CN103870807A - High spectrum mixed nuclear RX anomaly detection method - Google Patents
High spectrum mixed nuclear RX anomaly detection method Download PDFInfo
- Publication number
- CN103870807A CN103870807A CN201410064531.7A CN201410064531A CN103870807A CN 103870807 A CN103870807 A CN 103870807A CN 201410064531 A CN201410064531 A CN 201410064531A CN 103870807 A CN103870807 A CN 103870807A
- Authority
- CN
- China
- Prior art keywords
- function
- space
- kernel function
- data
- kernel
- 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
Images
Landscapes
- Image Processing (AREA)
Abstract
The invention provides a high spectrum mixed nuclear RX anomaly detection method. Original high spectrum data in a low dimension space are mapped to a high dimension feature space for processing in a nonlinear mode, and the method has well nonlinear anomaly detection capacity. Based on the Gauss radial basis kernel function, a new spectrum angle cosine kernel function is added, and a target is detected by the difference among similar coefficients among spectrum signals. According to the high spectrum mixed nuclear RX anomaly detection method, the inseparable data in the low dimension space can be mapped to the high dimension feature space for linear separation, so the target and the background can be more easily separated.
Description
Technical field
The method for detecting abnormality that the invention belongs to a kind of follow-on mixed nucleus RX operator, belongs to hyperspectral data processing method and applied technical field, is applicable to high-spectral data object detection field.
Background technology
Target in hyperspectral remotely sensed image has high spectral resolution, and it is that spatial image is tieed up to the data that the spectrum dimension information of information and atural object radiation combines together.Hyperspectral imaging can provide good environment for the automatic detection and Identification of target.In target detection, have many algorithms to obtain good result, but to have individual limitation be the prior imformation that need to be detected target for they.But, in current practical application, accurate not in default of complete spectra database and reflectivity inversion algorithm.So these prior imformations are difficult to obtain.Therefore,, without the Outlier Detection Algorithm of prior imformation, become the focus of remote sensing image application research.
The little target in high spectrum image is considered as the singular point under certain distribution occasion by these Outlier Detection Algorithms, and the impact point that makes to convert in rear data by particular procedure is given prominence to, and then automatically detects abnormal object.Wherein the most classical method is exactly the RX method based on generalized likelihood-ratio test that Reed and Yu propose.Due to the changeable complexity of atural object distribution, it is also complicated and changeable making high-spectral data, traditional RX operator has only utilized the low order statistical property of high-spectral data, and ignore the abundant nonlinear transformations comprising in a high-spectral data hundreds of wave band, between them, there are very strong correlation properties, in lower dimensional space, these data are linearly inseparables, can pass through a nonlinear mapping function, the spectroscopic data of low-dimensional is mapped in high-dimensional feature space, can linear separability thereby they are become from linearly inseparable, and then implementation pattern analysis, cluster etc., target and background can be separated to the full extent, improve accuracy rate, reduce false alarm rate.
Due to the unknown of Nonlinear Mapping function, cannot directly carry out data operation at feature space, at this moment need to be out of shape by kernel function technology " dimension disaster " that the computing that is then able to can to carry out data in high-dimensional feature space has existed while having avoided again computing to above formula.In this case, the people such as Kwon has proposed the RX algorithm based on core.Aspect selection kernel function, what great majority were selected is gaussian radial basis function kernel function, but, in the time that gaussian radial basis function kernel function is changed feature space data, only consider the difference of the energy of spectrum vector, and ignored because of the impact of atmosphere other factors on curve of spectrum difference.
Summary of the invention
The present invention is directed to the deficiencies in the prior art, a kind of mixed nucleus RX Hyperspectral imaging method for detecting abnormality is provided.
The concrete steps of the inventive method are:
1) pre-service high-spectral data.
Airborne-remote sensing is subject to the effect of the factors such as Atmospheric Absorption, refraction and scattering, and because atural object on the sunny side makes spectroscopic data produce the abnormal wave band of considerable influence with the natural cause such as in the shade, these abnormal wave bands cannot show characters of ground object in imaging, and the detection of impact to target, so these abnormal wave bands are done to direct rejecting processing.
2) mixed nucleus RX algorithm foreign matter detects
If x is input data luv space X (x ∈ X ∈ R
p) in an input vector of data set, F is the high-dimensional feature space (a Hilbert space) that luv space X obtains being associated by mapping function φ.φ (X) is that x is mapped to the correspondence vector in F.Mapping mode is:
φ:X→F,x→φ(x) (1)
I.e. data X in lower dimensional space
b=[x (1), x (2) ... x (N)] be mapped to the data φ (X that higher dimensional space obtains
bφ)=[(X (1)), φ (X (2)) ... φ (X (N))].Use function phi by data-mapping to after in F space, the similarity between different pieces of information can be measured by the inner product in F space.In the feature space F being associated with mapping function φ according to Mercer nuclear theory, inner product between all data can be by the calculating of implicit expression, this is by using a Mercer kernel function K () who is defined in input data space to realize, and can be write as:
k(x
i,x
j)=<φ(x
i),φ(x
j)>=φ(x
i)·φ(x
j) (2)
According to the core theories of learning, in the feature space F of higher-dimension, all hypothesis of core RX algorithm are identical with RX algorithm, just for different spaces, the core RX algorithm of characteristic of correspondence space F is
Adopt the RX algorithm coring of kernel trick to be:
Can find out from formula (5), know concrete Nonlinear Mapping function and in high-dimensional feature space, carry out corresponding dot-product operation not needing, only need to the dot-product operation of higher dimensional space be converted into the foreign matter that reaches core RX operator in the kernel function computing of lower dimensional space by suitable kernel function and detect effect.Next be selection and the structure to kernel function:
2.1) selection of kernel function
The present invention selects suitable kernel function K (x
i, x
j), former Feature Space Transformation has been arrived to a certain new feature space.
A) adopt the gaussian radial basis function kernel function that has wider domain of convergence as kernel function, mathematic(al) representation is as follows:
Wherein γ is kernel functional parameter.
B) mathematical model of spectrum angle cosine is:
And polynomial kernel function
K(x,y)=(x·y+c)
d,c≥0,d∈Z
+ (8)
Work as d=1, when c=0, be called linear kernel K (x, y)=< xy >.When d ≠ 0, when c=0, be called homogeneous polynomial core K (x, y)=< xy >
d.Other claims Nonlinear Homogeneous polynomial kernel.
By the differentiation of spectrum angle cosine function is obtained
Formula (9) is a Nonlinear Homogeneous polynomial expression.Make d ≠ 0, c=0 obtains innovative spectrum of the present invention angle cosine kernel function
2.2) structure of mixed nucleus function
Closure property by kernel function is known:
k(x,y)=k
1(x,y)+k
2(x,y) (11)
k(x,y)=αk
1(x,y) (12)
Formula (11) and (12) are all kernel functions, so the present invention combines gaussian radial basis function kernel function kr and spectrum angle cosine kernel function ks, form the kernel function of a mixing:
Krs=αkr+(1-α)ks (13)
Wherein 0≤α≤1.According to reconciling the value definite kernel function of α shared proportion in algorithm, in some high spectrum image, possibility real background data relatively meet the distribution of Gaussian function, and otherness between spectrum is very little, at this moment the shared ratio of gaussian kernel function is great, and effect can be better; On the contrary, in some high spectrum image, may actual background distributions be morbid state, and diversity ratio between spectrum be larger, now spectrum angle cosine kernel function proportion is large, and effect can be better.So, select a suitable α value, most important to the result of experimental data.By image and the data result of test, the correct verification and measurement ratio that uses mixed nucleus function to obtain target is greatly improved.
The present invention can, the data of those linearly inseparables of lower dimensional space, be mapped to high-dimensional feature space linear separability, thereby can more easily target and background area be separated.
Accompanying drawing explanation
Fig. 1 is the inventive method process flow diagram;
Fig. 2 is measured data 30 wave band figure;
Fig. 3 is RX operator detection figure;
Fig. 4 is the RX operator detection figure of gaussian radial basis function kernel function;
Fig. 5 is the RX operator detection figure of mixed nucleus function.
Embodiment
Below in conjunction with accompanying drawing, the invention will be further described.
As shown in Figure 1, the inventive method comprises the following steps:
1) pre-service of high-spectral data.
Airborne-remote sensing is subject to the effect of the factors such as Atmospheric Absorption, refraction and scattering, and because atural object on the sunny side makes spectroscopic data produce the abnormal wave band of considerable influence with the natural cause such as in the shade, these abnormal wave bands cannot show characters of ground object in imaging, and the detection of impact to target, so these abnormal wave bands are done to direct rejecting processing.
2) RX foreign matter detects
RX algorithm can be regarded the inverse process of principal component analysis (PCA) (PCA) in essence as.Suppose that the pixel number that a panel height spectrum picture comprises is N, each pixel has P wave band, each pixel wave band can be expressed as x (n)=(x by vector form
1(n), x
2(n) ... x
p(n))
t.Definition X
bfor each pixel spectra to be detected is column vector, i.e. an X
b=[x (1), x (2) ... x (N)].The binary hypothesis test of RX algorithm is defined as:
H0 x=n (there is no target)
H1 x=γ s+n (having target) (1)
In formula: in the time of γ=0, H0 is that target does not exist, in the time of γ >0, H1 is that target exists, and n represents the vector of background and noise, s=[s
1, s
2... s
p]
tfor the spectrum vector of target, on the basis of two-value statistical model, the expression formula of utilizing generalized likelihood-ratio test method can deduce out RX operator is:
In above formula
with
be respectively the estimated value of background mean value and covariance matrix, η is the decision threshold of RX operator.
3) core RX foreign matter detection algorithm
If x is input data luv space X (x ∈ X ∈ R
p) in an input vector of data set, F is the high-dimensional feature space (a Hilbert space) that luv space X obtains being associated by mapping function φ.φ (X) is that x is mapped to the correspondence vector in F.Mapping mode is:
φ:X→F x→φ(x) (3)
I.e. data X in lower dimensional space
b=[x (1), x (2) ... x (N)] be mapped to the data φ (X that higher dimensional space obtains
bφ)=[(X (1)), φ (X (2)) ... φ (X (N))].Use function phi by data-mapping to after in F space, the similarity between different pieces of information can be measured by the inner product in F space.In the feature space F being associated with mapping function φ according to Mercer nuclear theory, inner product between all data can be by the calculating of implicit expression, this is by using a Mercer kernel function K () who is defined in input data space to realize, and can be write as:
k(x
i,x
j)=<φ(x
i),φ(x
j)>=φ(x
i)·φ(x
j) (4)
According to the core theories of learning, in the feature space F of higher-dimension, all hypothesis of core RX algorithm are identical with RX algorithm, just for different spaces, therefore the binary hypothesis test problem definition of core RX algorithm is:
H0
φφ (x)=n
φ(there is no target)
H1
φφ (x)=γ φ (s)+n
φ(having target) (5)
In formula: γ > 0; n
φwith φ (s) is respectively the spectrum vector of the spectrum vector sum abnormal object of background in feature space F and noise, the core RX algorithm of characteristic of correspondence space F is
In formula:
with
be respectively the covariance matrix and the mean vector that in feature space, estimate from background sample, expression formula is respectively:
Adopt the RX algorithm coring of kernel trick to be:
Can find out from formula (9), know concrete Nonlinear Mapping function and in high-dimensional feature space, carry out corresponding dot-product operation not needing, only need to the dot-product operation of higher dimensional space be converted into the foreign matter that reaches core RX operator in the kernel function computing of lower dimensional space by suitable kernel function and detect effect.
4) selection of kernel function
The present invention selects suitable kernel function K (x
i, x
j), former Feature Space Transformation has been arrived to a certain new feature space.
A) adopt the gaussian radial basis function kernel function that has wider domain of convergence as kernel function, mathematic(al) representation is as follows:
Wherein γ is kernel functional parameter.
B) mathematical model of spectrum angle cosine is:
And polynomial kernel function
K(x,y)=(x·y+c)
d,c≥0,d∈Z
+ (12)
Work as d=1, when c=0, be called linear kernel K (x, y)=< xy >.When d ≠ 0, when c=0, be called homogeneous polynomial core K (x, y)=< xy >
d.Other claims Nonlinear Homogeneous polynomial kernel.
By the differentiation of spectrum angle cosine function is obtained
Formula (13) is a Nonlinear Homogeneous polynomial expression.Make d ≠ 0, c=0 obtains novel spectrum of the present invention angle cosine kernel function
5) structure of mixed nucleus function
Closure property by kernel function is known:
k(x,y)=k
1(x,y)+k
2(x,y) (15)
k(x,y)=αk
1(x,y) (16)
Formula (15) and (16) are all kernel functions, so the present invention combines gaussian radial basis function kernel function and spectrum angle cosine kernel function ks, form the kernel function of a mixing:
Krs=αkr+(1-α)ks (17)
Wherein 0≤α≤1.According to reconciling the value definite kernel function of α shared proportion in algorithm, in some high spectrum image, possibility real background data relatively meet the distribution of Gaussian function, and otherness between spectrum is very little, at this moment the shared ratio of gaussian kernel function is great, and effect can be better; On the contrary, in some high spectrum image, may actual background distributions be morbid state, and diversity ratio between spectrum be larger, now spectrum angle cosine kernel function proportion is large, and effect can be better.So, select a suitable α value, most important to the result of experimental data.By image and the data result of test, the correct verification and measurement ratio that uses mixed nucleus function to obtain target is greatly improved.
What in embodiment, adopt is one group of actual measurement high-spectral data, is that the visible and near infrared spectrum data of 80 wave bands good by registration to obtain the fused data of 155 wave bands with the short-wave infrared data of 75 wave bands after Band fusion and normalized.Size of data is the pixel of 226 Χ 500 Χ 155, and experiment scene has comprised take a large amount of vegetation (brushwood and harvested milpa) 1 van under overall background, 2 semitrailers and 4 irrelevant non-target targets.In scene, include 10 two class targets that scribble camouflage paint vehicle, above a row be 5 of coating military green iron plate targets, be that a row applies 5 of military green plank targets below.Fig. 2 is the measured data of 30 wave bands.
Experiment uses RX operator, the RX operator of gaussian radial basis function kernel function, the RX operator technology of mixed nucleus function respectively 155 wave band datas to be carried out to foreign matter detection, Fig. 3 (detection of RX operator), Fig. 4 (the RX operator of gaussian radial basis function kernel function detects), Fig. 5 (the RX operator of mixed nucleus function detects) of obtaining.Due to the changeable complexity of atural object distribution, it is also complicated and changeable making high-spectral data, traditional RX operator has only utilized the low order statistical property of high-spectral data, and ignore the abundant nonlinear transformations comprising in a high-spectral data hundreds of wave band, between them, there are very strong correlation properties, in lower dimensional space, these data are linearly inseparables, so some target and background is difficult for separating well at lower dimensional space in high-spectral data, so just likely target mistake is divided into background, background mistake has been divided into target, increase false alarm rate from reducing verification and measurement ratio.After using kernel function, can, the data of those linearly inseparables of lower dimensional space, be mapped to high-dimensional feature space linear separability, thereby can more easily target and background area be separated.Just can find that by Fig. 3 and Fig. 4 contrast the false alarm rate detecting with common RX operator is very large, and the image false alarm rate detecting with core RX operator is very little.
In order further to find out intuitively that the core RX of mixing detects the advantage of effect, reality is detected to data and add up, as table 1.
Table 1 actual measurement detects data statistic
Can find out that by the data statistics of table 1 no matter use which kind of method, the detection accuracy of target 1 is all the time between 70%-80%, the detection accuracy of target 2 is all the time more than 85%, and accuracy does not have target 2 height, and false alarm rate is really than target 2 height.Its reason may be that target 1 spectrum vector is more approaching with background spectrum vector.Or because atmospheric environment is larger on the impact of target 1, cause the reasons such as apparatus measures error.
By the Analysis of test results to overall goals, although use the accuracy of core RX operator testing result not have directly to use the accuracy of RX operator detection high, his false alarm rate reduces greatly, has dropped to 0.78% from 12.70%.Like this can be in actual central easier discovery target, false alarm rate height is as easy as rolling off a log obscures real goal and decoy, is difficult to analyze the position of real goal, does the judgement making mistake.After having used mixed nucleus RX operator to detect, although false alarm rate a little improves, bring up to 1.05% from 0.78%, that raising does not have great impact to resolution target true and false.But the accuracy of target has really improved a lot, has brought up to 83.50% from 78.49%.The image of Comprehensive Experiment and the analysis of data, visible mixed nucleus RX operator detection method no matter be target accuracy high aspect, should false alarm rate low aspect be all than common RX operator detection method and core RX operator detection method.
Claims (1)
1. a high spectral mixing core RX method for detecting abnormality, is characterized in that the method comprises the following steps:
1) pre-service high-spectral data: reject and be subject to Atmospheric Absorption, refraction and scattering process, and because of atural object on the sunny side and the in the shade abnormal wave band that makes spectroscopic data produce considerable influence;
2) mixed nucleus RX algorithm foreign matter detects, specifically:
If x is an input vector of data set in input data luv space X, F is the high-dimensional feature space that luv space X obtains being associated by mapping function φ; φ (X) is that x is mapped to the correspondence vector in F; Mapping mode is:
φ:X→F,x→φ(x) (1)
I.e. data X in lower dimensional space
b=[x (1), x (2) ... x (N)] be mapped to the data φ (X that higher dimensional space obtains
bφ)=[(X (1)), φ (X (2)) ... φ (X (N))]; Use function phi by data-mapping to after in F space, the similarity between different pieces of information can be measured by the inner product in F space; In the feature space F being associated with mapping function φ according to Mercer nuclear theory, inner product between all data can be by the calculating of implicit expression, this is by using a Mercer kernel function K () who is defined in input data space to realize, and can be write as:
k(x
i,x
j)=<φ(x
i),φ(x
j)>=φ(x
i)·φ(x
j) (2)
According to the core theories of learning, in the feature space F of higher-dimension, all hypothesis of core RX algorithm are identical with RX algorithm, just for different spaces, the core RX algorithm of characteristic of correspondence space F is
Adopt the RX algorithm coring of kernel trick to be:
Can find out from formula (5), know concrete Nonlinear Mapping function and in high-dimensional feature space, carry out corresponding dot-product operation not needing, only need to the dot-product operation of higher dimensional space be converted into the foreign matter that reaches core RX operator in the kernel function computing of lower dimensional space by suitable kernel function and detect effect; Next be selection and the structure to kernel function:
2.1) selection of kernel function, specifically:
Select suitable kernel function K (x
i, x
j), former Feature Space Transformation has been arrived to a certain new feature space;
A) adopt the gaussian radial basis function kernel function that has wider domain of convergence as kernel function, mathematic(al) representation is as follows:
Wherein γ is kernel functional parameter;
B) mathematical model of spectrum angle cosine is:
And polynomial kernel function
K(x,y)=(x·y+c)
d,c≥0,d∈Z
+ (8)
Work as d=1, when c=0, be called linear kernel K (x, y)=< xy >; When d ≠ 0, when c=0, be called neat
Order polynomial core K (x, y)=< xy >
d; Other claims Nonlinear Homogeneous polynomial kernel;
By the differentiation of spectrum angle cosine function is obtained
Formula (9) is a Nonlinear Homogeneous polynomial expression; Make d ≠ 0, c=0 obtains new spectrum angle cosine kernel function
2.2) structure of mixed nucleus function
Closure property by kernel function is known:
k(x,y)=k
1(x,y)+k
2(x,y) (11)
k(x,y)=αk
1(x,y) (12)
Formula (11) and (12) are all kernel functions, so gaussian radial basis function kernel function kr and spectrum angle cosine kernel function ks are combined, form the kernel function of a mixing:
Krs=αkr+(1-α)ks (13)
Wherein 0≤α≤1; The value of determining α according to the difference between real background data and spectrum in high spectrum image shared proportion in Kernels, thus realize method for detecting abnormality.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410064531.7A CN103870807A (en) | 2014-02-25 | 2014-02-25 | High spectrum mixed nuclear RX anomaly detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410064531.7A CN103870807A (en) | 2014-02-25 | 2014-02-25 | High spectrum mixed nuclear RX anomaly detection method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103870807A true CN103870807A (en) | 2014-06-18 |
Family
ID=50909323
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410064531.7A Pending CN103870807A (en) | 2014-02-25 | 2014-02-25 | High spectrum mixed nuclear RX anomaly detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103870807A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104504686A (en) * | 2014-12-04 | 2015-04-08 | 哈尔滨工程大学 | Hyper-spectral image abnormity detection method adopting local self-adaptive threshold segmentation |
CN104778706A (en) * | 2015-04-21 | 2015-07-15 | 西安电子科技大学 | Abnormity detection method and device on basis of non-negative matrix factorization |
CN106600602A (en) * | 2016-12-30 | 2017-04-26 | 哈尔滨工业大学 | Clustered adaptive window based hyperspectral image abnormality detection method |
CN107038436A (en) * | 2017-05-24 | 2017-08-11 | 哈尔滨工业大学 | A kind of high spectrum image object detection method based on tensor Spectral match filter |
CN107527043A (en) * | 2017-09-15 | 2017-12-29 | 湖南神帆科技有限公司 | A kind of variable close shot high spectrum image local anomaly detection method of exterior window |
-
2014
- 2014-02-25 CN CN201410064531.7A patent/CN103870807A/en active Pending
Non-Patent Citations (7)
Title |
---|
HEESUNG KWON ET AL: "HYPERSPECTRAL ANOMALY DETECTION USING KERNEL RX-ALGORITHM", 《2004 INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP)》 * |
HEESUNG KWON ET AL: "Kernel RX-Algorithm: A Nonlinear Anomaly Detector for Hyperspectral Imagery", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
史振威 等: "RX及其变种在高光谱图像中的异常检测", 《红外与激光工程》 * |
季亚新: "高光谱图像异常检测算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
尤佳: "基于核方法的高光谱图像异常检测算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
成宝芝 等: "基于粒子群优化聚类的高光谱图像异常目标检测", 《光电子· 激光》 * |
陈伟: "高光谱影像地物智能化提取理论与方法研究", 《中国优秀博士学位论文全文数据库 信息科技辑》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104504686A (en) * | 2014-12-04 | 2015-04-08 | 哈尔滨工程大学 | Hyper-spectral image abnormity detection method adopting local self-adaptive threshold segmentation |
CN104504686B (en) * | 2014-12-04 | 2017-06-23 | 哈尔滨工程大学 | A kind of hyperspectral image abnormal detection method of use local auto-adaptive Threshold segmentation |
CN104778706A (en) * | 2015-04-21 | 2015-07-15 | 西安电子科技大学 | Abnormity detection method and device on basis of non-negative matrix factorization |
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 |
CN107038436A (en) * | 2017-05-24 | 2017-08-11 | 哈尔滨工业大学 | A kind of high spectrum image object detection method based on tensor Spectral match filter |
CN107038436B (en) * | 2017-05-24 | 2020-08-25 | 哈尔滨工业大学 | Hyperspectral image target detection method based on tensor spectrum matched filtering |
CN107527043A (en) * | 2017-09-15 | 2017-12-29 | 湖南神帆科技有限公司 | A kind of variable close shot high spectrum image local anomaly detection method of exterior window |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107292317B (en) | Polarization SAR classification method based on shallow feature and T matrix deep learning | |
CN105069758B (en) | A kind of high spectrum image denoising method based on robust low-rank tensor | |
CN107451614B (en) | Hyperspectral classification method based on fusion of space coordinates and space spectrum features | |
CN104966085B (en) | A kind of remote sensing images region of interest area detecting method based on the fusion of more notable features | |
CN104374738B (en) | A kind of method for qualitative analysis improving identification result based on near-infrared | |
CN103247059B (en) | A kind of remote sensing images region of interest detection method based on integer wavelet and visual signature | |
CN104392251B (en) | Hyperspectral image classification method based on semi-supervised dictionary learning | |
Ren et al. | Automatic target recognition for hyperspectral imagery using high-order statistics | |
CN107103306B (en) | Winter wheat powdery mildew remote sensing monitoring method based on wavelet analysis and support vector machine | |
CN101144860A (en) | Hyperspectral image abnormal point detection method based on selective kernel principal component analysis | |
CN106886760B (en) | A kind of EO-1 hyperion Ship Detection combined based on empty spectrum information | |
CN103870807A (en) | High spectrum mixed nuclear RX anomaly detection method | |
CN102722892B (en) | SAR (synthetic aperture radar) image change detection method based on low-rank matrix factorization | |
CN103208011B (en) | Based on average drifting and the hyperspectral image space-spectral domain classification method organizing sparse coding | |
CN102540271B (en) | Semi-supervised hyperspectral sub-pixel target detection method based on enhanced constraint sparse regression method | |
CN103413292B (en) | Based on the hyperspectral image nonlinear abundance estimation method of constraint least square | |
CN104867150A (en) | Wave band correction change detection method of remote sensing image fuzzy clustering and system thereof | |
CN104299232B (en) | SAR image segmentation method based on self-adaptive window directionlet domain and improved FCM | |
CN109740631B (en) | OBIA-SVM-CNN remote sensing image classification method based on object | |
CN104751181A (en) | High spectral image Deming method based on relative abundance | |
CN107844751A (en) | The sorting technique of guiding filtering length Memory Neural Networks high-spectrum remote sensing | |
CN103456011A (en) | Improved hyperspectral RX abnormal detection method by utilization of complementary information | |
CN108197650A (en) | The high spectrum image extreme learning machine clustering method that local similarity is kept | |
CN106778680A (en) | A kind of hyperspectral image band selection method and device extracted based on critical bands | |
CN104809471B (en) | A kind of high spectrum image residual error integrated classification method based on spatial spectral information |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20140618 |