WO2018040387A1 - 基于支持向量数据描述的特征提取及分类方法及其系统 - Google Patents

基于支持向量数据描述的特征提取及分类方法及其系统 Download PDF

Info

Publication number
WO2018040387A1
WO2018040387A1 PCT/CN2016/110747 CN2016110747W WO2018040387A1 WO 2018040387 A1 WO2018040387 A1 WO 2018040387A1 CN 2016110747 W CN2016110747 W CN 2016110747W WO 2018040387 A1 WO2018040387 A1 WO 2018040387A1
Authority
WO
WIPO (PCT)
Prior art keywords
new feature
classification
support vector
sample
samples
Prior art date
Application number
PCT/CN2016/110747
Other languages
English (en)
French (fr)
Inventor
张莉
卢星凝
王邦军
李凡长
张召
Original Assignee
苏州大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 苏州大学 filed Critical 苏州大学
Priority to US15/738,066 priority Critical patent/US20180322416A1/en
Priority to EP16907682.5A priority patent/EP3346419A4/en
Publication of WO2018040387A1 publication Critical patent/WO2018040387A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the present invention relates to the field of feature extraction technologies, and in particular, to a feature extraction and classification method based on support vector data description and a system thereof.
  • Feature extraction is a commonly used dimension reduction method, mainly used to process tasks with a large number of objects.
  • the samples involved in these tasks are generally large-capacity data with fixed features, which can be binary, discrete multi-valued or continuous data.
  • the use of all the information of all the data is more conducive to obtaining accurate judgment decisions, but in actual operation, the original information of the data often contains relevant, noise or even redundant variables or attributes, wrong
  • the direct application of data processing results in a large amount of cost, including memory capacity size, time complexity, and decision accuracy.
  • feature extraction methods are needed to find compact sample information in the raw data.
  • Feature extraction is to construct a new feature subset by capturing key association information from the original input data.
  • each new feature is a function map of all the original features.
  • feature extraction methods based on Support Vector Machine (SVM) are often used.
  • SVM is a two-class method for constructing hyperplanes. It constructs a classification between multiple types of data through one-to-one and one-to-many modes, and calculates the distance from the sample to the hyperplane to construct new features. This method fully considers different types of data information, but when the amount of data is large, the computational complexity will be very obvious, especially the one-to-many method.
  • the object of the present invention is to provide a feature extraction and classification method based on support vector data description and a system thereof, which can reduce the calculation amount at the time of feature extraction and improve the speed of data classification.
  • the present invention provides a feature extraction and classification method based on support vector data description, including:
  • each of the Euclidean distances and the radius of the corresponding hypersphere model into a new feature relationship, to obtain a new feature sample corresponding to each of the samples; each set of the new feature samples is a new feature sample set;
  • the new feature sample set is classified by using a preset classification algorithm to obtain a classification result.
  • the obtaining process of the plurality of the supersphere models is specifically:
  • the support vector data description algorithm is used to train the J training subsets to obtain J supersphere models.
  • the new feature relationship is specifically:
  • the new feature sample is R j is a radius of the hypersphere model corresponding to the jth training subset, and a j is a spherical center of the hypersphere model corresponding to the jth training subset.
  • the preset classification algorithm comprises:
  • Neural network classification algorithm or support vector machine classification algorithm are Neural network classification algorithm or support vector machine classification algorithm.
  • the present invention also provides a feature extraction and classification system based on support vector data description, including:
  • a distance calculating unit for respectively calculating an Euclidean distance of each sample to a center of the plurality of hypersphere models corresponding to the various data types; wherein the plurality of the supersphere models are pre-trained by using a support vector data description algorithm;
  • a new feature generating unit configured to bring a radius of each of the Euclidean distances and the corresponding hypersphere model into a new feature relationship, to obtain a new feature sample corresponding to each of the samples; and each of the new feature samples
  • the collection is a new feature sample set
  • the classification unit is configured to perform classification processing on the new feature sample set by using a preset classification algorithm to obtain a classification result.
  • the classification unit is:
  • Neural network classifier or support vector machine classifier are Neural network classifier or support vector machine classifier.
  • the invention provides a feature extraction and classification method based on support vector data description, and calculates an Euclidean distance of a sample to a plurality of preset hypersphere models, and according to the Euclidean distance and the corresponding hypersphere model
  • the spherical core calculates the new feature samples corresponding to the samples, and then the new feature sample sets are obtained, and then classified. That is, the present invention adopts the hypersphere model in the support vector data description algorithm to perform feature extraction operation, and then classifies the extracted new feature samples, and the calculation amount is small compared with the SVM algorithm, thereby improving the speed of data classification.
  • the present invention also provides a feature extraction and classification system based on support vector data description, which also has the above effects, and details are not described herein again.
  • FIG. 1 is a flowchart of a process of feature extraction and classification based on support vector data description provided by the present invention
  • FIG. 2 is a schematic structural diagram of a feature extraction and classification system based on support vector data description provided by the present invention.
  • the core of the invention is to provide a feature extraction and classification method based on support vector data description and a system thereof, which can reduce the calculation amount in feature extraction and improve the speed of data classification.
  • FIG. 1 is a flowchart of a process for feature extraction and classification based on support vector data description according to the present invention. The method includes:
  • Step s101 respectively calculating an Euclidean distance of each sample to a center of a plurality of hypersphere models corresponding to various data types; wherein, the plurality of hypersphere models are pre-trained by using a support vector data description algorithm;
  • Step s102 Bring each Euclidean distance and the radius of the corresponding hypersphere model into a new feature relationship, and obtain a new feature sample corresponding to each sample; the set of each new feature sample is a new feature sample set;
  • Step s103 Perform a classification process on the new feature sample set by using a preset classification algorithm to obtain a classification result.
  • the preset classification algorithm here includes:
  • Neural network classification algorithm or support vector machine classification algorithm. Of course, other classifications can be used.
  • the algorithm is not limited by the present invention.
  • the obtaining process of the plurality of hypersphere models is specifically:
  • the support vector data description algorithm is used to train the J training subsets to obtain J hypersphere models.
  • the new feature relationship is specifically:
  • the new feature sample is R j is the radius of the hypersphere model corresponding to the jth training subset, and a j is the center of the hypersphere model corresponding to the jth training subset.
  • how many hypersphere models are specifically included are determined according to the actual number of data categories, and the present invention does not limit the number of categories and contents of the training subset.
  • the data dimension of the original training sample is m, that is, when the new feature sample of the sample is calculated by using the hypersphere model, the dimension of the original sample is m, and the new feature relation is known, and the dimension of the new feature sample is J, Moreover, the general category number J is smaller than m, so the feature extraction method based on the support vector data description adopted by the present invention can achieve the purpose of data dimensionality reduction.
  • Table 1 is a description of the Isolet data set in a specific embodiment
  • Table 2 is a comparison result of the classification effect of the present invention and the SVM algorithm
  • Table 3 It is the result of comparing the execution time of the invention with the SVM algorithm.
  • the invention provides a feature extraction and classification method based on support vector data description, and calculates an Euclidean distance of a sample to a plurality of preset hypersphere models, and according to the Euclidean distance and the corresponding hypersphere model
  • the spherical core calculates the new feature samples corresponding to the samples, and then the new feature sample sets are obtained, and then classified. That is, the present invention adopts the hypersphere model in the support vector data description algorithm to perform feature extraction operation, and then classifies the extracted new feature samples.
  • the calculation amount is small, the classification effect is good, the execution time is short, and the improvement is improved. The speed of data classification.
  • FIG. 2 is a schematic structural diagram of a feature extraction and classification system based on support vector data description.
  • the system includes:
  • a distance calculating unit 11 for respectively calculating an Euclidean distance of each sample to a center of a plurality of hypersphere models corresponding to various data types; wherein the plurality of hypersphere models are pre-trained by using a support vector data description algorithm;
  • a new feature generating unit 12 is configured to bring a radius of each Euclidean distance and its corresponding hypersphere model into a new feature relationship, to obtain a new feature sample corresponding to each sample; each set of new feature samples is a new feature sample set;
  • the classification unit 13 is configured to classify the new feature sample set by using a preset classification algorithm to obtain a classification result.
  • the classification unit 13 here is:
  • Neural network classifier or support vector machine classifier are Neural network classifier or support vector machine classifier. Of course, the present invention does not limit this.
  • the invention provides a feature extraction and classification system based on support vector data description, and calculates an Euclidean distance of a sample to a plurality of preset hypersphere models, and according to the Euclidean distance and the corresponding hypersphere model
  • the spherical core calculates the new feature samples corresponding to the samples, and then the new feature sample sets are obtained, and then classified. That is, the present invention adopts the hypersphere model in the support vector data description algorithm to perform feature extraction operation, and then classifies the extracted new feature samples.
  • the calculation amount is small, the classification effect is good, the execution time is short, and the improvement is improved. The speed of data classification.

Abstract

一种基于支持向量数据描述的特征提取及分类方法，包括分别计算每个样本到对应于各种数据类型的多个超球体模型的球心的欧氏距离；其中多个超球体模型采用支持向量数据描述算法预先训练得到（s101）；将各个欧氏距离与其对应的超球体模型的半径带入新特征关系式，得到每个样本对应的新特征样本；各个新特征样本的集合为新特征样本集（s102）；采用预设分类算法对新特征样本集进行分类处理，得到分类结果（s103）。本方法能够减小特征提取时的计算量，提高数据分类的速度。一种采用上述方法的基于支持向量数据描述的特征提取及分类系统，具有上述优点。

Description

基于支持向量数据描述的特征提取及分类方法及其系统
本申请要求于2016年8月30日提交中国专利局、申请号为201610767804.3、发明名称为“基于支持向量数据描述的特征提取及分类方法及其系统”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及特征提取技术领域，特别是涉及一种基于支持向量数据描述的特征提取及分类方法及其系统。
背景技术
特征提取作为一种常用的降维方法，主要用来处理包含大量对象的任务。这些任务中涉及的样本一般都是有着固定特征的大容量数据，可以是二值的，离散多值的或者连续的数据。在进行数据处理过程中，使用所有数据的全部信息虽更有利于获得精准的判断决策，但是在实际操作时，数据的原始信息中往往会包含了相关、噪声甚至冗余的变量或者属性，不对数据进行处理而直接应用，会导致大量的成本支出，这些可能的成本包括内存容量大小、时间复杂度和决策精度等。为了提高数据存储和计算性能，需要采用特征提取方法来找到原始数据中紧凑的样本信息。
特征提取是通过从原始输入数据中捕获关键关联信息，来构建一个新的特征子集。在特征提取方法中，每一个新的特征都是所有原始特征的函数映射。目前多采用基于支持向量机(Support Vector Machine，SVM)的特征提取方法。SVM是一种构建超平面的二分类方法，通过一对一和一对多的模式来构建多类数据之间的分类，并计算样本到超平面的距离来构建新特征。该方法充分考虑了不同类别的数据信息，但是当数据量较大时，计算复杂度也会十分明显，尤其是一对多方法。
因此，如何提供一种计算量小的基于支持向量数据描述的特征提取及分类方法及其系统是本领域技术人员目前需要解决的问题。
发明内容
本发明的目的是提供一种基于支持向量数据描述的特征提取及分类方法及其系统，能够减小特征提取时的计算量，提高数据分类的速度。
为解决上述技术问题，本发明提供了一种基于支持向量数据描述的特征提取及分类方法，包括：
分别计算每个样本到对应于各种数据类型的多个超球体模型的球心的欧氏距离；其中多个所述超球体模型采用支持向量数据描述算法预先训练得到；
将各个所述欧氏距离与其对应的所述超球体模型的半径带入新特征关系式，得到每个所述样本对应的新特征样本；各个所述新特征样本的集合为新特征样本集；
采用预设分类算法对所述新特征样本集进行分类处理，得到分类结果。
优选地，所述多个所述超球体模型的获得过程具体为：
将预先获得的原始训练样本按照数据类别分为J个训练子集Xj＝{(xi,yi)|xi∈Rm,yi＝j,i＝1,…,nj}，其中，j为数据类别，j＝1,…,J；Rm为维度为m的实数集合，n为所述训练样本总数，nj表示第j个所述训练子集中的样本数目，
Figure PCTCN2016110747-appb-000001
采用所述支持向量数据描述算法分别对J个所述训练子集进行训练，得到J个所述超球体模型。
优选地，所述新特征关系式具体为：
Figure PCTCN2016110747-appb-000002
其中，新特征样本为
Figure PCTCN2016110747-appb-000003
Figure PCTCN2016110747-appb-000004
Rj为第j个所述训练子集对应的所述超球体模型的半径，aj为第j个所述训练子集对应的所述超球体模型的球心。
优选地，所述预设分类算法包括：
神经网络分类算法或支持向量机分类算法。
为解决上述技术问题，本发明还提供了一种基于支持向量数据描述的特征提取及分类系统，包括：
距离计算单元，用于分别计算每个样本到对应于各种数据类型的多个超球体模型的球心的欧氏距离；其中多个所述超球体模型采用支持向量数据描述算法预先训练得到；
新特征生成单元，用于将各个所述欧氏距离与其对应的所述超球体模型的半径带入新特征关系式，得到每个所述样本对应的新特征样本；各个所述新特征样本的集合为新特征样本集；
分类单元，用于采用预设分类算法对所述新特征样本集进行分类处理，得到分类结果。
优选地，所述分类单元为：
神经网络分类器或支持向量机分类器。
本发明提供了一种基于支持向量数据描述的特征提取及分类方法，计算样本到多个预设的超球体模型的球心的欧氏距离，并依据该欧氏距离以及对应的超球体模型的球心计算得到样本对应的新特征样本，进而得到新特征样本集，进而进行分类。即本发明采用支持向量数据描述算法中的超球体模型进行特征提取操作，进而对提取出来的新特征样本进行分类处理，相比SVM算法，计算量小，提高了数据分类的速度。本发明还提供了一种基于支持向量数据描述的特征提取及分类系统，也具有上述效果，在此不再赘述。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附 图。
图1为本发明提供的一种基于支持向量数据描述的特征提取及分类方法的过程的流程图；
图2为本发明提供的一种基于支持向量数据描述的特征提取及分类系统的结构示意图。
具体实施方式
本发明的核心是提供一种基于支持向量数据描述的特征提取及分类方法及其系统，能够减小特征提取时的计算量，提高数据分类的速度。
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
本发明提供了一种基于支持向量数据描述的特征提取及分类方法，参见图1所示，图1为本发明提供的一种基于支持向量数据描述的特征提取及分类方法的过程的流程图；该方法包括：
步骤s101：分别计算每个样本到对应于各种数据类型的多个超球体模型的球心的欧氏距离；其中，多个超球体模型采用支持向量数据描述算法预先训练得到；
步骤s102：将各个欧氏距离与其对应的超球体模型的半径带入新特征关系式，得到每个样本对应的新特征样本；各个新特征样本的集合为新特征样本集；
步骤s103：采用预设分类算法对新特征样本集进行分类处理，得到分类结果。
其中，这里的预设分类算法包括：
神经网络分类算法或支持向量机分类算法。当然，也可采用其他分类 算法，本发明对此不作限定。
作为优选地，多个超球体模型的获得过程具体为：
将预先获得的原始训练样本按照数据类别分为J个训练子集Xj＝{(xi,yi)|xi∈Rm,yi＝j,i＝1,…,nj}，其中，j为数据类别，j＝1,…,J；Rm为维度为m的实数集合，n为训练样本总数，nj表示第j个训练子集中的样本数目，
Figure PCTCN2016110747-appb-000005
采用支持向量数据描述算法分别对J个训练子集进行训练，得到J个超球体模型。
其中，新特征关系式具体为：
Figure PCTCN2016110747-appb-000006
其中，新特征样本为
Figure PCTCN2016110747-appb-000007
Figure PCTCN2016110747-appb-000008
Rj为第j个训练子集对应的超球体模型的半径，aj为第j个训练子集对应的超球体模型的球心。
在计算新特征样本集时，具体包括多少个超球体模型根据实际的数据类别个数决定，本发明不限定训练子集的类别数量以及内容。
可以理解的是，原始训练样本的数据维度为m，即采用超球体模型计算样本的新特征样本时，原始样本的维度为m，由上述新特征关系式可知，新特征样本的维度为J，并且一般类别数J小于m，故本发明采用的基于支持向量数据描述的特征提取方法能够实现数据降维的目的。
为进一步理解本发明的有益效果，参见表1-表3所示，表1为一种具体实施例中Isolet数据集的描述情况，表2为本发明与SVM算法的分类效果比较结果，表3为本发明与SVM算法的执行时间比较结果。
数据集 类别 特征数 样本总数 训练样本数 测试样本数
Isolet 26 617 7797 6238 1559
表1 一种具体实施例中Isolet数据集的描述情况
Figure PCTCN2016110747-appb-000009
表2 本发明与SVM算法的分类效果比较结果(％)
Figure PCTCN2016110747-appb-000010
表3 本发明与SVM算法的执行时间比较结果
本发明提供了一种基于支持向量数据描述的特征提取及分类方法，计算样本到多个预设的超球体模型的球心的欧氏距离，并依据该欧氏距离以及对应的超球体模型的球心计算得到样本对应的新特征样本，进而得到新特征样本集，进而进行分类。即本发明采用支持向量数据描述算法中的超球体模型进行特征提取操作，进而对提取出来的新特征样本进行分类处理，相比SVM算法，计算量小，分类效果好，执行时间短，提高了数据分类的速度。
本发明还提供了一种基于支持向量数据描述的特征提取及分类系统，参见图2所示，图2为本发明提供的一种基于支持向量数据描述的特征提取及分类系统的结构示意图。该系统包括：
距离计算单元11，用于分别计算每个样本到对应于各种数据类型的多个超球体模型的球心的欧氏距离；其中多个超球体模型采用支持向量数据描述算法预先训练得到；
新特征生成单元12，用于将各个欧氏距离与其对应的超球体模型的半径带入新特征关系式，得到每个样本对应的新特征样本；各个新特征样本的集合为新特征样本集；
分类单元13，用于采用预设分类算法对新特征样本集进行分类处理，得到分类结果。
具体的，这里的分类单元13为：
神经网络分类器或支持向量机分类器。当然，本发明对此不作限定。
本发明提供了一种基于支持向量数据描述的特征提取及分类系统，计算样本到多个预设的超球体模型的球心的欧氏距离，并依据该欧氏距离以及对应的超球体模型的球心计算得到样本对应的新特征样本，进而得到新特征样本集，进而进行分类。即本发明采用支持向量数据描述算法中的超球体模型进行特征提取操作，进而对提取出来的新特征样本进行分类处理，相比SVM算法，计算量小，分类效果好，执行时间短，提高了数据分类的速度。
需要说明的是，在本说明书中，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
对所公开的实施例的上述说明，使本领域专业技术人员能够实现或使用本发明。对这些实施例的多种修改对本领域的专业技术人员来说将是显而易见的，本文中所定义的一般原理可以在不脱离本发明的精神或范围的情况下，在其他实施例中实现。因此，本发明将不会被限制于本文所示的这些实施例，而是要符合与本文所公开的原理和新颖特点相一致的最宽的范围。

Claims (6)

  1. 一种基于支持向量数据描述的特征提取及分类方法，其特征在于，包括：
    分别计算每个样本到对应于各种数据类型的多个超球体模型的球心的欧氏距离；其中多个所述超球体模型采用支持向量数据描述算法预先训练得到；
    将各个所述欧氏距离与其对应的所述超球体模型的半径带入新特征关系式，得到每个所述样本对应的新特征样本；各个所述新特征样本的集合为新特征样本集；
    采用预设分类算法对所述新特征样本集进行分类处理，得到分类结果。
  2. 根据权利要求1所述的方法，其特征在于，所述多个所述超球体模型的获得过程具体为：
    将预先获得的原始训练样本按照数据类别分为J个训练子集Xj＝{(xi,yi)|xi∈Rm,yi＝j,i＝1,…,nj}，其中，j为数据类别，j＝1,…,J；Rm为维度为m的实数集合，n为所述训练样本总数，nj表示第j个所述训练子集中的样本数目，
    Figure PCTCN2016110747-appb-100001
    采用所述支持向量数据描述算法分别对J个所述训练子集进行训练，得到J个所述超球体模型。
  3. 根据权利要求2所述的方法，其特征在于，所述新特征关系式具体为：
    Figure PCTCN2016110747-appb-100002
    其中，新特征样本为
    Figure PCTCN2016110747-appb-100003
    Figure PCTCN2016110747-appb-100004
    Rj为第j个所述训练子集对应的所述超球体模型的半径，aj为第j个所述训练子集对应的所述超球体模型的球心。
  4. 根据权利要求3所述的方法，其特征在于，所述预设分类算法包 括：
    神经网络分类算法或支持向量机分类算法。
  5. 一种基于支持向量数据描述的特征提取及分类系统，其特征在于，包括：
    距离计算单元，用于分别计算每个样本到对应于各种数据类型的多个超球体模型的球心的欧氏距离；其中多个所述超球体模型采用支持向量数据描述算法预先训练得到；
    新特征生成单元，用于将各个所述欧氏距离与其对应的所述超球体模型的半径带入新特征关系式，得到每个所述样本对应的新特征样本；各个所述新特征样本的集合为新特征样本集；
    分类单元，用于采用预设分类算法对所述新特征样本集进行分类处理，得到分类结果。
  6. 根据权利要求5所述的系统，其特征在于，所述分类单元为：
    神经网络分类器或支持向量机分类器。
PCT/CN2016/110747 2016-08-30 2016-12-19 基于支持向量数据描述的特征提取及分类方法及其系统 WO2018040387A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US15/738,066 US20180322416A1 (en) 2016-08-30 2016-12-19 Feature extraction and classification method based on support vector data description and system thereof
EP16907682.5A EP3346419A4 (en) 2016-08-30 2016-12-19 METHOD OF CHARACTERISTIC EXTRACTION AND CLASSIFICATION BASED ON MEDIUM VECTOR DATA DESCRIPTION AND SYSTEM THEREOF

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610767804.3A CN106446931A (zh) 2016-08-30 2016-08-30 基于支持向量数据描述的特征提取及分类方法及其系统
CN201610767804.3 2016-08-30

Publications (1)

Publication Number Publication Date
WO2018040387A1 true WO2018040387A1 (zh) 2018-03-08

Family

ID=58091398

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/110747 WO2018040387A1 (zh) 2016-08-30 2016-12-19 基于支持向量数据描述的特征提取及分类方法及其系统

Country Status (4)

Country Link
US (1) US20180322416A1 (zh)
EP (1) EP3346419A4 (zh)
CN (1) CN106446931A (zh)
WO (1) WO2018040387A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111325227A (zh) * 2018-12-14 2020-06-23 深圳先进技术研究院 数据特征提取方法、装置及电子设备

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108960269B (zh) * 2018-04-02 2022-05-27 创新先进技术有限公司 数据集的特征获取方法、装置及计算设备
CN108960056B (zh) * 2018-05-30 2022-06-03 西南交通大学 一种基于姿态分析和支持向量数据描述的跌倒检测方法
CN109492664B (zh) * 2018-09-28 2021-10-22 昆明理工大学 一种基于特征加权模糊支持向量机的音乐流派分类方法及系统
CN111382210B (zh) * 2018-12-27 2023-11-10 中国移动通信集团山西有限公司 一种分类方法、装置及设备
CN109974782B (zh) * 2019-04-10 2021-03-02 郑州轻工业学院 基于大数据敏感特征优化选取的设备故障预警方法及系统
CN111639065B (zh) * 2020-04-17 2022-10-11 太原理工大学 一种基于配料数据的多晶硅铸锭质量预测方法及系统
CN111985152B (zh) * 2020-07-28 2022-09-13 浙江大学 一种基于二分超球面原型网络的事件分类方法
CN112632857A (zh) * 2020-12-22 2021-04-09 广东电网有限责任公司广州供电局 一种配电网的线损确定方法、装置、设备和存储介质
WO2022174436A1 (zh) * 2021-02-22 2022-08-25 深圳大学 分类模型增量学习实现方法、装置、电子设备及介质
CN114104666A (zh) * 2021-11-23 2022-03-01 西安华创马科智能控制系统有限公司 煤矸识别方法及煤矿运送系统

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140376804A1 (en) * 2013-06-21 2014-12-25 Xerox Corporation Label-embedding view of attribute-based recognition
CN104361342A (zh) * 2014-10-23 2015-02-18 同济大学 一种基于几何不变形状特征的在线植物物种识别方法
CN104750875A (zh) * 2015-04-23 2015-07-01 苏州大学 一种机器错误数据分类方法及系统

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140376804A1 (en) * 2013-06-21 2014-12-25 Xerox Corporation Label-embedding view of attribute-based recognition
CN104361342A (zh) * 2014-10-23 2015-02-18 同济大学 一种基于几何不变形状特征的在线植物物种识别方法
CN104750875A (zh) * 2015-04-23 2015-07-01 苏州大学 一种机器错误数据分类方法及系统

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
QIN, YUPING ET AL.: "Study on Multi-Clas Text Classification Algorithm Based on Hypersphere Support Vector Machines", vol. 44, no. 19, 1 July 2008 (2008-07-01), pages 166 - 168, XP009511510 *
See also references of EP3346419A4 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111325227A (zh) * 2018-12-14 2020-06-23 深圳先进技术研究院 数据特征提取方法、装置及电子设备
CN111325227B (zh) * 2018-12-14 2023-04-07 深圳先进技术研究院 数据特征提取方法、装置及电子设备

Also Published As

Publication number Publication date
EP3346419A1 (en) 2018-07-11
US20180322416A1 (en) 2018-11-08
CN106446931A (zh) 2017-02-22
EP3346419A4 (en) 2019-07-03

Similar Documents

Publication Publication Date Title
WO2018040387A1 (zh) 基于支持向量数据描述的特征提取及分类方法及其系统
US11593458B2 (en) System for time-efficient assignment of data to ontological classes
CN111079639B (zh) 垃圾图像分类模型构建的方法、装置、设备及存储介质
CN107209861B (zh) 使用否定数据优化多类别多媒体数据分类
JP6928371B2 (ja) 分類器、分類器の学習方法、分類器における分類方法
CN107430625B (zh) 通过集群对文档进行分类
CN107832663A (zh) 一种基于量子理论的多模态情感分析方法
Tuba et al. Adjusted bat algorithm for tuning of support vector machine parameters
Lo et al. Using support vector machine ensembles for target audience classification on Twitter
US11288300B2 (en) Techniques and components to find new instances of text documents and identify known response templates
WO2020114108A1 (zh) 聚类结果的解释方法和装置
Doan et al. Overcoming the challenge for text classification in the open world
WO2021238279A1 (zh) 数据分类方法、分类器训练方法及系统
Li A review of machine learning algorithms for text classification
Sidaoui et al. Binary tree multi-class SVM based on OVA approach and variable neighbourhood search algorithm
Elgeldawi et al. Hyperparameter Tuning for Machine Learning Algorithms Used for Arabic Sentiment Analysis. Informatics 2021, 8, 79
Aparna et al. Comprehensive study and analysis of partitional data clustering techniques
Baitharu et al. Comparison of Kernel selection for support vector machines using diabetes dataset
JP6004014B2 (ja) 学習方法、情報変換装置および学習プログラム
Chen et al. A novel feature selection-based sequential ensemble learning method for class noise detection in high-dimensional data
Ma et al. Research on policy text clustering algorithm based on LDA-Gibbs model
Fu et al. Group based non-sparse localized multiple kernel learning algorithm for image classification
CN110532384A (zh) 一种多任务字典单分类方法、系统、装置及存储介质
Desai An Exploration of the Effectiveness of Machine Learning Algorithms for Text Classification
Wong The use of Big Data in Machine Learning Algorithm

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16907682

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 15738066

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE