demonstrate that the proposed method outperforms various comparison methods, achieving optimal performance.

关键词： Adaptation models Attention mechanisms Target recognition Transfer learning Radar Pattern recognition

来源：评论

学校读者我要写书评

暂无评论

CPR-Net: A Novel Joint Learning Network for Pulmonary Nodule Evaluation

CPR-Net: A Novel Joint Learning Network for Pulmonary Nodule...

引用

IEEE International Conference on Systems, Man and Cybernetics

作者： Gong Chen Jun Liu Chen-Qian Li Wuhan University of Science and Technology Wuhan China Hubei Province Key Laboratory of Intelligent Information Processing and Real-Time Industrial System Wuhan China

Lung cancer is the most common malignant tumor worldwide, with high mortality rates. Pulmonary nodule is a common manifestation of lung cancer. Accurate segmentation and detection of pulmonary nodules from CT scans are essential for proper assessment of patient prognosis. However, this task remains challenging due to various factors, such as class imbalance and the need for detailed characterization of lung nodule segmentation. To address these issues, we propose a novel network, CPR-Net (Convolutional Pyramid Residuals-Net), that combines segmentation and benign-malignant classification for CT images of lung nodules. Our approach utilizes a PBMM module to expand the perception field and enhance the representation capability of the model, allowing it to learn more detailed information. We evaluate the classification performance using accuracy, precision, recall, F1 score, AUC, and the segmentation performance using the Dice coefficient. Experiments on both publicly available datasets and self-constructed datasets demonstrate that our proposed method in this paper outperforms other methods in terms of classification and segmentation performance under limited labeled data conditions. Moreover, Our results suggest that this model has great potential for improving lung nodule diagnosis in radiology for heterogeneous intranodal images.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Space-surface Bistatic SAR Tomography: A Promising Approach for Accurate Local 3D Reconstruction

Space-surface Bistatic SAR Tomography: A Promising Approach ...

引用

IET International Radar Conference 2023, IRC 2023

作者： Chen, Zhiyang Li, Yuanhao Chen, Xinyan Hu, Cheng Datcu, Mihai Monti-Guarnieri, Andrea Virgilio Hobbs, Stephen School of Information and Electronics Beijing Institute of Technology Beijing100084 China Beijing Key Laboratory of Embedded Real-time Information Processing Technology Beijing100084 China Advanced Technology Research Institute Beijing Institute of Technology Jinan250300 China CEOSpaceTech University Politehnica of Bucharest Bucharest050663 Romania The Dipartimento di Elettronica Informazione e Bioingegneria Politecnico di Milano Milan20133 Italy School of Aerospace Transport and Manufacturing Cranfield University BedfordMK43 0AL United Kingdom

ISBN: (纸本)9781839539954

With a flexibly deployed receiver, space-surface bistatic Synthetic Aperture Radar (SS-BiSAR) can increase the observation sensitivity in the north-south direction, which is a promising method for local high-precision three-dimensional deformation retrieval. However, in urban areas, the layover effect hinders the determination of the dominant scatterer in each resolution cell. Multi-baseline SAR tomography is an effective approach to solving overlaid scatterers. This paper presents the first experiment result of SS-BiSAR tomography. By estimating and removing the interferometric fringes, the repeat-track orbit errors are effectively corrected. The experiment was conducted based on the first L-band spaceborne twin SAR - the Chinese Lutan-1 (LT-1). A total of 11 repeat-track SAR images were acquired, and three-dimensional point clouds of a building area was recovered, with an accuracy of approximately 4 meters. © The Institution of Engineering & technology 2023.

关键词： Synthetic aperture radar

来源：评论

学校读者我要写书评

暂无评论

HCU-Net: An Efficient Fully Convolutional Neural Network for Thyroid Nodules Ultrasound Image Segmentation

HCU-Net: An Efficient Fully Convolutional Neural Network for...

引用

IEEE International Conference on Systems, Man and Cybernetics

作者： Mengxi Ren Jun Liu School of Computer Science and Technology Wuhan University of Science and Technology Wuhan China Hubei Province Key Laboratory of Intelligent Information Processing and Real-Time Industrial System Wuhan University of Science and Technology Wuhan China

ISBN: (数字)9781665410205

ISBN: (纸本)9781665410212

Thyroid nodules, which can manifest as solitary or multiple growths, are among the most prevalent diseases with a high incidence rate. When examining thyroid nodules, ultrasound is typically the first step. Ultrasound image segmentation of thyroid nodules can assist physicians in diagnosis. However, the following three factors limit the development of thyroid nodule segmentation: (1) most of the existing studies optimize the segmentation of single nodules, and the incidence of multiple nodules is higher, so the segmentation of multiple nodules is also very important; (2) the ultrasound image of thyroid nodules is difficult to segment because of its low resolution and fuzzy boundary; (3) Due to the small data set of thyroid nodule ultrasound, in such a small dataset, attention mechanisms may be prone to overfitting, resulting in poorer generalization performance. Therefore, the attention-based model may not be suitable for the segmentation of thyroid nodules. Hence, we propose Hybrid Convolutional U-Net (HCU-Net), a simple and strong fully convolutional neural network for the segmentation of single or multiple thyroid nodules. HCU-Net incorporates a large receptive field to ensure accurate and stable segmentation. Specifically, we design Hybrid Branch Block to enhance the network's feature extraction capability for single or multiple nodules in ultrasound images by integrating multi-scale features. Furthermore, we add Enhanced ConvNeXt Blocks to the U-Net's final layer, enabling the extraction of global contextual information for segmentation performance. The Experimental results show that our proposed method can effectively improve these issues on two publicly available datasets. In addition, compared to attention-based models, our network is simpler and more suitable for small datasets like ultrasound images.

关键词： Image segmentation Ultrasonic imaging Image resolution Medical services Feature extraction Convolutional neural networks Medical diagnostic imaging Thyroid Overfitting Diseases

来源：评论

学校读者我要写书评

暂无评论

Multi-Scale and Improved Skip Connections Network for Skin Lesion Segmentation

Multi-Scale and Improved Skip Connections Network for Skin L...

引用

6th International Conference on Intelligent Computing and Signal processing (ICSP)

作者： Botao Zhang Jun Chen College of Computer Science and Technology Wuhan University of Science and Technology Wuhan China Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System Wuhan University of Science and Technology Wuhan China

ISBN: (数字)9798350376548

ISBN: (纸本)9798350376555

To solve the problem that the current skin lesion segmentation networks have a poor segmentation accuracy when lesion areas are of different shapes and sizes and low contrast at the edge areas, based on U-Net, we present a skin lesion segmentation network based on multi-scale and improved skip connections. Firstly, a multi-scale feature enhancement module is designed to replace the U-Net's $\boldsymbol{3\times 3}$ convolution blocks, which can improve the feature extraction ability of the network about lesion areas of different shapes and sizes. Secondly, a lesion attention module is introduced into the skip connection to improve the semantic expression of low-level features and the location attention of high-level features, which can improve the fine-grained information transmission efficiency at the skip connection. Finally, a new hybrid loss function is designed to calculate the label and the edge loss of lesion areas, which improves the ability of the network to segment lesion areas and edges. Experiments are conducted on the ISIC2017 and ISIC2018 datasets, and the results prove that our method have a higher segmentation accuracy in skin lesion segmentation.

关键词： Accuracy Shape Convolution Semantics information processing Feature extraction Skin Lesions

来源：评论

学校读者我要写书评

暂无评论

SAR Target Recognition Using Improved Monogenic-Based Feature Extraction Framework

SAR Target Recognition Using Improved Monogenic-Based Featur...

引用

2021 CIE International Conference on Radar, Radar 2021

作者： Li, Feng Yao, Weijun Li, Yang Chen, Wei Radar Research Laboratory School of Information and Electronics Beijing Institute of Technology Beijing Institute of Technology Chongqing Innovation Center Beijing China School of Information and Electronics Beijing Institute of Technology Beijing Key Laboratory of Embedded Real-time Information Processing Technology Radar Research Lab Beijing China

ISBN: (纸本)9781665498142

Applying computer vision methods to synthetic aperture radar (SAR) image recognition is a research trend in recent years, and a series of valuable results have been achieved. In order to use machine learning classifiers for recognition, it is necessary to extract effective features, and most of these features are directly extracted based on grayscale SAR images. SAR data is usually rare, and more difficult to collect than optical images. Therefore, the problem of recognition using a small-size training set for SAR is more challenging to practical pattern recognition methods. The monogenic signal is an extended version of analytic signals in high-dimensional space which has attracted attention. In this paper, a new recognition framework which is based on feature dimensionality augmentation using combined multi-scale monogenic components and histogram of oriented gradient (HOG) feature is proposed. Proposed feature is named as MONO-HOG. This letter focuses on recognition under both standard operating condition (SOC) and small-sample scene. Experiments on moving and stationary target automatic recognition (MSTAR) data set show that our proposed framework has satisfying performance. © 2021 IEEE.

关键词： feature extraction monogenic signal small sample synthetic aperture radar (SAR) target recognition

来源：评论

学校读者我要写书评

暂无评论

Discovering DTI and DDI by Knowledge Graph with MHRW and Improved Neural Network

Discovering DTI and DDI by Knowledge Graph with MHRW and Imp...

引用

2021 IEEE International Conference on Bioinformatics and Biomedicine, BIBM 2021

作者： Zhang, Shuo Lin, Xiaoli Zhang, Xiaolong Wuhan University of Science and Technology Hubei Key Laboratory of Intelligent Information Processing and Real-time Industrial System School of Computer Science and Technology Wuhan China

ISBN: (纸本)9781665401265

Drug discovery is of great significance in medical and biological research, while the study of Drug-Target Interaction (DTI) and Drug-Drug Interaction (DDI) can help accelerate drug discovery progress. This paper proposes a new hybrid method for DTI prediction and DDI prediction, which is called MHRW2Vec-TBAN that combines graph representation learning and neural network. MHRW2VecTBAN first constructs knowledge graph KG-DTI and KG-DDI that integrate data related to drugs and targets. Then, an improved graph representation learning model, MHRW2Vec model, is used to obtain feature vectors of reflecting the network structure information for improving the performance of representation learning. Finally, the feature vectors obtained are input to the improved neural network model TextCNN-BiLSTM-Attention Network (TBAN). The experimental results show that, compared with other existing methods, our method could discover more deeper the relationship between drugs and their potential neighborhoods, and has great improvements in DTI prediction and DDI prediction. In addition, the case study of prediction COVID-19 DTI also shows that the proposed model has the potential for actual drug discovery. © 2021 IEEE.

关键词： Knowledge graph

来源：评论

学校读者我要写书评

暂无评论

Drug-Target Interaction Prediction via Multiple Output Graph Convolutional Networks 17th

Drug-Target Interaction Prediction via Multiple Output Graph...

引用

17th International Conference on Intelligent Computing, ICIC 2021

作者： Ye, Qing Zhang, Xiaolong Lin, Xiaoli Hubei Key Laboratory of Intelligent Information Processing and Real-Time Industrial System School of Computer Science and Technology Wuhan University of Science and Technology Wuhan China

ISBN: (纸本)9783030845315

Computational prediction of drug-target interaction (DTI) is very important for the new drug discovery. Currently, graph convolutional networks (GCNs) have been gained a lot of momentum, as its performance on non-Euclidean data. Although drugs and targets are two typical non-Euclidean data, many problems are also existed when use GCNs in DTI prediction. Firstly, most state-of-the-art GCN models are prone to the vanishing gradient problem, which is more serious in DTI prediction, as the number of interactions is limit. Secondly, a suitable graph is hardly defined for GCN in DTI prediction, as the relationship between the samples is not explicitly provided. To overcome the above problems, in this paper, a multiple output graph convolutional network (MOGCN) based DTI prediction method is designed. MOGCN enhances its learning ability with many new designed auxiliary classifier layers and a new designed graph calculation method. Many auxiliary classifier layers can increase the gradient signal that gets propagated back, utilize multi-level features to train the model, and use the features produced by the higher, middle or lower layers in a unified framework. The graph calculation method can use both the label information and the manifold. The conducted experiments validate the effectiveness of our MOGCN. © 2021, Springer Nature Switzerland AG.

关键词： Forecasting

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：