Nonlinear Optimization and Dimensional Reduction Via Spearma

发布时间：2021-10-22 11:34

　　在工业4.0时代,工业计算机视觉及其相关应用日益获得成功和普及。由于多媒体数据规模的增加和算法的复杂性更加,它在增强计算机视觉算法中起着至关重要的作用。本研究引入了多种模型,其中介绍了最新的Spearman相关分析算法（带Rank的典型相关分析）及其代数扩展和深度学习模型。此外,这其中的大多数模型都受到了迁移学习方法的启发。本模型引入了非线性多维数据集,由于数据的非线性,多维数据及其对应的应用程序面临多个挑战。本文提出的模型通过与问题相关的数据集的分析,提出了针对复杂性问题的非线性优化和降维的解决方案。本研究主要分为以下三个部分:首先介绍了Spearman相关算法的内核扩展,涉及多维数据集的非线性问题,从一维Spearman相关分析算法到扩展的二维Spearman相关分析算法,进而扩展为三维Spearman相关分析算法等。此外,还介绍了Spearman相关算法及其扩展算法在具有迁移学习方法模型中的运用。然后,第二部分将提出的Spearman相关算法进一步扩展并用于多维数据集的信息投影和多视图非线性问题的相关模型构建。最后,利用本文提出的Spearman相关扩展算法解决了数据及图像分辨率... 

【文章来源】：江苏大学江苏省

【文章页数】：166 页

【学位级别】：博士

【文章目录】：
摘要
Abstract
List of Abbreviations
Chapter 1 Introduction
    1.1.Background
    1.2.Problem Statements and Motivations of Proposed Models
    1.3.Contributions of the Dissertation
    1.4.Implementations of Spearman Correlation Analysis in Proposed Models
        1.4.1.Nonlinear Data Driven Processing Via Multi-Dimensional Spearman Correlation Analysis
        1.4.2.Multi-Dimension Projection for Non-Linear Data via Spearman Correlation Analysis(MD-SCA)
        1.4.3.Non-Linear Optimizations Among Low Resolution and High Resoulation Image Via Spearman Correlation Analysis
    1.5.Organization of the Dissertation
Chapter 2 Literature Review
    2.1.Correlation Analysis
        2.1.1.Pearson Correlation Analysis
        2.1.2.Spearman Correlation Analysis
        2.1.3.Kendall Rank Correlation
        2.1.4.Basic Difference between Spearman Correlation Analysis and Kendall Rank Correlation
    2.2.Multivariate Extension of Correlation Analysis
        2.2.1.Canonical Correlation Analysis
        2.2.2.Canonical Correlation Analysis with Ranks
    2.3.Spearman Correlation Analysis for Image-To-Video Data Driven
        2.3.1.Dimension Reduction Approaches for Image-To-Video
        2.3.2.Spearman Correlation Analysis Supporting Frameworks and Libraries
    2.4.Multilinear Subspace Learning Algorithms
    2.5.Systametic Literature Review Inspired by Presented Models and Motivation
        2.5.1.Nonlinear Data Driven Processing Via Multi-Dimensional Spearman Correlation Analysis
        2.5.2.Multi-Dimension Projection for Non-Linear Data via Spearman Correlation Analysis(MD-SCA)
        2.5.3.Non-Linear Optimizations Among Low Resolution and High Resoulation Image Via Spearman Correlation Analysis
        2.5.4.Conclusion and Summary
Chapter 3 Nonlinear Data Driven Processing Via Multi-Dimensional Spearman Correlation Analysis
    3.1.Transfer Learning
    3.2.Implementation of Classical One-Dimensional Spearman Correlation Analysis with Transfer Learning Approach for Background and Scene Modelling
    3.3.The Systametic Approach of Proposed Model
        3.3.1.Overlapped Camera”EPFL”Video Dataset
        3.3.2.Video Structural and Semantic Segmentation
        3.3.3.Video Discription of Overlapped Scene
        3.3.4.Videos’Deep Visual Features Extraction
        3.3.5.Video Analysis for Background and Foreground via Spearman Correlation Analysis
        3.3.6.Implementation of Pairwise Cosine Distance on Videos’Rank Correlations
        3.3.7.Results
        3.3.8.Disscussion and Analysis
    3.4.Implementation of Spearman Correlation Analysis on Multi-Domain Datasets
        3.4.1.Hyperparameters Optimization in Deep Learning Models
        3.4.2.Bench Mark Datasets for Learning Montone Conditions
        3.4.3.Heart Disease
        3.4.4.Breast Cancer Wisconsin
        3.4.5.Liver Disorders
        3.4.6.Vehicle Silhouettes
        3.4.7.Glass
        3.4.8.Titanic
        3.4.9.Result Optimization of the Bench Mark Datasets Under the ROC Curve
    3.5.Multivariate Extension as Two-Dimensional Spearman Correlation Analysis(2D-SCA)
        3.5.1.Introduced Two-Dimensional Spearman Correlation Analysis
        3.5.2.Implementation of Proposed Approach on Two-Dimensional Data
        3.5.3.Setting and Primitives of Implementations
        3.5.4.Results
        3.5.5.Disscusion and Analysis
    3.6.Deep Three Dimensional Spearman Correlation Analysis(D3D-SCA)
        3.6.1.Customized Inception-V3
        3.6.2.Contribution of Spearman Correlation Analysis for Video Analysis
        3.6.3.The Novel Three-Dimensional Spearman Correlation Analysis
    3.7.Customized Xception Classifiers
    3.8.Experiment
        3.8.1.Industrial Product Datasets Encoded as Mode Flattening3D(Video)Pattern Data
        3.8.2.Image Frame Blocks for Visual Feature Maps
        3.8.3.Implementation of Pairwise D3D-Spearman Correlation
        3.8.4.Classification and Auto-Update of Model
        3.8.5.Application Deployed Server and Framework
    3.9.Results
    3.10.Conclusion and Analysis
Chapter 4 Multi-Dimension Projection for Non-Linear Data Via Spearman Correlation Analysis(MD-SCA)
    4.1.Introduction
    4.2.Problem Statements and Motivation
    4.3.Extension for Multi-Dimensional Informative Projections via Spearman Correlation Analysis(MD-SCA)
    4.4.Preliminaries of Multi-Dimensional Projection via Spearman Correlation Analysis(MD-SCA)
        4.4.1.Introduced Extension of Spearman Correlation Analysis for Multi-Dimensional Informative Projections
        4.4.2.Transformation from the Dual Representation Theory
    4.5.Implementation
        4.5.1.Dataset
        4.5.2.Deep Visual Features Map
        4.5.3.Implementation of MD-SCA in Proposed Deep Learning Model
    4.6.Server and Framework Specification
    4.7.Results
    4.8.Discussion Analysis and Conclusion
Chapter 5 Non-Linear Optimizations Among Low Resolution and High Resoulation Image Via Spearman Correlation Analysis
    5.1.Introduction
    5.2.Introduce Systematic Architecture of Proposed Model and Implementation
        5.2.1.Rule Induction of Proposed Deep Learning Model
        5.2.2.Setting of Hyperpermters in Convolutional Neural Network
        5.2.3.Convolutional Layers
        5.2.4.Tuning of Pooling Layer
        5.2.5.Tuning of Normalization Layers
        5.2.6.Notation of Spearman Correlation Analysis
        5.2.7.Extension of Spearman Correlation Analysis for Correlation between LR and HR Images
        5.2.8.Mapping Of Correlational Features from LR to Relative HR Correlational Features
        5.2.9.Radial Basis Function Network(RBFN)
        5.2.10.Xception Classifier
    5.3.Implementation of Deep Spearman Correlation Analysis
        5.3.1.MNIST Writing Dataset
        5.3.2.Reid Vehicle License Number Plate Dataset
    5.4.Server Specification
    5.5.Conclusion and Future Work
Chapter 6 Conclusion,Discussion and Analysis
    6.1.Conclusion with Consequences of Studies
    6.2.Future Work
References
Publications
Acknowledgement



本文编号：3451017