CN101806898A - Hyperspectral remote sensing image target detecting method based on variable end members - Google Patents
Hyperspectral remote sensing image target detecting method based on variable end members Download PDFInfo
- Publication number
- CN101806898A CN101806898A CN 201010130276 CN201010130276A CN101806898A CN 101806898 A CN101806898 A CN 101806898A CN 201010130276 CN201010130276 CN 201010130276 CN 201010130276 A CN201010130276 A CN 201010130276A CN 101806898 A CN101806898 A CN 101806898A
- Authority
- CN
- China
- Prior art keywords
- pixel
- remote sensing
- end member
- sensing image
- detected
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 230000003595 spectral effect Effects 0.000 claims abstract description 53
- 238000001514 detection method Methods 0.000 claims abstract description 42
- 239000000523 sample Substances 0.000 claims description 9
- 238000001228 spectrum Methods 0.000 abstract description 24
- 238000005516 engineering process Methods 0.000 abstract description 7
- 238000000354 decomposition reaction Methods 0.000 abstract description 6
- 239000011159 matrix material Substances 0.000 description 15
- VMXUWOKSQNHOCA-UKTHLTGXSA-N ranitidine Chemical compound [O-][N+](=O)\C=C(/NC)NCCSCC1=CC=C(CN(C)C)O1 VMXUWOKSQNHOCA-UKTHLTGXSA-N 0.000 description 11
- 239000000203 mixture Substances 0.000 description 9
- 238000012360 testing method Methods 0.000 description 5
- 238000004611 spectroscopical analysis Methods 0.000 description 4
- 230000000007 visual effect Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 239000000463 material Substances 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 239000004744 fabric Substances 0.000 description 2
- 230000008676 import Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000003657 Likelihood-ratio test Methods 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000012467 final product Substances 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 230000000631 nonopiate Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Images
Landscapes
- Image Processing (AREA)
Abstract
本发明公开了基于端元可变的高光谱遥感影像目标探测方法，包括：选择所需进行目标探测的待探测遥感影像；获取探测时需要的先验信息，所述先验信息包括目标端元光谱信息和背景端元光谱信息；利用交叉相关匹配技术遍历待探测遥感影像，确定待探测遥感影像中每个像元内的背景端元种类；对待探测遥感影像进行全限制性最小二乘的光谱分解，得到待探测遥感影像中每个像元内目标端元和各种背景端元的组分信息；建立基于广义似然比的探测器；采用探测器遍历待探测遥感影像，得到待探测遥感影像中每个像元的探测函数值判定待探测遥感影像中每个像元内是否含有目标。本发明的方法具有结构化强、适应度高、自组织、自学习的特点。
The invention discloses a hyperspectral remote sensing image target detection method based on variable endmembers, which includes: selecting a remote sensing image to be detected for target detection; Spectral information and background endmember spectral information; use cross-correlation matching technology to traverse the remote sensing image to be detected, and determine the background endmember type in each pixel in the remote sensing image to be detected; perform full restricted least squares spectrum of the remote sensing image to be detected Decomposition, to obtain the component information of the target endmember and various background endmembers in each pixel in the remote sensing image to be detected; establish a detector based on the generalized likelihood ratio; use the detector to traverse the remote sensing image to be detected, and obtain the remote sensing image to be detected The detection function value of each pixel in the image determines whether there is a target in each pixel in the remote sensing image to be detected. The method of the invention has the characteristics of strong structure, high adaptability, self-organization and self-learning.
Description
Technical field
The invention belongs to the remote sensing image processing technology field, especially a kind of target in hyperspectral remotely sensed image object detection method.
Background technology
In the target in hyperspectral remotely sensed image, atural object distribution situation more complicated, the extraction of interesting targets such as culture, moving vehicle is a difficult point problem wherein.Since the restriction of spatial resolution, in the target in hyperspectral remotely sensed image, mixed pixel phenomenon ubiquity.The mixed pixel phenomenon is meant, the pixel on the image is not that the reflected signal by single classification atural object correspondence constitutes, but the unlike signal acting in conjunction of plurality of classes atural object and constituting.So the spectrum that mixed pixel reflects on target in hyperspectral remotely sensed image is formed by multiple spectrum mixed together.In this case, the object detection method based on spectral signature is difficult to detect interesting target.
Target in hyperspectral remotely sensed image can reflect the SPECTRAL DIVERSITY of different atural objects meticulously.In feature space, unusual target has the distribution characteristics different with background atural object, and promptly target and background lay respectively at the zones of different of feature space.The target detection problem of target in hyperspectral remotely sensed image is a classification problem in essence.(D.Manolakis and G.Shaw, " Detection Algorithms for Hyperspectral ImagingApplications; " IEEE Signal Processing Magazine, vol.19, no.1, pp.29-43, Jan.2002.) still, but different with the general category problem is, the target in hyperspectral remotely sensed image target detection has a distinguishing feature: promptly the number of target is fewer, proportion shared in whole image is very low, and most pixels all are regarded as non-target or background in the image.In this case, Chang Gui sorting technique is difficult to set up by the less pixel of number.In addition, be the criteria for classification of the best with minimum mistake branch rate in the general classification method, this can cause mistake that pixel in the image all is divided into background or all is divided into target.Therefore, need the method for development at the target detection practical application, the target detection problem is considered as in each pixel of high-spectrum remote-sensing seeking the Existence problems of certain atural object or material, i.e. a binary hypothesis test problem that will pixel be investigated be judged to be target or non-target.(C.I.Chang, S.S.Chiang.Anomaly detectionand classification for hyperspectral imagery[J] .IEEE Trans.Geosci.Remote Sensing, 2002,40 (6): 1314-1325.) development is necessary at the target detection technology of abnormal target in hyperspectral remotely sensed image target characteristic.
The target detection of target in hyperspectral remotely sensed image, the main difficult point problem that exists is the mixed pixel problem.The mixed pixel problem is meant that because the restriction of spatial resolution, the pixel in the target in hyperspectral remotely sensed image is formed by multiple atural object mixed together, and therefore target to be detected will be present in the inferior pixel.This gives based on spectral signature coupling etc. and has brought challenge based on the detection method of pure spectra.In addition, owing to the influence of factors such as propagation in atmosphere, sensor noise, electromagnetic wave repeatedly reflect, the spectrum with a kind of material in mixed pixel may present certain difference.This carries out target detection for the priori spectral information that utilizes target and has brought difficulty.For the sub-pixel target detection problem in the mixed pixel, existing method mainly is based on the detection method of linear mixed model, be about to pixel and be considered as that spectrum by multiple pure atural object correspondence mixes, the target pixel is linear synthetic according to the component in pixel by target optical spectrum and diversity of settings spectrum, and the target detection problem is converted into the problem of seeking each pixel internal object spectrum existence on the image.For example shown in Figure 1, four pixels in the width of cloth raw video, by two kinds of pure atural objects for spectrum mix, therefore can be divided into two width of cloth component images, linear synthetic coefficient is respectively 1.0,0.0,1.0,0.5 and 0.0,1.0,0.0,0.0.In fact spectrum mixing situation may be more complicated.At the target optical spectrum variation issue, main method has based on method of subspace model etc., and pixel is considered as being made of target subspace and the linear combination of background subspace, and its basis remains linear mixed model.But, in existing all methods, all be to use identical end member number and kind whether to exist target to survey to each pixel.In practice, the end member number that comprises in each pixel is normally different with kind, thereby is inaccurate with the structure that all end members carry out detector, therefore finally may cause the separability decline of target and background.Therefore be necessary to come the structural exploration device, but and then reflect the calibration of target and background more exactly at the actual composition situation of mixed pixel in the image.
Summary of the invention
The objective of the invention is, on the basis that end member is selected, provide a kind of object detection method that is used for target in hyperspectral remotely sensed image.
For achieving the above object, target in hyperspectral remotely sensed image object detection method provided by the invention may further comprise the steps:
Step 1 is selected the required remote sensing image to be detected that carries out target detection;
Step 2 is obtained the prior imformation that needs when surveying, and described prior imformation comprises target end member spectral information and background end member spectral information;
Step 3 according to step 2 gained prior imformation, utilizes the crosscorrelation matching technique to travel through remote sensing image to be detected, determines the background end member kind in each pixel in the remote sensing image to be detected;
Step 4, utilize the background end member kind in step 2 gained prior imformation and each pixel of step 3 gained, remote sensing image to be detected is carried out the spectral resolution of complete restricted least square, obtain the component information of each pixel internal object end member and diversity of settings end member in the remote sensing image to be detected;
Step 5 obtains the linear mixed model of pixel according to step 4 obtained component information and step 2 gained prior imformation, according to the detector of this modelling based on Generalized Likelihood Ratio;
Step 6 adopts step 5 gained detector to travel through remote sensing image to be detected, obtains the probe function value of each pixel in the remote sensing image to be detected; According to gained probe function value and preset threshold value, judge whether contain target in the remote sensing image to be detected in each pixel.
And the specific implementation of step 3 adopts following computing formula:
In the formula, R
rBe background end member spectral information, R
tBe the spectral information of certain pixel in the remote sensing image to be detected, establishing has n kind background end member spectral information in the prior imformation, establish R
rFor wherein a kind of, calculate gained r and represent whether contain this kind background end member in this pixel.
The present invention utilizes crosscorrelation Spectral matching technology to try to achieve the contained background end member of each pixel classification information, result according to end member classification information and complete restricted decomposition constructs Adaptive matching subspace detection operator then, utilize end member classification information Dynamic Selection end member in detection, reduce of the influence of end member number estimated bias, improve the separability of detector target and background to result of detection.Method of the present invention has that structuring is strong, fitness is high, the characteristics of self-organization, self study, during enforcement computing very fast, be fit to the pixel design feature of target in hyperspectral remotely sensed image, be applicable to the target detection of target in hyperspectral remotely sensed image.
Description of drawings
Fig. 1 spectral resolution synoptic diagram;
The spectral resolution process flow diagram of the complete restricted least square of Fig. 2 embodiment of the invention;
The principal function flow chart of Fig. 3 embodiment of the invention.
Embodiment
The statistical nature of background is a key issue in the target detection.The target detection problem can be considered in each pixel of high-spectrum remote-sensing the Existence problems of seeking certain atural object or material, i.e. a binary hypothesis test problem that will pixel be investigated be judged to be target or non-target.Generally target is existed with not existing to be considered as the hypothesis that covariance is identical, average is different, so will utilize under the non-existent hypothesis of target, find the solution common statistical nature under two kinds of hypothesis.In the present invention, utilize the subspace model to eliminate the influence of target optical spectrum variation to target detection.In this model, background end member spectrum and its component information during the background statistical information of our employing utilize the detector of these information foundation based on generalized likelihood-ratio test.Therefore, the accuracy of background end member kind and its component directly influences the result that ideal is surveyed.In addition, in detector, the component information of end member is that the spectral resolution by complete restricted least square obtains, and the spectral resolution of complete restricted least square, general all spectral informations that also adopt carry out.Under the interference of factors such as noise, propagation in atmosphere, the end member that does not contain in the pixel can't be zero through its result of spectral resolution.Therefore also need before spectral resolution, determine the end member kind in the mixed pixel.
The present invention has taken into full account the influence of background structure information to target detection.Understanding to background structure information is accurate more, then to the description of background more near truth, but and then can reflect the calibration of target and background more accurately.Background end member kind is more accurate, then can utilize the spectral resolution of complete restricted least square to obtain comparatively accurate component information; Secondly, utilize correct background end member type and end-member composition structuring to express background information, then can utilize detector that echo signal is separated better.
Different with the object detection method of existing subspace model, the present invention utilizes the component with physical significance after the spectral resolution to carry out the target detector structure of maximum likelihood ratio test.What spectral resolution was used is the method for complete restricted least square.Complete restricted referring to wherein, the non-negative and component of the component after the decomposition and be.These two constraint conditions guarantee that the linear mixed model hypothesis to image is rational.Because target pixel number is less, therefore a reliable target detection algorithm generally utilizes target existence and target not to have two kinds of assumed statistical inspections development.The present invention adopts the subspace model to express this assumed statistical inspection.Under the non-existent situation of target, meet following normal distribution: average is that background end member matrix multiply by corresponding component matrix, and variance is that average is zero Gaussian noise; Under the situation that target exists, meet following normal distribution: average is that target end member, background end member multiply by corresponding component matrix, and it is zero Gaussian noise that variance is similarly average.Under this assumption, can set up two kinds of likelihood ratio operators under the hypothesis.Different with general likelihood ratio operator is, parameter wherein is not to obtain by maximal possibility estimation, but the component information with physical significance that utilizes spectral resolution to obtain.
Based on the spectral resolution of end member kind selection strategy and complete restricted least square, the invention provides the unusual object detection method of remote sensing image:
Step 1 is selected the required remote sensing image that carries out target detection, and the present invention is called remote control image to be detected.Embodiment adopts the remote sensing image handling procedure to implement the inventive method.The remote sensing image handling procedure can adopt software programming technique to provide by those skilled in the art according to technical scheme of the present invention, and visual c++6.0 exploitation is adopted in suggestion.During concrete enforcement, can be set in after the remote sensing image handling procedure ejects the image parameters dialog box, by input image width, highly, wave band number and data type select to open the required remote sensing image to be detected that carries out target detection of input.
Step 2, according to the actual demand of surveying, the input prior imformation.Prior imformation comprises target end member spectral information and background end member spectral information.The mode of the training field prior imformation of obtaining before can selecting to survey according to the actual demand of surveying obtains by the method for manually choosing or extracting automatically.If known some pixel contains target, some pixel is the background pixel that does not contain target, then can directly import target pixel to be detected and background pixel.Perhaps when surveying, unknown which be target, which is the background pixel, but can be by visual selection image top pixel as target, selection part pixel is a background, imports detector then.If the pure spectra of known target can be with its target prior imformation as input; If the part target pixel in the known image then can be averaged them, as the target priori spectral information of input.During concrete enforcement, if visual apparent in view, when actual ground cover type data are perhaps arranged, the method that suitable employing is manually chosen.If visual and not obvious, can adopt prior aries such as automatic end member extraction method such as adaptive iteration error analysis, projection pursuit.The mode that general employing is manually chosen obtains background end member spectral information and gets final product.
Step 3 according to step 2 gained prior imformation, utilizes the crosscorrelation matching technique to travel through remote sensing image to be detected, determines the background end member kind in each pixel in the remote sensing image to be detected.Crosscorrelation Spectral matching technology is that a kind of being used for carried out the existing method that spectral similarity is judged, judges the background end member kind that each pixel contains with it among the present invention, and matching process does not repeat them here.Wherein judge the specific implementation that whether contains certain background end member in certain pixel, then can adopt following computing formula:
In the formula, R
rBe background end member spectral information, R
tBe the spectral information of certain pixel in the remote sensing image to be detected, establishing has n kind background end member spectral information in the prior imformation, establish R
rFor wherein a kind of, calculate gained r and represent whether contain this kind background end member in this pixel.
Step 4, utilize the background end member kind in step 2 gained prior imformation and each pixel of step 3 gained, remote sensing image to be detected is carried out the spectral resolution of complete restricted least square, obtain the component information of each pixel internal object end member and diversity of settings end member in the remote sensing image to be detected.Particularly, the decomposition method of the complete restricted least square of Cai Yonging is existing numerical computation method based on active set here, and this method is to handle a kind of effective ways of optimization problem.
Step 5 obtains the linear mixed model of pixel according to step 4 obtained component information and step 2 gained prior imformation, according to the detector of this modelling based on Generalized Likelihood Ratio.During concrete enforcement, linear mixed model based on mixed pixel, adopt the detector of the method foundation of existing test of hypothesis based on Generalized Likelihood Ratio, the method of wherein utilizing Generalized Likelihood Ratio to detect is set up target respectively and is existed and the likelihood operator that does not exist under the situation, and component and end member information are obtained by the step of front.
Step 6 adopts step 5 gained abnormality detection operator to travel through remote sensing image to be detected, obtains the probe function value of each pixel in the remote sensing image to be detected; According to gained probe function value and preset threshold value, judge whether contain target in the remote sensing image to be detected in each pixel.During concrete enforcement,, when the whole image of traversal, press respectively and select corresponding end member number and kind, and then use detector to survey according to step 4 obtained component information.Preset threshold value is a relatively more subjective index, and more generally way is that a series of threshold value is set, and obtains corresponding detectivity and false alarm rate, selects threshold value according to the needs of false alarm rate.According to threshold value, whether reach threshold value according to the probe function value of each pixel in the remote sensing image to be detected, promptly whether contain target in each pixel of decidable.
In step 4, need carry out the spectral resolution of complete restricted least square to remote sensing image to be detected.Restricted linear hybrid decomposition analysis mainly consider the mixed pixel end-member composition than and be 1, the end-member composition ratio can not be two restrictive conditions of negative, the typical FCLS algorithm of realizing being based on the optimum constraint of non-negative least square.This method is divided into initiatively set and passive set two parts with the component of estimating.Initiatively set is made up of the end member sequence number of all non-positive component correspondences, and passive set is that positive corresponding end member sequence number is formed by component then.
Referring to Fig. 2, in the embodiment of the invention, (be meant the end member spectroscopic data of having obtained before known in advance or the detection with the end member spectroscopic data, it is the vector that gray-scale value constitutes on each wave band of the pure pixel of certain atural object, the method that end member extracts is existing) and image data that remote sensing image to be detected provides be input, use following step to find the solution:
1, the passive set of initialization P
(0)=1,2 ... p} and initiatively set
The sequence number of element numerical marker end member wherein makes k=0, begins to find the solution
When the k time computing carried out in expression, the component result of non-negative least square.
2, utilize end member matrix M and pixel vector r, by least-squares calculation component vector
Order
The end member matrix M is formed the end member matrix of its column vector by all spectral informations, and r is certain pixel on the image.For ease of understanding, illustrate for example at this: 100*100 automobile arranged, and they are made of wheel, vehicle body, chassis, five kinds of end members of engine, and each end member all is a vector.The end member matrix M is made of these five end members exactly so, and each row is exactly a kind of vector of end member correspondence.Take out any r (vector) in 10000 automobiles, it all is to constitute like this: r=M*a, a are the mixing constants of end member, are exactly component.One of them pixel may contain 10% A end member, 30% B end member, 40% C end member, 20% D end member, 0% E end member.Therefore, we have judged that at first E does not exist.Like this, the result of spectral resolution will be higher, more tally with the actual situation.Detector with this component information is set up just can separate the detection of a target in other words more exactly.
3, begin to carry out the k time circulation, promptly carry out the following the 4th and went on foot for the 12nd step.
4, make k=k+1.
5, adjustment obtains new passive set P
(k)With initiatively gather R
(k)Implementation procedure is for moving on to non-positive component end member initiatively set, the passive set P before being about to
(k-1)In all end member sequence numbers corresponding with non-positive component initiatively gather R before moving on to
(k-1)In, and with P
(k)And R
(k)Indicate new passive set and initiatively set respectively.
6, with the component vector
In with initiatively gather R
(k)The component of contained element correspondence is taken out, and constitutes new component vector
7, from (M
TM)
-1The passive set P of middle deletion
(k)The end member of middle element correspondence constitutes new matrix Φ
α (k)
8, calculating parameter
If λ
(k)In all components all be negative, then carry out the 11st the step, otherwise continuation carry out in proper order the 9th the step.
9, adjust passive set P
(k)With initiatively gather R
(k), this time specific implementation is for calculating λ
(k)Middle maximum component, the end member sequence number that it is corresponding is from initiatively gathering R
(k)In move to passive set P
(k)
10, pass through passive set P
(k)The end member of current correspondence is from (M
TM)
-1Middle rejecting obtains new matrix Ψ
λ (k)Calculate
If
Interior component has negative value, and then that it is corresponding end member is from passive set P
(k)Move on to and initiatively gather R
(k), and turned to for the 6th step.Otherwise continuation carried out for the 11st step in proper order.
11, adjust passive set P
(k), this step is with passive set P
(k)The end member of current correspondence is from matrix Ψ
λ (k)Middle rejecting obtains another matrix Ψ
λ (k)'.
12, calculate
Judge then.If all component deals are non-negative, then explanation traversal image finishes, and algorithm stops, and obtains spectral resolution result and output.Otherwise turned to for the 3rd step, continue to carry out circulation next time.
Describe technical solution of the present invention in detail below in conjunction with the concrete implementation step of embodiment, referring to Fig. 3:
(1) at first be that data are read in, promptly open the required remote sensing image to be detected that carries out target detection by the remote sensing image handling procedure, deposit this image in a two-dimensional array DataArray[bds] [num] (simple identification is the DataArray array among Fig. 3), bds, num correspond respectively to the wave band number and the pixel number of image;
(2) according to the actual demand of surveying, the input prior imformation.Part target pixel in the known image of embodiment is then got authority with them, and input is as the target end member spectral information of priori, and these target end member spectroscopic datas deposit array S[bds in] (simple identification is the S array among Fig. 3).Embodiment is after obtaining background end member spectral information by the method for manually choosing or extracting automatically, diversity of settings end member spectroscopic data is saved as matrix B, and wherein each classifies a kind of spectrum B[i of background end member as] [bds], i=1, en, en are background end member number.
(3) utilize the crosscorrelation matching technique, the traversal image is determined the background end member kind in each pixel.Crosscorrelation Spectral matching technology is a kind of existing spectral similarity decision method, judges the background end member kind that each pixel contains with it among the present invention, and embodiment defines CCSM () function and finishes this process.For the sake of ease of implementation, provide the function implementation procedure as follows:
At first, the crosscorrelation Spectral matching is calculated the projection value of spectrum on reference spectra of all pixels, by relatively obtaining maximal projection value r
MaxWith corresponding end member spectrum vector A
Max, A so
MaxAs the end member spectrum the highest, as the first-selected end member of this pixel with this pixel similarity.If r
MaxRegard end member A as
MaxTo the contribution of mixed pixel ρ, remain end member so and can be expressed as the contribution of ρ:
ρ
r＝ρ-r
maxA
max
Then, with formula
In R
rUse ρ
rReplace, continue to compare the residue end member in the pixel, find out maximum projection value and end member spectrum vector, successively projection calculating, following formula are carried out iteration.This process is actually the contribution rate of end member to the pixel response is sorted, and finds out the different end member composition that comprises in the pixel, and iteration satisfies certain condition ends, that is: ρ
rCertain component be negative value; Or Δ ρ changes very little.
Δρ＝ρ
r (k+1)-ρ
r (k)
ρ
r (k+1), ρ
r (k)It is respectively the pixel reacting value after k+1, k the iteration.
Many mixed pixels are only through an iteration ρ
rJust satisfy the termination condition.This is owing to be nonopiate between the end member spectrum vector of choosing.Therefore, redefine the contribution margin of residue pixel:
ρ
r=ρ-η r
MaxA
MaxThe η here adjusts coefficient as one, and value is [0,1].It affects end member number N to a certain extent, and when η obtains excessively, pixel only just satisfies the termination condition through fewer iterations, and this is obviously undesirable; When η obtains too smallly, the end member number remains unchanged, and the calculating of spectral response value is meaningless.In the experiment, generally get it between [0.55,0.86].According to last result, obtain the kind of background end member in each pixel, deposit array ES[en in] [num], element wherein distributes to characterize in certain pixel with 0,1 and does not contain or contain certain background end member.
(4) utilize various end member spectral informations, image is carried out the spectral resolution of complete restricted least square, obtain the component information of various end members.Particularly, the decomposition method of the complete restricted least square that embodiment adopts is the numerical computation method based on active set shown in Figure 2, is defined as function F CLS (), obtains the end-member composition a[en in each pixel at last] [num].
(5) obtain the linear mixed model of pixel according to step 4 obtained component information and step 2 gained prior imformation,, adopt the method for test of hypothesis to set up detector based on Generalized Likelihood Ratio then based on the linear mixed model of mixed pixel.
(6) according to end member number and kind in each pixel, when the whole remote sensing image to be detected of traversal,, and then use detector to survey respectively by selecting corresponding end member number and kind, obtain the probe function value of each pixel in the remote sensing image to be detected.Embodiment defines probe function Detector ();
Wherein in traversal during image, to each pixel, background end member matrix B and all end member matrix E will utilize two-dimensional array ES[en] [num] be that the end member selection matrix is done end member and selected,
With
For the end-member composition result that mixed pixel in (4) decomposes, corresponding with B and E respectively.
Be the array a[en that is tried to achieve in (4)] [num] certain component.For example, when investigating i pixel,
I=1 ... num;
J=2 ..., en, i=1 ... num, promptly
Vector for the end-member composition formation of having powerful connections in the pixel of being investigated.Cut apart according to preset threshold value, judge whether contain target in each pixel, can obtain the target detection result.
Claims (2)
1. one kind based on the variable target in hyperspectral remotely sensed image object detection method of end member, it is characterized in that, may further comprise the steps:
Step 1 is selected the required remote sensing image to be detected that carries out target detection;
Step 2 is obtained the prior imformation that needs when surveying, and described prior imformation comprises target end member spectral information and background end member spectral information;
Step 3 according to step 2 gained prior imformation, utilizes the crosscorrelation matching technique to travel through remote sensing image to be detected, determines the background end member kind in each pixel in the remote sensing image to be detected;
Step 4, utilize the background end member kind in step 2 gained prior imformation and each pixel of step 3 gained, remote sensing image to be detected is carried out the spectral resolution of complete restricted least square, obtain the component information of each pixel internal object end member and diversity of settings end member in the remote sensing image to be detected;
Step 5 obtains the linear mixed model of pixel according to step 4 obtained component information and step 2 gained prior imformation, according to the detector of this modelling based on Generalized Likelihood Ratio;
Step 6 adopts step 5 gained detector to travel through remote sensing image to be detected, obtains the probe function value of each pixel in the remote sensing image to be detected; According to gained probe function value and preset threshold value, judge whether contain target in the remote sensing image to be detected in each pixel.
2. target in hyperspectral remotely sensed image object detection method as claimed in claim 1 is characterized in that: the specific implementation of step 3 adopts following computing formula:
In the formula, R
rBe background end member spectral information, R
tBe the spectral information of certain pixel in the remote sensing image to be detected, establishing has n kind background end member spectral information in the prior imformation, establish R
rFor wherein a kind of, calculate gained r and represent whether contain this kind background end member in this pixel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201010130276 CN101806898A (en) | 2010-03-19 | 2010-03-19 | Hyperspectral remote sensing image target detecting method based on variable end members |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201010130276 CN101806898A (en) | 2010-03-19 | 2010-03-19 | Hyperspectral remote sensing image target detecting method based on variable end members |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN101806898A true CN101806898A (en) | 2010-08-18 |
Family
ID=42608777
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201010130276 Pending CN101806898A (en) | 2010-03-19 | 2010-03-19 | Hyperspectral remote sensing image target detecting method based on variable end members |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101806898A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101950361A (en) * | 2010-09-06 | 2011-01-19 | 中国科学院遥感应用研究所 | Adaptive extraction method of remote sensing image thematic information based on spectrum matching degree |
| CN102540271A (en) * | 2011-12-27 | 2012-07-04 | 南京理工大学 | Semi-supervised hyperspectral sub-pixel target detection method based on enhanced constraint sparse regression method |
| CN103115679A (en) * | 2013-01-18 | 2013-05-22 | 中国科学院对地观测与数字地球科学中心 | Obtaining method of ground object spectrum and device thereof and detection method of hyperspectral image and device thereof |
| CN103335968A (en) * | 2013-06-25 | 2013-10-02 | 中国科学院遥感与数字地球研究所 | Hyperspectral unmixing method based on background spectrum removal |
| CN104463897A (en) * | 2014-12-30 | 2015-03-25 | 电子科技大学 | Hyperspectral target detection method |
| CN105160642A (en) * | 2015-08-29 | 2015-12-16 | 西安科技大学 | Latent fingerprint image enhancement method for hyperspectral imaging |
| CN105761273A (en) * | 2016-03-18 | 2016-07-13 | 武汉大学 | Hyperspectral remote sensing image abnormal target detecting method based on graph construction |
| CN106600602A (en) * | 2016-12-30 | 2017-04-26 | 哈尔滨工业大学 | Clustered adaptive window based hyperspectral image abnormality detection method |
| CN107238591A (en) * | 2017-05-27 | 2017-10-10 | 大连海事大学 | The spectrum solution of living body fluorescent mixes detection method |
| CN109801226A (en) * | 2018-12-13 | 2019-05-24 | 中国科学院西安光学精密机械研究所 | Waterborne target detection method based on SAM and weighting auto-correlation CEM algorithm |
| CN112417934A (en) * | 2019-08-23 | 2021-02-26 | 华为技术有限公司 | Image detection method and related equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020012451A1 (en) * | 2000-06-13 | 2002-01-31 | Ching-Fang Lin | Method for target detection and identification by using proximity pixel information |
| 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 |
| CN101221243A (en) * | 2007-11-01 | 2008-07-16 | 复旦大学 | Mixed Pixel Decomposition Method of Remote Sensing Image Based on Non-negative Matrix Factorization |
| CN101504315A (en) * | 2009-02-23 | 2009-08-12 | 北京航空航天大学 | Expansion morphology and orthogonal subspace projection combined end member automatic extraction method |
-
2010
- 2010-03-19 CN CN 201010130276 patent/CN101806898A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020012451A1 (en) * | 2000-06-13 | 2002-01-31 | Ching-Fang Lin | Method for target detection and identification by using proximity pixel information |
| 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 |
| CN101221243A (en) * | 2007-11-01 | 2008-07-16 | 复旦大学 | Mixed Pixel Decomposition Method of Remote Sensing Image Based on Non-negative Matrix Factorization |
| CN101504315A (en) * | 2009-02-23 | 2009-08-12 | 北京航空航天大学 | Expansion morphology and orthogonal subspace projection combined end member automatic extraction method |
Non-Patent Citations (1)
| Title |
|---|
| 《遥感学报》 20090715 杜博等 一种自适应匹配子空间亚像元目标探测方法 第597-599页第1-5节 1-2 第13卷, 第04期 2 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101950361A (en) * | 2010-09-06 | 2011-01-19 | 中国科学院遥感应用研究所 | Adaptive extraction method of remote sensing image thematic information based on spectrum matching degree |
| CN102540271A (en) * | 2011-12-27 | 2012-07-04 | 南京理工大学 | Semi-supervised hyperspectral sub-pixel target detection method based on enhanced constraint sparse regression method |
| 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 |
| CN103115679A (en) * | 2013-01-18 | 2013-05-22 | 中国科学院对地观测与数字地球科学中心 | Obtaining method of ground object spectrum and device thereof and detection method of hyperspectral image and device thereof |
| CN103335968B (en) * | 2013-06-25 | 2015-04-01 | 中国科学院遥感与数字地球研究所 | Hyperspectral unmixing method based on background spectrum removal |
| CN103335968A (en) * | 2013-06-25 | 2013-10-02 | 中国科学院遥感与数字地球研究所 | Hyperspectral unmixing method based on background spectrum removal |
| CN104463897B (en) * | 2014-12-30 | 2017-11-10 | 电子科技大学 | A kind of method of hyper-spectral target detection |
| CN104463897A (en) * | 2014-12-30 | 2015-03-25 | 电子科技大学 | Hyperspectral target detection method |
| CN105160642A (en) * | 2015-08-29 | 2015-12-16 | 西安科技大学 | Latent fingerprint image enhancement method for hyperspectral imaging |
| CN105160642B (en) * | 2015-08-29 | 2017-10-03 | 西安科技大学 | The latent enhancement method of fingerprint image of high light spectrum image-forming |
| CN105761273A (en) * | 2016-03-18 | 2016-07-13 | 武汉大学 | Hyperspectral remote sensing image abnormal target detecting method based on graph construction |
| CN106600602A (en) * | 2016-12-30 | 2017-04-26 | 哈尔滨工业大学 | Clustered adaptive window based hyperspectral image abnormality detection method |
| CN106600602B (en) * | 2016-12-30 | 2019-08-23 | 哈尔滨工业大学 | Based on cluster adaptive windows hyperspectral image abnormal detection method |
| CN107238591A (en) * | 2017-05-27 | 2017-10-10 | 大连海事大学 | The spectrum solution of living body fluorescent mixes detection method |
| CN107238591B (en) * | 2017-05-27 | 2020-07-21 | 大连海事大学 | Spectral unmixing detection method for living body fluorescence |
| CN109801226A (en) * | 2018-12-13 | 2019-05-24 | 中国科学院西安光学精密机械研究所 | Waterborne target detection method based on SAM and weighting auto-correlation CEM algorithm |
| CN112417934A (en) * | 2019-08-23 | 2021-02-26 | 华为技术有限公司 | Image detection method and related equipment |
| WO2021036267A1 (en) * | 2019-08-23 | 2021-03-04 | 华为技术有限公司 | Image detection method and related device |
| CN112417934B (en) * | 2019-08-23 | 2024-05-14 | 华为技术有限公司 | Image detection method and related equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101806898A (en) | Hyperspectral remote sensing image target detecting method based on variable end members | |
| Halme et al. | Utility of hyperspectral compared to multispectral remote sensing data in estimating forest biomass and structure variables in Finnish boreal forest | |
| CN107292317B (en) | Polarimetric SAR classification method based on shallow features and T-matrix deep learning | |
| Gutiérrez et al. | Ground based hyperspectral imaging for extensive mango yield estimation | |
| Moreno et al. | Extreme learning machines for soybean classification in remote sensing hyperspectral images | |
| CN102110227B (en) | Compound method for classifying multiresolution remote sensing images based on context | |
| Kumar et al. | Deep remote sensing methods for methane detection in overhead hyperspectral imagery | |
| Uezato et al. | A novel endmember bundle extraction and clustering approach for capturing spectral variability within endmember classes | |
| CN105844279A (en) | Depth learning and SIFT feature-based SAR image change detection method | |
| CN107491793B (en) | Polarized SAR image classification method based on sparse scattering complete convolution | |
| Asadi et al. | Crop mapping through a hybrid machine learning and deep learning method | |
| CN103955711B (en) | A kind of mode identification method in imaging spectral target identification analysis | |
| Suarez et al. | Learning image vegetation index through a conditional generative adversarial network | |
| Fakhri et al. | A novel vegetation index-based workflow for semi-arid, sparse woody cover mapping | |
| López et al. | Classification of grapevine varieties using UAV hyperspectral imaging | |
| Aharoni et al. | Spectral light-reflection data dimensionality reduction for timely detection of yellow rust | |
| CN109447009B (en) | Hyperspectral image classification method based on subspace nuclear norm regularization regression model | |
| CN113378912A (en) | Forest area illegal reclamation land block detection method based on deep learning target detection | |
| CN108038440A (en) | A kind of Hyperspectral Remote Sensing Imagery Classification method based on overlapping pool | |
| Jiao et al. | Multiple instance hybrid estimator for learning target signatures | |
| Tahraoui et al. | Land change detection in sentinel-2 images using ir-mad and deep neural network | |
| CN109815825B (en) | Hyperspectral abnormal target detection method based on similarity constrained convex-nonnegative matrix decomposition | |
| Khanday et al. | Change detection in hyper spectral images | |
| Gao et al. | Study on deep learning model for online estimation of chlorophyll content based on near ground multispectral feature bands | |
| Vikström | A comparison of different machine learning algorithms applied to hyperspectral data analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20100818 |