检索结果-内蒙古大学图书馆

Optimized Parameter-Efficient Deep Learning Systems via Reversible Jump Simulated Annealing

IEEE JOURNAL OF SELECTED TOPICS IN signal processing 2024年第6期18卷 1010-1023页

作者： Marsh, Peter Kuruoglu, Ercan Engin Univ Town Shenzhen Tsinghua Berkeley Shenzhen Inst Tsinghua Shenzhen Int Grad Sch Shenzhen 518055 Peoples R China

We utilize the non-convex optimization method simulated annealing enriched with reversible jumps to enable a model selection capacity for deep learning models in a model size aware context. By using simulated annealing enriched with reversible jumps, we can yield a robust stochastic learning of the hidden posterior distribution of the structure, simultaneously constructing a more focused and certain estimate of the structure, all while making use of all the data. Being based upon Markov-chain learning methods, we constructed our priors to favor smaller and simpler architectures, allowing us to converge on the set of globally optimal models that are additionally parameter-efficient, seeking low parameter count deep models that retain good predictive accuracy. We demonstrate the capability on standard image recognition with CIFAR-10, as well as performing model selection on time-series tasks, realizing networks with competitive performance as compared to competing non-convex optimization methods such as genetic algorithms, random search, and Gaussian process based Bayesian optimization, while being less than half the size.

关键词： Simulated annealing Optimization neural networks Long short term memory image recognition Task analysis Data models Deep learning systems model selection reversible jump Markov chain Monte Carlo simulated annealing

来源：评论

学校读者我要写书评

暂无评论

A machine learning-based feature extraction method for image classification using ResNet architecture

引用

DIGITAL signal processing 2025年 160卷

作者： Liao, Jing Guo, Linpei Jiang, Lei Yu, Chang Liang, Wei Li, Kuanching Pop, Florin Hunan Univ Sci & Technol Sch Comp Sci & Engn Xiangtan 411100 Peoples R China Hunan Univ Sci & Technol Sanya Res Inst Sanya 572024 Peoples R China Hunan Key Lab Serv Comp & Novel Software Technol Xiangtan 411201 Peoples R China Natl Univ Sci & Technol Politehn Bucharest Bucharest 060042 Romania Natl Inst Res & Dev Informat ICI Bucharest 011455 Romania Acad Romanian Scientists Bucharest 050044 Romania

With the development of Deep Learning, Convolutional neural Networks (CNNs) have become a mainstream method for image classification, and the emergence of the ResNet architecture has significantly accelerated this process. However, as model depth increases, feature redundancy limits model performance. Although traditional machine learning methods like Principal Component Analysis (PCA) can effectively remove redundancy features, there is no effective method to integrate PCA as a feature extraction technique into different convolutional neural network architectures. This work proposes a Selective Principal Component Layer (SPCL), a feature extraction method that effectively incorporates PCA into convolutional neural networks to filter essential features and improve the feature representation ability of deep learning models. SPCL is applied to ResNet architecture models to reduce redundant features and enhance generalization performance in image classification tasks. Evaluations on CIFAR-10 and Tiny imageNet datasets demonstrate its effectiveness. The results show SPCL can be generally applied to ResNet architecture models and improve model accuracy, balancing the improvement of model performance and stability without adding significant computational overhead, demonstrating its potential to enhance performance in complex image classification tasks.

关键词： Deep learning Machine learning Principal component analysis Feature extraction image classification

来源：评论

学校读者我要写书评

暂无评论

MSAByNet: A multiscale subtraction attention network framework based on Bayesian loss for medical image segmentation

引用

BIOMEDICAL signal processing AND CONTROL 2025年 103卷

作者： Zhao, Longxuan Wang, Tao Chen, Yuanbin Zhang, Xinlin Tang, Hui Zong, Ruige Tan, Tao Chen, Shun Tong, Tong Fuzhou Univ Coll Phys & Informat Engn Fuzhou Peoples R China Fuzhou Univ Fujian Key Lab Med Instrumentat & Pharmaceut Techn Fuzhou Peoples R China Imperial Vis Technol Fuzhou Peoples R China Macao Polytech Univ Macau Peoples R China Fujian Med Univ Union Hosp Dept Ultrasound Fuzhou Peoples R China Fujian Med Ultrasound Res Inst Fuzhou Peoples R China

Medical image segmentation is a critical and complex process in medical image processing and analysis. With the development of artificial intelligence, the application of deep learning in medical image segmentation is becoming increasingly widespread. Existing techniques are mostly based on the U-shaped convolutional neural network and its variants, such as the U-Net framework, which uses skip connections or element-wise addition to fuse features from different levels in the decoder. However, these operations often weaken the compatibility between features at different levels, leading to a significant amount of redundant information and imprecise lesion segmentation. The construction of the loss function is a key factor in neural network design, but traditional loss functions lack high domain generalization and the interpretability of domain-invariant features needs improvement. To address these issues, we propose a Bayesian loss-based Multi-Scale Subtraction Attention Network (MSAByNet). Specifically, we propose an inter-layer and intra-layer multi-scale subtraction attention module, and different sizes of receptive fields were set for different levels of modules to avoid loss of feature map resolution and edge detail features. Additionally, we design a multi-scale deep spatial attention mechanism to learn spatial dimension information and enrich multi-scale differential information. Furthermore, we introduce Bayesian loss, re-modeling the image in spatial terms, enabling our MSAByNet to capture stable shapes, improving domain generalization performance. We have evaluated our proposed network on two publicly available datasets: the BUSI dataset and the Kvasir-SEG dataset. Experimental results demonstrate that the proposed MSAByNet outperforms several state-of-the-art segmentation methods. The codes are available at https://***/zlxokok/MSAByNet.

关键词： Deep learning Medical image segmentation Multi-scale processing Deep convolutional neural networks Bayesian loss

来源：评论

学校读者我要写书评

暂无评论

Enhanced function approximation and applications to image scaling: A new family of exponential sampling neural network Kantorovich operators

引用

APPLIED SOFT COMPUTING 2025年 174卷

作者： Agrawal, P. N. Baxhaku, Behar Berisha, Artan Inst Technol Roorkee Dept Math Roorkee 247667 India Univ Prishtina Dept Math Prishtina Kosovo

This paper introduces a novel family of exponential sampling type neural network Kantorovich operators, extending the work of Bajpeyi and Kumar (2021) and Bajpeyi (2023). Unlike previous research focused on approximating continuous functions, our operators are designed to handle Lebesgue integrable functions, offering enhanced versatility. We establish convergence theorems, analyze asymptotic behavior, and demonstrate the effectiveness of linear combinations for improving convergence rates. Our analysis extends to the multivariate setting, highlighting the operators' capability in approximating a wide range of functions. To evaluate the practical performance of our proposed operators, we conducted numerical experiments with different sigmoidal functions and parameter values. Our findings reveal that operators activated by the Parametric sigmoid function consistently outperform those activated by other sigmoidal functions, achieving up to 20.70% reduction in maximum absolute error and 10.03% reduction in root mean squared errors. When applied to image scaling, our operators demonstrated superior performance compared to state-of-the-art methods like nearest neighbor, bilinear, and bicubic interpolation. For the 'Baboon' image, we observed up to 5.62% increase in Peak signal- to-Noise Ratio (PSNR) and 5.25% increase in Structural Similarity Index Measure (SSIM). Similar enhancements were observed for the 'Flowers' and 'Retina' images. The paper includes a detailed description of the image processing algorithm, along with a flowchart illustrating the implementation. These results underscore the operators' potential in various machine learning tasks, motivating further research into their applications and optimization.

关键词： Sigmoidal function Mellin-Taylor formulae Logarithmic modulus of continuity Asymptotic expansion neural network kantorovich operators Digital image processing Peetre's K- functional

来源：评论

学校读者我要写书评

暂无评论

Unsupervised Angularly Consistent 4D Light Field Segmentation Using Hyperpixels and a Graph neural Network

IEEE OPEN JOURNAL OF SIGNAL PROCESSING

引用

IEEE OPEN JOURNAL OF signal processing 2025年 6卷 333-347页

作者： Hamad, Maryam Conti, Caroline Nunes, Paulo Soares, Luis Ducla Inst Univ Lisboa ISCTE IUL Inst Telecomunicacoes P-1649026 Lisbon Portugal

image segmentation is an essential initial stage in several computer vision applications. However, unsupervised image segmentation is still a challenging task in some cases such as when objects with a similar visual appearance overlap. Unlike 2D images, 4D Light Fields (LFs) convey both spatial and angular scene information facilitating depth/disparity estimation, which can be further used to guide the segmentation. Existing 4D LF segmentation methods that target object level (i.e., mid-level and high-level) segmentation are typically semi-supervised or supervised with ground truth labels and mostly support only densely sampled 4D LFs. This paper proposes a novel unsupervised mid-level 4D LF Segmentation method using Graph neural Networks (LFSGNN), which segments all LF views consistently. To achieve that, the 4D LF is represented as a hypergraph, whose hypernodes are obtained based on hyperpixel over-segmentation. Then, a graph neural network is used to extract deep features from the LF and assign segmentation labels to all hypernodes. Afterwards, the network parameters are updated iteratively to achieve better object separation using backpropagation. The proposed segmentation method supports both densely and sparsely sampled 4D LFs. Experimental results on synthetic and real 4D LF datasets show that the proposed method outperforms benchmark methods both in terms of segmentation spatial accuracy and angular consistency.

关键词： image segmentation Semantics Feature extraction Semantic segmentation Graph neural networks Training Measurement Deep learning Accuracy Light fields Light field unsupervised segmentation deep learning angular consistency graph neural network

来源：评论

学校读者我要写书评

暂无评论

Learning-Based image Compression With Parameter-Adaptive Rate-Constrained Loss

引用

IEEE signal processing LETTERS 2024年 31卷 1099-1103页

作者： Guerin Jr, Nilson D. da Silva, Renam Castro Macchiavello, Bruno Univ Brasilia Dept Comp Sci BR-70910900 Brasilia Brazil Univ Brasilia Fac UnB Gama BR-72444240 Gama Leste Brazil

In recent years, the crucial task of image compression has been addressed by end-to-end neural network methods. However, achieving fine-grained rate control in this new paradigm has presented challenges. In our previous work, we explored mismatches in rate estimation during target-rate-oriented training and proposed heuristics involving costly parameter searches as a solution. This work proposes a lightweight approach, which dynamically adapts loss parameters to mitigate rate estimation issues, ensuring precise target rate attainment. Inspired by Reinforcement Learning, our method exhibits performance comparable to preceding approaches on the Kodak dataset in terms of PSNR. Additionally, it reduces computational training costs.

关键词： Convolutional neural network learning-based image compression rate control variational autoencoder

来源：评论

学校读者我要写书评

暂无评论

Dual-step reconstruction algorithm to improve microscopy resolution by deep learning

引用

APPLIED OPTICS 2023年第13期62卷 3439-3444页

作者： Deng, Qisheng Zhu, Zece Shu, Xuewen Huazhong Univ Sci & Technol Wuhan Natl Lab Optoelect Wuhan 430074 Peoples R China Huazhong Univ Sci & Technol Sch Opt & Elect Informat Wuhan 430074 Peoples R China Wuhan Text Univ Dept Chem & Chem Engn Wuhan 430200 Peoples R China

Deep learning plays an important role in the field of machine learning, which has been developed and used in a wide range of areas. Many deep-learning-based methods have been proposed to improve image resolution, most of which are based on image-to-image translation algorithms. The performance of neural networks used to achieve image translation always depends on the feature difference between input and output images. Therefore, these deep-learning-based methods sometimes do not have good performance when the feature differences between low-resolution and high-resolution images are too large. In this paper, we introduce a dual-step neural network algorithm to improve image resolution step by step. Compared with conventional deep-learning methods that use input and output images with huge differences for training, this algorithm learning from input and output images with fewer differences can improve the performance of neural networks. This method was used to reconstruct high-resolution images of fluorescence nanoparticles in cells. (c) 2023 Optica Publishing Group

关键词： image processing image reconstruction image resolution neural networks Reconstruction algorithms stochastic gradient descent

来源：评论

学校读者我要写书评

暂无评论

MoCoDiff: Momentum context diffusion model for low-dose CT denoising

引用

DIGITAL signal processing 2025年 156卷

作者： Zhao, Shaoting Jiang, Ailian Ding, Jianguo Taiyuan Univ Technol Coll Comp Sci & Technol Taiyuan 030600 Peoples R China Blekinge Inst Technol Dept Comp Sci Karlskrona Sweden

Low-Dose Computed Tomography (LDCT) has gradually replaced Normal-Dose Computed Tomography (NDCT) due to its lower radiation exposure. However, the reduction in radiation dose has led to increased noise and artifacts in LDCT images. To date, many methods for LDCT denoising have emerged, but they often struggle to balance denoising performance with reconstruction efficiency. This paper presents a novel Momentum Context Diffusion model for low-dose CT denoising, termed MoCoDiff. First, MoCoDiff employs a Mean-Preserving stochastic Degradation (MPSD) operator to gradually degrade NDCT to LDCT, effectively simulating the physical process of CT degradation and greatly reducing sampling steps. Furthermore, the stochastic nature of the MPSD operator enhances the diversity of samples in the training space and calibrates the deviation between network inputs and time-step embedded features. Second, we propose a Momentum Context (MoCo) strategy. This strategy uses the most recent sampling result from each step to update the context information, thereby narrowing the noise level gap between the sampling results and the context data. This approach helps to better guide the next sampling step. Finally, to prevent issues such as over-smoothing of image edges that can arise from using the mean square error loss function, we develop a dual-domain loss function that operates in both the image and wavelet domains. This approach leverages wavelet domain information to encourage the model to preserve structural details in the images more effectively. Extensive experimental results show that our MoCoDiff model outperforms competing methods in both denoising and generalization performance, while also ensuring fast training and inference.

关键词： Low-dose CT Denoising Diffusion model stochastic degradation operator Momentum context

来源：评论

学校读者我要写书评

暂无评论

image Compression Using stochastic-AFD Based Multisignal Sparse Representation

引用

IEEE TRANSACTIONS ON image processing 2022年 31卷 5317-5331页

作者： Dai, Lei Zhang, Liming Li, Hong Univ Macau Fac Sci & Technol Taipa Macau Peoples R China Huazhong Univ Sci & Technol Sch Math & Stat Wuhan 430074 Hubei Peoples R China

Adaptive Fourier decomposition (AFD) is a newly developed signal processing tool that can adaptively decompose any single signal using a Szego kernel dictionary. To process multiple signals, a novel stochastic-AFD (SAFD) theory was recently proposed. The innovation of this study is twofold. First, a SAFD-based general multi-signal sparse representation learning algorithm is designed and implemented for the first time in the literature, which can be used in many signal and image processing areas. Second, a novel SAFD based image compression framework is proposed. The algorithm design and implementation of the SAFD theory and image compression methods are presented in detail. The proposed compression methods are compared with 13 other state-of-the-art compression methods, including JPEG, JPEG2000, BPG, and other popular deep learning-based methods. The experimental results show that our methods achieve the best balanced performance. The proposed methods are based on single image adaptive sparse representation learning, and they require no pre-training. In addition, the decompression quality or compression efficiency can be easily adjusted by a single parameter, that is, the decomposition level. Our method is supported by a solid mathematical foundation, which has the potential to become a new core technology in image compression.

关键词： image coding Dictionaries signal processing algorithms Transform coding Kernel Training Discrete cosine transforms stochastic adaptive Fourier decomposition sparse representation image compression image reconstruction

来源：评论

学校读者我要写书评

暂无评论

FLQ: Design and implementation of hybrid multi-base full logarithmic quantization neural network acceleration architecture based on FPGA

引用

signal processing-image COMMUNICATION 2025年 134卷

作者： Zhang, Longlong Hu, Xiang Liao, Xuan Zhou, Tong Peng, Yuanxi Natl Univ Def Technol Coll Comp Sci & Technol Changsha 410073 Peoples R China Natl Univ Def Technol Beijing Inst Adv Study Beijing 100000 Peoples R China

As deep neural network (DNN) models become more accurate, problems such as large model parameters and high computational complexity have become increasingly prominent, leading to a bottleneck in deploying them on resource-limited embedded platforms. In recent years, logarithm-based quantization techniques have shown great potential in reducing the inference cost of neural networks. However, current single-model log-quantization has reached an upper limit of classification performance, and little work has investigated hardware implementation of neural network quantization. In this paper, we propose a full logarithmic quantization (FLQ) mechanism that quantizes both weights and activation values into the logarithmic domain, compressing the parameters on AlexNet and VGG16 model by >6.4 times while maintaining an accuracy loss of within 2.5 % compared with benchmarking. Furthermore, we propose two optimization solutions for FLQ: activation segmented full logarithmic quantization (ASFLQ) and multi-ratio activation segmented full logarithmic quantization (Multi-ASFLQ), which can better balance the numerical representation range and quantization step. Under the condition of weight quantization of 5 bits and activation value quantization of 4 bits, the optimization methods proposed in this paper can improve the TOP1 of the VGG16 network model by 1 % and 1.6 %, respectively. Subsequently, we propose an implementation scheme of computing unit corresponding to the optimized FLQ mechanism above, which can not only convert multiplication operations into a shift operation but also integrate functions such as different ratio logarithmic bases and sparsity processing for activation, minimizing resource consumption as well as avoiding unnecessary calculations. Finally, we experiment with VGG19, Retnet50, and Densenet169 models, proving that the proposed method can achieve good performance under lower bit quantization. (c) 2001 Elsevier Science. All rights reserved

关键词： Deep neural network Logarithmic quantization Computing units FPGAs

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：