检索结果-内蒙古大学图书馆

3rd International Conference on VLSI, Communication and signal processing, VCAS 2020

作者： Mohan, Amrita Dwivedi, Ramji Kumar, Basant MNNIT Allahabad Uttar Pradesh Prayagraj India

ISBN: (纸本)9789811627606

In the proposed work, a CNN-based Super-Resolution Convolutional neural Network (SRCNN) framework for landslide photographs is used for resolution improvement. SRCNN method is based on a deep convolutional neural network, and it is a fast-moving field with several practical applications in different areas. The main aim of SRCNN is to construct a high-resolution image from low-resolution landslide images. For conducting this work, free landslide photographs are used from various search engines. By using Peak signal to Noise Ratio (PSNR), Mean squared error (MSE), and Structural Similarity (SSIM) index, performance of the proposed algorithm is evaluated on landslide photographs in a meaningful way. The proposed work involves following steps: (i) Pre-processing: Upscaling low-resolution (LR) landslide image resized to the desired size using bilinear interpolation, (ii) Activation Feature Extraction: Set of convolved feature map is extracted from the upscaled LR landslide image, (iii) Multidimensional Scaling: Maps the activation map representing LR landslide image to high-resolution (HR) image, (iv) Reconstruction: Producing the HR landslide image from HR patches. In this work for quantitative evaluation of super-resolution quality, two most commonly used parameters PSNR and SSIM are evaluated. Higher values of PSNR and SSIM suggest that the reconstructed photographs are of enhanced quality. The experimental results on landslide photographs demonstrate that using bilinear interpolation for low-resolution landslide photographs to intensify its target size and then utilizing three CNN layers to do nonlinear mapping significantly improve the reconstruction quality of landslide photographs. In terms of PSNR and SSIM values, satisfactory reconstruction result is achieved as compared to the traditional methods. The obtained improved resolution landslide photographs manifest noticeable qualitative enhancement. © 2022, The Author(s), under exclusive license to Springer Nature

关键词： image reconstruction

来源：评论

学校读者我要写书评

暂无评论

Recognition of Radar Emitter signal images Using encoding signal methods

Recognition of Radar Emitter Signal Images Using encoding si...

引用

International Conference on neural Networks, Information and Communication Engineering

作者： Zhang, Shengli Pan, Jifei Han, Zhenzhong Qiu, Risheng Natl Univ Def Technol Elect Countermeasure Inst Hefei 230037 Anhui Peoples R China

ISBN: (数字)9781510647381

ISBN: (纸本)9781510647381;9781510647374

We innovatively apply three different methods of encoding signal as 2-D plots to radar emitter signal recognition: Recurrence Plots (RP). Gramian Angular Field (GAF) and Markov Transition Field (MTF), thus the radar emitter signal recognition problem is converted into image processing problem. We build a 2-stage convolutional neural network (CNN) model to make use of its mature technology in the field of computer vision and image processing for signal recognition. These pipeline offers the following advantages: i) Encoding signal methods enable us to visualize certain aspects of the radar emitter signals through 2D images, and ii) CNN can automatically learn different levels of radar signal features and achieve a high recognition rate. In order to get a best encoding method, we compare the recognition accuracy of the three methods mentioned above, and combine the images encoded by the first two methods (RP and GAF) with high recognition accuracy into 2-channel RP-GAF images and get a better recognition rate. At last, we analyze the reasons why three encoding methods achieve different recognition rates.

关键词： radar emitter signal recognition encoding signal methods deep learning

来源：评论

学校读者我要写书评

暂无评论

Deep neural Network Models Trained with a Fixed Random Classifier Transfer Better Across Domains

Deep Neural Network Models Trained with a Fixed Random Class...

引用

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

作者： Hafiz Tiomoko Ali Umberto Michieli Ji Joong Moon Daehyun Kim Mete Ozay Samsung Research UK Samsung Research Korea

The recently discovered neural collapse (NC) phenomenon states that the last-layer weights of Deep neural Networks (DNN), converge to the so-called Equiangular Tight Frame (ETF) simplex, at the terminal phase of their training. This ETF geometry is equivalent to vanishing within-class variability of the last layer activations. Inspired by NC properties, we explore in this paper the transferability of DNN models trained with their last layer weight fixed according to ETF. This enforces class separation by eliminating class covariance information, effectively providing implicit regularization. We show that DNN models trained with such a fixed classifier significantly improve transfer performance, particularly on out-of-domain datasets. On a broad range of fine-grained image classification datasets, our approach outperforms i) baseline methods that do not perform any covariance regularization (up to 22%), as well as ii) methods that explicitly whiten covariance of activations throughout training (up to 19%). Our findings suggest that DNNs trained with fixed ETF classifiers offer a powerful mechanism for improving transfer learning across domains.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Micro-Expression Recognition Method Based on Dual-Stream Network and Fused Differential

Micro-Expression Recognition Method Based on Dual-Stream Net...

引用

IEEE International Conference on signal and image processing (ICSIP)

作者： Shicheng Lu Runcai Huang Shanghai University of Engineering Science Shanghai China

ISBN: (数字)9798350350920

ISBN: (纸本)9798350350937

Micro-expressions have the characteristics of short duration and low range of motion, making them difficult to recognize with the naked eye. Traditional networks only focus on image features and do not learn information about changes in faces. Using a 3D convolutional neural network to add the time dimension can perform spatiotemporal feature learning. However, the number of micro-expression data sets is small, and the 3D convolutional neural network has limited learning ability for small sample data sets, and usually falls into the local optimum, resulting in poor model performance and affecting recognition efficiency. In order to effectively utilize the spatiotemporal feature information of the data and the small amount of data, we propose a method based on a dual-stream network and fusion difference. This method effectively integrates optical flow time information and top-frame spatial information using a dual-stream network. Experiments were conducted on CASME ii and SAMM datasets. The results show that the values of UF1 and UAR of our proposed method are 0.9304 and 0.9306 respectively on CASME ii, and 0.7932 and 0.7948 respectively on SAMM, outperforming several popular methods in terms of recognition performance.

关键词： Representation learning Solid modeling Three-dimensional displays image recognition Training data Feature extraction Spatiotemporal phenomena

来源：评论

学校读者我要写书评

暂无评论

Model Based and Physics Informed Deep Learning neural Network Structures

arXiv

引用

arXiv 2024年

作者： Mohammad-Djafari, Ali Chu, Ning Wang, Li Cai, Caifang Yu, Liang Bures sur Yvette91440 France Zhejiang Shangfeng special blower company Shaoxing312352 China Central South University Changsha China Khalifa University Abu Dhabi United Arab Emirates School of Civil Aviation Northwestern Polytechnical Univ. Xian710072 China State Key Laboratory of Airliner Integration Technology and Flight Simulation Shanghai200126 China

neural Networks (NN) has been used in many areas with great success. When a NN’s structure (Model) is given, during the training steps, the parameters of the model are determined using an appropriate criterion and an optimization algorithm (Training). Then, the trained model can be used for the prediction or inference step (Testing). As there are also many hyperparameters, related to the optimization criteria and optimization algorithms, a validation step is necessary before its final use. One of the great difficulties is the choice of the NN’s structure. Even if there are many"on the shelf" networks, selecting or proposing a new appropriate network for a given data, signal or image processing, is still an open problem. In this work, we consider this problem using model based signal and image processing and inverse problems methods. We classify the methods in five classes, based on: i) Explicit analytical solutions, ii) Transform domain decomposition, iii) Operator Decomposition, iv) Optimization algorithms unfolding, and v) Physics Informed NN methods (PINN). Few examples in each category are explained. © 2024, CC BY.

关键词： neural network models

来源：评论

学校读者我要写书评

暂无评论

ADVERSARIAL TRAINING WITH stochastic WEIGHT AVERAGE

ADVERSARIAL TRAINING WITH STOCHASTIC WEIGHT AVERAGE

引用

IEEE International Conference on image processing (ICIP)

作者： Hwang, Joong-won Lee, Youngwan Oh, Sungchan Bae, Yuseok Elect & Telecommun Res Inst Daejeon South Korea

ISBN: (纸本)9781665441155

Although adversarial training is the most reliable method to train robust deep neural networks so far, adversarially trained networks still show large gap between their accuracies on clean images and those on adversarial images. In conventional classification problem, one can gain higher accuracy by ensembling multiple networks. However, in adversarial training, there are obstacles to adopt such ensemble method. First, as inner maximization is expensive, training multiple networks adversarially becomes overburden. Moreover, the naive ensemble faces dilemma on choosing target model to generate adversarial examples with. Training adversarial examples of the members causes covariate shift, while training those of ensemble diminishes the benefit of ensembling. With these insights, we adopt stochastic weight average methods and improve it by considering overfitting nature of adversarial training. Our method take the benefit of ensemble while avoiding the described problems. Experiments on CIFAR10 and CIFAR100 shows our method improves the robustness effectively.

关键词： Adversarial training stochastic Weight Average Ensemble Robustness Hard Example Mining

来源：评论

学校读者我要写书评

暂无评论

Source localization in resource-constrained sensor networks based on deep learning

引用

neural COMPUTING & APPLICATIONS 2021年第9期33卷 4217-4228页

作者： Javadi, S. Hamed Guerrero, Angela Mouazen, Abdul M. Univ Ghent Fac Biosci Engn Dept Environm B-9000 Ghent Belgium

Source localization with a network of low-cost motes with limited processing, memory, and energy resources is considered in this paper. The state-of-the-art methods are mostly based on complicated signal processing approaches in which motes send their (processed) data to a fusion center (FC) wherein the source is localized. These methods are resource-demanding and mostly do not meet the limitations of motes and network. In this paper, we consider distributed detection where each mote performs a binary hypothesis test to detect locally the existence of a desired source and sends its (potentially erroneous) decision to FC during just one bit (1 indicates source existence and 0 otherwise). Hence, both processing and bandwidth constraints are met. We propose to use an artificial neural network (ANN) to correct erroneous local decisions. After error correction, the region affected by the source is specified by nodes with decision 1. Moreover, we propose to localize the source by deep learning in FC which converts the network of decisions 1 and 0 to a black and white image with white pixels in the locations of motes with decision 1. The proposed schemes of error correction by ANN (ECANN) and source localization with deep learning (SoLDeL) were evaluated in a fire detection application. We showed that SoLDeL performs appropriately and scales well into large networks. Moreover, the applicability of ECANN in delineation of farm management zones was illustrated.

关键词： Artificial neural network (ANN) Decentralized detection Deep learning Error type I Error type ii Internet of things (IoT) Source localization Target tracking Wireless sensor networks (WSN)

来源：评论

学校读者我要写书评

暂无评论

Correction of geometric artifact in cone-beam computed tomography through a deep neural network

引用

APPLIED OPTICS 2021年第7期60卷 1843-1850页

作者： Xiao, Kai Yan, Bin Natl Key Lab Sci & Technol Blind Signal Proc Chengdu 610000 Peoples R China Natl Digital Switching Syst Engn & Technol Res Ct Zhengzhou 450001 Peoples R China

Cone-beam computed tomography is a noninvasive detection system that can obtain the three-dimensional structure of objects in a way that does not damage the object. It is widely applied in precision instruments, medical detection, and other fields. However, in the actual process, if a geometric artifact appears in the results, it will affect the quality of reconstructed images, including detail loss and decreased spatial resolution, which leads to inaccurate distinction of defects in detection. We propose a method for correcting a geometric artifact by means of data-driven projection and neural networks. The network designed is a deep neural network with six convolutional layers and six deconvolutional layers that can correct a geometric artifact by training a large number of labeled data and unlabeled data. Compared with other networks that require prior information for reconstructed images, the proposed method uses a projection data-driven approach that can avoid the requirement for prior information. The simulation data have been tested under varying degrees of noise, and satisfactory geometric artifact correction results have been obtained. Meanwhile, we use the actual data of line pairs and ball grid array solder joints to conduct experiments. The results obtained by our method are compared with two other phantom-based method and the U-net method, respectively. The results of similarity and spatial resolution show that the proposed method can achieve the comparable results as the two types of methods. At the same time, we apply a projection data-driven approach to avoid the requirement for prior information, which is more conducive to the correction of the geometric artifact in practical situations where prior information is lacking. (C) 2021 Optical Society of America

关键词： Computed tomography image processing image quality neural networks Spatial resolution stochastic gradient descent

来源：评论

学校读者我要写书评

暂无评论

Bio-medical image Denoising using Autoencoders 2

Bio-medical Image Denoising using Autoencoders

引用

2nd International Conference on Next Generation Intelligent Systems, ICNGIS 2022

作者： Thomas, Jintu Merin Ameenudeen, P.E. Electronics and Communication Engineering College of Engineering Trivandrum Trivandrum India

ISBN: (纸本)9781665467926

image denoising, in which the original image is reconstructed by eliminating noise from a noise-corrupted version of the original image, is one of the greatest requirements in the image processing field. The sources of noise in image may be in-trinsic (such as sensors) and extrinsic (such as environment) that are often practically unavoidable. One of the crucial initial steps in biomedical image processing and analysis is image denoising as improved visual quality of images may increase the accuracy of medical diagnosis and image preprocessing is required at the input for analytical methods like segmentation and content recognition. Autoencoder is an artificial neural network that is trained in the unsupervised manner to reconstruct its input at the output. This work aims at the development of a denoising convolutional autoencoder which will first generate the encoded lower-dimensional representation of the image and then decodes the image back from the lower-dimensional representation. The denoising autoencoders build corrupted copies of the input images by adding random noise and then attempt to eliminate the noise from the noisy input to reconstruct the output that is very similar to the original input. Medical images like X-Ray, Magnetic Resonance Imaging(MRI) etc. which are corrupted by various types of noise like Gaussian noise, Rician noise etc. are denoised using the convolutional denoising autoencoder and their denoising efficiency is measured using the Structural Similarity Index Measure(SSIM) parameter of both the noisy image and the denoised image compared to the original noise free image and the Peak signal to Noise Ratio(PSNR) of the denoised image with the noise free image as reference. © 2022 IEEE.

关键词： image denoising

来源：评论

学校读者我要写书评

暂无评论

Physics-informed deep learning for fringe pattern analysis

引用

Opto-Electronic Advances 2024年第4期7卷 4-15页

作者： Wei Yin Yuxuan Che Xinsheng Li Mingyu Li Yan Hu Shijie Feng Edmund Y.Lam Qian Chen Chao Zuo Smart Computational Imaging Laboratory(SCILab) School of Electronic and Optical EngineeringNanjing University of Science and TechnologyNanjing 210094China Smart Computational Imaging Research Institute(SCIRI)of Nanjing University of Science and Technology Nanjing 210019China Jiangsu Key Laboratory of Spectral Imaging&Intelligent Sense Nanjing 210094China Department of Electrical and Electronic Engineering The University of Hong KongPokfulamHong Kong SAR 999077China

Recently,deep learning has yielded transformative success across optics and photonics,especially in optical *** neural networks (DNNs) with a fully convolutional architecture (e.g.,U-Net and its derivatives) have been widely implemented in an end-to-end manner to accomplish various optical metrology tasks,such as fringe denoising,phase unwrapping,and fringe ***,the task of training a DNN to accurately identify an image-to-image transform from massive input and output data pairs seems at best naive,as the physical laws governing the image formation or other domain expertise pertaining to the measurement have not yet been fully exploited in current deep learning *** this end,we introduce a physics-informed deep learning method for fringe pattern analysis (PI-FPA) to overcome this limit by integrating a lightweight DNN with a learning-enhanced Fourier transform profilometry (Le FTP) *** parameterizing conventional phase retrieval methods,the Le FTP module embeds the prior knowledge in the network structure and the loss function to directly provide reliable phase results for new types of samples,while circumventing the requirement of collecting a large amount of high-quality data in supervised learning *** by the initial phase from Le FTP,the phase recovery ability of the lightweight DNN is enhanced to further improve the phase accuracy at a low computational cost compared with existing end-to-end *** results demonstrate that PI-FPA enables more accurate and computationally efficient single-shot phase retrieval,exhibiting its excellent generalization to various unseen objects during *** proposed PI-FPA presents that challenging issues in optical metrology can be potentially overcome through the synergy of physics-priors-based traditional tools and data-driven learning approaches,opening new avenues to achieve fast and accurate single-shot 3D imaging.

关键词： optical metrology deep learning physics-informed neural networks fringe analysis phase retrieval

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：