检索结果-内蒙古大学图书馆

High Accurate and Efficient 3D Network for image Reconstruction of Diffractive-Based Computational Spectral Imaging

IEEE ACCESS 2024年 12卷 120720-120728页

作者： Fan, Hao Li, Chenxi Xu, Huangrong Zhao, Lvrong Zhang, Xuming Jiang, Heng Yu, Weixing Chinese Acad Sci Xian Inst Opt & Precis Mech Key Lab Spectral Imaging Technol Xian 710119 Peoples R China Univ Chinese Acad Sci Ctr Mat Sci & Optoelect Engn Beijing 100049 Peoples R China Hong Kong Polytech Univ Dept Appl Phys Hong Kong Peoples R China

Diffractive optical imaging spectroscopy as a promising miniaturized and high throughput portable spectral imaging technique suffers from the problem of low precision and slow speed, which limits its wide use in various applications. To reconstruct the diffractive spectral image more accurately and fast, a three-dimensional spectrum recovery algorithm is proposed in this paper. The algorithm takes advantage of a neural network for image reconstruction which consists of a U-Net architecture with 3D convolutional layers to improve the processing precision and speed. Numerical experiments are conducted to prove its effectiveness. It is shown that the mean peak signal-to-noise ratio (MPSNR) of the recovered image relative to the original image is improved by 1.8 dB in comparison to other traditional methods. In addition, the obtained mean structural similarity (MSSIM) of 0.91 meets the standard of discrimination to human eyes. Moreover, the algorithm runs in just 0.36 s, which is faster than other traditional methods. 3D convolutional networks play a critical role in performance improvement. Improvements in processing speed and accuracy have greatly benefited the realization and application of diffractive optical imaging spectroscopy. The new algorithm with high accuracy and fast speed has a great potential application in diffraction lens spectroscopy and paves a new way for emerging more portable spectral imaging technique.

关键词： Imaging Lenses image reconstruction Three-dimensional displays Accuracy Optical imaging Spectral analysis Inverse problems Diffraction Computational imaging spectral imaging inverse problems diffractive lenses

来源：评论

学校读者我要写书评

暂无评论

Deep Learning and Fuzzy Logic Based Intelligent Technique for the image Enhancement and Edge Detection Framework

引用

TRAITEMENT DU signal 2023年第1期40卷 351-359页

作者： Jawdekar, Anand Dixit, Manish RGPV Dept CSE Bhopal 462033 MP India MITS Dept CSE Gwalior 474005 MP India

Medical imaging is the promising area in digital image processing. Medical images are useful for all types of medical treatment and diagnostics. Medical images are captured through the medical devices, consists some kind of noises and it requires efficient enhancement techniques. Medical imaging also useful in the image segmentation and object detection purposes. Various researcher proposed several types of enhancement techniques and edge detection techniques, but still accuracy and noise are challenge for the enhanced image. So, it is the need of some intelligent techniques to address these issues. In this work we proposed deep learning-based convolution neural network for the image denoising and image enhancement and for the edge detection fuzzy logic-based approach used. The model of DnCNN used here for the image denoising and image enhancement, this model comprises several convolution layers along with input and output layer, this model learns according to the weights and bias. Also, fuzzy logic technique implemented fuzzy inference rules which can give more accurate edges of the image. The result obtained through this hybrid approach is very interesting and effective as compare with previous approaches like histogram-based approach and linear filtering approach. Proposed methods give the promising results as compare with existing methods. All types of simulation performed in MATLAB 2020.

关键词： CNN DnCNN FIS PSNR MSE SSIM image enhancement edge detection

来源：评论

学校读者我要写书评

暂无评论

Vision Transformers for Single image Specular Highlight Detection 32

Vision Transformers for Single Image Specular Highlight Dete...

引用

32nd IEEE signal processing and Communications Applications Conference (SIU)

作者： Karacan, Levent Gaziantep Univ Bilgisayar Muhendisligi Bolumu Gaziantep Turkiye

ISBN: (纸本)9798350388978;9798350388961

Specular highlight detection is a useful task influencing applications such as image analysis and scene understanding. This study investigates using a multi-scale patch-based self-attention mechanism in a deep neural network model for specular highlight detection. Multi-scale patch-based self-attention enhances the model's ability to capture intricate patterns and global dependencies. The proposed method produces highly accurate results on real images with complex specular highlights and is highly competitive with state-of-the-art methods.

关键词： specular highlight detection vision transformers

来源：评论

学校读者我要写书评

暂无评论

An end-to-end deep learning approach for real-time single image dehazing

引用

JOURNAL OF REAL-TIME image processing 2023年第1期20卷 12页

作者： Jeong, Chi Yoon Moon, KyeongDeok Kim, Mooseop Elect & Telecommun Res Inst Artificial Intelligence Res Lab 218 Gajeong Ro Daejeon 34129 South Korea Univ Sci & Technol Artificial Intelligence 217 Gajeong Ro Daejeon 34113 South Korea

image dehazing methods can restore clean images from hazy images and are popularly used as a preprocessing step to improve performance in various image analysis tasks. In recent times, deep learning-based methods have been used to sharply increase the visual quality of restored images, but they require a long computation time. The processing time of image-dehazing methods is one of the important factors to be considered in order not to affect the latency of the main image analysis tasks such as detection and segmentation. We propose an end-to-end network model for real-time image dehazing. We devised a zoomed convolution group that processes computation-intensive operations with low resolution to decrease the processing time of the network model without performance degradation. Additionally, the zoomed convolution group adopts an efficient channel attention module to improve the performance of the network model. Thus, we designed a network model using a zoomed convolution group to progressively recover haze-free images using a coarse-to-fine strategy. By adjusting the sampling ratio and the number of convolution blocks that make up the convolution group, we distributed small and large computational complexities respectively in the early and later operational stages. The experimental results with the proposed method on a public dataset showed a real-time performance comparable to that of another state-of-the-art (SOTA) method. The proposed network's peak-signal-to-noise ratio was 0.8 dB lower than that of the SOTA method, but the processing speed was 10.4 times faster.

关键词： Single-image dehazing Lightweight dehazing networks image restoration End-to-end learning Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Batch-Ensemble stochastic neural Networks for Out-of-Distribution Detection 48

Batch-Ensemble Stochastic Neural Networks for Out-of-Distrib...

引用

48th IEEE International Conference on Acoustics, Speech and signal processing, ICASSP 2023

作者： Chen, Xiongjie Li, Yunpeng Yang, Yongxin University of Surrey United Kingdom Queen Mary University of London United Kingdom

ISBN: (纸本)9781728163277

Out-of-distribution (OOD) detection has recently received much attention from the machine learning community because it is important for deploying machine learning models in real-world applications. In this paper we propose an uncertainty quantification approach by modeling data distributions in feature spaces. We further incorporate an efficient ensemble mechanism, namely batch-ensemble, to construct the batch-ensemble stochastic neural networks (BE-SNNs) and overcome the feature collapse problem. We compare the performance of the proposed BE-SNNs with the other state-of-the-art approaches and show that BE-SNNs yield superior performance on several OOD detection benchmarks, such as the Two-Moons dataset, the FashionMNIST vs MNIST dataset, Fashion-MNIST vs NotMNIST dataset, and the CIFAR10 vs SVHN dataset. © 2023 IEEE.

关键词： ensemble methods Out-of-distribution detection stochastic neural networks

来源：评论

学校读者我要写书评

暂无评论

RCT: Resource Constrained Training for Edge AI

引用

IEEE TRANSACTIONS ON neural NETWORKS AND LEARNING SYSTEMS 2024年第2期35卷 2575-2587页

作者： Huang, Tian Luo, Tao Yan, Ming Zhou, Joey Tianyi Goh, Rick ASTAR Inst High Performance Comp Singapore 138632 Singapore ASTAR Ctr Frontier AI Res Singapore 138632 Singapore

Efficient neural network training is essential for in situ training of edge artificial intelligence (AI) and carbon footprint reduction in general. Train neural network on the edge is challenging because there is a large gap between limited resources on edge and the resource requirement of current training methods. Existing training methods are based on the assumption that the underlying computing infrastructure has sufficient memory and energy supplies. These methods involve two copies of the model parameters, which is usually beyond the capacity of on-chip memory in processors. The data movement between off-chip and on-chip memory incurs large amounts of energy. We propose resource constrained training (RCT) to realize resource-efficient training for edge devices and servers. RCT only keeps a quantized model throughout the training so that the memory requirement for model parameters in training is reduced. It adjusts per-layer bitwidth dynamically to save energy when a model can learn effectively with lower precision. We carry out experiments with representative models and tasks in image classification, natural language processing, and crowd counting applications. Experiments show that on average, 8-15-bit weight update is sufficient for achieving SOTA performance in these applications. RCT saves 63.5%-80% memory for model parameters and saves more energy for communications. Through experiments, we observe that the common practice on the first/last layer in model compression does not apply to efficient training. Also, interestingly, the more challenging a dataset is, the lower bitwidth is required for efficient training.

关键词： Training Quantization (signal) Memory management neural networks Computational modeling System-on-chip Program processors Efficient training memory efficient neural network quantization weight update

来源：评论

学校读者我要写书评

暂无评论

Land use/land cover change classification and prediction using deep learning approaches

引用

signal image AND VIDEO processing 2024年第1期18卷 223-232页

作者： Ebenezer, P. Adlene Manohar, S. SRM Inst Sci & Technol Coll Engn & Technol Dept Comp Sci & Engn Vadapalani Campus Chennai Tamil Nadu India

Nowadays, land use and land cover (LULC) change is a major problem for decision-makers and ecologists on account of its impact on natural ecosystems. In this manuscript, LU/LC change classification and prediction using deep convolutional spiking neural network (DCSNN) and enhanced Elman spike neural network (EESNN) (LU/LC-DCSNN-EESNN) is proposed. The input images are gathered from IRS Satellite Resourcesat-1 LISS-III with Cartosat-1 digital elevation model (DEM) satellite imagery of the Javadi Hills, Tamil Nadu. After that, the images are pre-processed using the fast discrete curvelet transform and wrapping (FDCT-WRP) method is used for extracting the region of interest (ROI) coordinates of Javadi Hills satellite image. Then, for categorizing the area of forest and non-forest, the DCSNN is used. The categorized images are given to post-classification process for eradicating the noise and misclassification errors by Markov chain random field (MCRF) co-simulation approach. The LU/LC changes are predicted using EESNN method. The performance metrics, like precision, accuracy, f1 score, error rate, specificity, recall, kappa coefficient and ROC, are analyzed. The proposed LU/LC-DCSNN-EESNN method has attained 19.45%, 20.56% and 23.67% higher accuracy, 19.45%, 32.56% and 17.45% higher F-measure, and 16.78%, 22.09% and 32.39% lower error rate compared with the existing methods.

关键词： Deep convolutional spiking neural network (DCSNN) Enhanced Elman spike neural network (EESNN) Fast discrete curvelet transform with wrapping method LU/LC prediction Markov chain random field (MCRF)

来源：评论

学校读者我要写书评

暂无评论

TransUMobileNet: Integrating multi-channel attention fusion with hybrid CNN-Transformer architecture for medical image segmentation

引用

BIOMEDICAL signal processing AND CONTROL 2025年 107卷

作者： Cai, Sijing Jiang, Yukun Xiao, Yuwei Zeng, Jian Zhou, Guangming Fujian Univ Technol Sch Transportat Fuzhou 350108 Peoples R China Fujian Univ Technol Sch Elect Elect Engn & Phys Fuzhou 350108 Peoples R China

To address the common issue of strong similarity and blurred boundaries between lesion and normal tissues in medical images, we propose the TransUMobileNet model, which employs a symmetrical encoder-decoder structure. First, the feature encoder uses a hybrid CNN-Transformer architecture, where the Transformer encodes tokenized image patches from convolutional neural network (CNN) feature maps as input sequences for global context extraction. The Transformer's sequence prediction attention mechanism enhances the encoding of long-range dependencies and expressive learning, strengthening global information representation. Second, the feature decoder uses a fully symmetrical encoding form. Through symmetrical skip connections, the loss of positional information in the Transformer decoding path is mitigated, improving the depiction of target boundaries. The feature decoder utilizes cascaded upsampling to restore local spatial information and enhance finer details. Additionally, a Multi-Channel Attention Fusion (MCAF) module is incorporated into the decoding section. This module, characterized by a structure with small channels at both ends and a large one in the middle, along with an attention mechanism, enriches feature information and automatically adjusts weights for key regions, enhancing focus on target areas. TransUMobileNet was evaluated on three different public medical image segmentation datasets and a custom thyroid nodule segmentation dataset. The results show that TransUMobileNet achieves a recall rate of 82.23% and a mean average precision of 95.62%, outperforming current mainstream methods for medical image segmentation.

关键词： Transformer Unet Medical image segmentation Attention Mechanism

来源：评论

学校读者我要写书评

暂无评论

stochastic DEEP image PRIOR FOR MULTISHOT COMPRESSIVE SPECTRAL image FUSION

STOCHASTIC DEEP IMAGE PRIOR FOR MULTISHOT COMPRESSIVE SPECTR...

引用

IEEE International Conference on Acoustics, Speech, and signal processing (ICASSP)

作者： Jacome, Roman Monroy, Brayan Bacca, Jorge Arguello, Henry Univ Ind Santander Bucaramanga 680002 Colombia

ISBN: (纸本)9798350302615

The Deep image Prior (DIP) technique has been successfully employed in Compressive Spectral Imaging (CSI) as a non-data-driven deep model approach. DIP methodology updates the deep network's weights by minimizing a loss function that considers the difference between the measurements and the forward operator of the network's output. However, this method often yields local minima as all the measurements are evaluated at each iteration. This paper proposes a stochastic deep image prior (SDIP) approach, which stochastically trains DIP networks using random subsets of measurements from different CSI sensors in a CSI fusion (CSIF) setting, resulting in the improvement of the convergence through stochastic gradient descent optimization. The proposed SDIP method improves upon the deterministic DIP and requires less computational time since fewer forward operators are required per iteration. The SPID method provides comparable performance against the state-of-the-art CSIF techniques based on supervised data-driven and unsupervised methods, achieving up to 5 dB in the reconstruction.

关键词： Multishot Acquisition Synthetic Aperture Imaging stochastic Gradient Descent Deep image Prior

来源：评论

学校读者我要写书评

暂无评论

A multi-branch multi-scale convolutional neural network using automatic detection of fetal arrhythmia

引用

signal image AND VIDEO processing 2024年第SUPPL 1期18卷 87-96页

作者： Kanna, S. K. Rajesh Shajin, Francis H. Rajesh, P. Mannepalli, Kasiprasad Rajalakshmi Inst Technol Dept Mech Engn Chennai India Xpertmindz Innovat Solut Pvt Ltd Dept Elect & Commun Engn Kuzhithurai Tamil Nadu India Xpertmindz Innovat Solut Pvt Ltd Dept Elect & Elect Engn Kuzhithurai Tamil Nadu India Koneru Lakshmaiah Educ Fdn Dept Elect & Commun Engn Vijayawada Andhra Pradesh India

Fetal arrhythmia can manifest as irregular cardiac rhythm, abnormal heart rate or both irregular cardiac rhythm and abnormal heart rate. Fetal testing without proper care is very dangerous. Therefore, a multi-branch multi-scale convolutional neural network (MMSCNN) using automatic detection of fetal arrhythmia is proposed in this paper. Here, the input ECG signals are amassed from fetal ECG dataset. Then the input signals are preprocessed using the multivariate iterative filtering for removing noise and artifacts. Sparse regularization-based fuzzy C-means clustering and the pre-treated AECG signal sectioned into frames of 100-ms period are considered in the segmentation process. The classification method using multi-branch convolution neural network is classified as normal and arrhythmia. The weight parameter of the MMSCNN is optimized using bald eagle search optimization algorithm. The performance of the proposed method is analyzed with the help of metrics, such as accuracy, precision, ROC, F-score, specificity, recall and error rate analysis. Finally, the proposed MMSCNN-AD-FA method attains higher accuracy compared with existing methods.

关键词： Fetal arrhythmia Multivariate iterative filtering Morlet scattering transform Multi-branch multi-scale convolutional neural network Bald eagle search optimization

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：