检索结果-内蒙古大学图书馆

Low-dose CT image denoising using deep convolutional neural networks with extended receptive fields

signal image AND VIDEO processing 2022年第7期16卷 1963-1971页

作者： Nguyen Thanh Trung Dinh-Hoan Trinh Nguyen Linh Trung Luong, Marie Vietnam Natl Univ Univ Engn & Technol Adv Inst Engn & Technol Hanoi Vietnam Viettel Grp Viettel Cyberspace Ctr Hanoi Vietnam Thai Nguyen Univ Univ Informat & Commun Technol Thai Nguyen Vietnam Univ Paris 13 L2TI Lab Galillee Inst Villetaneuse France

How to reduce radiation dose while preserving the image quality as when using standard dose is an important topic in the computed tomography (CT) imaging domain because the quality of low-dose CT (LDCT) images is often strongly affected by noise and artifacts. Recently, there has been considerable interest in using deep learning as a post-processing step to improve the quality of reconstructed LDCT images. This paper provides, first, an overview of learning-based LDCT image denoising methods from patch-based early learning methods to state-of-the-art CNN-based ones and, then, a novel CNN-based method is presented. In the proposed method, preprocessing and post-processing techniques are integrated into a dilated convolutional neural network to extend receptive fields. Hence, large distance pixels in input images will participate in enriching feature maps of the learned model, leading to effective denoising. Experimental results showed that the proposed method is light, while its denoising effectiveness is competitive with well-known CNN-based models.

关键词： Low-dose CT denoising Dilated convolution Receptive field

来源：评论

学校读者我要写书评

暂无评论

A deep neural network-based end-to-end 3D medical abdominal segmentation and reconstruction model

引用

signal image AND VIDEO processing 2024年第SUPPL 1期18卷 513-522页

作者： Cui, Jin Jiang, Yuhan Kings Coll London Fac Life Sci & Med London England Fujian Univ Chinese Tradit Med Sch Sci Acupuncture Fuzhou Fujian Peoples R China

In recent years, medical imaging-based disease detection and diagnosis have become a mainstream practice in the medical industry. Various computer-aided reconstruction models assist medical practitioners to detect tumors and polyps, thereby diagnosing them more effectively. Deep learning is an emerging computer vision method that has seen tremendous growth in this field, but labeling data in medical imaging is difficult and inefficient, making it extremely expensive to train for supervised learning methods. For unsupervised learning models, the intrinsic logic limits its effectiveness on unlabeled data. In this paper, in order to address this challenge, traditional geometric 3D model reconstruction methods are combined with the latest supervised deep learning models to propose a novel GeoVNet, a semi-supervised model. The proposed GeoVNet model is provided to the intestinal dataset for reconstruction and segmentation. Finally, the experimental evaluations are conducted for various performance metrics, and the precision rate attained is 98.16%, recall rate achieved is 97.9%, IoU as well as dice coefficient obtained are 83.5% as well as 90.9%, respectively. This shows that proposed approach outperforms unlabeled multi-device MRI intestinal imaging methods, and common models such as 3DUNet and DeepLabV3. In addition to this, the ablation analysis is conducted in order to validate and determine the effect of model hyperparameters on segmentation results.

关键词： Deep learning Medical image segmentation VNet, GeoVNet Dataset, 3D model Reconstruction Segmentation

来源：评论

学校读者我要写书评

暂无评论

A Novel CNN-Based Framework for Detection and Classification of Power Quality Disturbances: Exploring Multi-Class Versus Multi-Label Classification

引用

IEEE ACCESS 2025年 13卷 38878-38888页

作者： Zlatkova, Aleksandra Taskovski, Dimitar Ss Cyril & Methodius Univ Skopje Fac Elect Engn & Informat Technol Skopje 1000 North Macedonia

The detection and classification of power quality (PQ) disturbances remains a significant challenge because of the rapid integration of renewable energy sources (RES), widespread use of power electronics, and increasing prevalence of sensitive microcontrollers. These evolving PQ issues necessitate the development of accurate and reliable methods for identifying and classifying PQ disturbances. In this paper, we propose a novel model based on a deep convolutional neural network (DCNN) for the feature extraction and classification of PQ disturbances. The architecture of the model was inspired by the visual geometry group (VGG), which is known for its effectiveness in image processing. The extracted features are highly suitable for both multi-class (MC) and multi-label (ML) classification tasks, effectively addressing the complexity of PQ disturbance signals. The ML approach proved its excellence in the classification of complex PQ disturbances. The performance of the model was rigorously evaluated using various metrics across different scenarios, which demonstrated exceptional accuracy and robustness. The model was trained, validated, and tested using synthetically generated data under different signal-to-noise ratio (SNR) scenarios ensuring its effectiveness in practical applications.

关键词： Convolutional neural network deep learning multi-class classification multi-class classification multi-label classi- fication multi-label classi- fication power quality power quality

来源：评论

学校读者我要写书评

暂无评论

COMPRESSION-AWARE TUNING FOR COMPRESSING VOLUMETRIC RADIANCE FIELDS 31

COMPRESSION-AWARE TUNING FOR COMPRESSING VOLUMETRIC RADIANCE...

引用

2024 International Conference on image processing

作者： Tang, Luyang Zhai, Yongqi Wang, Ronggang Peking Univ Sch Elect & Comp Engn Beijing Peoples R China Peng Cheng Lab Shenzhen Peoples R China

ISBN: (纸本)9798350349405;9798350349399

Reconstructing neural radiance fields with explicit volumetric grids can significantly improve training and inference efficiency. However, these methods store features on voxels, inevitably increasing the cost of storage and transmission for scenes. To solve this limitation, some recent works are inspired by model compression methods, reducing storage costs through pruning and quantization. In this paper, we propose a simple and effective framework called compression-aware vector quantized radiance fields (CA-VQRF), aiming to improve the view synthesis performance of the compressed grid while ensuring the compression ratio. We discover the short-comings of the existing work and propose corresponding solutions. Specifically, we introduce a pruning-aware tuning strategy to shield the impact of the pruned-off voxels in the joint finetuning stage. In addition, a noise-aware tuning strategy is proposed to further compensate for the performance loss caused by vector quantization. Extensive experiments demonstrate that compared with the state-of-the-art compression method, our CA-VQRF can achieve better view synthesis results at the same compression ratio.

关键词： neural Radiance Fields Real-time Rendering Compact Radiance Fields

来源：评论

学校读者我要写书评

暂无评论

LRPS-GCN: A millimeter wave sparse imaging algorithm based on graph signal

引用

IET RADAR SONAR AND NAVIGATION 2024年第10期18卷 1652-1669页

作者： Che, Li Wu, Yongman Jiang, Liubing Mu, Yujie Guilin Univ Elect Technol Sch Informat & Commun Guilin Peoples R China Guilin Univ Elect Technol Key Lab Wireless Broadband Commun & Signal Proc Gu Guilin Peoples R China Guilin Univ Elect Technol Sch Comp & Informat Secur Guilin Peoples R China

Aiming at the problems of slow speed and poor accuracy of traditional millimeter wave sparse imaging, a sparse imaging algorithm based on graph convolution model is proposed from the perspective of sparse signal recovery. The graph signal model is constructed by combining the low-rank and piecewise smoothing(LRPS) regular terms, based on which the proximal operator is replaced by the denoising graph convolution network, and the graph convolution sparse reconstruction network LRPS-GCN is constructed, and the recovered target image is obtained by iterating with the optimal non-linear sparse variation. For the proposed algorithm, simulation experiments are carried out using synthetic datasets under different target densities, iteration times and noise environments, and compared with the traditional graph signal reconstruction algorithm and the deep compressed sensing reconstruction algorithm, and then use the measured data with varying degrees of sparsity to validate. The experimental results show that the reconstructed images by this algorithm have better performance in terms of normalised mean square error, target to background ratio, reconstruction time and memory usage. To address the slow speed and poor accuracy issues of traditional millimeter wave sparse imaging, a novel sparse imaging algorithm based on graph convolution model is introduced, utilising low-rank and piecewise smoothing regular terms to construct the graph signal model and employing a denoising graph convolution network for sparse reconstruction. Simulation experiments on synthetic datasets and real measured data demonstrate that this algorithm significantly outperforms traditional and deep compressed sensing reconstruction methods in normalised mean square error, target to background ratio, reconstruction time, and memory usage. image

关键词： convolutional neural nets graph theory microwave imaging signal reconstruction

来源：评论

学校读者我要写书评

暂无评论

DBCTfuse: A Dual-Branch CNN-Transformer feature extraction network for Infrared and Visible image Fusion 17

DBCTfuse: A Dual-Branch CNN-Transformer feature extraction n...

引用

17th IEEE International Conference on signal processing, ICSP 2024

作者： Li, Yujing Ma, Xiaole Chen, Xiaoxuan Hu, Shaohai Beijing Jiaotong University School of Computer Science and Technology Beijing100044 China

ISBN: (纸本)9798350387384

The process of fusing infrared and visible images necessitates integrating thermal radiation information from infrared image with the edge and texture detail captured by visible images. Most current fusion methods are based on convolutional neural networks, which can effectively extract local features from source images but are somewhat lacking in global features. To address this, we propose a dual-branch feature extraction fusion network based on convolutional neural network and Transformer - DBCTfuse, to effectively synthesize both local details and global contextual information from source images. Comprehensive experiments conclusively demonstrate the superiority of our DBCTfuse method in infrared and visible image fusion, surpassing contemporary state-of-the-art algorithms in both qualitative and quantitative assessments. © 2024 IEEE.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Investigating the Use of neural Networks in image and Video processing for Improved Media Quality

Investigating the Use of Neural Networks in Image and Video ...

引用

2024 OPJU International Technology Conference on Smart Computing for Innovation and Advancement in Industry 4.0, OTCON 2024

作者： Jaiswal, Manuraj Choubey, Abha Choubey, Siddhartha Jaiswal, Dipti Galgotias University Department of Computer Science and Engineering Uttarpradesh Greater Noida201310 India Shri Shankaracharya Technical Campus Department of Computer Science and Engineering Chhattisgarh Durg491001 India Noida International University Department of Computer Science and Engineering Uttar Pradesh Greater Noida201310 India

ISBN: (纸本)9798350373783

Today, in the digital age, high-quality video and pictures are more important than ever in many areas, from medical images and monitoring to fun and multimedia. This study looks into how neural networks can be used in picture and video processing to make media better. New technologies like deep learning and neural network designs have made it possible for computers to fix and improve video content to a whole new level. Getting rid of image noise, improving sharpness, and denoising videos are all common issues that this study aims to look at how well neural networks deal with them. Part of this research is coming up with new ways to use generative adversarial networks (GANs), reinforcement learning techniques, convolutional and recurrent neural networks, and more. As part of the investigation, we compare the performance of the suggested methods to a group of well-known image and video processing methods. These include Bicubic Interpolation, Histogram Equalization, Wavelet Denoising, Non-Local Means Denoising, Sharpening Filters, and Dehazing and Contrast Enhancement. As part of the performance measures, signal-to-noise ratios, picture quality indices, and studies on how people see things are used. The big steps forward made possible by neural networks are shown by our results. image and video quality, noise reduction, and clarity are all much better when the suggested methods are used instead of more common ones in a number of test cases. There are many ways that neural networks could completely change how picture and video processing is done. This study shows some of these ways. © 2024 IEEE.

关键词： Deep reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

ECA-ATCNet: Efficient EEG Decoding with Spike Integrated Transformer Conversion for BCI Applications

引用

APPLIED SCIENCES-BASEL 2025年第4期15卷 1894-1894页

作者： Li, Xuhang Shen, Qianzi Wang, Haitao Wang, Zijian Donghua Univ Sch Comp Sci & Technol Shanghai 201620 Peoples R China China Mobile Shanghai Ind Res Inst Shanghai 201206 Peoples R China

The Brain-Computer Interface (BCI) has applications in smart homes and healthcare by converting EEG signals into control commands. However, traditional EEG signal decoding methods are affected by individual differences, and although deep learning techniques have made significant breakthroughs, challenges such as high energy consumption and the processing of raw EEG data remain. This paper introduces the Efficient Channel Attention Temporal Convolutional Network (ECA-ATCNet) to enhance feature learning by applying Efficient Channel Attention Convolution (ECA-conv) across spatial and spectral dimensions. The model outperforms state-of-the-art methods in both within-subject and between-subject classification tasks on MI-EEG datasets (BCI-2a and PhysioNet), achieving accuracies of 87.89% and 71.88%, respectively. Additionally, the proposed Spike Integrated Transformer Conversion (SIT-conversion) method, based on Spiking-Softmax, converts the Transformer's self-attention mechanism into Spiking neural Networks (SNNs) in just 12 time steps. The accuracy loss of the converted ECA-ATCNet model is only 0.6% to 0.73%, while its energy consumption is reduced by 52.84% to 53.52%. SIT-conversion enables ultra-low-latency, near-lossless ANN-to-SNN conversion, with SNNs achieving similar accuracy to their ANN counterparts on image datasets. Inference energy consumption is reduced by 18.18% to 45.13%. This method offers a novel approach for low-power, portable BCI applications and contributes to the advancement of energy-efficient SNN algorithms.

关键词： brain-computer interface MI-EEG efficient channel attention spiking neural networks ANN-to-SNN conversion

来源：评论

学校读者我要写书评

暂无评论

Automated epilepsy seizure detection from EEG signal based on hybrid CNN and LSTM model

引用

signal image AND VIDEO processing 2023年第4期17卷 1113-1122页

作者： Pandey, Saroj Kumar Janghel, Rekh Ram Mishra, Pankaj Kumar Ahirwal, Mitul Kumar GLA Univ Mathura Chaumuhan India Natl Inst Technol Raipur Madhya Pradesh India Rungta Coll Engn & Technol Bhilai Bhilai India Maulana Azad Natl Inst Technol Bhopal Bhopal India

Epilepsy is a neurological disorder that affects the normal functioning of the brain. More than 10% of the population across the globe is affected by this disorder. Electroencephalogram (EEG) is prominently employed to accumulate information about the brain's electrical activity. This study proposes an end-to-end system using a combination of two deep learning models Convolutional neural Networks (CNNs) and Long Short-Term Memory Networks (LSTM) for the classification of EEG signals of epilepsy disordered subjects into three classes, namely preictal, normal, and seizure. The experimental results are obtained using the publicly available and popular Bonn University dataset. In this CNN-LSTM classification model the feature extraction, selection, and classification tasks are performed automatically without using handcrafted feature extraction methods. The performance of the CNN-LSTM model is examined and evaluated in terms of specificity, sensitivity, and accuracy using the tenfold cross-validation approach. The experiments performed and the obtained results show the accuracy of 99.33%, sensitivity of 99.33%, and specificity of 99.66%, respectively. Our results highlight that deep learning methods are best suited for classification in comparison to other existing state-of-the-art methods.

关键词： Long short-term memory Epilepsy EEG signals Convolutional-neural-network Classification Deep learning Seizure Accuracy

来源：评论

学校读者我要写书评

暂无评论

Deep learning approach for identification of H II regions during reionization in 21-cm observations - II. Foreground contamination

引用

MONTHLY NOTICES OF THE ROYAL ASTRONOMICAL SOCIETY 2024年第3期528卷 5212-5230页

作者： Bianco, Michele Giri, Sambit K. Prelogovic, David Chen, Tianyue Mertens, Florent G. Tolley, Emma Mesinger, Andrei Kneib, Jean-Paul Ecole Polytech Fed Lausanne EPFL Lab Astrophys Observ Sauverny CH-1290 Versoix Switzerland Univ Sussex Astron Ctr Dept Phys & Astron Pevensey 3 Bldg Brighton BN1 9QH England Nordita KTH Royal Inst Technol Hannes Alfvens Vag 12 SE-10691 Stockholm Sweden Stockholm Univ Hannes Alfvens Vag 12 SE-10691 Stockholm Sweden Univ Zurich Inst Computat Sci Winterthurerstr 190 CH-8057 Zurich Switzerland Scuola Normale Super Pisa Piazza Cavalieri 7 I-56126 Pisa Italy Sorbonne Univ PSL Res Univ CNRS LERMAObserv Paris F-75014 Paris France

The upcoming Square Kilometre Array Observatory will produce images of neutral hydrogen distribution during the epoch of reionization by observing the corresponding 21-cm signal. However, the 21-cm signal will be subject to instrumental limitations such as noise and galactic foreground contamination that pose a challenge for accurate detection. In this study, we present the SegU-Net v2 framework, an enhanced version of our convolutional neural network, built to identify neutral and ionized regions in the 21-cm signal contaminated with foreground emission. We trained our neural network on 21-cm image data processed by a foreground removal method based on Principal Component Analysis achieving an average classification accuracy of 71 per cent between redshift z = 7 and 11. We tested SegU-Net v2 against various foreground removal methods, including Gaussian Process Regression, Polynomial Fitting, and Foreground-Wedge Removal. Results show comparable performance, highlighting SegU-Net v2's independence on these pre-processing methods. Statistical analysis shows that a perfect classification score with AUC = 95 is possible for 8 < z < 10. While the network prediction lacks the ability to correctly identify ionized regions at higher redshift and differentiate well the few remaining neutral regions at lower redshift due to low contrast between 21-cm signal, noise, and foreground residual in images. Moreover, as the photon sources driving reionization are expected to be located inside ionized regions, we show that SegU-Net v2 can be used to correctly identify and measure the volume of isolated bubbles with V-ion > (10cmpc)(3 )at z > 9, for follow-up studies with infrared/optical telescopes to detect these sources.

关键词： techniques: image processing techniques: interferometric dark ages, reionization, first stars early Universe

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：