检索结果-内蒙古大学图书馆

symposium of image, signal processing, and Artificial Vision (STSIVA)

作者： Xia Lin Haomiao Li Xin Jiang Yuchao Gao Jinran Wu Yang Yang College of Automation and College of Artificial Intelligence Nanjing University of Posts and Telecommunications Nanjing China School of Mathematical Sciences Queensland University of Technology Brisbane Australia

ISBN: (纸本)9781665426565

An improved version of the arithmetic optimization algorithm (AOA) based on the opposition-based learning (OBL) strategy called OBLAOA is proposed in this paper. The proposed OBLAOA algorithm consists of two stages, and in the second stage we adds OBL to update the AOA population in each iteration. The improved AOA is compared with the original AOA by using 12 benchmark functions in different dimensions to validate the improvement on exploration with the OBL. Eventually ,we get a conclusion that the OBLAOA is committed to take both candidate solutions and their opposite solutions into consideration, which shows greater opportunity to reach the global optimal and faster convergence acceleration than AOA.

关键词： image segmentation Conferences Sociology Benchmark testing Feature extraction Statistics Informatics

来源：评论

学校读者我要写书评

暂无评论

SwinFuse: A Residual Swin Transformer Fusion Network for Infrared and Visible images

引用

ieee TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT 2022年 71卷

作者： Wang, Zhishe Chen, Yanlin Shao, Wenyu Li, Hui Zhang, Lei Taiyuan Univ Sci & Technol Sch Appl Sci Taiyuan 030024 Peoples R China Jiangnan Univ Sch Artificial Intelligence & Comp Sci Wuxi 214122 Jiangsu Peoples R China Nanyang Normal Univ Sch Mechatron Engn Nanyang 473061 Peoples R China

The existing deep learning fusion methods mainly concentrate on convolutional neural networks (CNNs), and few attempts are made with transformer. Meanwhile, the convolutional operation is a content-independent interaction between the image and the convolution kernel, which may lose some important contexts and further limit fusion performance. Toward this end, we present a simple and strong fusion baseline for infrared and visible images, namely, residual Swin Transformer fusion network, termed SwinFuse. Our SwinFuse includes three parts: the global feature extraction, fusion layer, and feature reconstruction. In particular, we build a fully attentional feature encoding backbone to model the long-range dependencies, which is a pure transformer network and has a stronger representation ability compared with the CNNs. Moreover, we design a novel feature fusion strategy based on the L-1-norm for sequence matrices and measure the corresponding activity levels from row and column vector dimensions, which can well retain competitive infrared brightness and distinct visible details. Finally, we testify our SwinFuse with nine state-of-the-art traditional and deep learning methods on three different datasets through subjective observations and objective comparisons, and the experimental results manifest that the proposed SwinFuse obtains surprising fusion performance with strong generalization ability and competitive computational efficiency. The code will be available at https://***/Zhishe-Wang/SwinFuse.

关键词： Deep learning feature normalization image fusion self-attention mechanism Swin Transformer

来源：评论

学校读者我要写书评

暂无评论

Sparse-view X-ray spectral CT reconstruction using annihilating filter-based low rank hankel matrix approach

Sparse-view X-ray spectral CT reconstruction using annihilat...

引用

ieee International symposium on Biomedical Imaging

作者： Yo Seob Han Kyong Hwan Jin Kyungsang Kim Jong Chul Ye Dept. Bio & Brain Engineering Korea Advanced Institute of Science & Technology (KAIST) Center for Advanced Medical Imaging Sciences Massachusetts General Hospital and Harvard Medical School Bio-Imaging and Signal Processing Lab. (BISPL) Dept. Bio & Brain Engineering Korea Advanced Institute of Science & Technology (KAIST)

ISBN: (纸本)9781479923519

In a kVp switching-based sparse view spectral CT, each spectral image cannot be reconstructed separably by an analytic reconstruction method, because the projection views for each spectral band is too sparse. However, the underlying structure is common between the spectral bands, so there exists inter-spectral redundancies that can be exploited by the recently proposed annihilating filter-based low rank Hankel matrix approach (ALOHA). More specifically, the sparse view projection data are first rebinned in the Fourier space, from which Hankel structured matrix with missing elements are constructed for each spectral band. Thanks to the inter-spectral correlations as well as transform domain sparsity of underlying images, the concatenated Hankel structured matrix is low-ranked, and the missing Fourier data for each spectral band can be simultaneously estimated using a low rank matrix completion. To reduce the computational complexity furthermore, we exploit the Hermitian symmetry of Fourier data. Numerical experiments confirm that the proposed method outperforms the existing ones.

关键词： Computed tomography image reconstruction Transforms TV Redundancy Switches Sparse matrices

来源：评论

学校读者我要写书评

暂无评论

Exploring Model Compression Techniques for Efficient 1D CNN-Based Hand Gesture Recognition on Resource-Constrained Edge Devices

Exploring Model Compression Techniques for Efficient 1D CNN-...

引用

International Conference on Advanced Technologies for signal and image processing (ATSIP)

作者： Mahdi Mnif Salwa Sahnoun Mahdi Djemaa Ahmed Fakhfakh Olfa Kanoun National School of Electronics and Telecommunications of Sfax University of Sfax Tunisia Laboratory of Signals Systems Artificial Intelligence and Networks (SM@RTS) Digital Research Center of Sfax (CRNS) Tunisia Measurements and Sensor Technology Faculty of Electrical Engineering and Information Technology Chemnitz University of Technology Germany

ISBN: (数字)9798350351484

ISBN: (纸本)9798350351491

Tiny Machine Learning is undergoing rapid evolution in the context of edge computing and intelligent Internet of Things (IoT) devices. This paper investigates the potential of model compression techniques for enhancing energy efficiency within IoT edge devices, as an application, hand gesture recognition based on Electrical Impedance Tomography data was used. To achieve efficient and accurate machine learning on battery-powered devices, three model compression techniques: Global pruning, Knowledge Distillation (KD), and quantization were investigated and applied on a 1D convolutional neural network. The global pruning technique resulted in a slight improvement in test data accuracy while maintaining the model size at 284.45 kB, though it increased the model’s sparsity. With KD, the model size significantly decreased to 75.042 kB, with negligible impact on accuracy. In contrast, 8-bit quantization reduced the model size to 78.434 kB, but this came at the cost of a substantial 13.75% decrease in accuracy. Each model compression technique contributes to reducing model size or increasing sparsity. This is particularly beneficial for deploying deep learning models on resource-constrained IoT devices.

关键词： Electrical impedance tomography Accuracy image coding Quantization (signal) computational modeling Tiny machine learning image edge detection

来源：评论

学校读者我要写书评

暂无评论

Iterative scatter correction for digital tomosynthesis using composition ratio update and GPU based Monte Carlo simulation

Iterative scatter correction for digital tomosynthesis using...

引用

ieee International symposium on Biomedical Imaging

作者： Kyung Sang Kim Young Hoon Seong Jongha Lee Kwang Eun Jang Jong Chul Ye Bio Imaging & Signal Processing Lab. Dept. of Bio& Brain Engineering KAIST Daejon Republic of Korea Samsung Advanced Institute of Technology Samsung Electronics Yongin Kyunggi Republic of Korea

ISBN: (纸本)9781457718571

In digital tomosynthesis (DTS), accurate scatter correction is often necessary for quantitative analysis. This is especially important because low energy x-ray of 10-40 keV, which is widely used for the breast imaging to enhance the contrast between adipose and glandular, results in high scatter fraction. In this paper, we propose an iterative scatter correction for digital tomopsynthesis using composition ratio update and GPU based Monte Carlo simulation (MCS). One of the technical difficulty in scatter estimation using MCS for tomosynthesis is that accurate segmentation of 3D volume is very difficult due to the low resolution of the reconstruction object. Thus, an intermediate surrogate object is introduced to represent composition ratio between adipose and glandular. We show that the composition ratio can be calculated using average attenuation coefficients. Another technical challenge is extremely high computational cost of MCS. We overcome this using GPU based ultra-fast MCS. Our results demonstrate that our iterative scatter correction using composition ratio update is indeed effective in improving the quality of the reconstruction object in a reasonable time frame.

关键词： Graphics processing unit Attenuation Photonics image reconstruction Monte Carlo methods Materials Phantoms

来源：评论

学校读者我要写书评

暂无评论

Performance Evaluation of Beam Multiply and Sum Beamforming with Coherent Plane Wave Compounding: In-vitro Results

Performance Evaluation of Beam Multiply and Sum Beamforming ...

引用

South Asian Ultrasonics symposium (SAUS), ieee

作者： A. N. Madhavanunni Mahesh Raveendranatha Panicker Dept. of Electrical Engineering Indian Institute of Technology Palakkad Palakkad Kerala India Infocomm Technology Cluster Singapore Institute of Technology Singapore

ISBN: (数字)9798350384017

ISBN: (纸本)9798350384024

In the recent past, a novel beamforming scheme known as beam multiply and sum (BMAS) which combines the delay and sum (DAS) and filtered delay multiply and sum (F-DMAS) and thus provides a trade-off between computational complexity and output image quality has been proposed by us. The BMAS, despite being shown to provide superior performance in contrast enhancement and lesion detection with single nonsteered plane wave imaging, its performance has not been evaluated with coherent plane wave compounding (CPWC). To this end, an experimental quantitative performance investigation of BMAS with CPWC for different sets of transmit plane wave (PW) angles is presented in this work. The resolution and contrast characteristics of BMAS beamforming with CPWC on a standard in-vitro resolution-distortion and contrast dataset and its efficacy is shown in-vivo for a carotid artery dataset from the plane-wave imaging challenge for medical ultrasound (PICMUS) dataset. We also report a qualitative and quantitative performance comparison between BMAS, DAS, F-DMAS, and minimum variance distortionless response (MVDR) algorithms with CPWC. From the experimental results, it was observed that BMAS with CPWC improves the image quality with fewer PWs which signifies its potential to provide a better trade-off between frame rate and image quality in high frame rate imaging.

关键词： image quality Performance evaluation Ultrasonic imaging image resolution Array signal processing Delays Lesions

来源：评论

学校读者我要写书评

暂无评论

image Classification CNNs using the FINN Engine for SRAM-based APSoC in Satellite Applications

Image Classification CNNs using the FINN Engine for SRAM-bas...

引用

symposium on Integrated Circuits and Systems Design (SBCCI)

作者： Arthur F. Ely Fabio Benevenuti Guilherme Korol Antonio Carlos S. Beck Jose Rodrigo Azambuja Fernanda Lima Kastensmidt Institute of Informatics – PPGC – PGMICRO Federal University of Rio Grande do Sul (UFRGS) Porto Alegre Brazil

ISBN: (数字)9798350391695

ISBN: (纸本)9798350391701

This study investigates the performance of two Convolutional Neural Networks (CNNs) designed for aerial image classification. We modify to these CNNs by adjusting parameters such as quantization and parallelism to facilitate their implementation using the FINN engine optimized for the Zynq-7000 APSoC. The goal is to assess the effects of different quantization levels, network topologies, and architectural parallelism on the area, computing performance, and power consumption within the Zynq APSoC. Results illustrate various trade-offs between resource usage, computational performance, and power consumption, all while meeting predefined accuracy requirements.

关键词： Quantization (signal) Power demand Space missions Parallel processing Throughput Topology Convolutional neural networks Engines System analysis and design image classification

来源：评论

学校读者我要写书评

暂无评论

Mapping DCNN to a Three Layer Modular Architecture: A Systematic Way for Obtaining Wider and More Effective Network

Mapping DCNN to a Three Layer Modular Architecture: A System...

引用

ieee International Conference on signal and image processing (ICSIP)

作者： Chowdhury Chowdhury Kai Su Huitao Wang Qiangfu Zhao The University of Aizu System Intelligence Laboratory Aizu-Wakamatsu Japan

ISBN: (数字)9781728168968

ISBN: (纸本)9781728168975

We propose a modular Deep Convolutional Neural Network (DCNN) architecture which has the property of block-like design and re-usage of parameters by certain blocks. We leverage networks from the ResNet family as the backbone for our proposed architecture. The proposed network architecture is composed of three primary blocks, i) common block ii) hidden blocks and iii) output block. The common block mainly focuses on learning low-level features which are afterward fed to ρ hidden blocks at a time, where ρ is the branching factor. The hidden blocks are placed in a row-wise manner and replicated ρ times depending on the computational availability. The output block takes features as an input from the common block and all the hidden blocks during both training and testing phases. In addition, each of the hidden blocks together with the common block is a stand-alone neural network, which implies that we have ρ neural networks coupled in a single framework. During the training phase, in every single epoch, each of the stand-alone neural networks (i.e. common block + hidden block) is trained in Round Robin fashion followed by fine-tuning the whole network graph. Inference can be performed using either the whole network graph or just any sub-graph of it. To substantiate the effectiveness of our proposed neural architecture we performed experiment on CIFAR-10 and CIFAR-100 datasets.

关键词： Training Systematics Neural networks Computer architecture Network architecture Feature extraction Testing

来源：评论

学校读者我要写书评

暂无评论

3D Point Cloud Enhancement using Unsupervised Anomaly Detection

3D Point Cloud Enhancement using Unsupervised Anomaly Detect...

引用

ieee International symposium on Systems Engineering (ISSE)

作者： Yousra Regaya Fodil Fadli Abbes Amira Department of Computer Science and Engineering Qatar University Doha Qatar Department of Architecture and Urban Planning Qatar University Doha Qatar Faculty of Computing De Montfort University Leicester UK

3D point cloud is increasingly getting attention for perceiving 3D environment which is needed in many emerging applications. This data structure is challenging due to its characteristics and the limitation of the acquisition step which adds a considerable amount of noise. Therefore, enhancing 3D point clouds is a very crucial and critical step. In this paper, we investigate two promising unsupervised techniques which are One-Class SVM (OCSVM) and Isolation Forest (IF). These two techniques optimize the separation between relevant/normal points and irrelevant/noisy points. For evaluation, three metrics are computed, which are the processing time, the number of detected noisy points, and Peak signal-to-Noise (PSNR) in order to compare the both proposed techniques with one of the recommended filters in the literature which is Moving Least Square (MLS) filter. The obtained results reveal promising capability in terms of effectiveness. However, OCSVM technique suffers from high computational time; therefore, its efficiency is enhanced using modern Graphics processing Unit (GPU) with an average rate improvement of 1.8.

关键词： graphics processing units image enhancement least squares approximations object detection support vector machines

来源：评论

学校读者我要写书评

暂无评论

Implementing a self-checking neural system for photon event identification by SRAM-based FPGAs

Implementing a self-checking neural system for photon event ...

引用

ieee International symposium on Defect and Fault Tolerance in VLSI Systems

作者： M. Alderighi S. D'Angelo V. Piuri G.R. Sechi Istituto di Fisica Cosmica G. Occhialini Consiglio Nationale delle Ricerche Milan Italy Dipartimento di Elettronica e Informazione Politecnico di Milano Milan Italy

The paper presents and evaluates the design and the implementation of a self-checking neural system for photon event identification in intensified charge-coupled device detectors. The neural approach reveals more effective than classical algorithmic approaches thanks to its learning through example ability. Implementation is accomplished by SRAM-based FPGAs, which have generated increasing interest in the space community. The adoption of suitable on-line fault detection techniques is illustrated taking into account in a specific way SEU induced faults. The techniques are based on AN coding, particularly 3N coding, which constitutes a reasonable trade-off between circuit complexity and computational delay. Estimations of circuit area overhead and fault coverage are reported.

关键词： Field programmable gate arrays Circuit faults Single event upset Event detection Detectors Complexity theory Charge coupled devices image storage signal processing algorithms Artificial neural 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：