CN100483147C - High spectrum sub-pixel target detection method and device - Google Patents

High spectrum sub-pixel target detection method and device Download PDF

Info

Publication number
CN100483147C
CN100483147C CN 200710176782 CN200710176782A CN100483147C CN 100483147 C CN100483147 C CN 100483147C CN 200710176782 CN200710176782 CN 200710176782 CN 200710176782 A CN200710176782 A CN 200710176782A CN 100483147 C CN100483147 C CN 100483147C
Authority
CN
Grant status
Grant
Patent type
Prior art keywords
vector
represents
target
spectrum
pixel
Prior art date
Application number
CN 200710176782
Other languages
Chinese (zh)
Other versions
CN101144861A (en )
Inventor
李庆波
张广军
鑫 聂
Original Assignee
北京航空航天大学
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
Grant date

Links

Abstract

本发明公开了一种高光谱亚像元目标探测方法,该方法包括:建立目标光谱和待测图像像元光谱二维矩阵的逆模型;获取该逆模型的回归系数向量;根据得到的回归系数向量获取各像元的马氏距离;判定马氏距离大于阀值的回归系数所对应的像元为亚像元目标点。 The present invention discloses a high-spectral target detection sub-pixel, the method comprising: establishing an inverse model of the target spectrum and measured spectrum of a two-dimensional pixel image matrix; obtaining the regression coefficient vector of the inverse model; the regression coefficient obtained vector obtaining the Mahalanobis distance of each of pixels; Mahalanobis distance greater than a threshold determined regression coefficients corresponding to the pixel of the target point subpixel. 本发明还提供了一种高光谱亚像元目标探测装置,采用本发明的亚像元目标探测方法及装置,不需要背景端元光谱的先验信息,对背景光谱的复杂度不敏感,具有较高的目标探测准确度和较快的运算速度。 The present invention also provides a hyperspectral Subpixel target detection apparatus according to the present invention, the target sub-pixel detection method and apparatus, does not require prior information of the background spectrum element end, the complexity of the background spectrum is not sensitive, having higher target detection accuracy and faster processing speed.

Description

高光谱亚像元目标探测方法及装置 Hyperspectral Subpixel Target detection method and apparatus

技术领域 FIELD

本发明涉及高光谱遥感探测技术领域,尤其涉及一种高光谱亚像元目标探测方法及装置。 The present invention relates to the field of remote sensing hyperspectral detection technology, and particularly to a Hyperspectral Subpixel Target detection method and apparatus.

背景技术 Background technique

高光谱遥感是二十世纪末地球观测系统中最重要的技术突破之一,它克服了传统单波段、多光谱遥感在波段数、波段范围、精细信息表达等方面的局限性,以较窄的波段区间、较多的波段数量提供遥感信息,能够从光谱空间中对地物予以细分和鉴别,在资源遥感、环境遥感、生态遥感等领域得到了广泛应用。 Hyperspectral remote sensing is one of the most important twentieth century Earth observation system technology breakthrough, it has overcome the traditional single-band, multispectral remote sensing limitations in terms of the number of bands, wavelength range, fine expression information, etc., in order to narrow band range, the number of bands to provide more information on remote sensing, can be broken down and identification of surface features from spectral space, has been widely applied in the fields of remote sensing, environmental sensing, ecology and remote sensing. 高光谱遥感数据最主要的特点是将传统的图像维与光谱维信息融合为一体,在获取地表空间图像的同时,得到每个地物的连续光谱信息,该光谱信息能够反映出复杂背景下弱小目标和背景的细微差别,因此可借助丰富的光谱信息发现空间特征难以或无法探测的地面目标。 Hyperspectral Data main feature is the dimension of a conventional image information for the integrated spectral dimension, the acquired aerial image surface obtained while continuous spectrum information of each feature of the spectral information can reflect weak complex background the nuances of target and background, so you can find ground targets difficult or impossible to detect spatial characteristics of the aid abundant spectral information. 由于目前高光谱遥感空间分辨率有限,单一像元很难只包括一种地物成分,其光谱往往是多种地物光谱的混叠,因此对于此类体积尺寸小于像元空间分辨率的目标探测称为亚像元目标探测。 Hyperspectral due to the current limited spatial resolution, a single pixel includes only one hard component feature, which is often more spectral aliasing spectral feature, and therefore for such volume is smaller than the image size of the target spatial resolution element probe called sub-pixel target detection. 前述的空间分辨率即是指图像上所能辨别的地面物体最小尺寸。 The spatial resolution which means the minimum dimension of land object image can be discriminated. 高光谱目标探测技术被广泛的应用于军事目标探测、矿物勘探、植被分布评估、环境污染检测等领域。 Hyper-spectral target detection technique is widely used to detect a military target, mineral exploration, evaluation field vegetation distribution, environmental pollution detection.

目前常见的亚像元目标探测方法主要包括:复合光谱最小二乘分离的探测方法、基于正交子空间投影(OSP,Orthogonal Subpixel Projection)的探测方法和基于广义似然比的探测方法等。 Current common subpixel main target detection method comprising: detecting a composite spectral separation method of least squares, the orthogonal subspace projection (OSP, Orthogonal Subpixel Projection) detection methods and detection methods based on the generalized likelihood ratio based on the like.

其中,复合光谱最小二乘分离的探测方法为:由于待测图像中的任一像元光谱均可以看成多种基本物质的光谱的加权和,权值即对应像元中各组成物质所占的混合比例,如果图像地物中各像元组分的纯光谱可从已知的光谱数据库中得到,则可以利用有约束最小二乘方法计算出目标对应的组分光谱在混合像元光谱中所占的比例,从而探测小于地面像元的目标。 Wherein the composite spectral separation detection least squares method: test image due to any element of a spectral image can be viewed as a weighted spectrum and the plurality of basic substances, i.e., the weight corresponding to each pixel in the composition of the material occupies mixing ratio, if the image feature of each image element pure spectral components may be obtained from a database of known spectra, the spectral components can be calculated by using the target corresponding to the constrained least squares method in a mixed pixel spectrum proportion so as to detect ground targets smaller than a pixel. 该方法可以对像元光谱进行定量分析,但前提是需要各像元组分光谱的先验信息,因此很难应用到背景未知的目标探测领域。 The method may be for pixel spectra analysis, but only requires a priori information of each pixel of the spectral components, it is difficult to apply to the background detection field of the unknown target.

基于OSP的探测方法,主要是将像元光谱向量投影到背景特征的正交子空间,从而消除背景信息,突出目标光谱信息。 OSP-based detection method, the main spectral vectors pixels projected onto orthogonal subspaces background characteristics, thereby eliminating the background information, projection target spectral information. 现有技术中存在一种无监督的正交子空间投影方法对亚像元目标进行检测,该方法首先利用一种迭代的方法找到一组背景地物的端元光谱,也即地物中各组成物质的纯光谱,而后利用这些背景地物端元光谱构造检测算子,动态的对亚像元目标进行检测。 The prior art present an unsupervised orthogonal subspace projection method Subpixel detection target, the method first found using an iterative method endmember set of background feature, i.e. each feature pure spectral composition of matter, then use these background feature detection configured endmember operator, the dynamic subpixel target detection. 该方法可以在背景地物端元未知的情况下构造出OSP检测算子,对背景噪声有一定的抑制作用,但由于计算所得到的背景地物端元光谱精确度不高,导致探测准确度不高,而且计算耗时较长。 The method may be configured in the context of end-membered feature OSP unknown detecting an operator, it has a certain extent of background noise, but the feature values ​​calculated by the background endmember accuracy is not high, resulting in detection accuracy It is not high, and the computer takes a long time. 现有技术中还存在一种基于广义似然比的探测方法,主要是先假设背景光谱信息满足某种多维分布,然后构造一定的检测算子通过假设检验的方法判定像元中是否含有潜在的目标光谱。 There is also prior art method of detecting method based on generalized likelihood ratio, is the first major and background information satisfies certain spectral multivariate then constructed by certain detection operator hypothesis test determines whether the pixel potential containing target spectrum. 这种方法能够给出理论上的虚警率,从而自适应地调节阈值,但该方法只能检测出背景中的奇异点,且需要一定的图像信息,训练样本的好坏对探测结果的影响较大。 This method is theoretically capable of giving false alarm rate, thereby adaptively adjusting the threshold value, but the method can only detect a singular point in the background and requires a certain image information, influence the quality of the detection results of training samples of larger. 由于该探测方法需要有背景地物的先验信息,且对背景复杂度较为敏感,不能在背景变化复杂的小目标探测领域进行。 Since the method of detecting the background feature requires a priori information, and more sensitive to the complexity of the background, not in a complex background art small change in target detection.

综上所述,现有技术中的亚像元目标探测方法大都需要背景地物端元光谱的先验信息,对背景光谱的复杂度较为敏感,而且目标探测的准确度较低、速度较慢。 In summary, the prior art method of sub-pixel target detection requires mostly background feature endmember a priori information, it is sensitive to the complexity of the background spectrum, and low target detection accuracy, slower .

发明内容 SUMMARY

有鉴于此,本发明的主要目的在于提供一种高光谱亚像元目标探测方法及目标探测准确度较低和速度较慢的缺陷。 In view of this, the main object of the present invention is to provide a high spectral subpixel target detection and target detection method less accurate and slower defects.

为达到上述目的,本发明的技术方案是这样实现的: To achieve the above object, the technical solution of the present invention is implemented as follows:

本发明一种高光谱亚像元目标探测方法,包括以下步骤: . Subpixel hyperspectral target detection method of the present invention, comprising the following steps:

建立目标光谱和待测图像像元光谱二维矩阵的逆模型; Inverse model to establish a target spectrum and measured spectrum image pixel two-dimensional matrix;

获取所述逆模型的回归系数向量; Obtaining the regression coefficient vector of the inverse model;

根据所述回归系数向量获取各像元的马氏距离; Get Mahalanobis distance for each pixel based on the regression coefficient vector;

判定马氏距离大于阀值的回归系数所对应的像元为亚像元目标点。 Regression coefficient determining the Mahalanobis distance is greater than the threshold value corresponding to the pixel of the target point subpixel.

其中,所述建立目标光谱和待测图像像元光谱二维矩阵的逆模型,具体包括: Wherein said inverse model to establish a target spectrum and measured spectrum of a two-dimensional pixel image matrix, comprises:

将获取的待测图像像元的三维高光谱数据表示为高光谱反射率的二维矩阵: The acquired three-dimensional test image pixel data representing a high spectral two-dimensional matrix of high spectral reflectance:

其中,表示像元光谱的二维矩阵,[PpP2WwPPa',]和[P1, P2... Phxx,..P,”.]表示像元的光谱矢量,m表示波段数,η表示图像像元总数,X表示图像像元行数,表示图像像元列数,《 = xx_y; Wherein, as a two-dimensional matrix elements represents the spectrum, [PpP2WwPPa ',] and [P1, P2 ... Phxx, .. P, ".] Denotes an image vector elements spectrum, m represents the number of bands, [eta] represents a pixel image Total, X represents the number of pixels the image lines, the image represents the number of pixel columns, "= xx_y;

建立所述目标光谱和所述待测图像像元光谱二维矩阵的线性关系: Establishing the linear relationship between the target spectrum and the spectral image of the test image element two-dimensional matrix:

其中,5;表示目标光谱向量,尺·表示像元光谱的二维矩阵,Cks表示回归系数,民…表示噪声向量。 Among them, 5; vector indicates the target spectrum, foot-pixel represents a two-dimensional matrix spectrum, Cks represent the regression coefficients, the people ... represents the noise vector.

其中,所述将三维高光谱数据表示为二维矩阵和建立线性关系之间,还包括:对所述目标光谱和所述待测图像像元光谱二维矩阵进行预处理。 Wherein the three-dimensional hyper-spectral data are presented as between two-dimensional matrix and the linear relation is established, further comprising: an image of the target spectrum and the measured spectrum of a two-dimensional matrix of picture elements preprocessing.

其中,所述预处理包括标准正交变换处理或附加散射校正处理。 Wherein said orthonormal transform processing including pretreatment or additional scatter correction process.

其中,所述标准正交变换处理具体包括: Wherein said orthonormal transform processing comprises:

其中,/¾胃表示经过正交变换处理后待测图像中第A个像元在第A个波段的反射率值,k表示待测图像中第/Z个像元在各个波段处反射率的平均值,《表示波段数,w-1表示自由度; Wherein, / ¾ represents stomach after orthogonal transform processing in the A-th test image pixel A reflectance bands of values, k represents the first test image / Z th pixel reflectivity at each band of average "represents a number of bands, w-1 represents a degree of freedom;

所述附加散射校正处理具体包括: The additional scatter correction process comprises:

计算平均光谱矢量: Calculating an average vector of the spectrum:

对每一个像元光谱进行线性回归: Spectrum for each pixel by linear regression:

进行附加散射校正: Additional scatter correction:

其中,P表示平均光谱矢量, Wherein, P represents the average spectral vectors,

表示对所有像元光谱矢量的求和,? It represents the sum of all pixel spectrum vector? /1表示第个像元的光谱矢量,mh、A分别表示第A个像元光谱矢量ρΛ与所有像元平均光谱的线性回归的斜率与截距,pA(Msn表示经过附加散射校正后的像元光谱矢量。 / 1 represents one pixel of the spectral vector, mh, A respectively represent the A-th pixel and the slope of the spectral vector ρΛ linear regression intercept all the pixels averaged spectra, pA (Msn image represented through the additional scatter correction element spectral vector.

其中,通过偏最小二乘迭代方法获取所述逆模型的回归系数向量,具体包括: Wherein obtaining the regression coefficient vector of the inverse model by partial least squares iterative method comprises:

a、根据目标光谱向量获取初始权重向量:wn = RS.;.,其中,S,表示目标光谱向量,表示初始权重向量,W表示像元光谱的二维矩阵; a, obtaining the target spectral vector according to the initial weight vector: wn = RS;, wherein, S, represents the target spectral vector represents the initial weight vector, W represents a two-dimensional matrix of pixels spectra;.

b、根据所述初始权重向量计算得分向量:其中,(表示得分向量,i?表示像元光谱的二维矩阵,表示初始权重向量; B, score vector calculated according to the initial weight vector: wherein, (a score indicating a vector, i denotes a two-dimensional image matrix elements spectrum represents the initial weight vector;?

C、根据所述得分向量计算所述目标光谱的载荷向量:其中,%表示所述目标光谱的载荷向量,表示目标光谱向量,t表示得分向量; C, the load vector is calculated based on the spectrum of the target score vector: where% represents the load vector of the target spectrum, the target spectral vector represents, t represents the score vector;

d、根据所述得分向量计算所述像元光谱二维矩阵的载荷向量:p„=Rt„,其中,&表示所述像元光谱二维矩阵的载荷向量,A表示像元光谱的二维矩阵,t„表示得分向量; d, the load is calculated as the vector elements in accordance with the spectral two-dimensional matrix of score vectors: p "= Rt", wherein the image represents & spectrum load vector element two-dimensional matrix, A represents the spectrum of a two-dimensional pixel matrix, t "represents the score vector;

e、获取所述回归系数向量:Cm =WiP1Wy'Qr ,其中, E, obtaining the regression coefficient vector: Cm = WiP1Wy'Qr, wherein

f、计算残差平方和: f, and calculate the residual sum of squares:

,其中,切㈨表示残差平方和, Wherein cutting ix represents residual sum of squares,

m表示波段数,凡表示原始的像元光谱二维矩阵,St表示目标光谱向量,设定阀值为G,若&外7-1)-怒(《)^,则取该观>)对应的Cf^s为回归系数向量;否则,令St H凡,R^R-tnPn,返回步骤a重复上述操作,直到5S0-1)-SS⑷SG,然后获取回归系数向量qs。 m represents the number of bands, where the spectral representation of the original two-dimensional matrix of pixels, St represents the target spectral vector, the threshold is set to G, when the outer & 7-1) - anger ( ") ^, then take the concept>) corresponding to the Cf ^ s regression coefficient vector; otherwise, let where St H, R ^ R-tnPn, returns to step a above operations are repeated until 5S0-1) -SS⑷SG, then get the regression coefficient vector qs.

本发明还提供了一种高光谱亚像元目标探测装置,包括: The present invention also provides a hyperspectral Subpixel target detection apparatus, comprising:

模型建立单元,用于建立目标光谱和待测图像像元光谱二维矩阵的逆模型; 回归系数向量获取单元,用于获取所述逆模型的回归系数向量; Model means for establishing an inverse model of the target spectrum and measured spectrum of a two-dimensional pixel image matrix; regression coefficient vector obtaining unit, configured to obtain the regression coefficient vector of the inverse model;

马氏距离获取单元,用于根据所述回归系数向量获取各像元的马氏距离;判定单元,用于判定马氏距离大于阀值的回归系数所对应的像元为亚像元目标点。 Mahalanobis distance obtaining unit, configured to obtain each of the Mahalanobis distance according to the pixel regression coefficient vector; determining means for determining a Mahalanobis distance greater than a threshold regression coefficient corresponding pixel of the target point subpixel.

其中,所述模型建立单元包括: Wherein said model establishing unit comprises:

矩阵生成子单元,用于将待测图像像元的三维高光谱数据表示为高光谱反射率的二维矩阵; Matrix generating sub-unit, three-dimensional image of an image to be tested for hyperspectral data element is represented as a two-dimensional matrix of high spectral reflectance;

线性关系建立子单元,用于建立所述目标光谱和所述待测图像像元光谱二维矩阵的线性关系。 Linear relationship establishing subunit, for establishing a linear relationship between the target spectrum and the spectral image of the test image element two-dimensional matrix.

其中,所述模型建立单元还包括:预处理子单元,用于对所述目标光谱和所述矩阵生成子单元生成的待测图像像元光谱二维矩阵进行预处理并提供给线性关系建立子单元。 Wherein said model establishing unit further comprises: pre-processing sub-unit, used for measuring the spectrum of the target image and the sub-matrix generating unit generates a two-dimensional matrix of pixel spectra pretreated and supplied to the linear relationship established sub unit.

其中,所述装置还包括:数据获取单元,用于获取待测图像像元的三维高光谱数据并提供给模型建立单元。 Wherein said apparatus further comprises: a data acquisition unit for acquiring three-dimensional data of hyperspectral image picture elements to be measured and provided to the model unit.

与现有技术相比,本发明具有以下优点: Compared with the prior art, the present invention has the following advantages:

本发明所用到的逆模型不需要背景端元光谱的先验信息,仅需要先验的目标光谱信息,因此对背景光谱的复杂度不敏感;探测准确度仅仅依赖于目标光谱的精确程度,从而目标探测准确度高;又由于偏最小二乘中的迭代计算方法法,使得本发明的探测方法具有较快的运算速度。 Inverse model does not need to use the background of the present invention endmember prior information, only certain spectral information a priori, thus the complexity of the background spectrum is insensitive; detection accuracy depends only on the accuracy of the target spectrum, whereby target detection accuracy is high; and because the partial least squares iterative calculation method, so that the detection method of the present invention has a faster processing speed.

附图说明 BRIEF DESCRIPTION

图I为本发明一种高光谱亚像元目标探测方法的流程图; Figure I a flowchart of a high pixel target detection method of the present invention, the spectral alkylene;

图2为本发明实施例的原始高光谱图像的一示意图; FIG 2 is an original hyper-spectral image according to an embodiment schematic view of the invention;

图3为本发明实施例的原始高光谱图像的另一示意图; Figure 3 a schematic view of another embodiment of the hyperspectral image of the original embodiment of the present invention;

图4为本发明实施例的目标光谱示意图; FIG 4 is a schematic diagram of the spectrum of the target of the present embodiment of the invention;

图5为本发明实施例的混入目标光谱后的像元点示意图; FIG 5 a schematic view of the pixel points mixed target spectra embodiment of the present invention;

图6为本发明实施例的回归系数三维示意图; FIG regression coefficients Example 6 three-dimensional schematic embodiment of the invention;

图7为本发明实施例的马氏距离示意图; FIG 7 Mahalanobis distance schematic diagram of an embodiment of the invention;

图8为本发明实施例的亚像元目标探测结果示意图; FIG 8 alkylene embodiment of the present invention showing the results of the detection target pixel;

图9为本发明一种高光谱亚像元目标探测装置的组成结构示意图。 A structure diagram of sub-pixel hyperspectral target detection apparatus of FIG. 9 of the present invention.

具体实施方式 detailed description

下面结合附图和具体实施例对本发明的技术方案做进一步详细阐述: In conjunction with the accompanying drawings and the following specific examples further elaborate the technical solution of the present invention:

如图I所示,图I为本发明一种高光谱亚像元目标探测方法的流程图,主要包括以下步骤: FIG I, the flowchart of FIG hyperspectral subpixel I target detection method of the present invention includes the following steps:

步骤101,建立目标光谱和待测图像像元光谱二维矩阵的逆模型。 Step 101, the inverse model spectra pixel two-dimensional matrix, and establishing the target spectra measured image.

首先,将获取到待测图像像元的三维高光谱数据表示为高光谱反射率的二维矩阵。 First, the acquired three-dimensional data measured hyperspectral image pixel is represented as two-dimensional matrix of high spectral reflectance. 该待测图像像元的三维高光谱数据是由成像光谱仪获得,成像光谱仪在每个波段可获得一幅感光图像,而由一些列连续的波段的图像叠加而成的立方体即为高光谱图像,该高光谱图像包括图像空间的两维信息和光谱波段方向的一维信息,因此为三维的高光谱数据。 The three-dimensional test image hyperspectral image data element is obtained by the imaging spectrometer, imaging spectrometer can be obtained a photosensitive image of each band, by a number of superimposed image sequence obtained by a continuous band is the hyper-spectral image cube, the hyper-spectral image comprising a two-dimensional image information and the spectral band direction one-dimensional space information, a three-dimensional hyperspectral data. 获取待测图像像元的三维高光谱数据,并将该三维的高光谱数据表示为如下的二维矩阵: Acquiring three-dimensional test image hyperspectral image metadata and the three-dimensional data of hyperspectral two-dimensional matrix is ​​represented as follows:

R_ = [ΡρΡ2···Ρ^+/···Ρχχ.κ]^<i<x,0<j<y ( I ) R_ = [ΡρΡ2 ··· Ρ ^ + / ··· Ρχχ.κ] ^ <i <x, 0 <j <y (I)

或R_ = [Pi^P2-P,+,x/-P,xJ,]^< i<x,0<j<y (2)元按列展开的表示形式,Rmxil表示像元光凊的二维矩阵,X表示高光谱图像像元行数,y表示高光谱图像像元列数,m表示波段数,η表示图像像元总数,《 =[PpP2..-PywiJ和Lpi,P2..·ρ,+Λ·χ7...ρ~]表示高光谱图像像元的光谱矢量,在按行展开的L,,中,ρ_,表示待测图像中第i行第j列对应像元的光谱矢量;同理,在按列展开的中,P,+w则表示待测图像中第i行第j列对应像元的光谱矢量。 Or R_ = [Pi ^ P2-P, +, x / -P, xJ,] ^ <i <x, 0 <j <y (2) represents in the form of expanded columns, Rmxil represents two pixel light crispy dimensional matrix, X represents a hyperspectral image pixel rows, y represents a hyperspectral image pixel column number, m represents the number of bands, η represents the image total number of cells, "= [ppP2 ..- PywiJ and Lpi, P2 .. · ρ, + Λ · χ7 ... ρ ~] shows the spectrum vector hyperspectral image pixel in row L ,, the expanded, ρ_, representing an image to be measured in the i-th row j-th column of the pixel corresponding to the spectrum vector; Similarly, in the expanded columns, P, + w represents an image to be measured in the i-th row j-th column vector of the corresponding pixel of the spectrum. 每个像元的光谱矢量包括该像元在各个波段处的反射率值,例如:假设第^个像元的光谱矢量为Pa,则Pa = [Ph',Ph2"-PMr"PhJT,[Ph',Ph2…Pmt..PhJr 代表I Al,Ph2 ■ ·-Phk ·■ -Phm J的转置矩阵,其中/¾表示该第h个像元在第左个波段处的反射率值。 Each pixel includes a spectral vector of the pixel in the reflectance value at each band, for example: assume that the first image spectrum ^ th vector element is Pa, the Pa = [Ph ', Ph2 "-PMr" PhJT, [Ph ', Ph2 ... Pmt..PhJr representatives I Al, Ph2 ■ · -Phk · ■ -Phm J transposed matrix, where / ¾ represents the h-th pixel value in the reflectivity of the bands at the left. 由于成像光谱仪获取到图像像元点在每个波段处的值即是该像元点在该波段处的反射率值,因此该反射率值为巳知量。 Since the imaging spectrometer to acquire the image pixel value for each band in a point that is at the pixel point value in the reflectance of the wavelength band, so that the reflectance value of a known amount Pat.

然后,对目标光谱和待测图像像元光谱二维矩阵进行预处理,以校正因大气散射引起的光谱误差。 Then, the image of the target spectrum and the measured spectrum of a two-dimensional pixel matrix preprocessed to correct errors due to atmospheric scattering spectra caused. 该预处理可为标准正交变换处理,也可为附加散射校正处理。 This pretreatment can be orthonormal transform processing, scatter correction may also be an additional processing. 当然,本发明的预处理方法并不仅仅局限于上述两种处理方法,其他任何能校正因大气散射引起的光谱误差的处理方法也应属于本发明的保护范围。 Of course, the pretreatment method of the present invention is not limited to the above-described two methods, any other method of processing spectral error can be corrected due to atmospheric scattering should also fall within the scope of the present invention.

标准正交变换处理的公式如下: Orthonormal transform processing equation is as follows:

其中,Ac.表示经过正交变换处理后待测图像中第〃个像元在第*个波段的反射率值,表示待测图像中第/2个像元在各个波段处反射率的平均值,W表示波段数,m-Ι表示自由度。 Wherein, Ac. After denotes orthogonal transform processing the first test image 〃 th pixel value of reflectivity * bands, showing the first image of the measurement / 2 pixels in each of the average reflectance at a wavelength band , W represents the number of bands, m-Ι represented freedom.

附加散射校正处理的过程包括: Additional scatter correction processing process comprises:

首先,计算平均光谱矢量: First, calculate the average spectral vectors:

然后,对每一个像元光谱进行线性回归: Then, for each pixel spectra linear regression:

再进行附加散射校正: Then additional scatter correction:

上述(4)到(6)式中,F表示平均光谱矢量, To (6) above (4) wherein, F represents the average vector of the spectrum,

表示对所有像元光谱矢量的求和,ρΑ表示第A个像元的光谱矢量,mh、匕分别表示第/2个像元光谱矢量Pa与所有像元平均光谱的线性回归的斜率与截距,VhiMSO表示经过附加散射校正后的像元光谱矢量。 Represents a summation of all pixels of the spectral vector, ρΑ A represents the spectral vectors of picture elements, mh, respectively dagger represents the slope of the linear regression intercept / 2 spectral vectors Pa and the pixels of all pixels averaged spectra , VhiMSO pixel represents elapsed after the additional spectrum vector scatter correction.

分别对目标光谱和待测图像中各像元光谱进行预处理之后,则可得到消除大气散射误差之后较精确的光谱信息。 After each of the target spectrum and the measured spectrum image preprocessing each cell can be obtained after the elimination of atmospheric scattering errors more accurate spectral information.

最后,建立目标光谱和待测图像像元光谱二维矩阵的线性关系,该线性关系表示如下: Finally, create the image as the target spectrum and measured spectrum membered linear two-dimensional matrix, the linear relationship is expressed as follows:

其中,&表示同标光谱向量,表示像元光谱的二维矩阵,Cks表示回归系数向量,值的大小可以反映目标光谱对各像元光谱的贡献程度,值越大,则表明目标光谱对像元光谱的贡献越大,£wxl表示噪声向量,为一维噪声矩阵。 Where & represents the same standard spectral vectors represent two-dimensional matrix of pixels of the spectrum, Cks represent the regression coefficient vector, the size of the target value may reflect the spectrum of the spectrum as the degree of contribution dollars, the greater the value of each, indicates that the spectrum of the target image the larger the yuan spectral contributions, £ WXL represents a noise vector, a one-dimensional noise matrix.

步骤102,获取该逆模型的回归系数向量。 Step 102, obtaining the regression coefficient vector of the inverse model.

由于7?·的秩小于波段数,中各像元光谱存在高度相关性,釆用通常的最小二乘方法对回归系数进行估计时需要对进行求逆,JC1代表t的转置矩阵,而当中的变量高度相关时,行列式几乎接近于零,对RLR_求逆会产生严重的舍入误差,因此釆用通常的最小二乘方法求取该逆模型的回归系数向量,具体包括: Since 7? · Rank is less than the number of bands, each pixel there is a high correlation spectroscopy, it is necessary to perform inverse Bian when estimation of the regression coefficient with a conventional least squares, t JC1 is representative of the transposed matrix, and which when highly correlated variables, the determinant is almost close to zero, inversion of RLR_ have severe rounding error, thus preclude obtaining the regression coefficient vector of the inverse model by a conventional least squares method, comprises:

a、根据目标光谱向量计算初始权重向量: a, initial weight vector is calculated based on the target spectral vectors:

其中,&表示目标光谱向量,\为已知量,51/'代表5;的转置矩阵,表示初始权重向量,K表示像元光谱的二维矩阵。 Wherein, & represents a target spectral vectors, \ known quantity, 51 / 'represents 5; a transposed matrix, represents the initial weight vector, K represents a two-dimensional matrix of pixel spectra.

b、根据初始权重向量计算得分向量: b, The weight vector calculated score vector initial weights:

其中,表示得分向量,i?表示像元光谱的二维矩阵,T?7'代表i?的转置矩阵,w;,表不初始权重向量。 Wherein the score represents the vector, I? Represents a two-dimensional matrix of pixel spectra, T? 7 'representative of I? Transposed matrix, w ;, the table is not the initial weight vector.

C、根据得分向量计算目标光谱的载荷向量: C, is calculated based on the target spectral score vector load vector:

其中,\表示目标光谱向量,表示载荷向量,&表示得分向量。 Wherein, \ represents the target spectral vector, represents the load vector, & represents a score vector.

d、根据得分向量计算像元光谱二维矩阵的载荷向量: d, the score is calculated according to the vector load vector pixel spectral two-dimensional matrix:

其中,A表示像元光谱二维矩阵的载荷向量,i?表示像元光谱的二维矩阵,表示得分向量。 Wherein, A represents the load vector element as spectral two-dimensional matrix, I? Represents a two-dimensional matrix of pixel spectra, showing score vector.

e、计算回归系数向量: e, calculation of the regression coefficient vector:

其中,e = k,,%.·.%}, P = {p”p2〜p„h W = IwllW2...^], (PVr1 代表对(P7I) Wherein, e = k ,,%. ·.%}, P = {p "p2~p" h W = IwllW2 ... ^], (PVr1 representative (P7I)

进行求逆运算。 Perform the inverse operation.

f、计算残差平方和: f, and calculate the residual sum of squares:

其中,姑(〃)表示残差平方和,m表示波段数,&表示原始的像元光谱二维矩阵,若515(/7-I) -怒(《) SG,则表明收敛,取该怒0)对应的Cpi^s为回归系数向量;否则,令άΆ-α,R = Rt„pn ,返回步骤a重复上述操作,直到55(/7-l)-&^)<G,然后获取回归系数向量qs。在实际应用中,G值可根据实步骤103,根据得到的回归系数向量获取各像元的马氏距离。 Wherein Gu (〃) represents residual sum of squares, m represents a number of bands, & spectrum representation of the original two-dimensional matrix of pixels, if the 515 (/ 7-I) - anger ( ") SG, indicates convergence, to take the anger 0) corresponding Cpi ^ s is the regression coefficient vector; otherwise, let άΆ-α, R = Rt "pn, return to step a above operation is repeated until 55 (/ 7-l) - & ^) <G, then obtain the regression QS coefficient vector. in practice, G values, obtaining the Mahalanobis distance of each pixel of the regression coefficient vector obtained in step 103 according to the real.

根据得到的回归系数向量并利用下式计算各像元的马氏距离,公式如下: Vector using the calculated regression coefficients for each pixel according to the Mahalanobis distance obtained using the following formula:

其中,C,,表示回归系数向量中对应第/2个像元的回归系数值,《表示像元总数,表示回归系数的标准差。 Wherein, C ,, represents / 2 regression coefficient values ​​of picture elements corresponding to the regression coefficient vector "indicates total number of cells, represents the difference between the standard regression coefficients. & (Ca)表示第A个像元对应的回归系数的马氏距离。 & (Ca) represents the Mahalanobis distance as the A-th regression coefficient corresponding to the element. 前述⑷可通过下式求得: ⑷ by the following formula:

其中,W(C)表示回归系数的标准差,Ca表示第/Z个像元对应的回归系数值,Z表示所有像元回归系数的平均值 Wherein, W (C) indicates the difference between the standard regression coefficients, Ca represents / Z th image element corresponding to the regression coefficient values, Z represents the average of all pixels of the regression coefficients

步骤104,判定马氏距离大于阀值的回归系数所对应的像元为亚像元目标点。 Step 104, the Mahalanobis distance greater than a threshold determined regression coefficients corresponding to sub-pel pixel is the target point.

计算出各像元回归系数所对应的马氏距离后,检测阀值可根据如下方式来快速获得:即可以认为回归系数小于O的像元必为背景像元,则该些已知的背景像元必存在一个马氏距离最大的像元点,可取该像元所对应的马氏距离为背景与目标的分割阈值,即马氏距离检测阈值。 After calculating the Mahalanobis distance of each pixel corresponding to the regression coefficient, the detection threshold may be quickly obtained according to the following manner: i.e. the regression coefficient of less than O can be considered as the background pixel element will, those known in the background image there must be a Mahalanobis distance membered maximum pixel point, it is desirable that the image segmentation threshold element corresponding to the Mahalanobis distance for the background and object, i.e., the Mahalanobis distance detection threshold.

下面结合具体实施例对上述本发明的高光谱亚像元目标探测方法进一步详细阐述。 In conjunction with the specific embodiments set forth in further detail below hyperspectral Subpixel target detection method of the present invention. 本实例所用的高光谱遥感数据来源于机载成像光谱仪,机载成像光谱仪为釆用推扫成像方式的成像光谱仪,在0.4微米-2.45微米的波长范围获取224个波长处的空间图像信息,波长间隔为10纳米,当飞机在20千米高空飞行时,图像空间分辨率可达20x20米。 Hyperspectral Data in this example was derived from airborne imaging spectrometer, Airborne Imaging Spectrometer to preclude the use of pushbroom imaging spectrometer imaging modality acquires spatial image information 224 at a wavelength in the wavelength range 0.4 microns -2.45 microns wavelength interval of 10 nm, when the aircraft is flying at 20 km altitude, image spatial resolution up to 20x20 meters. 本实施例使用的高光谱图像如图2所示,大小为614 X 512像元,每个像元光谱包括224个波长,波长范围从369.85纳米到2506.81纳米。 Hyperspectral image used in the present embodiment shown in Figure 2, a size of 614 X 512 pixels, each pixel 224 comprises spectrum wavelengths, ranging from 369.85 nm to 2506.81 nm. 具体探测过程如下: Specific detection process is as follows:

Α、取图2中所示白色方框内64x64像元的图像为本实施例的待测图像,除224个波段中的1-6、33、107-114、153-168、222-224波段等35个信噪比较低的坏波段,其佘190波段为本实施例的使用波段,也即波段数m=190。 Α, 2 shown in FIG 64x64 image taken in white pixel image of the measurement block embodiment of the present embodiment, band 224 in addition to the band 1-6,33,107-114,153-168,222-224 35 and other bad low SNR bands, bands of a band 190 which She embodiment of the present embodiment, i.e., band number m = 190. 取图2右下角黑色圆圈内箭头所指的反射能量值较高的屋顶光谱为本实施例的目标光谱,该目标光谱的光谱图如图4所示。 Lower right corner of FIG. 2 taken the black arrow in the circle of the reflected energy value higher spectral target spectra roof embodiment of the present embodiment, the spectrum of the target spectrum as shown in FIG. 提取64x64图像中坐标为(10,32)、(10,42)、(10,52)、(32, 32)、(32,42)、(32, 52)、(42, 32)、(42, 42)、(42,52)的九个像元点,分别混入5%的目标光谱,混入目标光谱后的像元点如图5所示。 64x64 extracted image coordinates (10, 32), (10, 42), (10, 52), (32, 32), (32, 42), (32, 52), (42, 32), (42 , 42), (42, 52) of the nine pixel points were mixed with 5% of the target spectrum, pixel 5 after mixing point target spectrum shown in FIG. 如果将图像中的像元按行进行展开,则(10,32)的像元点对应的光谱矢量即Sp64x_2,其他像元点也同理,在此不再一一描述。 If the pixels in the image are expanded by row, the (10, 32) corresponding to the pixel point i.e. the spectral vector Sp64x_2, other pixel points are also the same reason, which is not described one by one.

需要指出的是,本发明实施例的待测图像像元点是任意选取的,选取像元点的数量也是任意的,此处同时选取九个像元点是为了表明本发明的探测方法可同时对多个像元点进行探测。 It should be noted that the measured image pixel from the embodiment of the present invention is arbitrarily selected, to select the number of pixel dots is arbitrary, while nine pixel select point is to show that the detection method of the present invention herein may simultaneously to detect multiple pixel points. 对前述的九个像元点混入目标光谱后,该九个像元点中即存在了目标光谱的信息,则通过本发明的探测方法即可将存在目标光谱的该九个点都探测出来,这也是本发明实施例的最终目的。 After the nine pixel mixing point target spectrum, the target spectrum of the nine pixel information in that there is a point, then the method of the present invention by detecting the presence of the target spectrum to detect all the nine points out, this is the ultimate object of the present invention embodiments.

B、将所获取的九个像元点的三维高光谱数据表示为二维的高光谱反射率矩阵,表示方法与前述相同,在此不再多述。 B, and the acquired three-dimensional image of nine membered hyperspectral data points represented as a two-dimensional matrix of high spectral reflectance, the same representation, not to repeat here.

C、对目标光谱和像元光谱二维矩阵分别进行预处理,处理方法与前述相同,在此也不再多述。 C, the target spectra, and the spectral element are two-dimensional matrix image preprocessing, the processing method of the same, this is not described more. 然后,建立目标光谱和像元光谱之间的逆模型。 Then, establish goals and spectral inverse model between the yuan spectral image.

D、利用单因变量偏最小二乘方法对上述逆模型的回归系数向量进行求解,在该偏最小二乘方法中设定阀值为10'求取回归系数的三维图如图6所示,图中的X、Y轴分别代表待测图像的横、纵坐标,Z轴代表回归系数的值。 D, using a single dependent variable partial least squares regression coefficient vector of the above-described method of inverse model is solved, the threshold is set to 10 'regression coefficient obtaining three-dimensional view shown in Figure 6 in the partial least squares method, figure X, Y represent the horizontal axis of the image to be measured, and the ordinate, Z axis represents the value of regression coefficient. 从图6中可看出,混有5%目标光谱的像元点与未混入目标光谱的背景像元点的回归系数值差异较大,可见回归系数的大小可以反映目标光谱对像元光谱的贡献程度。 As can be seen from Figure 6, mixed with the background pixel point not mixed regression coefficient values ​​of the target spectral target spectra of 5% points quite different pixel size regression coefficient may reflect the visible spectrum of the target pixel spectrum the degree of contribution.

E、计算各像元点回归系数的马氏距离。 E, Mahalanobis distance of each pixel point of the regression coefficients. 计算结果如图7所示,图中的横轴代表像元数,纵轴代表马氏距离值。 The results shown in Figure 7, the horizontal axis in the figure represents the number of pixels, the vertical axis represents the value of the Mahalanobis distance. 由上述阈值选取方法可得检验阀值为 By the threshold selection method available for the threshold test

—Λ . ^II γ) j _ j — j- -^fL ■. II Cd NiL Jt 人/At ■L- I—t , j—» AiI JLL4 Tl rl"* Ϊ3ΠΓ 4y>7 IT" 、/"L值。因此,检测结果如图8所示,所选九个像元点都被判定为亚像元目标点,检测结果与实际情况相符。 -Λ ^ II γ) j _ j -. J- - ^ fL ■ II Cd NiL Jt person / At ■ L- I-t, j- »AiI JLL4 Tl rl" * Ϊ3ΠΓ 4y> 7 IT ", /". L value. Therefore, as shown in FIG detection result, the selected points are nine pixels 8 pixels determined target point, the detection result is consistent with the fact alkylene.

本发明还提供了一种高光谱亚像元目标探测装置,如图9所示,该装置包括:数据获取单元100、模型建立单元200、回归系数向量获取单元300、马氏距离获取单元400和判定单元500。 The present invention also provides a hyperspectral Subpixel target detection apparatus shown in Figure 9, the apparatus comprising: a data acquisition unit 100, a model establishing unit 200, the regression coefficient vector obtaining unit 300, acquisition unit 400, and Mahalanobis distance determination unit 500. 其中,数据获取单元100,用于获取待测图像像元的三维高光谱数据。 Wherein the data acquisition unit 100 for acquiring a three-dimensional hyper-spectral test image pixel data. 模型建立单元200,连接数据获取单元100,用于根据数据获取单元100获取的像元高光谱数据建立目标光谱和待测图像像元光谱的逆模型。 Model establishing unit 200, connected to the data acquisition unit 100 for acquiring pixel spectrum inverse model unit 100 acquires the measured spectra and the establishment of a target image pixel data based on hyperspectral data. 回归系数向量获取单元300,连接模型建立单元200,用于获取逆模型的回归系数向量。 Regression coefficient vector obtaining unit 300, model connection unit 200, configured to obtain the regression coefficient vector inverse model. 马氏距离获取单元400,连接回归系数向量获取单元300,用于根据得到的回归系数向量获取各像元的马氏距离。 Mahalanobis distance obtaining unit 400, connected to the regression coefficient vector obtaining unit 300 for obtaining each of the Mahalanobis distance based on picture elements obtained regression coefficient vector. 判定单元500,连接马氏距离获取单元400,用于判定马氏距离大于阀值的回归系数所对应的像元为亚像元目标点。 Determination unit 500, acquisition unit 400 is connected to the Mahalanobis distance, the Mahalanobis distance for determining the regression coefficient greater than a threshold corresponding to the pixel of the target point subpixel.

其中,模型建立单元200还包括:矩阵生成子单元210、预处理子单元220和线性关系建立子单元230。 Wherein the model establishing unit 200 further includes: the sub-matrix generating unit 210, pre-processing sub-unit 220 and the linear relationship establishing subunit 230. 矩阵生成子单元210,用于将待测图像像元的三维高光谱数据表示为高光谱反射率的二维矩阵。 Matrix generation subunit 210, configured to test the three-dimensional image data of hyperspectral image element is represented as a two-dimensional matrix of high spectral reflectance. 预处理子单元220,连接矩阵生成子单元2]0,用于对目标光谱和矩阵生成子单元210生成的待测图像像元光谱二维矩阵进行预处理并提供给线性关系建立子单元230。 Pre-processing sub-unit 220, the connection matrix generating unit 2] 0, for an image object to be measured and the spectral matrix generation subunit 210 generates a preprocessed image element two-dimensional matrix and spectra to the linear relationship establishing subunit 230. 线性关系建立子单元230,连接预处理子单元220,用于建立目标光谱和待测像元光谱二维矩阵的线性关系。 Linear relationship establishing subunit 230, pre-processing sub-unit 220 is connected, for establishing a linear relationship between a target spectrum and the measured spectrum of a two-dimensional pixel matrix.

综上所述,本发明一种高光谱亚像元目标探测的方法及装置,所用到的逆模型不需要背景端元光谱的先验信息,仅需要先验的目标光谱信息,因此对背景光谱的复杂度不敏感;探测准确度仅仅依赖于目标光谱的精确程度,从而目标探测准确度高;又由于偏最小二乘中的迭代计算方法仅需要少量矩阵求逆运算,且在检测部分釆用了速度较快的马氏奇异值检测方法,使得本发明的探测方法具有较快的运算速度。 In summary, the present invention hyperspectral imaging method and apparatus membered alkylene target detection, the inverse model does not need to use the background spectrum prior information element end, only certain spectral priori information, the background spectrum complexity insensitive; detection accuracy depends only on the accuracy of the target spectrum, so that the target detection accuracy is high; and because the partial least squares iterative calculation only a small amount of matrix inversion, and preclude the use of the detection section the Mahalanobis singular value faster detection methods, such detection method of the present invention has a faster processing speed.

以上所述,仅为本发明的较佳实施例,并非用于限定本发明的保护范围。 The above are only preferred embodiments of the present invention is not intended to limit the scope of the present invention.

Claims (10)

  1. 1、一种高光谱亚像元目标探测方法,其特征在于,包括以下步骤: 建立目标光谱和待测图像像元光谱二维矩阵的逆模型; 获取所述逆模型的回归系数向量; 根据所述回归系数向量获取各像元的马氏距离; 判定马氏距离大于阀值的回归系数所对应的像元为亚像元目标点。 1, hyperspectral subpixel target detection method comprising the steps of: establishing a target spectrum and the measured spectrum of pixel image model inverse two-dimensional matrix; obtaining the regression coefficient vector of the inverse model; in accordance with the said regression coefficient vector obtaining the Mahalanobis distance of each of pixels; Mahalanobis distance greater than a threshold determined regression coefficients corresponding to the pixel of the target point subpixel.
  2. 2、如权利要求I所述高光谱亚像元目标探测方法,其特征在于,所述建立目标光谱和待测图像像元光谱二维矩阵的逆模型,具体包括: 将获取的待测图像像元的三维高光谱数据表示为高光谱反射率的二维矩阵: 2, as claimed in claim I Hyperspectral Subpixel Target detection method, wherein said inverse model to establish a target spectrum and measured spectrum of a two-dimensional pixel image matrix, comprises: an image test image acquired element dimensional hyperspectral data represented as two-dimensional matrix of high spectral reflectance:
    or
    其中,尺>><„ 表示像元光谱的二维矩阵,[P1, P2.. -Vy^J · ·-Vxxy ]和[P1, P2.. .P— · · -Pxxy ]表示像元的光谱矢量,W表示波段数,^表示图像像元总数,X表示图像像元行数,少表不图像像兀列数,n = xxy; 建立所述目标光谱和所述待测图像像元光谱二维矩阵的线性关系: St=R— xcPLS+Ε—, 其中,St表示目标光谱向量,表示像元光谱的二维矩阵,^8表示回归系数,£^表示噪声向量。 Wherein the foot >> < "indicates two-dimensional matrix of pixel spectra, [P1, P2 .. -Vy ^ J · · -Vxxy] and [P1, P2 .. .P- · · -Pxxy] represents picture elements spectrum vector, W represents the number of bands, ^ represents the image total number of cells, X represents the number of pixels the image lines, the image is not a small number of tables like Wu columns, n = xxy; establishing a target pixel spectrum and the spectrum of the test image linear two-dimensional matrix: St = R- xcPLS + Ε-, where, St represents the target spectral vector, expressed as a two-dimensional matrix elements spectrum, ^ 8 represents a regression coefficient, £ ^ denotes the noise vector.
  3. 3、如权利要求2所述高光谱亚像元目标探测方法,其特征在于,所述将三维高光谱数据表示为二维矩阵和建立线性关系之间,还包括:对所述目标光谱和所述待测图像像元光谱二维矩阵进行预处理。 3, as claimed in Hyperspectral Subpixel Target detection method according to claim 2, wherein said data representing the three-dimensional hyper-spectral two-dimensional matrix, and between the linear relation is established, further comprising: the target spectra, and the said measured spectra dimensional matrix pixel image preprocessing.
  4. 4、如权利要求3所述高光谱亚像元目标探测方法,其特征在于,所述预处理包括标准正交变换处理或附加散射校正处理。 4, as claimed in Hyperspectral Subpixel Target detection method according to claim 3, wherein said orthonormal transform processing including pretreatment or additional scatter correction process.
  5. 5、如权利要求4所述高光谱亚像元目标探测方法,其特征在于,所述标准正交变换处理具体包括: 5. Hyperspectral Subpixel Target detection method of claim 4, wherein said orthonormal transform processing comprises:
    其中,Phk„w,,表示经过正交变换处理后待测图像中第&个像元在第*个波段的反射率值,只表示待测图像中第A个像元在各个波段处反射率的平均值,m表示波段数,m-1表示自由度; 所述附加散射校正处理具体包括: 计算平均光谱矢量: Wherein, Phk "w ,, represents the orthogonal transform processing after the first test image was like Element & reflectance of bands * value, only the first test image represents one pixel A reflectivity at each band the average, m represents a number of bands, m-1 represents a degree of freedom; said additional scatter correction process comprises: calculating an average vector of the spectrum:
    对每一个像元光谱进行线性回归: Spectrum for each pixel by linear regression:
    进行附加散射校正: Additional scatter correction:
    其中,^表示平均光谱矢量,表示对所有像元光谱矢量的求和,Pa表示第个像元的光谱矢量,rnh、&分别表示第/ζ个像元光谱矢量P,与所有像元平均光谱的线性回归的斜率与截距,pMMS.n表示经过附加散射校正后的像元光谱矢量。 Wherein ^ represents the average spectral vector represents the summation of all pixels of the spectral vector, Pa represents the spectral vectors of picture elements, rnh, & respectively denote / ζ th pixel vector P spectra, averaged spectra from all pixels linear regression slope and intercept, pMMS.n represented after additional spectral vector scatter correction pixels.
  6. 6、如权利要求I所述高光谱亚像元目标探测方法,其特征在于,通过偏最小二乘迭代方法获取所述逆模型的回归系数向量,具体包括: a、根据目标光谱向量获取初始权重向量 6, I as claimed in Hyperspectral Subpixel Target detection method according to claim, wherein obtaining the regression coefficient vector of the inverse model by an iterative partial least squares method, comprises: a, obtaining an initial weight vector from the target spectrum vector
    ',其中,5;表示目标光谱向量,表示初始权重向量,i?表示像元光谱的二维矩阵; b、根据所述初始权重向量计算得分向量:L=R1Wn,其中,表示得分向量,A表示像元光谱的二维矩阵,w;,表示初始权重向量; C、根据所述得分向量计算所述目标光谱的载荷向量:其中,&表示所述目标光谱的载荷向量,&表示目标光谱向量,^表示得分向量; d、根据所述得分向量计算所述像元光谱二维矩阵的载荷向量:Pn = Rtll,其由„类元靳祙伤读一維妬陡的栽益向I. 类元德谱的一一維祐眭.,表示得分向量; e、获取所述回归系数向量: ', Wherein 5; represents a target spectrum vector represents the initial weight vector, i denotes a two-dimensional matrix of pixels of the spectrum; B, calculates a score vector from the weight vector of the initial weight:? L = R1Wn, which represents the score vector, A like a two-dimensional matrix elements represents the spectrum, w ;, represents the initial weight vector; C, calculating the load vector of the target spectrum based on the score vector: wherein & vector representing the load of the target spectrum, the target spectral vector represents & , ^ represents a score vector; D, is calculated according to the score vector of the image vector element load spectrum two-dimensional matrix: Pn = Rtll, which is read by a steep dimensional jealous "classifier Jin Sork beneficial to plant injury class I. Pedigrees of eleven membered Weiyou Sui, represents score vector; E, obtaining the regression coefficient vector:
    ,其中, ,among them,
    f、计算残差平方和 f, and calculates a residual square
    其中,SS⑷表示残差平方和, m表示波段数,A表示原始的像元光谱二维矩阵,&表示目标光谱向量,设定阀值为G,若 Wherein, SS⑷ residual sum of squares represents, m represents a number of bands, A spectral representation of the original two-dimensional matrix of pixels, & represents a target spectral vectors, the threshold is set to G, if
    SG,则取该货⑷对应的c%s为回归系数向量;否则,^Sl=StH R = R-tnPn,返回步骤a重复上述操作,直到姑(η-l)-双(《)<G,然后获取回归系数向量qs。 SG, then take the goods ⑷ corresponding to c% s regression coefficient vector; otherwise, ^ Sl = StH R = R-tnPn, returns to step a above operations are repeated until Gu (η-l) - bis ( ") <G , then get the regression coefficient vector qs.
  7. 7、一种高光谱亚像元目标探测装置,其特征在于,包括: 模型建立单元,用于建立目标光谱和待测图像像元光谱二维矩阵的逆模型; 回归系数向量获取单元,用于获取所述逆模型的回归系数向量; 马氏距离获取单元,用于根据所述回归系数向量获取各像元的马氏距离;判定单元,用于判定马氏距离大于阀值的回归系数所对应的像元为亚像元目标点》 7. A hyperspectral Subpixel Target detection device, characterized by comprising: model means for establishing an inverse model of the target spectrum and measured spectrum of a two-dimensional pixel image matrix; regression coefficient vector obtaining means for obtaining the regression coefficient vector of the inverse model; Mahalanobis distance obtaining means for obtaining respective pixel Mahalanobis distance based on the regression coefficient vector; determining means for determining the regression coefficients of the Mahalanobis distance is greater than the threshold value corresponding to the sub-pixel to pixel target point. "
  8. 8、如权利要求7所述高光谱亚像元目标探测装置,其特征在于,所述模型建立单元包括: 矩阵生成子单元,用于将待测图像像元的三维高光谱数据表示为高光谱反射率的二维矩阵; 线性关系建立子单元,用于建立所述目标光谱和所述待测图像像元光谱二维矩阵的线性关系。 8, according to claim 7 Hyperspectral Subpixel Target detection device, characterized in that the model unit comprises: a matrix generating sub-unit, three-dimensional image for the test image data element is represented hyperspectral hyperspectral a two-dimensional matrix of reflectivity; linear relationship establishing subunit, for establishing a linear relationship between the target spectrum and the spectral image of the test image element two-dimensional matrix.
  9. 9、如权利要求8所述高光谱亚像元目标探测装置,其特征在于,所述模型建立单元还包括:预处理子单元,用于对所述目标光谱和所述矩阵生成子单元生成的待测图像像元光谱二维矩阵进行预处理并提供给线性关系建立子单元。 9, according to claim 8 Hyperspectral Subpixel Target detection device, characterized in that the model unit further comprises: pre-processing sub-unit, for the target spectra, and the sub-matrix generating unit generates the spectra measured image pixel dimensional matrix pretreated and supplied to the linear relationship establishing subunit.
  10. 10、如权利要求7至9中任一项所述高光谱亚像元目标探测装置,其特征-a- JX- 壮苗杜.ίό 田工抑杜卻丨丨阳成;沾二祐吉;数据并提供给模型建立单元。 10, as a Hyperspectral Subpixel Target detection device according to any of claims 7-9, characterized in -a- JX- seedlings Du Du .ίό field work was suppressed to the male Shushu; James two Sukeyoshi; data and provided to the model unit.
CN 200710176782 2007-11-02 2007-11-02 High spectrum sub-pixel target detection method and device CN100483147C (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 200710176782 CN100483147C (en) 2007-11-02 2007-11-02 High spectrum sub-pixel target detection method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 200710176782 CN100483147C (en) 2007-11-02 2007-11-02 High spectrum sub-pixel target detection method and device

Publications (2)

Publication Number Publication Date
CN101144861A true CN101144861A (en) 2008-03-19
CN100483147C true CN100483147C (en) 2009-04-29

Family

ID=39207485

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 200710176782 CN100483147C (en) 2007-11-02 2007-11-02 High spectrum sub-pixel target detection method and device

Country Status (1)

Country Link
CN (1) CN100483147C (en)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN100573193C (en) 2008-04-28 2009-12-23 北京航空航天大学 High light spectrum small target detection method and apparatus
CN101504315B (en) 2009-02-23 2011-01-05 北京航空航天大学 Expansion morphology and orthogonal subspace projection combined end member automatic extraction method
CN101702021B (en) 2009-11-06 2011-10-26 华中科技大学 Method for obtaining spectrum-integrated time-varying target optical spectrum information and device thereof
CN101893552B (en) * 2010-07-06 2012-06-27 西安电子科技大学 Hyperspectral imager and imaging method based on compressive sensing
CN102331402B (en) * 2011-06-01 2013-05-01 哈尔滨工业大学 Chemical gas detection and identification method by hyperspectral imaging
CN102289673B (en) * 2011-06-22 2013-04-10 复旦大学 Method for selecting hyperspectral remote sensing image bands based on partial least squares
CN102540271B (en) * 2011-12-27 2014-03-19 南京理工大学 Semi-supervised hyperspectral sub-pixel target detection method based on enhanced constraint sparse regression method
CN102799896A (en) * 2012-06-29 2012-11-28 中山大学 POLSAR image unsupervised classification method based on target scattering identification
CN103150482B (en) * 2013-03-28 2016-03-16 湖南中烟工业有限责任公司 Drying pls determined based on the process parameters of the method Factors
CN103955711B (en) * 2014-05-20 2017-07-07 北京航空航天大学 A method of forming a pattern recognition analysis to identify the target spectrum
CN104463897B (en) * 2014-12-30 2017-11-10 电子科技大学 A high-spectral target detection method

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5627635A (en) * 1994-02-08 1997-05-06 Newnes Machine Ltd. Method and apparatus for optimizing sub-pixel resolution in a triangulation based distance measuring device
CN1696736A (en) * 2005-05-31 2005-11-16 中国科学院上海技术物理研究所 Vector filtering method of high spectrogram based on detection of neighborhood background
CN1710445A (en) * 2005-07-05 2005-12-21 华东师范大学 Aeronautical high-spectrum remote-sensing image spectrum zonc noise self-detecting and removing method
CN1900740A (en) * 2005-11-18 2007-01-24 北京航空航天大学 High spectrum full polarization imaging remote sensing system
CN101030299A (en) * 2007-03-29 2007-09-05 复旦大学 Method for decomposing remote-sensing-mixed image element based on data space orthogonality

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5627635A (en) * 1994-02-08 1997-05-06 Newnes Machine Ltd. Method and apparatus for optimizing sub-pixel resolution in a triangulation based distance measuring device
CN1696736A (en) * 2005-05-31 2005-11-16 中国科学院上海技术物理研究所 Vector filtering method of high spectrogram based on detection of neighborhood background
CN1710445A (en) * 2005-07-05 2005-12-21 华东师范大学 Aeronautical high-spectrum remote-sensing image spectrum zonc noise self-detecting and removing method
CN1900740A (en) * 2005-11-18 2007-01-24 北京航空航天大学 High spectrum full polarization imaging remote sensing system
CN101030299A (en) * 2007-03-29 2007-09-05 复旦大学 Method for decomposing remote-sensing-mixed image element based on data space orthogonality

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
High spectrum purity signal source applied in X-band radarfrequency criterion generator. Yang,Yumei.2001 CIE International Cenfevence on Radar Proceedings . 2001 *
地表参量遥感反演理论与方法研究. 唐世浩.万方数据资源系统--中国学位论文全文库. 2001 *
高光谱与多角度数据联合进行混合像元分解研究. 唐世浩等.遥感学报,第7卷第3期. 2003 *
高光谱遥感数据光谱曲线分形特征研究. 周子勇.全国国土资源与环境遥感技术应用交流会论文文集. 2004 *

Also Published As

Publication number Publication date Type
CN101144861A (en) 2008-03-19 application

Similar Documents

Publication Publication Date Title
Keshava A survey of spectral unmixing algorithms
Brando et al. Satellite hyperspectral remote sensing for estimating estuarine and coastal water quality
Dobson et al. Knowledge-based land-cover classification using ERS-1/JERS-1 SAR composites
Foody et al. An evaluation of some factors affecting the accuracy of classification by an artificial neural network
Liang et al. Atmospheric correction of Landsat ETM+ land surface imagery. II. Validation and applications
Baret et al. Estimating canopy characteristics from remote sensing observations: Review of methods and associated problems
Solberg et al. Texture fusion and feature selection applied to SAR imagery
Weiss et al. Evaluation of canopy biophysical variable retrieval performances from the accumulation of large swath satellite data
Hyyppä et al. Accuracy comparison of various remote sensing data sources in the retrieval of forest stand attributes
Schroeder et al. Atmospheric correction algorithm for MERIS above case‐2 waters
Farrah et al. HST/WFPC2 imaging of the QDOT ultraluminous infrared galaxy sample
North et al. Retrieval of land surface bidirectional reflectance and aerosol opacity from ATSR-2 multiangle imagery
Legleiter et al. Passive optical remote sensing of river channel morphology and in-stream habitat: Physical basis and feasibility
US6208752B1 (en) System for eliminating or reducing exemplar effects in multispectral or hyperspectral sensors
Fang et al. Retrieving leaf area index with a neural network method: Simulation and validation
Gal et al. The northern sky optical cluster survey. II. An objective cluster catalog for 5800 square degrees
Townsend Relationships between forest structure and the detection of flood inundation in forested wetlands using C-band SAR
Broadhurst et al. Mapping cluster mass distributions via gravitational lensing of background galaxies
Borak Feature selection and land cover classification of a MODIS-like data set for a semiarid environment
Guillaso et al. Building characterization using L-band polarimetric interferometric SAR data
Loisel et al. Estimation of the inherent optical properties of natural waters from the irradiance attenuation coefficient and reflectance in the presence of Raman scattering
Bréon et al. A comparison of cloud droplet radii measured from space
Ientilucci et al. Advances in wide-area hyperspectral image simulation
Abuelgasim et al. Evaluation of national and global LAI products derived from optical remote sensing instruments over Canada
Iordache et al. MUSIC-CSR: Hyperspectral unmixing via multiple signal classification and collaborative sparse regression

Legal Events

Date Code Title Description
C06 Publication
C10 Entry into substantive examination
C14 Grant of patent or utility model
CF01