WO2017190542A1 - 一种基于分块的vca端元提取方法 - Google Patents

一种基于分块的vca端元提取方法 Download PDF

Info

Publication number
WO2017190542A1
WO2017190542A1 PCT/CN2017/074294 CN2017074294W WO2017190542A1 WO 2017190542 A1 WO2017190542 A1 WO 2017190542A1 CN 2017074294 W CN2017074294 W CN 2017074294W WO 2017190542 A1 WO2017190542 A1 WO 2017190542A1
Authority
WO
WIPO (PCT)
Prior art keywords
block
vca
matrix
abundance
endmembers
Prior art date
Application number
PCT/CN2017/074294
Other languages
English (en)
French (fr)
Inventor
刘治
聂明钰
邱清晨
孙育霖
Original Assignee
山东大学
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 山东大学 filed Critical 山东大学
Priority to US16/099,140 priority Critical patent/US10984291B2/en
Publication of WO2017190542A1 publication Critical patent/WO2017190542A1/zh

Links

Images

Classifications

    • G06T3/067—
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T19/00—Manipulating 3D models or images for computer graphics
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • 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
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
    • G06F18/2178—Validation; Performance evaluation; Active pattern learning techniques based on feedback of a supervisor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23211—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with adaptive number of clusters
    • 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
    • G06F18/245—Classification techniques relating to the decision surface
    • G06F18/2451—Classification techniques relating to the decision surface linear, e.g. hyperplane
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB

Definitions

  • the present invention relates to the field of image processing technologies, and in particular, to a block-based VCA end-member extraction method.
  • Hyperspectral image is a three-dimensional data image that reflects the spectral information and spatial information of the object at the same time. It has the characteristics of wide coverage band, fast losslessness and sufficient spectral information content.
  • the hyperspectral imaging system utilizes an image spectrometer and sensor to acquire hyperspectral images in a series of relatively continuous bands of the imaged object in the ultraviolet, visible, near-infrared, and infrared (wavelengths between 300 nm and 2500 nm). Due to the spatial resolution limitations of sensors and the complex diversity of natural objects, some pixels often contain more than one substance, and such pixels are called mixed pixels. Due to the existence of mixed pixels, the traditional classification method is not suitable for accurate subdivision and discrimination of features. In order to improve the accuracy of remote sensing classification, the problem of mixed pixels must be solved, so the mixed pixel unmixing becomes a key step before the classification and recognition of features.
  • LMM linear mixed model
  • the pixels in the hyperspectral image are linearly combined by a certain ratio (abundance) of the spectral characteristics of the basic substances (end elements) constituting the image at different wavelength bands.
  • Hybrid pixel demixing is the inverse of the hyperspectral image linear mixture model. It is a process of extracting the endmembers in the image and calculating the corresponding abundance of the endmembers.
  • Vertex Component Analysis is a most basic geometry-based hyperspectral image endmember extraction method. Based on the spatial distribution characteristics of hyperspectral datasets, the geometry-based endmember extraction method considers that in geometric space, endmembers often exist in the vertices of monomorphs, polyhedrons or convex cones composed of hyperspectral image datasets. The VCA extracts the endmembers one by one by iteratively searching for orthogonal vectors in the data space and calculating the projection distance of the pixels on the orthogonal vectors.
  • the present invention discloses a block-based VCA end-member extraction method.
  • the present invention divides a complex environment of hyperspectral images into a plurality of relatively simple image parts by using a certain classification method. Then VCA endmember extraction is performed on a simple image, which must reduce the influence of the noise of the global image on the algorithm, avoid missing the main endmember, and improve the accuracy of endmember extraction.
  • a block-based VCA endmember extraction method comprising:
  • the unsupervised classification method is used to roughly block the hyperspectral image, and the pixels with similar substances are divided into the same block;
  • the VCA is used to extract the end regions of each block after the block, and the abundance is inferred by using the least squares method after extracting the end elements, and a main end element is determined according to the size of the abundance value in each block;
  • the main endmembers in all blocks are extracted and composed into an endmember matrix of the global image.
  • the PCA is required to perform dimensionality reduction on the hyperspectral image data.
  • ISODATA iterative self-organizing data analysis method
  • ISODATA has automatically merged and split categories in the unsupervised classification process.
  • the merging mechanism means that when the total number of classes is too large or the distance between two classes of classes is less than a certain threshold, the two classes are merged into a new class, and the number of samples in the class is less than a certain threshold. ;
  • the splitting mechanism means that when the total number of classes is too small or the number of samples in a certain class exceeds a certain threshold, and the standard deviation within the class is greater than the splitting threshold, it is divided into two categories, thereby obtaining a clustering result with a relatively reasonable number of classes.
  • the number of end elements r′ is set, where r′ ⁇ r, respectively, for VCA end element extraction.
  • VCA end-element extraction algorithm process first finds an initial unit vector, and then projects all the pixel points onto the vector, and the pixel point with the largest projection distance is recorded as an end point, and the end-element matrix set is added.
  • the new set of endmembers find a vector that is orthogonal to all the found endmembers, and proceed to the next round of loops, calculate the pixel projection distance, and find new endmembers until all endmembers are found.
  • the abundance inversion is performed on the block regions by the least squares method. According to the abundance size feedback in the block region, the main end elements in each block region are determined, and the main end elements in all blocks are extracted to form an end element matrix of the global image.
  • the least squares method can be divided into unconstrained least squares UCLS according to whether non-negative constraints and sums of 1 are considered in the solution process, and is 1 constrained least squares SCLS, non-negative constrained least squares NCLS, full constrained minimum two Multiplication FCLS.
  • the present invention Prior to the use of VCA to extract endmembers, the present invention first uses unsupervised classification to aggregate similar pixels in a hyperspectral image, excluding the effects of unrelated pixels, and reducing the complexity of the endmember extraction environment.
  • the present invention uses the VCA endmember extraction method in a relatively simple environment environment after partitioning, and then uses the intra-block abundance inversion result to feed back the main endmembers in the control block, thereby avoiding missing major endmembers.
  • Figure 1 is a flow chart of the present invention
  • Figure 2b Image of the Washington DC mall image after ISODATA segmentation
  • 3a-3e are comparison diagrams (Washington DC mall data) using a block-based VCA extraction end element, an original VCA extraction end element, and a theoretical end element;
  • Figures 5a-5f are comparisons (HYDICE Urban data) using block-based VCA extraction endmembers, original VCA extraction endmembers, and theoretical endmembers.
  • a block-based VCA endmember extraction method includes the following steps:
  • Step (1) input hyperspectral image data X ⁇ R m ⁇ n , where m is the number of bands of the hyperspectral image, n is the total number of pixels of the hyperspectral image, and the number of end elements is r, and the hyperspectral image data is performed using PCA. Dimensional processing.
  • Step (4) For all the block regions ⁇ i , set the number of end elements r', where r' ⁇ r, respectively perform VCA end element extraction.
  • Step (6) determines the main end elements in each of the block regions according to the abundance size feedback in the block region, and extracts the main end elements in all the blocks to form an end element matrix of the global image.
  • PCA principal component analysis
  • the iterative self-organizing data analysis method (ISODATA) algorithm is an unsupervised classification method that extracts features directly from samples without clustering knowledge.
  • the ISODATA algorithm improves the K-means clustering. After all the samples have been adjusted, the average of each sample is recalculated, and the categories are automatically merged and split, which has a certain self-organization.
  • the merging mechanism in the ISODATA algorithm means that when the total number of classes is too large or the distance between two classes of classes is less than a certain threshold, the two classes are merged into a new class, and the number of samples in the class is less than a certain threshold. This class is cancelled.
  • the splitting mechanism means that when the total number of classes is too small or the number of samples in a certain class exceeds a certain threshold, and the standard deviation within the class is greater than the splitting threshold, it is divided into two categories, thereby obtaining a clustering result with a relatively reasonable number of classes.
  • the VCA end-element extraction algorithm is based on a linear spectral model.
  • the end-members are extracted one by one by iteratively searching for orthogonal vectors in the data space and calculating the projection distance of the pixels on the orthogonal vectors.
  • the basic theory of VCA is that several vertices of a simplex can be expanded into a subspace, and the vertices of a simplex are the maximum points of projection length on a vector orthogonal to this subspace.
  • the VCA end-element extraction algorithm first finds an initial unit vector, and then projects all the pixel points onto this vector.
  • the pixel points with the largest projection distance are recorded as end points, and the end-element matrix set is added. According to the new set of endmembers, find a vector that is orthogonal to all the found endmembers, and proceed to the next round of loops to calculate the pixel projection distance and find new endmembers. Until all the endmembers are found.
  • Abundance matrix satisfies and Non-negative Restrictions.
  • the least squares method can be divided into unconstrained least squares (UCLS) and unconstrained least squares (SCLS), non-negative constrained least squares (NCLS) according to whether the non-negative constraint and the constraint of 1 are considered in the solution process. ), Full Constrained Least Squares (FCLS).
  • the specific implementation part uses the relatively simple spatial distribution of the features of the Washington DC mall data and spatial distribution of relatively complex HYDICE Urban data to test.
  • the block-based VCA endmember extraction method is compared with the original VCA endmember extraction method, and the artificially extracted pure endmember is used as the theoretical endmember.
  • Experimental data HYDICE Urban hyperspectral data contains 210 spectral bands with a dimension of 307 ⁇ 307.
  • the image data contains six substances: road, soil, tree, grass, roof, metal.
  • the band affected by water absorption was removed, and the data contained 178 bands.
  • a pseudo color map of the partial image is shown in Figure 4a.
  • test results are shown in Figures 3a-3e (Washington DC mall data) and Figures 5a-5f (HYDICE Urban data). It can be seen from the comparison results of the endmember spectral map that the endmember curve extracted by the block-based VCA endmember extraction method is very close to the theoretical spectral curve of the endmember.
  • the endmember spectrum obtained by different endmember extraction methods and the spectral angle of the theoretical endmember spectrum (SAD) are used. Measure the accuracy of the extracted endmembers.
  • the formula for the spectral angle is defined as Among them, A theo is the theoretical value of an endmember , and A unmix is the endmember spectral information value extracted by different endmember extraction methods. The smaller the spectral angle, the closer the two spectral vectors are.
  • the spectral angle comparison results are shown in Table 1 and Table 2. It can be clearly seen that for the relatively simple spatial distribution of the ground objects, the data and spatial distribution of the relatively complex HYDICE Urban data, the accuracy of the block-based VCA extraction endmembers is greatly improved.
  • Table 1 is a comparison table (Washington DC mall data) using the block-based VCA extraction end element, the original VCA extraction end element and the theoretical end element spectral angle;
  • Table 2 is the use of the block-based VCA extraction end element, the original VCA A comparison table of the spectral angles of the endmembers and the theoretical endmembers (HYDICE Urban data).
  • Table 1 is based on the block VCA, the VCA extracted end element and the theoretical end element SAD comparison table
  • Table 2 is based on the block VCA, the VCA extracted end element and the theoretical end element SAD comparison table
  • the invention divides the hyperspectral image of the complex environment into a plurality of relatively simple image parts by using a certain classification method, and then extracts the end elements by block, thereby excluding the influence of the unrelated pixels to some extent, and reducing the end element extraction.
  • the complexity of the environment reduces the impact of the noise of the global image on the algorithm and avoids missing major endmembers. Specific examples show that the invention greatly improves the accuracy of extracting endmembers.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Multimedia (AREA)
  • Mathematical Physics (AREA)
  • Remote Sensing (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Astronomy & Astrophysics (AREA)
  • Computer Graphics (AREA)
  • Computer Hardware Design (AREA)
  • General Health & Medical Sciences (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Medical Informatics (AREA)
  • Algebra (AREA)
  • Image Analysis (AREA)

Abstract

一种基于分块的VCA端元提取方法，包括：利用非监督分类方法对高光谱图像进行粗略分块，将具有相似物质的像元分为相同块内；对分块后的各块内区域分别使用VCA进行端元提取，提取端元之后使用最小二乘法反演丰度，每块内根据丰度值的大小确定一个主要端元；将所有块内的主要端元提取出并组成全局图像的端元矩阵。在分块后的相对简单的环境块内环境使用VCA端元提取方法，然后利用块内丰度反演结果反馈控制块内的主要端元，从而避免了遗漏主要端元。

Description

一种基于分块的VCA端元提取方法 技术领域
本发明涉及图像处理技术领域，具体涉及一种基于分块的VCA端元提取方法。
背景技术
高光谱图像是同时反映物体反射光谱信息和空间信息的三维数据图像，具有覆盖波段广，快速无损，光谱信息含量充分等特点。高光谱成像系统利用图像分光仪和传感器，在紫外线，可见光，近红外，红外波段(波长在300nm-2500nm之间)获取成像对象的一系列相对连续波段下的高光谱图像。由于传感器的空间分辨率限制以及自然界地物的复杂多样性，一些像元中常常不止含有一种物质，这样的像元被称为混合像元。由于混合像元的存在，使得传统的分类方法不适于地物的精确细分和鉴别。为了提高遥感分类的准确性，必须解决混合像元的问题，因此混合像元解混在地物分类和识别之前成为关键的一步。
在进行混合像元解混之前，首先需要建立高光谱图像的线性混合模型(LMM)。在LMM中，高光谱图像中的像元是由组成图像的基本物质(端元)在不同波段下的光谱特性以一定的比例(丰度)线性组合而成。混合像元解混是高光谱图像线性混合模型的逆过程，是使用一定手段提取图像中的端元以及计算端元对应丰度的过程。
顶点成分分析(VCA)是一种最基本基于几何的高光谱图像端元提取方法。基于几何的端元提取方法从高光谱数据集空间分布特征出发，认为在几何空间中，端元常存在于由高光谱图像数据集组成的单形体、多面体或者凸锥的顶点部分。VCA通过反复寻找数据空间中的正交向量并计算像元在正交向量上的投影距离逐一提取端元。
由于实际地物的复杂性以及噪声等影响，在利用VCA进行端元提取的过程中，常常会出现遗漏主要端元，提取端元不准确，容易受噪声影响等现象。
发明内容
为解决现有技术存在的不足，本发明公开了一种基于分块的VCA端元提取方法，本发明将高光谱图像复杂的环境用一定的分类方法，分为多个相对简单的图像部分，然后在简单的图像上进行VCA端元提取，这一定降低了全局图像的噪声对算法的影响，避免了遗漏主要端元，提高了端元提取的准确性。
为实现上述目的，本发明的具体方案如下：
一种基于分块的VCA端元提取方法，包括：
利用非监督分类方法对高光谱图像进行粗略分块，将具有相似物质的像元分为相同块内；
对分块后的各块内区域分别使用VCA进行端元提取，提取端元之后使用最小二乘法反演丰度，每块内根据丰度值的大小确定一个主要端元；
将所有块内的主要端元提取出并组成全局图像的端元矩阵。
进一步的，在利用非监督分类方法对高光谱图像进行粗略分块之前还需要使用PCA对高光谱图像数据进行降维处理。
进一步的，在PCA降维中，首先对输入的高维图像数据X＝(x1,x2,...,xm)T进行向量中心化，计算向量中心化之后的数据的协方差矩阵，并计算协方差矩阵的特征值矩阵Λ和特征向量矩阵A；
然后使用特征向量矩阵A对高维图像数据X进行主成分变换为Z＝ATX；
最后选取Z中的部分主成分作为原高维数据的低维特征，从而实现数据降维。
进一步的，对降维后的高光谱图像数据利用迭代自组织数据分析方法ISODATA进行非监督分类，分类类数l和已知的图像中的端元个数r相等，即l＝r，分块结果为Γi，其中i＝1,2,...,r。
进一步的，ISODATA在非监督分类过程中，加入了对类别进行自动的合并和分裂。其中合并机制是指，当总类数过多或者某两类类中心距离小于某一阈值时，将该两类进行合并为新的一类，类内样本数目小于某一阈值时将该类取消；
分裂机制是指，当总类数过少或者某类内样本数目超过某一阈值，类内标准差大于分裂阈值，则将其分为两类，从而得到类数比较合理的聚类结果。
进一步的，对于所有的分块区域Γi，设置端元个数r′，其中r′＜r，分别进行VCA端元提取。
进一步的，VCA端元提取算法过程是首先找到一个初始单位向量，然后将所有像元点投影到这一向量上，投影距离最大的像元点记为端元点，加入端元矩阵集合，依据新的端元集合，再找到一个和所有已经找到的端元都正交的向量，并进行下一轮循环，计算像元投影距离，找寻新的端元，直到找到所有端元为止。
进一步的，对于所有的分块区域Γi，提取块内端元之后，分别利用最小二乘法对分块区域进行丰度反演。根据分块区域内丰度大小反馈，确定每个分块区域中的主要端元，提取出所有块中的主要端元，构成全局图像的端元矩阵。
进一步的，在线性模型中，高光谱图像的像元X是端元矩阵E和丰度矩阵A的线性组合， 即满足公式X＝E×A，丰度矩阵元素aij满足和为一
Figure PCTCN2017074294-appb-000001
和非负性
Figure PCTCN2017074294-appb-000002
的约束条件；
最小二乘法根据求解过程中是否考虑非负约束和和为1的约束，可分为无约束最小二乘法UCLS，和为1约束最小二乘法SCLS，非负约束最小二乘法NCLS，全约束最小二乘法FCLS。
进一步的，在无约束最小二乘法UCLS中，不考虑丰度的和为1与非负性约束，求得r个端元{ej}后，j＝1,2,...,r,利用最小二乘法求解线性混合模型可得像元i的丰度估计为aUCLS(xi)＝(ETE)-1ETxi。
本发明的有益效果：
1.本发明在使用VCA提取端元之前，首先利用非监督分类将高光谱图像中相似像元聚合，排除不相关像元的影响，降低端元提取环境的复杂度。
2.本发明在分块后的相对简单的环境块内环境使用VCA端元提取方法，然后利用块内丰度反演结果反馈控制块内的主要端元，从而避免了遗漏主要端元。
附图说明
图1为本发明的流程图；
图2a Washington DC mall图像；
图2b Washington DC mall图像使用ISODATA分块之后的图像；
图3a-图3e为使用基于分块的VCA提取端元、原始VCA提取端元与理论端元的比较图(Washington DC mall数据)；
图4a HYDICE Urban图像；
图4b HYDICE Urban图像使用ISODATA分块之后的图像；
图5a-图5f为使用基于分块的VCA提取端元、原始VCA提取端元与理论端元的比较图(HYDICE Urban数据)。
具体实施方式：
下面结合附图对本发明进行详细说明：
一种基于分块的VCA端元提取方法，包括如下步骤：
步骤(1)输入高光谱图像数据X∈Rm×n，其中m为高光谱图像的波段数，n为高光谱图像的像元总数，端元个数r，使用PCA对高光谱图像数据进行降维处理。
步骤(2)对降维后的高光谱图像数据利用ISODATA进行非监督分类，分类类数为l，其中l＝r。
步骤(3)利用分类结果对高光谱图像进行分块为Γi，其中i＝1,2,...,r。
步骤(4)对于所有的分块区域Γi，设置端元个数r′，其中r′＜r，分别进行VCA端元提取。
步骤(5)对于所有的分块区域Γi，分别利用最小二乘法对分块区域进行丰度反演。
步骤(6)根据分块区域内丰度大小反馈，确定每个分块区域中的主要端元，提取出所有块中的主要端元，构成全局图像的端元矩阵。
所述步骤(1)的PCA降维处理：
在进行ISODATA非监督分类之前，需要对信号进行降维，本发明使用主成分分析(PCA)降维。PCA是一种线性变换，变换之后各主成分之间互不相关，并且按照包含的信息量从大到小排列。将高维数据经过PCA变化之后，前几个主成分涵盖了原数据的主要信息，因此可以用低维特征刻画原高维数据，从而实现数据降维。在PCA降维中，首先对输入的高维图像数据X＝(x1,x2,...,xm)T进行向量中心化，计算向量中心化之后的数据的协方差矩阵，并计算协方差矩阵的特征值矩阵Λ和特征向量矩阵A。然后使用主成分变换矩阵A进行主成分变换Z＝ATY。最后选取Z中的部分主成分作为原高维数据的低维特征，从而实现数据降维。
所述步骤(2)中的ISODATA非监督分类：
迭代自组织数据分析方法(ISODATA)算法，是一种无需先验知识，直接从样本中提取特征进行聚类的非监督分类方法。ISODATA算法改进了K均值聚类，把所有样本都调整完毕之后，再重新计算各类样本的均值，并且加入了对类别进行自动合并和分裂，具有一定的自组织性。ISODATA算法中的合并机制是指，当总类数过多或者某两类类中心距离小于某一阈值时，将该两类进行合并为新的一类，类内样本数目小于某一阈值时将该类取消。分裂机制是指，当总类数过少或者某类内样本数目超过某一阈值，类内标准差大于分裂阈值，则将其分为两类，从而得到类数比较合理的聚类结果。
所述步骤(4)中的VCA端元提取：
VCA端元提取算法是基于线性光谱模型的，通过反复寻找数据空间中的正交向量并计算像元在正交向量上的投影距离逐一提取端元。VCA的基础理论是，单形体若干个顶点可以张成一个子空间，而单形体的顶点是在某个与这个子空间正交的向量上的投影长度最大值点。
VCA端元提取算法首先找到一个初始单位向量，然后将所有像元点投影到这一向量上，投影距离最大的像元点记为端元点，加入端元矩阵集合。依据新的端元集合，再找到一个和所有已经找到的端元都正交的向量，并进行下一轮循环，计算像元投影距离，找寻新的端元， 直到找到所有端元为止。
所述步骤(5)中的最小二乘法：
在线性模型中，高光谱图像的像元X是端元矩阵E和丰度矩阵A的线性组合，即满足公式X＝E×A。丰度矩阵满足和为一
Figure PCTCN2017074294-appb-000003
和非负性
Figure PCTCN2017074294-appb-000004
约束条件。当已经求出端元矩阵之后，混合像元求解丰度的问题就变成了一个简单的线性问题，因此可以用最小二乘法求解。最小二乘法根据求解过程中是否考虑非负约束和和为1的约束，可分为无约束最小二乘法(UCLS)，和为1约束最小二乘法(SCLS)，非负约束最小二乘法(NCLS)，全约束最小二乘法(FCLS)。在UCLS中，不考虑丰度的和为1与非负性约束，求得r个端元{ej}(j＝1,2,...,r)后，利用最小二乘法求解线性混合模型可得像元i的丰度估计为aUCLS(xi)＝(ETE)-1ETxi。
具体实施例:
具体实施部分使用地物空间分布相对简单的Washington DC mall数据和空间分布相对复杂HYDICE Urban数据分别进行测试。在试验过程中，将基于分块的VCA端元提取方法与原始VCA端元提取方法进行试验比较，并将人工提取的纯净端元作为理论端元。
实验数据Washington DC mall是拍摄于美国华盛顿地区的高光谱数据。该数据共有210个波段，去除掉部分受噪声影响的波段，剩下191个波段。整个图像大小为1280×307，本试验选取该数据中地物空间分布相对简单的，大小为200×150的一部分图像。该部分图像的伪彩色图如图2a所示。该部分图像中含有五种物质，分别是水域、马路、草地、屋顶和树木。
实验数据HYDICE Urban高光谱数据包含210个光谱波段，维数为307×307。图像数据中包含六种物质：路，泥土，树，草，屋顶，金属。在实验中，去掉水吸收影响的波段，该数据剩下178个波段。该部分图像的伪彩色图如图4a所示。
试验结果如图3a-图3e(Washington DC mall数据)和图5a-图5f(HYDICE Urban数据)所示。由端元光谱图的比较结果可以看出，基于分块的VCA端元提取方法提取的端元曲线与端元的理论光谱曲线十分接近。
为了进一步得到不同端元提取方法提取端元准确性在数字量上的对比，使用由不同端元提取方法得到的端元光谱与理论端元光谱的光谱角(Spectral Angle Distance,SAD)的大小来衡量提取端元的准确性。光谱角的公式定义如
Figure PCTCN2017074294-appb-000005
其中，Atheo为一 个端元的理论值，Aunmix为利用不同的端元提取方法提取的端元光谱信息值，光谱角越小表示两个光谱向量越接近。光谱角比较结果如表1所示和表2所示。可以明确看出，对于地物空间分布相对简单的Washington DC mall数据和空间分布相对复杂HYDICE Urban数据，基于分块的VCA提取端元的精确度都有很大的提高。
表1为使用基于分块的VCA提取端元、原始VCA提取端元与理论端元的光谱角的比较表(Washington DC mall数据)；表2为使用基于分块的VCA提取端元、原始VCA提取端元与理论端元的光谱角的比较表(HYDICE Urban数据)。
表1基于分块VCA，VCA提取的端元与理论端元SAD比较表
SAD 水域 马路 草坪 屋顶 树木 平均值
原始VCA 0.58180 0.17582 0.21371 0.08858 0.74156 0.360294
分块VCA 0.30364 0.11693 0.20546 0.08059 0.18585 0.178497
表2基于分块VCA，VCA提取的端元与理论端元SAD比较表
SAD 泥土 路 树 草 屋顶 合金 平均值
原始VCA 0.14804 1.12343 0.32690 0.10448 0.17928 0.17943 0.34359
分块VCA 0.06507 0.22934 0.17016 0.28003 0.22107 0.36128 0.22116
本发明将复杂环境的高光谱图像用一定的分类方法，分为多个相对简单的图像部分，然后分块进行提取端元，从而在一定程度上排除不相关像元的影响，降低端元提取环境的复杂度，减少了全局图像的噪声对算法的影响，避免了遗漏主要端元。具体事例表明，该发明大大提高了提取端元的准确度。
上述虽然结合附图对本发明的具体实施方式进行了描述，但并非对本发明保护范围的限制，所属领域技术人员应该明白，在本发明的技术方案的基础上，本领域技术人员不需要付出创造性劳动即可做出的各种修改或变形仍在本发明的保护范围以内。

Claims (8)

  1. 一种基于分块的VCA端元提取方法，其特征是，包括：
    利用非监督分类方法对高光谱图像进行粗略分块，将具有相似物质的像元分为相同块内；
    对分块后的各块内区域分别使用VCA进行端元提取，提取端元之后使用最小二乘法反演丰度，每块内根据丰度值的大小确定一个主要端元；
    将所有块内的主要端元提取出并组成全局图像的端元矩阵。
  2. 如权利要求1所述的一种基于分块的VCA端元提取方法，其特征是，在利用非监督分类方法对高光谱图像进行粗略分块之前还需要使用PCA对高光谱图像数据进行降维处理。
  3. 如权利要求2所述的一种基于分块的VCA端元提取方法，其特征是，在PCA降维中，首先对输入的高维图像数据X＝(x1,x2,...,xm)T进行向量中心化，计算向量中心化之后的数据的协方差矩阵，并计算协方差矩阵的特征值矩阵Λ和特征向量矩阵A；
    然后使用特征向量矩阵A对高维图像数据X进行主成分变换为Z＝ATX；
    最后选取Z中的部分主成分作为原高维数据的低维特征，从而实现数据降维。
  4. 如权利要求3所述的一种基于分块的VCA端元提取方法，其特征是，对降维后的高光谱图像数据利用迭代自组织数据分析方法ISODATA进行非监督分类，分类类数l和已知的图像中的端元个数r相等，即l＝r，分块结果为Γi，其中i＝1,2,...,r。
  5. 如权利要求4所述的一种基于分块的VCA端元提取方法，其特征是，ISODATA在非监督分类过程中，加入了对类别进行自动的合并和分裂；
    其中，合并机制是指，当总类数过多或者某两类类中心距离小于某一阈值时，将该两类进行合并为新的一类，类内样本数目小于某一阈值时将该类取消；
    分裂机制是指，当总类数过少或者某类内样本数目超过某一阈值，类内标准差大于分裂阈值，则将其分为两类，从而得到类数比较合理的聚类结果。
  6. 如权利要求4所述的一种基于分块的VCA端元提取方法，其特征是，对于所有的分块区域Γi，设置端元个数r′，其中r′＜r，分别进行VCA端元提取；
    VCA端元提取算法过程是首先找到一个初始单位向量，然后将所有像元点投影到这一向量上，投影距离最大的像元点记为端元点，加入端元矩阵集合，依据新的端元集合，再找到一个和所有已经找到的端元都正交的向量，并进行下一轮循环，计算像元投影距离，找寻新的端元，直到找到所有端元为止。
  7. 如权利要求4所述的一种基于分块的VCA端元提取方法，其特征是，对于所有的分块 区域Γi，提取块内端元之后，分别利用最小二乘法对分块区域进行丰度反演；
    根据分块区域内丰度大小反馈，确定每个分块区域中的主要端元，提取出所有块中的主要端元，构成全局图像的端元矩阵。
  8. 如权利要求3所述的一种基于分块的VCA端元提取方法，其特征是，在线性模型中，高光谱图像的像元X是端元矩阵E和丰度矩阵A的线性组合，即满足公式X＝E×A，丰度矩阵元素aij满足和为一
    Figure PCTCN2017074294-appb-100001
    和非负性
    Figure PCTCN2017074294-appb-100002
    的约束条件；
    在无约束最小二乘法UCLS中，不考虑丰度的和为1与非负性约束，求得r个端元{ej}后，j＝1,2,...,r，利用最小二乘法求解线性混合模型可得像元i的丰度估计为aUCLS(xi)＝(ETE)-1ETxi。
PCT/CN2017/074294 2016-05-04 2017-02-21 一种基于分块的vca端元提取方法 WO2017190542A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/099,140 US10984291B2 (en) 2016-05-04 2017-02-21 End-member extraction method based on segmented vertex component analysis (VCA)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610289893.5 2016-05-04
CN201610289893.5A CN105976310B (zh) 2016-05-04 2016-05-04 一种基于分块的vca端元提取方法

Publications (1)

Publication Number Publication Date
WO2017190542A1 true WO2017190542A1 (zh) 2017-11-09

Family

ID=56993703

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/074294 WO2017190542A1 (zh) 2016-05-04 2017-02-21 一种基于分块的vca端元提取方法

Country Status (3)

Country Link
US (1) US10984291B2 (zh)
CN (1) CN105976310B (zh)
WO (1) WO2017190542A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110992273A (zh) * 2019-11-04 2020-04-10 中国科学院西安光学精密机械研究所 一种自相似性约束的高光谱影像解混方法
CN111008975A (zh) * 2019-12-02 2020-04-14 北京航空航天大学 一种空间人造目标线性模型的混合像元解混方法及系统
CN111612724A (zh) * 2020-05-29 2020-09-01 南阳理工学院 一种快速高光谱图像线性解混方法
CN112380967A (zh) * 2020-11-12 2021-02-19 北京航空航天大学 一种联合图像信息的空间人造目标光谱解混方法及系统
CN112504975A (zh) * 2020-12-14 2021-03-16 杭州电子科技大学 一种基于约束非负矩阵分解的高光谱解混方法
CN113777104A (zh) * 2021-08-24 2021-12-10 北京工商大学 一种单粒玉米种子高光谱检测成熟度的方法

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105976310B (zh) 2016-05-04 2018-01-12 山东大学 一种基于分块的vca端元提取方法
US20200364835A1 (en) * 2018-01-31 2020-11-19 Nec Corporation Image processing device, image processing method and storage medium
CN111007018B (zh) * 2019-12-18 2022-06-24 湖北久之洋红外系统股份有限公司 一种用于光谱气体检测的背景估计方法及系统
CN111508065B (zh) * 2020-04-15 2023-06-16 铜陵学院 基于空谱联合压缩感知的高光谱重建方法、设备及系统
CN113177188B (zh) * 2021-04-19 2021-12-07 中国农业科学院农业资源与农业区划研究所 一种基于光谱新特征的反演叶片叶绿素含量的方法
CN113743325B (zh) * 2021-09-07 2024-01-12 中国人民解放军火箭军工程大学 监督与非监督的高光谱混合像元分解方法
CN113807273B (zh) * 2021-09-18 2023-12-19 哈尔滨工程大学 基于增强全变分约束表示模型与端元背景字典的高光谱异常检测方法
CN113902759B (zh) * 2021-10-13 2022-04-22 自然资源部国土卫星遥感应用中心 一种空谱信息联合的星载高光谱影像分割与聚类方法
CN114002165B (zh) * 2021-10-29 2022-08-30 中国科学院新疆生态与地理研究所 一种基于铜元素光谱指数反演的铜元素丰度预测方法
CN115829886B (zh) * 2022-12-21 2023-08-11 哈尔滨师范大学 基于端元自适应不相干和空间约束的盲高光谱解混方法
CN116091832B (zh) * 2023-02-16 2023-10-20 哈尔滨工业大学 基于自编码器网络的肿瘤细胞切片高光谱图像分类方法
CN116297497B (zh) * 2023-05-23 2023-08-15 武汉大学 一种基于高光谱遥感的手机面板质量检测方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090144350A1 (en) * 2007-09-28 2009-06-04 Chein-I Chang Maximum simplex volume criterion-based endmember extraction algorithms
CN102592134A (zh) * 2011-11-28 2012-07-18 北京航空航天大学 一种高光谱与红外数据多级决策融合分类方法
CN104008574A (zh) * 2014-06-16 2014-08-27 浙江大学 一种基于无限高斯混合模型的高光图图像解混方法
CN105976310A (zh) * 2016-05-04 2016-09-28 山东大学 一种基于分块的vca端元提取方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7321791B2 (en) * 2003-09-23 2008-01-22 Cambridge Research And Instrumentation, Inc. Spectral imaging of deep tissue
EP2476008B1 (en) * 2009-09-10 2015-04-29 Rudjer Boskovic Institute Underdetermined blind extraction of components from mixtures in 1d and 2d nmr spectroscopy and mass spectrometry by means of combined sparse component analysis and detection of single component points
US9449244B2 (en) * 2013-12-11 2016-09-20 Her Majesty The Queen In Right Of Canada, As Represented By The Minister Of National Defense Methods for in-scene atmospheric compensation by endmember matching
CN104463247B (zh) * 2014-12-09 2016-08-31 山东大学 一种高光谱图像分类中光谱向量互相关特征的抽取方法
CN104751181B (zh) * 2015-04-02 2018-01-12 山东大学 一种基于相对丰度的高光谱图像解混方法
CN105046648B (zh) * 2015-06-25 2019-01-22 北京师范大学 一种构建高时空遥感数据的方法
WO2017055895A1 (en) * 2015-09-30 2017-04-06 Arcelormittal Method of online characterization of a layer of oxides on a steel substrate

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090144350A1 (en) * 2007-09-28 2009-06-04 Chein-I Chang Maximum simplex volume criterion-based endmember extraction algorithms
CN102592134A (zh) * 2011-11-28 2012-07-18 北京航空航天大学 一种高光谱与红外数据多级决策融合分类方法
CN104008574A (zh) * 2014-06-16 2014-08-27 浙江大学 一种基于无限高斯混合模型的高光图图像解混方法
CN105976310A (zh) * 2016-05-04 2016-09-28 山东大学 一种基于分块的vca端元提取方法

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110992273A (zh) * 2019-11-04 2020-04-10 中国科学院西安光学精密机械研究所 一种自相似性约束的高光谱影像解混方法
CN110992273B (zh) * 2019-11-04 2023-04-18 中国科学院西安光学精密机械研究所 一种自相似性约束的高光谱影像解混方法
CN111008975A (zh) * 2019-12-02 2020-04-14 北京航空航天大学 一种空间人造目标线性模型的混合像元解混方法及系统
CN111008975B (zh) * 2019-12-02 2022-09-09 北京航空航天大学 一种空间人造目标线性模型的混合像元解混方法及系统
CN111612724A (zh) * 2020-05-29 2020-09-01 南阳理工学院 一种快速高光谱图像线性解混方法
CN111612724B (zh) * 2020-05-29 2023-02-07 南阳理工学院 一种快速高光谱图像线性解混方法
CN112380967A (zh) * 2020-11-12 2021-02-19 北京航空航天大学 一种联合图像信息的空间人造目标光谱解混方法及系统
CN112380967B (zh) * 2020-11-12 2022-05-24 北京航空航天大学 一种联合图像信息的空间人造目标光谱解混方法及系统
CN112504975A (zh) * 2020-12-14 2021-03-16 杭州电子科技大学 一种基于约束非负矩阵分解的高光谱解混方法
CN112504975B (zh) * 2020-12-14 2022-12-30 杭州电子科技大学 一种基于约束非负矩阵分解的高光谱解混方法
CN113777104A (zh) * 2021-08-24 2021-12-10 北京工商大学 一种单粒玉米种子高光谱检测成熟度的方法
CN113777104B (zh) * 2021-08-24 2023-08-29 北京工商大学 一种单粒玉米种子高光谱检测成熟度的方法

Also Published As

Publication number Publication date
CN105976310A (zh) 2016-09-28
US20190392261A1 (en) 2019-12-26
US10984291B2 (en) 2021-04-20
CN105976310B (zh) 2018-01-12

Similar Documents

Publication Publication Date Title
WO2017190542A1 (zh) 一种基于分块的vca端元提取方法
Dong et al. Spectral–spatial joint sparse NMF for hyperspectral unmixing
Tong et al. Region-based structure preserving nonnegative matrix factorization for hyperspectral unmixing
CN107563442B (zh) 基于稀疏低秩正则图张量化嵌入的高光谱图像分类方法
Wang et al. Adaptive ${L} _ {\bf 1/2} $ sparsity-constrained NMF with half-thresholding algorithm for hyperspectral unmixing
CN110427997B (zh) 面向复杂遥感影像背景的改进cva变化检测方法
Guo et al. Template matching via $ l_1 $ minimization and its application to hyperspectral data
Ren et al. Mineral mapping with hyperspectral image based on an improved k-means clustering algorithm
Kowkabi et al. A fast spatial–spectral preprocessing module for hyperspectral endmember extraction
Hasanlou et al. A sub-pixel multiple change detection approach for hyperspectral imagery
CN115272861A (zh) 一种基于光谱相关性的子空间稀疏表征高光谱目标检测方法
CN108280486B (zh) 一种基于端元簇的高光谱图像解混方法
CN108229426B (zh) 一种基于差分描述子的遥感图像变化向量变化检测法
Beaven et al. Comparison of Gaussian mixture and linear mixture models for classification of hyperspectral data
Sun et al. Trilateral filter on multispectral imagery for classification and segmentation
Xie et al. SMRD: A Local Feature Descriptor for Multi-modal Image Registration
Merugu et al. A Review of Some Information Extraction Methods, Techniques and their Limitations for Hyperspectral Dataset
Ding et al. Isomap-based subspace analysis for the classification of hyperspectral data
Chang et al. Finding endmembers in hyperspectral imagery
CN112651295A (zh) 一种城市绿地树木识别系统及方法
Zhang et al. Extraction of earth surface texture features from multispectral remote sensing data
Lin et al. Bayesian information criterion based feature filtering for the fusion of multiple features in high-spatial-resolution satellite scene classification
Qiao et al. Adaptive thematic object extraction from remote sensing image based on spectral matching
Oguslu et al. Hyperspectral image classification using a spectral-spatial sparse coding model
CN113723335B (zh) 基于稀疏表示的高光谱混合像元分解方法

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

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

Ref document number: 17792371

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 17792371

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 27/05/2019)

122 Ep: pct application non-entry in european phase

Ref document number: 17792371

Country of ref document: EP

Kind code of ref document: A1