检索结果-内蒙古大学图书馆

arXiv 2024年

作者： De Rosa, Pasquale Bromberg, Yérom-David Felber, Pascal Mvondo, Djob Schiavoni, Valerio University of Neuchâtel Neuchâtel Switzerland University of Rennes CNRS INRIA IRISA Rennes France

In machine learning (ML), the inference phase is the process of applying pre-trained models to new, unseen data with the objective of making predictions. During the inference phase, end-users interact with ML services to gain insights, recommendations, or actions based on the input data. For this reason, serving strategies are nowadays crucial for deploying and managing models in production environments effectively. These strategies ensure that models are available, scalable, reliable, and performant for real-world applications, such as time series forecasting, image classification, natural language processing, and so on. In this paper, we evaluate the performances of five widely-used model serving frameworks (TensorFlow Serving, TorchServe, MLServer, MLflow, and BentoML) under four different scenarios (malware detection, cryptocoin prices forecasting, image classification, and sentiment analysis). We demonstrate that TensorFlow Serving is able to outperform all the other frameworks in serving deep learning (DL) models. Moreover, we show that DL-specific frameworks (TensorFlow Serving and TorchServe) display significantly lower latencies than the three general-purpose ML frameworks (BentoML, MLFlow, and MLServer). © 2024, CC BY.

关键词： deep learning

来源：评论

学校读者我要写书评

暂无评论

SPPNet: An Approach For real-time Encrypted Traffic Classification Using deep learning

SPPNet: An Approach For Real-Time Encrypted Traffic Classifi...

引用

IEEE Global Communications Conference (GLOBECOM)

作者： Meslet-Millet, Fabien Chaput, Emmanuel Mouysset, Sandrine Univ Toulouse IRIT Toulouse INP ENSEEIHT Toulouse France Univ Toulouse IRIT UPS Toulouse France

ISBN: (纸本)9781728181042

Data flow management has become a key network activity, strengthening the need for efficient data flow classification tools. However, pervasive encryption of communication has dramatically jeopardised the legacy tools. Recent advances in deep learning offer a wide variety of architectures that seem relevant for this purpose. These architectures are based on different data representations as input of their classification process. In this paper, we show the need for a deeper understanding of the features used by deep learning models to perform such classification. Our objective is to exploit this knowledge for defining a better data processing so that the chosen architecture will significantly improve the classification process. We will show that some information carried by packet headers need to be analyzed through a separate process. This analysis highlight that current deep learning approaches in the literature fail to classify encrypted flows in practice. We therefore propose a new modular deep learning architecture called Servername Protocol Packet Network (SPPNet) to overcome this drawback. We will show by a proof of concept that SPPNet allows to perform real-time network flow classification at packet level.

关键词： deep learning Encrypted Network traffic classification

来源：评论

学校读者我要写书评

暂无评论

CAD system for epileptic seizure detection from EEG through image processing and SURF-BOF technique

引用

MACHINE learning-SCIENCE AND TECHNOLOGY 2023年第4期4卷 045029页

作者： Alshayeji, Mohammad H. Kuwait Univ Coll Engn & Petr Comp Engn Dept POB 5969 Kuwait 13060 Kuwait

Epilepsy is one of the most debilitating neurological diseases that abruptly alters a person's way of life. Manual diagnosis is a laborious and time-consuming task prone to human error. Therefore, automating this task by developing an intelligent system is necessary. Existing deep learning (DL) models require high training time, large datasets, and machines with more memory and processing power. In addition, owing to the black-box nature of DL models, no one can determine the features that the network prefers for classification decisions. To overcome these challenges, this study proposes an accurate, automatic, and fast-intelligent system for epilepsy detection using a computer-aided diagnosis (CAD) -two-dimensional machine learning (ML) framework. Existing ML models struggle to produce reliable and acceptable diagnostic results owing to the low amplitude and nonstationary nature of electroencephalograms (EEGs), particularly in clinical situations where environmental influences are almost impossible to eliminate. The proposed model was built using the Children's Hospital Boston and the Massachusetts Institute of Technology dataset, and represents the first study that employs the speeded-up robust feature (SURF) bag of features technique for this application, which generates local features from spectrogram images of the respective one-dimensional EEG signal inputs. In addition, DL features were extracted from the spectrogram images for model performance comparison. Both features were used separately to train the ML classifiers. Implementing SURF offers fast computation and makes the model invariant to distortions, noise, scaling, and so on. Therefore, the proposed model is more suitable for real-time applications, and this ML framework provides an enhanced accuracy of 99.78% compared to the support vector machine-RBF classifier, along with 99.56% sensitivity, 100% specificity, and an error rate of 0.22%. The higher detection accuracy demonstrates the effectiveness of

关键词： EEG computer-aided diagnosis machine learning epilepsy bag of features deep learning

来源：评论

学校读者我要写书评

暂无评论

Angular Joint Loss for image Classification 13

Angular Joint Loss for Image Classification

引用

13th International Conference on Digital image processing (ICDIP)

作者： Ai, Hao Liao, Qingmin Tsinghua Univ Shenzhen Key Lab Informat Sci & Tech Shenzhen Engn Lab IS & DCPDept Elect EngnGrad Sch & Shenzhen Beijing Peoples R China

ISBN: (数字)9781510646018

ISBN: (纸本)9781510646018

Current CNN-based deep feature learning methods for image classification are mainly focused on learning with two types of loss functions: the softmax loss and its variants with class-level labels, or the metric-learning losses with pairwise labels. The former regards the weight vectors of the last fully connected layer of CNNs as representations of the class centres and focuses on the similarity between a single feature and each weight vector. The latter aims at seeking appropriate distance metrics between features pairs to boost the CNNs by learning a discriminative embedding space. In this work, we propose a simple yet effective new loss function called angular joint loss (AJL) to exploit the best of both worlds. By integrating feature-weight angles with feature-feature angles, AJL inherits the simplicity of softmax loss and the power of pair-wise metric learning and enhances both the intra-class compactness and the inter-class separability. The proposed AJL is simple to implement and has no time-consuming mining of pair or triple samples. On some popular image classification datasets (CIFAR-10, CIFAR-100 and imageNet), extensive experimental results show that the proposed AJL can achieve the state-of-the-art performance.

关键词： deep feature learning metric learning CNN angular joint loss image classification loss

来源：评论

学校读者我要写书评

暂无评论

deep learning-Based Generic Automatic Surface Defect Inspection (ASDI) With Pixelwise Segmentation

引用

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT 2021年 70卷 1-10页

作者： Wu, Xiaojun Qiu, LingTeng Gu, Xiaodong Long, Zhili Harbin Inst Technol Shenzhen Sch Mech Engn & Automat Shenzhen 518055 Guangdong Peoples R China Alibaba AILab Hangzhou 311121 Zhejiang Peoples R China

Automatic surface defect inspection (ASDI) is a crucial and challenging problem in industry because it affects the quality and efficiency of production greatly. deep learning-based methods achieve promising improvements for surface detection, but they rely on a massive training data set that is impractical in industry. In this study, we propose a generic method that works effectively even on the small size training data set. Especially, we introduce a ResMask generative adversarial network (GAN) framework that is a residual GAN to expand the insufficient defect data sets. Meanwhile, the existing inspection data sets are so much easier to detect than the data sets in the real industrial scenarios that new industrial surface defect data sets containing more diverse and challenging images are established. Then, a coarse-to-fine module (CFM) that consists of a coarse detection subnetwork and a fine segmentation subnetwork is proposed for the needs of fast detection for high-resolution images. In the coarse detection stage, spatial pyramid pooling (SPP) is utilized to increase the receptive field of the network, reduce the false detection rate, and determine the approximate location of defects. In the fine segmentation stage, the receptive field of the network is enlarged by atrous SPP (ASPP), and skip links that incorporate low-level with high-level features achieve pixelwise precision on defect segmentation. Finally, our algorithm has achieved state-of-the-art results in DAGM, HR, and WB data sets (0.859, 0.761, and 0.805, respectively) according to MIoU. It achieves an average processing time of 44.4 ms on the test images with a resolution of 512 x 512 and 131.6 ms for 2048 x 2048 images. On the DAGM data set, the detection accuracy of mean intersection over union (MIoU) reaches 0.859.

关键词： Automatic surface defect inspection (ASDI) deep learning defect image generation generative adversarial networks (GANs) pixelwise segmentation

来源：评论

学校读者我要写书评

暂无评论

deep-learning-assisted and GPU-accelerated vector Doppler imaging with aliasing-resistant velocity estimation

引用

ULTRASONICS 2023年第1期134卷 107050页

作者： Nahas, Hassan Yiu, Billy Y. S. Chee, Adrian J. Y. Au, Jason S. Yu, Alfred C. H. Univ Waterloo Schlegel Res Inst Aging Waterloo ON Canada Univ Waterloo Dept Elect & Comp Engn Waterloo ON Canada Univ Waterloo Dept Kinesiol Waterloo ON Canada

Vector flow imaging is a diagnostic ultrasound modality that is suited for the visualization of complex blood flow dynamics. One popular way of realizing vector flow imaging at high frame rates over 1000 fps is to apply multi -angle vector Doppler estimation principles in conjunction with plane wave pulse-echo sensing. However, this approach is susceptible to flow vector estimation errors attributed to Doppler aliasing, which is prone to arise when a low pulse repetition frequency (PRF) is inevitably used due to the need for finer velocity resolution or because of hardware constraints. Existing dealiasing solutions tailored for vector Doppler may have high computational demand that makes them unfeasible for practical applications. In this paper, we present the use of deep learning and graphical processing unit (GPU) computing principles to devise a fast vector Doppler esti-mation framework that is resilient against aliasing artifacts. Our new framework works by using a convolutional neural network (CNN) to detect aliased regions in vector Doppler images and subsequently applying an aliasing correction algorithm only at these affected regions. The framework's CNN was trained using 15,000 in vivo vector Doppler frames acquired from the femoral and carotid arteries, including healthy and diseased conditions. Re-sults show that our framework can perform aliasing segmentation with an average precision of 90 % and can render aliasing-free vector flow maps with real-time processing throughputs (25-100 fps). Overall, our new framework can improve the visualization quality of vector Doppler imaging in real-time.

关键词： Vector Doppler imaging Multi -angle Doppler estimation Aliasing correction deep learning GPU computing

来源：评论

学校读者我要写书评

暂无评论

Feasibility Study of a Collimator-less 4π FOV Imaging System using Pixelated GAGG(Ce) Scintillator and deep learning Method

Feasibility Study of a Collimator-less 4π FOV Imaging Syste...

引用

IEEE Symposium on Nuclear Science (NSS/MIC)

作者： S. Han D. Hong A. Jo W. Lee Department of Health and Safety Convergence Science Korea University Seoul Republic of Korea Transdisciplinary Major in Learning Health Systems Korea University Seoul Republic of Korea Health Science Research Center Korea University Seoul Republic of Korea School of Health and Environmental Science Korea University Seoul Republic of Korea

ISBN: (数字)9798350388152

ISBN: (纸本)9798350388169

This study explores the feasibility of a collimator-less 4 p field of view (FOV) imaging system using pixelated GAGG(Ce) scintillator and deep learning method. We proposed pixelated detection system that reduced the time for training the fully convolutional network (FCN) model because of its small data size and provided convenience in processing as training data, enabling real-time imaging applications. The FCN model was configured using the TensorFlow library 2.0. The train data and label data for FCN model were acquired by using Monte Carlo N-Particle (MCNP) simulations, and the standard isotopes were ${ }^{57} \mathrm{Co},{ }^{133} \mathrm{Ba},{ }^{22} \mathrm{Na}$, and ${ }^{137} \mathrm{Cs}$. The trained FCN model was validated using test data, and the images reconstructed by the FCN model were compared with the labelled data through the structural similarity index measure (SSIM) method. The results confirmed that the model could predict up to five sources with varying positions and isotopes. The method of estimating radiation source positions using deep learning offers substantial advantages in real-time imaging. Once trained, the model can rapidly and efficiently compute source positions, providing significant benefits in terms of speed and simplicity for various industrial applications.

关键词： image quality deep learning Training Scintillators Computational modeling Training data Data models

来源：评论

学校读者我要写书评

暂无评论

A Single-Shot Autofocus Approach for Surface Plasmon Resonance Microscopy

引用

ANALYTICAL CHEMISTRY 2021年第4期93卷 2433-2439页

作者： Xu, Ying Wang, Xu Zhai, Chunhui Wang, Jingan Zeng, Qiang Yang, Yuting Yu, Hui Hangzhou Dianzi Univ Coll Automat Hangzhou 310018 Zhejiang Peoples R China Shanghai Jiao Tong Univ Sch Biomed Engn Shanghai 200030 Peoples R China Shanghai Jiao Tong Univ Sch Elect Informat & Elect Engn Shanghai 200030 Peoples R China

Surface plasmon resonance microscopy (SPRM) has been widely used as a sensitive imaging platform for chemical and biological analysis. The SPRM system inevitably suffers from focus inhomogeneity and drifts, especially in long-term recordings, leading to distorted images and inaccurate quantification. Traditional focus correction approaches require additional optical parts to detect and adjust focal conditions. Herein, we propose a deep-learning-based image processing method to gain autofocused SPRM images, without increasing the complexity of the optical systems. We trained a generative adversarial network (GAN) model with thousands of SPRM images of nanoparticles acquired at different focal distances. The trained model was able to directly generate focused SPRM images from single-shot defocused images, with no prior knowledge of the focus conditions during recording. Experiments using Au nanoparticles show that this method is effective in both static and time-lapse monitoring. The proposed autofocus technique thus provides an approach for improving the consistency among SPRM studies and for long-term monitoring.

关键词： Generative adversarial network Digital image processing Surface plasmon resonance microscopy Surface plasmon resonance deep learning Plasmon Autofocus Surface plasmon

来源：评论

学校读者我要写书评

暂无评论

Indoor Human Counting Based on Range-Doppler Spectrum in the Presence of Material Penetration and Occlusion

Indoor Human Counting Based on Range-Doppler Spectrum in the...

引用

International Conference on Communications, Signal processing, and their Applications (ICCSPA)

作者： Ilhan Aytutuldu Zehra Bilici Yakup Genc Yusuf Sinan Akgul Department of Computer Engineering Gebze Technical University Kocaeli Turkey

ISBN: (数字)9798350384819

ISBN: (纸本)9798350384826

This paper introduces an ensemble approach for indoor human counting method based on millimeter-wave (mmWave) radar technology. Through advanced spatio-temporal range-Doppler analysis, the proposed model classifies room occupancy (empty, single-person, multiple people) with 90% accuracy. Beyond precision counting, the system's real-time data facilitates smart applications, such as optimizing air conditioning during pandemics or other dynamic scenarios. The most common technologies such as image or wearable inertial sensors are not preferable due to privacy issues or subject discomfort. This compact solution combines mmWave radar and deep learning for efficient, adaptable, and secure indoor environment management.

关键词： Privacy Accuracy Pandemics Radar detection Radar imaging Millimeter wave radar real-time systems Radar signal processing Doppler radar Millimeter wave communication

来源：评论

学校读者我要写书评

暂无评论

Sustaining Solar Panel Cleanliness with Multiclass Classification of Panel images Using deep learning & ML Classifiers Trained on SVD Embeddings

Sustaining Solar Panel Cleanliness with Multiclass Classific...

引用

International Conference on Computer Research and Development (ICCRD)

作者： Supratik Sekhar Bhattacharya Anusuya B School of Computing NUS Singapore Singapore School of Electrical Engineering (SELECT) Vellore Institute of Technology (VIT University) Vellore India

ISBN: (数字)9798331531881

ISBN: (纸本)9798331531898

Maintaining the optimal performance of solar panels poses a complex challenge, especially in regions abundant in solar energy like deserts and tropical areas. This research paper explores the utilization of machine learning and deep learning techniques in categorizing the states of solar panels through visual image processing. In regions where solar farms are increasingly prevalent, challenges such as dust accumulation, cracks, and delamination hinder energy production and diminish solar panel efficiency. Our study employs various machine learning classifier models and deep learning to a unique dataset comprising real-world images of solar panels categorized into distinct groups: clean panels, dusty panels, and panels exhibiting breaks and delamination. The aim is to distinguish between surfaces that are clean, cleanable (dirty), and uncleanable (surface breakage and delamination).Given the high dimensionality of our novel image data (120000), we leverage dimensionality reduction with singular value decomposition (SVD) in our study to train the machine learning classifier models. An SVD dimensionality reduction approach shows excellent compression with a compressed 10-dimensional SVD embedding obtaining high accuracies and F1 scores when used to train our ten machine learning algorithms. We obtain a peak F1 score as well as a balanced accuracy of 0.96 with the trained support vector machine classifier (SVC) model using our ten-dimensional SVD embeddings as the input. Several decision tree-based models perform well too with a F1 score of 0.85 for an Adaboost classifier, 0.9 for an extra tree classifier, 0.93 for our decision tree classifier, and 0.94 for the bagging classifier model. We also create and train a custom deep convolutional neural network (DCNN) with 2566129 parameters on the uncompresses data that performs extremely well. We obtain a micro precision, recall, and F1-score of 0.98 as well as a macro precision of 0.99, macro recall of 0.98, and macro F1-sco

关键词： Dimensionality reduction deep learning Surface cleaning Support vector machines Accuracy image coding Static VAr compensators Solar panels Convolutional neural networks time complexity

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：