WO2022166363A1 - 一种基于近邻子空间划分高光谱影像波段选择方法及系统 - Google Patents

一种基于近邻子空间划分高光谱影像波段选择方法及系统 Download PDF

Info

Publication number
WO2022166363A1
WO2022166363A1 PCT/CN2021/135928 CN2021135928W WO2022166363A1 WO 2022166363 A1 WO2022166363 A1 WO 2022166363A1 CN 2021135928 W CN2021135928 W CN 2021135928W WO 2022166363 A1 WO2022166363 A1 WO 2022166363A1
Authority
WO
WIPO (PCT)
Prior art keywords
bands
band
subspace
hyperspectral image
correlation coefficient
Prior art date
Application number
PCT/CN2021/135928
Other languages
English (en)
French (fr)
Inventor
朱信忠
徐慧英
王俊
唐厂
赵建民
Original Assignee
浙江师范大学
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 浙江师范大学 filed Critical 浙江师范大学
Publication of WO2022166363A1 publication Critical patent/WO2022166363A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00Scenes; Scene-specific elements
    • G06V20/10Terrestrial scenes
    • G06V20/194Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17Systems in which incident light is modified in accordance with the properties of the material investigated
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/01Arrangements or apparatus for facilitating the optical investigation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • G06F18/241Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24147Distances to closest patterns, e.g. nearest neighbour classification
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N2021/1793Remote sensing
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02ATECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A40/00Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
    • Y02A40/10Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture

Definitions

  • the present application relates to the technical field of hyperspectral remote sensing image band selection, and in particular, to a method and system for selecting a hyperspectral image band based on neighbor subspace division.
  • hyperspectral images Compared with traditional RGB images, hyperspectral images have the characteristics of rich information, large number of bands, and high resolution, so they are widely used in target detection, environmental monitoring, mineral exploration, agricultural resources survey and marine research. Because it contains a large amount of band information, and the feature similarity between adjacent bands is high, this increases the computational complexity of hyperspectral image classification to a certain extent, and also affects the final classification accuracy of the classifier. Therefore, it is necessary to reduce the dimensionality of hyperspectral images to solve the problem of dimensional disaster.
  • Feature extraction and feature selection have always been research hotspots in the field of data dimensionality reduction, and these two methods are also commonly used in hyperspectral image dimensionality reduction. Among them, in hyperspectral images, feature selection can also be called band selection.
  • the purpose of this application is to provide a method and system for selecting a hyperspectral image band based on neighbor subspace division, aiming at the defects of the prior art.
  • a method for selecting a hyperspectral image band based on neighbor subspace division comprising:
  • step S1 the correlation coefficient between any two adjacent bands included in the input hyperspectral image cube is calculated to obtain the vector of the correlation coefficient, which is expressed as:
  • X ⁇ R W ⁇ H ⁇ L represents the hyperspectral image cube
  • L represents the number of bands contained in the hyperspectral image cube
  • W and H represent the width and height of each band, respectively
  • X i represents the number of bands in the hyperspectral image dataset.
  • X j represents the j-th band in the hyperspectral image dataset
  • X i (m,n) represents the value of the m-th row and n-th column of X i
  • R represents the vector of correlation coefficients
  • V i represents the i-th band
  • V j represents the pixel difference of the j-th band
  • V o represents the pixel difference of the o-th band
  • X o represents the o-th band in the hyperspectral image dataset
  • o ⁇ i,j ⁇ .
  • the minimum value points of the correlation coefficients are screened by using the toolbox of MATLAB.
  • findpeaks represents a built-in MATLAB function
  • R represents the vector of correlation coefficients between adjacent bands in the entire dataset
  • y represents a vector containing all minima.
  • step S3 the information entropy of the band in each subspace is calculated, which is expressed as:
  • represents the entire sample space
  • p(i) represents the probability of sample i appearing in the image
  • H represents the information entropy of sample i.
  • step S3 select the required number of characteristic bands from the subspace, and be expressed as:
  • Z represents the number of bands contained in each band subspace
  • L represents the number of bands contained in the entire data set
  • K represents the number of characteristic bands to be selected
  • S represents the selected bands.
  • a hyperspectral image band selection system based on neighbor subspace division including:
  • the calculation module is used to input the hyperspectral image cube, and calculate the correlation coefficient between any two adjacent bands contained in the input hyperspectral image cube, and obtain the vector of the correlation coefficient;
  • the screening module is used to find all the correlation coefficient extreme points according to the obtained correlation coefficient vector, and filter out the correlation coefficient minimum value points from all the found correlation coefficient extreme value points, and pass the selected correlation coefficient minimum value points. point to determine the optimally divided hyperspectral band subspace;
  • the selection module is used to sort the subspaces according to the size of the number of bands in the subspace, calculate the information entropy of the bands in each subspace, and select the required number of characteristic bands from the subspace according to the calculated information entropy.
  • the calculation module calculates the correlation coefficient between any two adjacent bands contained in the input hyperspectral image cube, and obtains the vector of the correlation coefficient, which is expressed as:
  • X ⁇ R W ⁇ H ⁇ L represents the hyperspectral image cube
  • L represents the number of bands contained in the hyperspectral image cube
  • W and H represent the width and height of each band, respectively
  • X i represents the number of bands in the hyperspectral image dataset.
  • X j represents the j-th band in the hyperspectral image dataset
  • X i (m,n) represents the value of the m-th row and n-th column of X i
  • R represents the vector of correlation coefficients
  • V i represents the i-th band
  • V j represents the pixel difference of the j-th band
  • V o represents the pixel difference of the o-th band
  • X o represents the o-th band in the hyperspectral image dataset
  • o ⁇ i,j ⁇ .
  • represents the entire sample space
  • p(i) represents the probability of sample i appearing in the image
  • H represents the information entropy of sample i.
  • a required number of characteristic bands are selected from the subspace, which is expressed as:
  • Z represents the number of bands contained in each band subspace
  • L represents the number of bands contained in the entire data set
  • K represents the number of characteristic bands to be selected
  • S represents the selected bands.
  • the present application finds the grouping critical points according to the curvature change of the correlation coefficients by calculating the correlation coefficient between adjacent bands, and then takes the number of critical points as the final number of clusters.
  • the information entropy is used as the evaluation index for band selection.
  • the information entropy is used as the weight to sort each band, and then the band with the largest information entropy is selected as the characteristic band, which can ensure that the final selected band contains relatively complete information.
  • FIG. 1 is a flowchart of a method for selecting a hyperspectral image band based on neighbor subspace division provided by Embodiment 1;
  • Fig. 2 is a kind of hyperspectral image band selection block diagram based on neighbor subspace division provided by Embodiment 1;
  • FIG. 3 is a schematic diagram of an example of a neighbor subspace division provided by Embodiment 1;
  • Embodiment 4 is a schematic diagram of an example of information entropy of all bands on the Indian Pines data set provided by Embodiment 1;
  • FIG. 5 is a schematic diagram of the Overall-Accuracy curve of different hyperspectral image band selection methods provided in Embodiment 2 on the Botswana dataset;
  • FIG. 6 is a schematic diagram of the Overall-Accuracy curve of different hyperspectral image band selection methods provided in Embodiment 2 on the Salinas data set;
  • FIG. 7 is a schematic diagram of the Overall-Accuracy curve of different hyperspectral image band selection methods provided in the second embodiment on the Indian Pines dataset.
  • the present application provides a method and system for selecting a hyperspectral image band based on neighbor subspace division.
  • a method for selecting a hyperspectral image band based on neighbor subspace division includes:
  • this embodiment proposes a hyperspectral image band selection method based on neighbor subspace division, which is referred to as SEASP for short.
  • SEASP adopts a combination of clustering and sorting to solve the problem of low precision when the number of cluster centers is small.
  • step S11 a hyperspectral image cube is input, and a correlation coefficient between any two adjacent bands included in the input hyperspectral image cube is calculated to obtain a vector of correlation coefficients.
  • a hyperspectral image can be regarded as a two-dimensional matrix composed of pixels with different spectral values, so let X i represent the ith band in the hyperspectral image dataset. Then the correlation coefficient between the bands X i and X j is calculated as follows:
  • X ⁇ R W ⁇ H ⁇ L represents the hyperspectral image cube
  • L represents the number of bands contained in the hyperspectral image cube
  • W and H represent the width and height of each band, respectively
  • X i represents the number of bands in the hyperspectral image dataset.
  • X j represents the j-th band in the hyperspectral image dataset
  • X i (m,n) represents the value of the m-th row and n-th column of X i
  • R represents the vector of correlation coefficients
  • V i represents the i-th band
  • V j represents the pixel difference of the j-th band
  • V o represents the pixel difference of the o-th band
  • X o represents the o-th band in the hyperspectral image dataset
  • the band image contained in it is regarded as a two-dimensional matrix composed of pixels with different spectral values, and then the above calculation method is used to obtain any arbitrary Correlation coefficients between two adjacent bands, resulting in a correlation coefficient vector.
  • step S12 find all the correlation coefficient extreme points according to the obtained correlation coefficient vector, and filter out the correlation coefficient minimum value points from all the found correlation coefficient extreme value points, and pass the selected correlation coefficient minimum value points. point to determine the optimally divided hyperspectral band subspace.
  • the criterion for measuring the division is usually to make the intra-class correlation the highest and the inter-class correlation the lowest.
  • SEASP the correlation between groups is determined mainly according to the change rate of the correlation coefficient. Specifically, if the correlation between the two bands is the smallest in a certain interval, that is, the rate of change of the correlation coefficient is the largest in the interval, it means that the two bands do not belong to the same group with a high probability, and they are two bands. Split point between adjacent groups. Considering the orderliness between the bands and the weak correlation of non-neighboring spaces, the embodiment only calculates the correlation coefficients of adjacent bands. At this time, the clustering problem is transformed into finding discontinuities between groups in an ordered dataset. question. Therefore, according to the obtained correlation coefficient vector, all the minimum values of the correlation coefficients contained in it are calculated, so as to obtain the discontinuous points of all subspaces.
  • findpeaks represents a built-in MATLAB function
  • R represents the vector of correlation coefficients between adjacent bands in the entire dataset
  • y represents a vector containing all minima.
  • the findpeaks function judges whether a point is a peak point, it mainly compares the function value at the point with its adjacent function values. If it is the largest, it is judged as one of the waveforms. crest point. Therefore, when the input parameters are inverted, the obtained peak point is the trough point of the original data.
  • the function value at point x is F(x)
  • F(x) satisfies both F(x) ⁇ F(x-1) and F(x) ⁇ F(x+1)
  • the discontinuous points between the ordered bands are determined by the selection of the minimum point, so as to obtain the divided subspace of the bands. The specific example is shown in Figure 3.
  • step S13 the subspaces are sorted according to the number of bands in the subspace, the information entropy of the bands in each subspace is calculated, and a required number of characteristic bands are selected from the subspace according to the calculated information entropy.
  • Information entropy is used to measure the average amount of information contained in an image. The greater the information entropy of an image, the richer the information contained in the image. Therefore, information entropy is more appropriate as a measure of the importance of the band.
  • the specific calculation method is as follows:
  • represents the entire sample space
  • p(i) represents the probability of sample i appearing in the image
  • H represents the information entropy of sample i.
  • the band with the largest information entropy is selected from each subspace as the characteristic band; when K is greater than the number of divided subspaces, the information entropy is selected from each subspace according to the size of the information entropy.
  • Select S bands are as follows:
  • Z represents the number of bands contained in each band subspace
  • L represents the number of bands contained in the entire data set
  • K represents the number of characteristic bands to be selected
  • S represents the selected bands.
  • the desired feature band subset Y can be finally obtained.
  • the correlation coefficient between adjacent bands is calculated, and the grouping critical points are found according to the curvature change of the correlation coefficient, and then the number of critical points is used as the final number of clusters.
  • the information entropy is used as the evaluation index for band selection. When selecting bands for each cluster, the information entropy is used as the weight to sort each band, and then the band with the largest information entropy is selected as the characteristic band, which can ensure that the final selected band contains relatively complete information.
  • the selected classifiers are KNN, SVM and LDA.
  • the parameter k of the KNN classifier is set to 5.
  • the distance function uniformly used by SVM is a Gaussian kernel function.
  • the factor is 100. Since the above three classifiers are all supervised classification methods, 10% of the entire data set is randomly selected as training samples and the rest of the data is used as test samples when conducting experiments.
  • the present embodiment is also compared with several relatively advanced algorithms at present, which are ASPS_MN, ASPS_IE, TOF, UBS and FNGBS respectively.
  • the number of selected bands ranges from 5 to 50.
  • a total of 12 bands were selected in the experiment, namely 5, 7, 10, 15, 26, 30, 36, 39, 42, 44, 47, and 49.
  • three metrics are used to analyze the classification results, namely overall accuracy (OA), average overall accuracy (AOA) and Kappa coefficient (Kappa). All experiments are run on MATLAB 2016a, the CPU is i7-5500U, 2.40GHz, and the memory is 8Gb.
  • Table 1 shows the AOA and Kappa of competing algorithms on three public datasets. Among them, in the test comparison, the number of bands selected for the three datasets is 36, 5, and 5, respectively, and AOA is the range of 10 running results. To clearly demonstrate the performance of the compared algorithms on the three datasets, the two better classification results are marked in bold in Table 1.
  • the times of all competing algorithms are also calculated on the band selection of the three datasets.
  • the number of bands selected in the three datasets is set to 10, 15, and 20, respectively, showing an increasing trend.
  • Table 2 shows the time it takes for different algorithms to select the same number of bands on different datasets. It can be seen from the results in the table that the time consumed by the algorithm proposed in this embodiment is relatively small, and is comparable to other algorithms to a certain extent.
  • the algorithm proposed in this embodiment is not only simple in principle, but also has certain comparability with other algorithms in the classification performance of the three public data sets, or even better than other algorithms, and its execution speed is also relatively high. Fast, thus verifying the effectiveness and feasibility of the algorithm.
  • This embodiment provides a hyperspectral image band selection system based on neighbor subspace division, including:
  • the calculation module is used to input the hyperspectral image cube, and calculate the correlation coefficient between any two adjacent bands contained in the input hyperspectral image cube, and obtain the vector of the correlation coefficient;
  • the screening module is used to find all the correlation coefficient extreme points according to the obtained correlation coefficient vector, and filter out the correlation coefficient minimum value points from all the found correlation coefficient extreme value points, and pass the selected correlation coefficient minimum value points. point to determine the optimally divided hyperspectral band subspace;
  • the selection module is used to sort the subspaces according to the size of the number of bands in the subspace, calculate the information entropy of the bands in each subspace, and select the required number of characteristic bands from the subspace according to the calculated information entropy.
  • calculation module calculates the correlation coefficient between any two adjacent bands contained in the input hyperspectral image cube, which is expressed as:
  • X ⁇ R W ⁇ H ⁇ L represents the hyperspectral image cube
  • L represents the number of bands contained in the hyperspectral image cube
  • W and H represent the width and height of each band, respectively
  • X i represents the number of bands in the hyperspectral image dataset.
  • X j represents the j-th band in the hyperspectral image dataset
  • X i (m,n) represents the value of the m-th row and n-th column of X i
  • represents the entire sample space
  • p(i) represents the probability of sample i appearing in the image
  • H represents the information entropy of sample i.
  • a required number of characteristic bands are selected from the subspace, which is expressed as:
  • Z represents the number of bands contained in each band subspace
  • L represents the number of bands contained in the entire data set
  • K represents the number of characteristic bands to be selected
  • S represents the selected bands.
  • the correlation coefficient between adjacent bands is calculated, and the grouping critical points are found according to the curvature change of the correlation coefficient, and then the number of critical points is used as the final number of clusters.
  • the information entropy is used as the evaluation index for band selection. When selecting bands for each cluster, the information entropy is used as the weight to sort each band, and then the band with the largest information entropy is selected as the characteristic band, which can ensure that the final selected band contains relatively complete information.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Evolutionary Computation (AREA)
  • Biochemistry (AREA)
  • Analytical Chemistry (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Chemical & Material Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Remote Sensing (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Image Analysis (AREA)

Abstract

一种基于近邻子空间划分高光谱影像波段选择方法及系统,该方法包括:S11.输入高光谱影像立方体,并计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量;S12.根据得到的相关系数的向量寻找所有相关系数极值点,并在寻找到的所有相关系数极值点中筛选出相关系数极小值点,通过筛选出的相关系数极小值点确定最优划分的高光谱波段子空间;S13.根据子空间中波段数的大小对子空间进行排序,并计算每个子空间中波段的信息熵,根据计算得到的信息熵从子空间中选取所需数量的特征波段。该方法可以确保最终选取的波段包含相对完整的信息。

Description

一种基于近邻子空间划分高光谱影像波段选择方法及系统 技术领域
本申请涉及高光谱遥感影像波段选择技术领域,尤其涉及一种基于近邻子空间划分高光谱影像波段选择方法及系统。
背景技术
相比于传统的RGB图像,高光谱影像具有信息量丰富、波段数目多、分辨率高等特点,因此它被广泛应用于目标检测、环境监测、矿物勘探、农业资源调查和海洋研究等方面。由于其包含大量的波段信息,且相邻波段之间的特征相似度较高,这在一定程度上增加了高光谱影像分类的计算复杂度,并且还会影响分类器最终的分类精度。所以需要对高光谱影像进行降维处理,从而解决维度灾难问题。
特征提取和特征选择一直是数据降维领域的研究热点,这两种方法也通常应用于高光谱影像降维。其中,在高光谱影像中,特征选择也可以称为波段选择。
目前来看,大多数基于聚类的波段选择算法都存在对聚类中心数的强敏感性问题。具体表现在,当选取的聚类中心数较少时,精度较低;当选取的聚类中心数超过一定数量时,精度反而会随着聚类中心数的增加呈现出一种下降的趋势。这两种情况显然不满足实际需要。
发明内容
本申请的目的是针对现有技术的缺陷,提供了一种基于近邻子空间划分高光谱影像波段选择方法及系统。
为了实现以上目的,本申请采用以下技术方案:
一种基于近邻子空间划分高光谱影像波段选择方法,包括:
S1.输入高光谱影像立方体,并计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量;
S2.根据得到的相关系数的向量寻找所有相关系数极值点,并在寻找到的 所有相关系数极值点中筛选出相关系数极小值点,通过筛选出的相关系数极小值点确定最优划分的高光谱波段子空间;
S3.根据子空间中波段数的大小对子空间进行排序,并计算每个子空间中波段的信息熵,根据计算得到的信息熵从子空间中选取所需数量的特征波段。
进一步的,所述步骤S1中计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量,表示为:
Figure PCTCN2021135928-appb-000001
Figure PCTCN2021135928-appb-000002
其中,X∈R W×H×L表示高光谱影像立方体,L表示高光谱影像立方体所包含的波段数目,W和H分别表示每个波段的宽度和高度;X i表示高光谱影像数据集中第i个波段;X j表示高光谱影像数据集中第j个波段;X i(m,n)表示X i第m行第n列的值;R表示相关系数的向量;V i表示第i个波段的像素差;V j表示第j个波段的像素差;V o表示第o个波段的像素差;X o表示高光谱影像数据集中第o个波段;
Figure PCTCN2021135928-appb-000003
表示第o个波段的像素均值;o={i,j}。
进一步的,所述步骤S2中筛选出相关系数极小值点是通过MATLAB的工具箱进行筛选的。
进一步的,所述通过MATLAB的工具箱进行筛选,表示为:
y=findpeaks(-R)
其中,findpeaks表示MATLAB内置函数;R表示整个数据集相邻波段之间的相关系数向量;y表示包含所有极小值的向量。
进一步的,所述步骤S3中计算每个子空间中波段的信息熵,表示为:
Figure PCTCN2021135928-appb-000004
其中,Ω表示整个样本空间,p(i)表示样本i出现在图像中的概率;H表示样本i的信息熵。
进一步的,所述步骤S3中根据计算得到的信息熵从子空间中选取所需数 量的特征波段,表示为:
Figure PCTCN2021135928-appb-000005
其中,Z表示每个波段子空间所包含的波段数目;L表示整个数据集所包含的波段数;K表示需要选取的特征波段数;S表示选取的波段。
相应的,还提供一种基于近邻子空间划分高光谱影像波段选择系统,包括:
计算模块,用于输入高光谱影像立方体,并计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量;
筛选模块,用于根据得到的相关系数的向量寻找所有相关系数极值点,并在寻找到的所有相关系数极值点中筛选出相关系数极小值点,通过筛选出的相关系数极小值点确定最优划分的高光谱波段子空间;
选取模块,用于根据子空间中波段数的大小对子空间进行排序,并计算每个子空间中波段的信息熵,根据计算得到的信息熵从子空间中选取所需数量的特征波段。
进一步的,所述计算模块中计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量,表示为:
Figure PCTCN2021135928-appb-000006
Figure PCTCN2021135928-appb-000007
其中,X∈R W×H×L表示高光谱影像立方体,L表示高光谱影像立方体所包含的波段数目,W和H分别表示每个波段的宽度和高度;X i表示高光谱影像数据集中第i个波段;X j表示高光谱影像数据集中第j个波段;X i(m,n)表示X i第m行第n列的值;R表示相关系数的向量;V i表示第i个波段的像素差;V j表示第j个波段的像素差;V o表示第o个波段的像素差;X o表示高光谱影像数据集中第o个波段;
Figure PCTCN2021135928-appb-000008
表示第o个波段的像素均值;o={i,j}。
进一步的,所述选取模块中计算每个子空间中波段的信息熵,表示为:
Figure PCTCN2021135928-appb-000009
其中,Ω表示整个样本空间,p(i)表示样本i出现在图像中的概率;H表示样本i的信息熵。
进一步的,所述选取模块中根据计算得到的信息熵从子空间中选取所需数量的特征波段,表示为:
Figure PCTCN2021135928-appb-000010
其中,Z表示每个波段子空间所包含的波段数目;L表示整个数据集所包含的波段数;K表示需要选取的特征波段数;S表示选取的波段。
与现有技术相比,本申请通过计算相邻波段之间的相关系数,根据相关系数的曲率变化情况找到分组临界点,然后以临界点数目作为最终确定的聚类簇个数。此外以信息熵作为波段选取的评价指标。在对每个聚类簇进行波段选取时,将信息熵作为权重对每个波段排序,然后选择信息熵最大的波段作为特征波段,这样可以确保最终选取的波段包含相对完整的信息。
附图说明
图1是实施例一提供的一种基于近邻子空间划分高光谱影像波段选择方法流程图;
图2是实施例一提供的一种基于近邻子空间划分高光谱影像波段选择框图;
图3是实施例一提供的近邻子空间划分示例示意图;
图4是实施例一提供的Indian Pines数据集上所有波段的信息熵示例示意图;
图5是实施例二提供的不同的高光谱影像波段选择方法在Botswana数据集上的Overall-Accuracy曲线示意图;
图6是实施例二提供的不同的高光谱影像波段选择方法在Salinas数据集上的Overall-Accuracy曲线示意图;
图7是实施例二提供的不同的高光谱影像波段选择方法在Indian Pines数 据集上的Overall-Accuracy曲线示意图。
具体实施方式
以下通过特定的具体实例说明本申请的实施方式,本领域技术人员可由本说明书所揭露的内容轻易地了解本申请的其他优点与功效。本申请还可以通过另外不同的具体实施方式加以实施或应用,本说明书中的各项细节也可以基于不同观点与应用,在没有背离本申请的精神下进行各种修饰或改变。需说明的是,在不冲突的情况下,以下实施例及实施例中的特征可以相互组合。
本申请针对现有缺陷,提供了一种基于近邻子空间划分高光谱影像波段选择方法及系统。
实施例一
本实施例提供的一种基于近邻子空间划分高光谱影像波段选择方法,如图1、2所示,包括:
S11.输入高光谱影像立方体,并计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量;
S12.根据得到的相关系数的向量寻找所有相关系数极值点,并在寻找到的所有相关系数极值点中筛选出相关系数极小值点,通过筛选出的相关系数极小值点确定最优划分的高光谱波段子空间;
S13.根据子空间中波段数的大小对子空间进行排序,并计算每个子空间中波段的信息熵,根据计算得到的信息熵从子空间中选取所需数量的特征波段。
在本实施例中,基于波段之间是有序排列,并且距离较远的波段之间相关性较低的思想,因此认为按照有序波段进行划分不仅可以降低波段子空间之间的相关性,同时也能避免最后选取的特征波段之间相关性较高的问题。基于此,本实施例提出了一种基于近邻子空间划分的高光谱影像波段选择方法,简称为SEASP。其中,SEASP采用了聚类和排序相结合的方式来解决当聚类中心数较少时精度较低的问题。
在步骤S11中,输入高光谱影像立方体,并计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量。
对于给定的输入高光谱影像立方体,首先计算其包含的任意两个相邻波段 之间的相关系数,得到对应的相关系数向量。
其中,计算高光谱影像立方体中所包含的任意两个波段之间的相关系数的具体步骤为:
给定一个高光谱影像立方体X∈R W×H×L,其中L代表高光谱影像立方体所包含的波段数目,W和H则分别代表每个波段的宽度和高度。
一般来说,可将高光谱波段影像看作一个由具有不同光谱值的像素点组成的二维矩阵,所以令X i代表高光谱影像数据集中第i个波段。那么波段X i和X j之间的相关系数计算方法如下:
Figure PCTCN2021135928-appb-000011
Figure PCTCN2021135928-appb-000012
其中,X∈R W×H×L表示高光谱影像立方体,L表示高光谱影像立方体所包含的波段数目,W和H分别表示每个波段的宽度和高度;X i表示高光谱影像数据集中第i个波段;X j表示高光谱影像数据集中第j个波段;X i(m,n)表示X i第m行第n列的值;R表示相关系数的向量;V i表示第i个波段的像素差;V j表示第j个波段的像素差;V o表示第o个波段的像素差;X o表示高光谱影像数据集中第o个波段;
Figure PCTCN2021135928-appb-000013
表示第o个波段的像素均值;o={i,j}。
Figure PCTCN2021135928-appb-000014
的计算方法如下:
Figure PCTCN2021135928-appb-000015
考虑到高光谱波段之间的有序性和非近邻空间的弱相关性,只计算相邻波段之间的相关系数,最后得到相关系数向量。
本实施例对于输入的高光谱影像立方体X∈R W×H×L,将其包含的波段影像看作一个由具有不同光谱值的像素点组成的二维矩阵,然后按照上述计算方法求出任意两个相邻波段之间的相关系数,得到相关系数向量。
在步骤S12中,根据得到的相关系数的向量寻找所有相关系数极值点,并 在寻找到的所有相关系数极值点中筛选出相关系数极小值点,通过筛选出的相关系数极小值点确定最优划分的高光谱波段子空间。
从相关系数向量中寻找所有相关系数极值点,通过极小值点的选取确定最优划分的高光谱波段子空间。
在聚类算法的思想中,衡量划分的标准通常是使得类内相关度最高,类间相关度最低。基于此,在SEASP中,主要是根据相关系数的变化率来确定分组之间的相关性大小。具体来说,若两个波段之间的相关性在某个区间内最小,即相关系数的变化率在该区间内最大,说明这两个波段在很大概率上不属于同一组,为两个相邻分组之间的分割点。考虑到波段之间的有序性和非近邻空间的弱相关性,实施例只计算相邻波段的相关系数,此时聚类划分问题便转化为在有序数据集中寻找分组之间的间断点问题。所以根据得到的相关系数向量计算其包含的所有相关系数极小值,从而得到所有子空间的间断点。
从相关系数向量中寻找所有相关系数极小值点的具体步骤为:
在有序的高光谱波段中,聚类划分便转化为在有序数据集中寻找分组之间的间断点。对于寻找相关系数极小值点问题,可以采用MATLAB现有的工具箱进行解决,具体计算方法如下所示:
y=findpeaks(-R)
其中,findpeaks表示MATLAB内置函数;R表示整个数据集相邻波段之间的相关系数向量;y表示包含所有极小值的向量。
具体来说,findpeaks函数在判断某一点是否为峰值点时,主要是将该点处的函数值和其左右相邻的函数值进行比较,若其为最大,则将其判断为该波形的一个波峰点。所以当对输入参数进行取反操作时,求得的波峰点便是原始数据的波谷点。现假设在点x处的函数值为F(x),那么当F(x)同时满足F(x)<F(x-1)和F(x)<F(x+1)时,便将其作为一个波谷点。通过极小值点的选取来确定有序波段之间的间断点,从而得到划分的波段子空间,具体实例如图3所示。
在步骤S13中,根据子空间中波段数的大小对子空间进行排序,并计算每个子空间中波段的信息熵,根据计算得到的信息熵从子空间中选取所需数量的特征波段。
按照每个子空间所包含的波段数大小对其进行排序,并计算每个波段的信息熵,最后按照信息熵大小从这若干个子空间中选取所需数量的特征波段。
计算每个波段的信息熵大小并从中选择所需数量的特征波段的具体步骤为:
信息熵用来衡量图像中包含的平均信息量的大小,一个图像的信息熵越大,那么该图像所包含的信息也就越丰富。所以信息熵作为衡量波段重要性程度的一个度量指标是较为合适的。关于其具体的计算方法如下所示:
Figure PCTCN2021135928-appb-000016
其中,Ω表示整个样本空间,p(i)表示样本i出现在图像中的概率;H表示样本i的信息熵。
如图4给出了Indian Pines数据集中所有波段的信息熵示例。
在根据信息熵对所有波段进行排序的过程中,考虑到传入参数K,即所需波段数量的变化,将划分的每个波段子空间按照其包含的波段数大小进行排序,得到排序后的波段子空间。
当所需波段的数量小于或等于划分的子空间个数时,则从排序后的子空间中选出对应数量的波段子空间,然后从选出的子空间中选取信息熵最大的波段作为特征波段;当所需波段的数量大于划分的子空间个数时,则从每个子空间中选出S个波段作为特征波段。
具体为:当K小于或等于划分子空间个数时,从每个子空间中选取信息熵最大的波段作为特征波段;当K大于划分子空间个数时,从每个子空间中按照信息熵的大小选取S个波段。其中,关于S的具体计算方法如下:
Figure PCTCN2021135928-appb-000017
其中,Z表示每个波段子空间所包含的波段数目;L表示整个数据集所包含的波段数;K表示需要选取的特征波段数;S表示选取的波段。
通过上式的计算,可以最终得到所需特征波段子集Y。
与现有技术相比,本实施例通过计算相邻波段之间的相关系数,根据相关系数的曲率变化情况找到分组临界点,然后以临界点数目作为最终确定的聚类 簇个数。此外以信息熵作为波段选取的评价指标。在对每个聚类簇进行波段选取时,将信息熵作为权重对每个波段排序,然后选择信息熵最大的波段作为特征波段,这样可以确保最终选取的波段包含相对完整的信息。
实施例二
本实施例提供的一种基于近邻子空间划分高光谱影像波段选择方法与实施例一的区别在于:
本实施例为了验证所提出的SEASP算法在高光谱波段选择上的有效性,进行了大量的相关实验。
实验中主要选取了三个公共数据集,分别为Botswana数据集、Salinas数据集和Indian Pines数据集。选取的分类器有KNN、SVM和LDA,KNN分类器的参数k设置为5,SVM统一采用的距离函数为高斯核函数,Indian Pines和Botswana数据集的惩罚系数均为10000,Salinas数据集的惩罚系数为100。由于上述三种分类器都是有监督的分类方法,所以在进行实验的时候,随机选取整个数据集的10%作为训练样本,剩下的数据作为测试样本。此外,本实施例还和当下几种比较先进的算法进行了对比,它们分别是ASPS_MN、ASPS_IE、TOF、UBS和FNGBS。对于三个公共的高光谱影像数据集,由于目前其最佳选择的波段数目未知,所以在本次实验中,选取波段数目的范围为5到50。为了充分展示波段数目设置的随机性,实验选取的波段数目一共有12个,分别为5,7,10,15,26,30,36,39,42,44,47,49。本次实验采用了三种度量标准来对分类的结果进行分析,分别为总体精度(OA),平均总体精度(AOA)以及Kappa系数(Kappa)。所有的实验均在MATLAB 2016a上运行,CPU为i7-5500U,2.40GHz,内存8Gb。
表1展示了竞争算法在三个公共数据集上所表现出的AOA和Kappa。其中,在进行测试比较时,三个数据集选取的波段数目分别为36、5和5,AOA为10次运行结果的范围。为了清晰地展现出所比较算法在三个数据集上的性能,在表1中以粗体的形式将两种较好的分类结果标出。
Figure PCTCN2021135928-appb-000018
Figure PCTCN2021135928-appb-000019
表1
从图5、图6、图7可以看出,本实施例所提出的算法在一些特定的波段数目上所表现出来的性能优于其他算法。在图5中,当选择的波段数目大于25时,SEASP始终保持一个稳定性,而其他算法则处于上下波动状态。此外,从所有分类器的结果可以看出,当选择的波段数目较少时,有些算法的结果不如其他算法,这也反映出它们对选择波段数目的强敏感性。只有当选择的波段数目达到一定数量时,其才能够表现出一个更好的性能。从整体来看,随着选择波段数目的不断增加,SEASP所表现出的结果变化幅度并没有其他算法那么大,这也侧面验证了其对选择波段数目的敏感性是较小的,同时说明了其在少量选择的波段数目上也能够表现出不错的性能。在图6和图7中,随着选择波段数目的增加,SEASP基本上是处于稳定上升的状态,没有出现明显的反弹现象。而其他算法则不太稳定,当选择的波段数目变多时,相反结果还会出现下降的趋势。这反映了随着选取波段数目的增加,这些算法都不可避免的选取了一些冗余波段,这就导致了分类精度的下降。而本实施例所提出的算法则不会出现这个明显的现象,也说明了在选取波段时,SEASP充分考虑到最后选择的目标波段之间强相关性的问题,从而改正了大多数基于聚类的波段选择算法的不足。
为了进一步验证所提出方法的可行性,在三个数据集的波段选择上还计算了所有竞争算法的时间。其中,在三个数据集的波段选取数目上面,分别设置为10,15,20,呈现出一种递增的趋势。表2给出了不同算法在不同数据集上选择同样数目波段所耗费的时间。从表中结果可以看出,本实施例提出的算法所耗费时间是较小的,并且与其他算法相比也具有一定的可比性。
Figure PCTCN2021135928-appb-000020
Figure PCTCN2021135928-appb-000021
表2
从上述一系列实验可知,本实施例所提出的算法不仅原理简单,而且在三个公共数据集的分类性能上与其他算法具有一定的可比性,甚至优于其他算法,另外其执行速度也是较快的,从而验证了该算法的有效性和可行性。
实施例三
本实施例提供一种基于近邻子空间划分高光谱影像波段选择系统,包括:
计算模块,用于输入高光谱影像立方体,并计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量;
筛选模块,用于根据得到的相关系数的向量寻找所有相关系数极值点,并在寻找到的所有相关系数极值点中筛选出相关系数极小值点,通过筛选出的相关系数极小值点确定最优划分的高光谱波段子空间;
选取模块,用于根据子空间中波段数的大小对子空间进行排序,并计算每个子空间中波段的信息熵,根据计算得到的信息熵从子空间中选取所需数量的特征波段。
进一步的,所述计算模块中计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,表示为:
Figure PCTCN2021135928-appb-000022
Figure PCTCN2021135928-appb-000023
其中,X∈R W×H×L表示高光谱影像立方体,L表示高光谱影像立方体所包含的波段数目,W和H分别表示每个波段的宽度和高度;X i表示高光谱影像数据集中第i个波段;X j表示高光谱影像数据集中第j个波段;X i(m,n)表示X i第m行第n列的值;R(X i,X j)表示两个相邻波段之间的相关系数;V i表示第i个波段的像素差;V j表示第j个波段的像素差;V o表示第o个波段的像素差;X o表示高光谱影像数据集中第o个波段;
Figure PCTCN2021135928-appb-000024
表示第o个波段的像素均值;o={i,j}。进一步的,所述选取模块中计算每个子空间中波段的信息熵,表示 为:
Figure PCTCN2021135928-appb-000025
其中,Ω表示整个样本空间,p(i)表示样本i出现在图像中的概率;H表示样本i的信息熵。
进一步的,所述选取模块中根据计算得到的信息熵从子空间中选取所需数量的特征波段,表示为:
Figure PCTCN2021135928-appb-000026
其中,Z表示每个波段子空间所包含的波段数目;L表示整个数据集所包含的波段数;K表示需要选取的特征波段数;S表示选取的波段。
需要说明的是,本实施例提供的一种基于近邻子空间划分高光谱影像波段选择系统与实施例一类似,在此不多做赘述。
与现有技术相比,本实施例通过计算相邻波段之间的相关系数,根据相关系数的曲率变化情况找到分组临界点,然后以临界点数目作为最终确定的聚类簇个数。此外以信息熵作为波段选取的评价指标。在对每个聚类簇进行波段选取时,将信息熵作为权重对每个波段排序,然后选择信息熵最大的波段作为特征波段,这样可以确保最终选取的波段包含相对完整的信息。
注意,上述仅为本申请的较佳实施例及所运用技术原理。本领域技术人员会理解,本申请不限于这里所述的特定实施例,对本领域技术人员来说能够进行各种明显的变化、重新调整和替代而不会脱离本申请的保护范围。因此,虽然通过以上实施例对本申请进行了较为详细的说明,但是本申请不仅仅限于以上实施例,在不脱离本申请构思的情况下,还可以包括更多其他等效实施例,而本申请的范围由所附的权利要求范围决定。

Claims (10)

  1. 一种基于近邻子空间划分高光谱影像波段选择方法,其特征在于,包括:
    S1.输入高光谱影像立方体,并计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量;
    S2.根据得到的相关系数的向量寻找所有相关系数极值点,并在寻找到的所有相关系数极值点中筛选出相关系数极小值点,通过筛选出的相关系数极小值点确定最优划分的高光谱波段子空间;
    S3.根据子空间中波段数的大小对子空间进行排序,并计算每个子空间中波段的信息熵,根据计算得到的信息熵从子空间中选取所需数量的特征波段。
  2. 根据权利要求1所述的一种基于近邻子空间划分高光谱影像波段选择方法,其特征在于,所述步骤S1中计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量,表示为:
    Figure PCTCN2021135928-appb-100001
    Figure PCTCN2021135928-appb-100002
    其中,X∈R W×H×L表示高光谱影像立方体,L表示高光谱影像立方体所包含的波段数目,W和H分别表示每个波段的宽度和高度;X i表示高光谱影像数据集中第i个波段;X j表示高光谱影像数据集中第j个波段;X i(m,n)表示X i第m行第n列的值;R表示相关系数的向量;V i表示第i个波段的像素差;V j表示第j个波段的像素差;V o表示第o个波段的像素差;X o表示高光谱影像数据集中第o个波段;
    Figure PCTCN2021135928-appb-100003
    表示第o个波段的像素均值;o={i,j}。
  3. 根据权利要求2所述的一种基于近邻子空间划分高光谱影像波段选择方法,其特征在于,所述步骤S2中筛选出相关系数极小值点是通过MATLAB的工具箱进行筛选的。
  4. 根据权利要求3所述的一种基于近邻子空间划分高光谱影像波段选择方法,其特征在于,所述过MATLAB的工具箱进行筛选,表示为:
    y=findpeaks(-R)
    其中,findpeaks表示MATLAB内置函数;R表示整个数据集相邻波段之间的相关系数向量;y表示包含所有极小值的向量。
  5. 根据权利要求4所述的一种基于近邻子空间划分高光谱影像波段选择方法,其特征在于,所述步骤S3中计算每个子空间中波段的信息熵,表示为:
    Figure PCTCN2021135928-appb-100004
    其中,Ω表示整个样本空间,p(i)表示样本i出现在图像中的概率;H表示样本i的信息熵。
  6. 根据权利要求5所述的一种基于近邻子空间划分高光谱影像波段选择方法,其特征在于,所述步骤S3中根据计算得到的信息熵从子空间中选取所需数量的特征波段,表示为:
    Figure PCTCN2021135928-appb-100005
    其中,Z表示每个波段子空间所包含的波段数目;L表示整个数据集所包含的波段数;K表示需要选取的特征波段数;S表示选取的波段。
  7. 一种基于近邻子空间划分高光谱影像波段选择系统,其特征在于,包括:
    计算模块,用于输入高光谱影像立方体,并计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量;
    筛选模块,用于根据得到的相关系数的向量寻找所有相关系数极值点,并在寻找到的所有相关系数极值点中筛选出相关系数极小值点,通过筛选出的相关系数极小值点确定最优划分的高光谱波段子空间;
    选取模块,用于根据子空间中波段数的大小对子空间进行排序,并计算每个子空间中波段的信息熵,根据计算得到的信息熵从子空间中选取所需数量的特征波段。
  8. 根据权利要求7所述的一种基于近邻子空间划分高光谱影像波段选择系统,其特征在于,所述计算模块中计算输入的高光谱影像立方体中包含的任意两个相邻波段之间的相关系数,得到相关系数的向量,表示为:
    Figure PCTCN2021135928-appb-100006
    Figure PCTCN2021135928-appb-100007
    其中,X∈R W×H×L表示高光谱影像立方体,L表示高光谱影像立方体所包含的波段数目,W和H分别表示每个波段的宽度和高度;X i表示高光谱影像数据集中第i个波段;X j表示高光谱影像数据集中第j个波段;X i(m,n)表示X i第m行第n列的值;R表示相关系数的向量;V i表示第i个波段的像素差;V j表示第j个波段的像素差;V o表示第o个波段的像素差;X o表示高光谱影像数据集中第o个波段;
    Figure PCTCN2021135928-appb-100008
    表示第o个波段的像素均值;o={i,j}。
  9. 根据权利要求8所述的一种基于近邻子空间划分高光谱影像波段选择系统,其特征在于,所述选取模块中计算每个子空间中波段的信息熵,表示为:
    Figure PCTCN2021135928-appb-100009
    其中,Ω表示整个样本空间,p(i)表示样本i出现在图像中的概率;H表示样本i的信息熵。
  10. 根据权利要求9所述的一种基于近邻子空间划分高光谱影像波段选择系统,其特征在于,所述选取模块中根据计算得到的信息熵从子空间中选取所需数量的特征波段,表示为:
    Figure PCTCN2021135928-appb-100010
    其中,Z表示每个波段子空间所包含的波段数目;L表示整个数据集所包含的波段数;K表示需要选取的特征波段数;S表示选取的波段。
PCT/CN2021/135928 2021-02-07 2021-12-07 一种基于近邻子空间划分高光谱影像波段选择方法及系统 WO2022166363A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110174636.8A CN113075129B (zh) 2021-02-07 2021-02-07 一种基于近邻子空间划分高光谱影像波段选择方法及系统
CN202110174636.8 2021-02-07

Publications (1)

Publication Number Publication Date
WO2022166363A1 true WO2022166363A1 (zh) 2022-08-11

Family

ID=76609942

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/135928 WO2022166363A1 (zh) 2021-02-07 2021-12-07 一种基于近邻子空间划分高光谱影像波段选择方法及系统

Country Status (4)

Country Link
CN (1) CN113075129B (zh)
LU (1) LU502854B1 (zh)
WO (1) WO2022166363A1 (zh)
ZA (1) ZA202207737B (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117435940A (zh) * 2023-12-20 2024-01-23 龙建路桥股份有限公司 一种面向冬季混凝土养护过程中光谱检测方法

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113075129B (zh) * 2021-02-07 2023-03-31 浙江师范大学 一种基于近邻子空间划分高光谱影像波段选择方法及系统
CN113486876A (zh) * 2021-09-08 2021-10-08 中国地质大学(武汉) 一种高光谱影像波段选择方法、装置及系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8081825B2 (en) * 2008-02-27 2011-12-20 Ajou University Industry Cooperation Foundation Method for realtime target detection based on reduced complexity hyperspectral processing
CN103065293A (zh) * 2012-12-31 2013-04-24 中国科学院东北地理与农业生态研究所 相关性加权的遥感影像融合方法及该融合方法的融合效果评价方法
CN103886334A (zh) * 2014-04-08 2014-06-25 河海大学 一种多指标融合的高光谱遥感影像降维方法
CN104122210A (zh) * 2014-07-02 2014-10-29 中国林业科学研究院林业研究所 一种基于最佳指数-相关系数法的高光谱波段提取方法
CN107124612A (zh) * 2017-04-26 2017-09-01 东北大学 基于分布式压缩感知的高光谱图像压缩方法
US20190293620A1 (en) * 2018-03-20 2019-09-26 SafetySpect, Inc. Apparatus and method for multimode analytical sensing of items such as food
CN113075129A (zh) * 2021-02-07 2021-07-06 浙江师范大学 一种基于近邻子空间划分高光谱影像波段选择方法及系统

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101131734A (zh) * 2007-06-25 2008-02-27 北京航空航天大学 适用于高光谱遥感图像的自动波段选择方法
CN104751179B (zh) * 2015-04-01 2018-02-06 河海大学 一种基于博弈论的多目标高光谱遥感影像波段选择方法
CN108154094B (zh) * 2017-12-14 2020-04-24 浙江工业大学 基于子区间划分的高光谱图像非监督波段选择方法
CN110751142B (zh) * 2019-09-25 2022-07-26 河海大学 一种改进型的高光谱遥感影像波段选择方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8081825B2 (en) * 2008-02-27 2011-12-20 Ajou University Industry Cooperation Foundation Method for realtime target detection based on reduced complexity hyperspectral processing
CN103065293A (zh) * 2012-12-31 2013-04-24 中国科学院东北地理与农业生态研究所 相关性加权的遥感影像融合方法及该融合方法的融合效果评价方法
CN103886334A (zh) * 2014-04-08 2014-06-25 河海大学 一种多指标融合的高光谱遥感影像降维方法
CN104122210A (zh) * 2014-07-02 2014-10-29 中国林业科学研究院林业研究所 一种基于最佳指数-相关系数法的高光谱波段提取方法
CN107124612A (zh) * 2017-04-26 2017-09-01 东北大学 基于分布式压缩感知的高光谱图像压缩方法
US20190293620A1 (en) * 2018-03-20 2019-09-26 SafetySpect, Inc. Apparatus and method for multimode analytical sensing of items such as food
CN113075129A (zh) * 2021-02-07 2021-07-06 浙江师范大学 一种基于近邻子空间划分高光谱影像波段选择方法及系统

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
"Chinese Master Thesis Hunan University", 28 February 2019, HUNAN UNIVERSITY, CN, article JI CHUNRUI: "Research on Feature Extraction and Classification Recognition Algorithm for Hyperspectral", pages: 1 - 124, XP055957297 *
MA LI 、: "Band Selection Algorithm for Hyperspectral Images Using Weighted Band Index", HEILONGJIANG SCIENCE AND TECHNOLOGY INFORMATION, HEILONGJIANG SCIENCE POPULARIZATION CENTER, CN, no. 4, 5 February 2010 (2010-02-05), CN , pages 71, XP055957330, ISSN: 2096-4390 *
WANG QI, ET AL.: "Hyperspectral band selection via adaptive subspace partition strategy", IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, vol. 12, no. 12, 31 December 2019 (2019-12-31), pages 4940 - 4950, XP011770034, DOI: 10.1109/JSTARS.2019.2941454 *
WU YIQUAN, ZHOU YANG, SHENG DONG-HUI, YE XIAO-LAI: "Band Selection of Hyperspectral Image based on Optimal Linear Prediction of Principal Components in Subspace", HONGWAI YU HAOMIBO XUEBAO - JOURNAL OF INFRARED AND MILLIMETERWAVES, KEXUE CHUBANSHE, BEIJING, CN, vol. 37, no. 1, 28 February 2018 (2018-02-28), CN , pages 119 - 128, XP055957316, ISSN: 1001-9014 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117435940A (zh) * 2023-12-20 2024-01-23 龙建路桥股份有限公司 一种面向冬季混凝土养护过程中光谱检测方法
CN117435940B (zh) * 2023-12-20 2024-03-05 龙建路桥股份有限公司 一种面向冬季混凝土养护过程中光谱检测方法

Also Published As

Publication number Publication date
CN113075129A (zh) 2021-07-06
LU502854B1 (en) 2023-01-30
ZA202207737B (en) 2022-07-27
CN113075129B (zh) 2023-03-31

Similar Documents

Publication Publication Date Title
WO2022166363A1 (zh) 一种基于近邻子空间划分高光谱影像波段选择方法及系统
Zhang et al. Local density adaptive similarity measurement for spectral clustering
US9864928B2 (en) Compact and robust signature for large scale visual search, retrieval and classification
Chen et al. Model Metric Co-Learning for Time Series Classification.
CN103403704B (zh) 用于查找最近邻的方法和设备
WO2010035659A1 (ja) 入力データの分類に用いる特徴を選択するための情報処理装置
Yang et al. A feature-metric-based affinity propagation technique for feature selection in hyperspectral image classification
KR101178333B1 (ko) 웨이블릿 기반의 중심대칭-국부이진 패턴과 계층적 랜덤 포레스트를 이용한 사람 검출 방법
CN109583469B (zh) 基于互信息的k均值高光谱图像波段聚类方法
WO2020048145A1 (zh) 数据检索的方法和装置
CN110751027B (zh) 一种基于深度多示例学习的行人重识别方法
Duin et al. Mode seeking clustering by KNN and mean shift evaluated
Tavenard et al. Balancing clusters to reduce response time variability in large scale image search
US10671663B2 (en) Generation device, generation method, and non-transitory computer-readable recording medium
CN110188864B (zh) 基于分布表示和分布度量的小样本学习方法
CN114663770A (zh) 一种基于集成聚类波段选择的高光谱图像分类方法及系统
CN109871768B (zh) 基于共享最近邻的高光谱最优波段选择方法
CN110929801B (zh) 一种基于改进的Euclid距离KNN分类方法和系统
CN111027609B (zh) 一种图像数据加权分类方法和系统
CN107220661A (zh) 基于多模态融合的光谱波段选择方法
Chu et al. Logo recognition and localization in real-world images by using visual patterns
CN104123538B (zh) 一种基于视觉词袋的网络不良图像检测方法
Li et al. A local potential-based clustering algorithm for unsupervised hyperspectral band selection
US20230259756A1 (en) Graph explainable artificial intelligence correlation
CN115510959A (zh) 基于自然最近邻和多簇合并的密度峰值聚类方法

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: 21924356

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21924356

Country of ref document: EP

Kind code of ref document: A1