检索结果-内蒙古大学图书馆

Doctor’s Cursive Handwriting Recognition System Using Deep Learning

Doctor’s Cursive Handwriting Recognition System Using Deep ...

International Conference on Humanoid, Nanotechnology, Information Technology, Communication and Control, Environment and Management (HNICEM)

作者： Lovely Joy Fajardo Niño Joshua Sorillo Jaycel Garlit Cia Dennise Tomines Mideth B. Abisado Joseph Marvin R. Imperial Ramon L. Rodriguez Bernie S. Fabito College of Computing and Information Technologies National University Manila City Philippines College of Computer Studies Pamantasan ng Cabuyao Cabuyao City Philippines

ISBN: (数字)9781728130446

ISBN: (纸本)9781728130453

Handwriting is a skill to express thoughts, ideas, and language. Over the years, medical doctors have been well-known for having illegible cursive handwritings and has been a generally accepted matter. The datasets used in this paper are samples of doctors cursive handwriting collected from several clinics and hospitals of Metro Manila, Quezon City and Taytay, Rizal. In this paper, we present the Handwriting Recognition System using Deep Convolutional Recurrent neural network that is developed in order to identify the text in the image of prescriptions written by the doctors and show the readable text conversion of the cursive handwriting. In this study two models were evaluated and based on the experimentation CRNN with model-based normalization scheme than the CRNN alone. This study achieved 76% training accuracy rate and the developed model was found successfully implemented in a mobile application, having achieved a validation accuracy of 72% for the validation set from the remaining 540 images of prescription. The mobile application was validated for the second time using the captured 48 handwriting samples written by the researchers and correctly identified 17 images out of 48 this gives us a 35% validation accuracy.

关键词： Medical services Character recognition Handwriting recognition Biomedical imaging optical character recognition software Recurrent neural networks Training

来源：评论

学校读者我要写书评

暂无评论

Object Classification using Deep Learning on Extremely Low-Resolution Time-of-Flight Data

Object Classification using Deep Learning on Extremely Low-R...

引用

2018 International Conference on Digital Image computing: Techniques and Applications, DICTA 2018

作者： Ruvalcaba-Cardenas, Ana Daysi Scoleri, Tony Day, Geoffrey Dept. Elec. Eng. RMIT Melbourne Australia EOCM Group CEWD Defence Science and Technology Group Adelaide Australia WSTS Group WCSD Defence Science and Technology Group Adelaide Australia

ISBN: (纸本)9781538666029

This paper proposes two novel deep learning models for 2D and 3D classification of objects in extremely low-resolution time-of-flight imagery. The models have been developed to suit contemporary range imaging hardware based on a recently fabricated Single Photon Avalanche Diode (SPAD) camera with 64 χ 64 pixel resolution. Being the first prototype of its kind, only a small data set has been collected so far which makes it challenging for training models. To bypass this hurdle, transfer learning is applied to the widely used VGG-16 convolutional neural network (CNN), with supplementary layers added specifically to handle SPAD data. This classifier and the renowned Faster-RCNN detector offer benchmark models for comparison to a newly created 3D CNN operating on time-of-flight data acquired by the SPAD sensor. Another contribution of this work is the proposed shot noise removal algorithm which is particularly useful to mitigate the camera sensitivity in situations of excessive lighting. models have been tested in both low-light indoor settings and outdoor daytime conditions, on eight objects exhibiting small physical dimensions, low reflectivity, featureless structures and located at ranges from 25m to 700m. Despite antagonist factors, the proposed 2D model has achieved 95% average precision and recall, with higher accuracy for the 3D model. © 2018 IEEE.

关键词： optical radar

来源：评论

学校读者我要写书评

暂无评论

A Novel Feature-spanning Machine Learning Technology for Defect Inspection

A Novel Feature-spanning Machine Learning Technology for Def...

引用

International Microsystems, Packaging, Assembly and Circuits Technology (IMPACT)

作者： Yi-Chung Hsu Ping-Yen Kuo Wong-Shian Huang AI & Advanced Optical Technology Development Division Corporate R&D Advanced Semiconductor Engineering (ASE) Inc. Zhubei City Hsinchu County 302 Taiwan (R.O.C.)

ISBN: (数字)9781728160702

ISBN: (纸本)9781728160719

Defect inspection (to detect, classify, measure and analyze) has long been a challenging task in semiconductor manufacturing (MFG) domain. This paper discusses a new Machine Learning (ML) approach which can be used to assist defect inspection in different MFG scenarios. Associated solutions have been developed and applied to the surface defect inspection for substrate components, used in our IC packaging *** the past decade, the help from image-based Automated optical Inspection (AOI) equipment has significantly reduced manual efforts in substrate/PCB defect examination, but is still insufficient in defect classification automation. Recently, the adoption of ML and Convolution neural network (CNN) based Deep Learning technologies raised much hoping to advance the defect classification automation to a new level that acceptable for rigid MFG practices. Most of the published CNN models, however, tend to use large number of learning parameters (floating-point variables) during computing in order to gain high image recognition accuracy. The parameters' massive blow-up often causes heavy power-gobbling computation. While applying CNN to our substrate defect datasets, each of which contains hundreds of thousands of high-resolution images, collected across different MFG processes and produced products, a training run could take days or even weeks to finish. Also, more the learning variables for training are used, longer the inference time will be. The situations make these complex CNN models hard to meet our overnight retraining and in-line inference time constraint. Therefore, to develop a more efficient ML method is strongly preferred in our MFG *** this paper, we develop a feature-spanning ML approach which takes the feature of an image as a base. Through mathematical transformations the base is spanned to fit into an accumulated and gradually divided feature-of-classes space. During the feature-spanning process, the initialization of parameters is tigh

关键词： Inspection Substrates Benchmark testing Training Feature extraction Machine learning Production

来源：评论

学校读者我要写书评

暂无评论

基于字序列的非结构化简历信息解析方法

引用

计算机工程与设计 2019年第6期40卷 1769-1774页

作者：陈毅符磊张剑黄石磊重庆邮电大学光通信与网络重点实验室重庆400065 安徽大学计算智能与信号处理教育部重点实验室安徽合肥230601 北京大学深圳研究院广东深圳518057 深港产学研基地深圳市智能媒体和语音重点实验室广东深圳518057 深港产学研基地产业发展中心广东深圳518057

为有效解决传统简历解析方法效率低、成本高、泛化能力差的问题,提出一种基于字序列的非结构化文本简历解析方法。利用BLSTM对字序列进行建模,获得一个包含字序列信息的词表示;由BLSTM神经网络强大的学习能力对特征进行学习,获得相应的... 详细信息

为有效解决传统简历解析方法效率低、成本高、泛化能力差的问题,提出一种基于字序列的非结构化文本简历解析方法。利用BLSTM对字序列进行建模,获得一个包含字序列信息的词表示;由BLSTM神经网络强大的学习能力对特征进行学习,获得相应的特征;根据前后标签的约束,使用CRF获得最优标签序列(CBLSTM-CRF);利用梯度下降算法训练神经网络,使用预训练字向量、Dropout优化神经网络,完成对中文简历的解析工作。实验结果表明,CBLSTM-CRF方法对简历解析的效果优于传统方案,利用BLSTM对字序列进行建模的方法在其它模型上也取得了较好的效果。

关键词：中文简历字序列非结构化神经网络条件随机场

来源：评论

学校读者我要写书评

暂无评论

High-Power LED Photoelectrothermal Analysis Based on Backpropagation Artificial neural networks

引用

IEEE TRANSACTIONS ON ELECTRON DEVICES 2017年第7期64卷 2867-2873页

作者： Liu, Hongwei Guo, Kai Zhang, Zanyun Yu, Dandan Zhang, Jianxin Ning, Pingfan Cheng, Junchao Li, Xiaoyun Niu, Pingjuan Tianjin Polytech Univ Sch Elect & Informat Engn Tianjin 300387 Peoples R China Tianjin Key Lab Optoelect Detect Technol & Syst Tianjin 300387 Peoples R China Tianjin Key Lab Adv Elect Engn & Energy Technol Tianjin 300387 Peoples R China

As an electroluminescent device, the coupling relationship between light-emitting diode (LED) input currents, optical power, and LED junction temperature is a complicated multiphysics process. In this paper, a simplified LED photoelectron-thermal (PET) model by artificial neural network (ANN), which can translate multiphysics field issue into a single physics field problem, is mentioned to study the coupling relationship. In the first, an LED lumens, optical power, and electric power at different temperatures are monitored in a temperature controlling integrating sphere. Then, a backpropagation (BP) ANN is trained by these data to construct an LED Photo-Electron-Thermal (PET) relationship. In addition, LED luminaire thermal analyzing is performed using a finite-element method on the outputs of the BP ANN. Finally, the advantage of this method in terms of saving computing resources and computing time is analyzed by comparing the degrees of freedom in different models. The result shows that at least 6.7 times computing resource are saved by this method, which will reduce the LED thermal management system analyzing time greatly. Finally, the application and extension of this model are discussed.

关键词： Backpropagation (BP) artificial neural network (ANN) finite-element method (FEM) light-emitting diode (LED) photoelectron-thermal (PET) couple thermal management

来源：评论

学校读者我要写书评

暂无评论

Dynamic Spatio-Temporal Feature Learning via Graph Convolution in 3D Convolutional networks

Dynamic Spatio-Temporal Feature Learning via Graph Convoluti...

引用

IEEE International Conference on Data Mining Workshops (ICDM Workshops)

作者： Jun Li Xianglong Liu Jun Xiao Hainan Li Shuo Wang Liang Liu State Key Lab of Software Development Environment Beihang University China Beijing Advanced Innovation Center for Big Data-Based Precision Medicine Beihang University Beijing China

ISBN: (纸本)9781728148977

Video data owns strong dynamic features in both spatial and temporal domains. In the literature, 3D Convolutional neural networks (3D CNNs) serve as a successful technique to simultaneously learn the spatio-temporal features. However, due to the expensive computations, usually the kernel size of the convolutions used in 3D CNNs is rather small, and thus largely limits their learning capability. To address this issue, in this paper we attempt to enhance the learning capability of 3D CNNs for extracting the dynamic features. We capture the long-distance information by modeling the temporal and spatial features as graphs, and then learn the dynamic graph structure information from the feature maps of 3D CNNs. This corresponds to the powerful Graph Convolutional networks (GCNs), whose adjacent matrix is determined dynamically based on the feature maps. With the learnt dynamic graph, we introduce and fuse a framewise GCN and a channel-wise GCN to enhance the temporal and spatial feature learning of 3D CNNs. Our proposed spatiotemporal graph convolutional network (STGCN) works as a general module that can be embedded into the popular 3D CNNs architectures (e.g., ResNeXt, P3D). Extensive experiments on two video datasets (UCF-101 and HMDB-51) for action recognition task demonstrate the state-of-the-art models with our STGCN module can achieve significant performance improvement.

关键词： Three-dimensional displays Convolution Solid modeling Convolutional codes Mathematical model Computational modeling optical computing

来源：评论

学校读者我要写书评

暂无评论

Handwritten Yi Character Recognition with Density-based Clustering Algorithm and Convolutional neural network 20

Handwritten Yi Character Recognition with Density-based Clus...

引用

20th IEEE International Conference on Computational Science and Engineering (CSE) / 15th IEEE/IFIP International Conference on Embedded and Ubiquitous computing (EUC)

作者： Jia Xiaodong Gong Wendong Yuan Jie Minzu Univ China Sch Informat & Engn Beijing Peoples R China

ISBN: (纸本)9781538632215

A great deal of research has focused on using convolutional neural network for optical character recognition. However we encountered two typical problem in this field when applied convolutional neural network to handwritten Yi character recognition. First, since convolutional neural network is a kind of supervised deep learning model, the manual training data labeling is a very time consuming and labor intensive work. Second, because the theory is not well studied, the structure design and parameter adjustment of convolutional neural network depend heavily on experience, and our recognition accuracy was not satisfactory at the beginning. To address these two problems, in this paper, for one thing, we use entropy theory improved a density-based clustering algorithm, which is proved very effective in data labeling. For another, as to the problem of structure design and parameter adjustment, we compared performance of models with different scales and different parameters, and gave some experience about this problem. Finally we achieved 99.65% accuracy on the test set. We hope that this paper will inspire more researches on convolutional neural network applied to dataset-lacked optical character recognition problems.

关键词： convolutional neural network optical character recognition clustering algorithm

来源：评论

学校读者我要写书评

暂无评论

Converged Recommendation System Based on RNN and BP neural networks

Converged Recommendation System Based on RNN and BP Neural N...

引用

International Conference on Big Data and Smart computing (BIGCOMP)

作者： ZhaoWei Qu Shuqiang Zheng Xiaoru Wang Xiaomin Song Baiwei Li Xiaohui Song Institute of Network Technology Beijing University of Posts and Telecommunications Beijing China School of Computer Science Beijing University of Posts and Telecommunications Beijing China Institute of Information Photonics and Optical Communication Beijing University of Posts and Telecommunications School of Cyberspace Security Beijing University of Posts and Telecommunications Beijing China

ISBN: (纸本)9781538636503

Recommendation systems based on rating prediction model didn't consider temporal context and recurrent pattern. And past user behavior analysis didn't simultaneously consider long-term behavior, short-term behavior, and behavior with popular elements. In this paper, we propose a new model RNN-BPNNCM to predict user's next consumption behavior and then we recommend predicted item to the user. First, we use recurrent neural networks to analyze three kinds of behavior sequences to figure out three kinds of probabilities for every possible item. Then we use back propagation neural networks to figure out final probability. Finally, we take the top items with highest final probabilities as items that user will most likely consume next time and we recommend them to the user. RNN-BPNNCM solves the above problems of temporal context, recurrent pattern, and multiple user behaviors. Experiment shows RNN-BPNNCM more accurately predict the user's next consumption behavior and has more excellent performance for recommendation systems.

关键词： Hidden Markov models Recurrent neural networks Predictive models Telecommunications Electronic mail Social network services

来源：评论

学校读者我要写书评

暂无评论

Deep optical neural network by living tumour brain cells

arXiv

引用

arXiv 2018年

作者： Pierangeli, D. Palmieri, V. Marcucci, G. Moriconi, C. Perini, G. de Spirito, M. Papi, M. Conti, C. Department of Physics University Sapienza Piazzale Aldo Moro 5 Rome00185 Italy Institute of Physics Fondazione Policlinico Universitario A. Gemelli IRCCS ‐ Università Cattolica del Sacro Cuore Rome00168 Italy School of Pharmacy and Pharmaceutical Sciences Cardiff University CardiffCF10 3NB United Kingdom Via dei Taurini 19 Rome00185

The new era of artificial intelligence demands large-scale ultrafast hardware for machine learning.1 optical artificial neural networks process classical and quantum information at the speed of light, and are compatible with silicon technology, but lack scalability and need expensive manufacturing of many computational layers2–6. New paradigms, as reservoir computing7–9 and the extreme learning machine10, suggest that disordered11–16 and biological materials may realize artificial neural networks with thousands of computational nodes trained only at the input and at the readout. Here we employ biological complex systems, i.e., living three-dimensional tumour brain models, and demonstrate a random neural network (RNN) trained to detect tumour morphodynamics via image transmission17,18. The RNN, with the tumour spheroid19 as a three-dimensional deep computational reservoir, performs programmed optical functions and detects cancer morphodynamics from laser-induced hyperthermia inaccessible by optical imaging. Moreover, the RNN quantifies the effect of chemotherapy inhibiting tumour growth. We realize a non-invasive smart probe for cytotoxicity assay, which is at least one order of magnitude more sensitive with respect to conventional imaging. Our random and hybrid photonic/living system is a novel artificial machine for computing and for the real-time investigation of tumour dynamics. Copyright © 2018, The Authors. All rights reserved.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

感知器残差网络和超限学习机融合的3D物体识别

引用

中国图象图形学报 2019年第10期24卷 1738-1749页

作者：黄强王永雄上海理工大学光电信息与计算机工程学院上海200093 上海康复器械工程技术研究中心上海200093

目的随着3D扫描技术和虚拟现实技术的发展,真实物体的3D识别方法已经成为研究的热点之一。针对现有基于深度学习的方法训练时间长,识别效果不理想等问题,提出了一种结合感知器残差网络和超限学习机(ELM)的3D物体识别方法。方法以超限学... 详细信息

目的随着3D扫描技术和虚拟现实技术的发展,真实物体的3D识别方法已经成为研究的热点之一。针对现有基于深度学习的方法训练时间长,识别效果不理想等问题,提出了一种结合感知器残差网络和超限学习机(ELM)的3D物体识别方法。方法以超限学习机的框架为基础,使用多层感知器残差网络学习3D物体的多视角投影特征,并利用提取的特征数据和已知的标签数据同时训练了ELM分类层、K最近邻(KNN)分类层和支持向量机(SVM)分类层识别3D物体。网络使用增加了多层感知器的卷积层替代传统的卷积层。卷积网络由改进的残差单元组成,包含多个卷积核个数恒定的并行残差通道,用于拟合不同数学形式的残差项函数。网络中半数卷积核参数和感知器参数以高斯分布随机产生,其余通过训练寻优得到。结果提出的方法在普林斯顿3D模型数据集上达到了94.18%的准确率,在2D的NORB数据集上达到了97.46%的准确率。该算法在两个国际标准数据集中均取得了当前最好的效果。同时,使用超限学习机框架使得本文算法的训练时间比基于深度学习的方法减少了3个数量级。结论本文提出了一种使用多视角图识别3D物体的方法,实验表明该方法比现有的ELM方法和深度学习等最新方法的识别率更高,抗干扰性更强,并且其调节参数少,收敛速度快。

关键词：多层感知器残差网络多通道分类器超限学习机 3D物体识别特征提取

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：