CN106096663A - A kind of grader construction method of the target in hyperspectral remotely sensed image being grouped based on sparse isomery - Google Patents
A kind of grader construction method of the target in hyperspectral remotely sensed image being grouped based on sparse isomery Download PDFInfo
- Publication number
- CN106096663A CN106096663A CN201610472840.7A CN201610472840A CN106096663A CN 106096663 A CN106096663 A CN 106096663A CN 201610472840 A CN201610472840 A CN 201610472840A CN 106096663 A CN106096663 A CN 106096663A
- Authority
- CN
- China
- Prior art keywords
- sample
- target
- band
- sensed image
- remotely sensed
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB
Abstract
The grader construction method of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery, the present invention relates to the grader construction method of the target in hyperspectral remotely sensed image being grouped based on sparse isomery.The technology that the present invention is to solve needs the problem that substantial amounts of training sample and nicety of grading reduce.Step of the present invention is: one, one target in hyperspectral remotely sensed image to be sorted of input;Two, structure sample set L;Three, sparse description set of vectors V is obtained;Four, isomery Groups List GL is obtained;Five, wave band list TBList to be trained is obtained;Six, according to TBList and have tag along sort training sample set LS construct training sample subset, by algorithm of support vector machine study obtain target in hyperspectral remotely sensed image grader.The present invention utilizes less sample to choose one group of target in hyperspectral remotely sensed image wave band having enough diversityes, structural classification device, it is thus achieved that preferably Hyperspectral Remote Sensing Imagery Classification quality.The present invention is applied to remote sensing images analysis and processing technology field.
Description
Technical field
The present invention relates to the grader construction method of the target in hyperspectral remotely sensed image being grouped based on sparse isomery.
Background technology
Target in hyperspectral remotely sensed image is by EO-1 hyperion sensor, the ultraviolet of electromagnetic spectrum, visible ray, near-infrared and in red
Exterior domain, target area imaging simultaneously is produced by continuous with tens of to hundreds of and segmentation spectral band.Utilize EO-1 hyperion
Image can react the feature of atural object on deeper level, for identifying that more indistinguishable atural object tool plays a very important role, and mesh
Before be widely used in the fields such as Surface classification, agricultural monitoring, environmental management.
Target in hyperspectral remotely sensed image to be utilized needs to classify each pixel, it is thus achieved that the Land_use change that each pixel is corresponding
Type.Build grader Hyperspectral imaging can be classified automatically and can give corresponding class to each pixel in image
Mesh, therefore builds grader particularly significant for Hyperspectral imaging application.Owing to target in hyperspectral remotely sensed image comprises hundreds of wave bands,
Data dimension is higher, and direct structural classification model can cause transition fitting phenomenon, so the technology used at present the most first profit
With principal component analysis, attribute based on decision-making capability choose, chromosome based on genetic algorithm is chosen etc., and technology carries out EO-1 hyperion
Choosing of wave band, then utilizes nerve net, support vector machine, decision tree scheduling algorithm structural classification device according to the wave band chosen,
And then utilize this grader to carry out mechanized classification.This type of method has following two limitation, and one is, this type of method needs big
The training sample of amount, and certain areas obtainable training sample quantity is considerably less, it is difficult to ensure to choose valuable wave band;Two
It is, it is more difficult to ensure that the wave band chosen has the target that enough diversityes comprehensively react to be sorted so that dividing of acquisition
Relatively large deviation is there is between class model and class object.Two kinds of limitation all can cause nicety of grading to reduce so that EO-1 hyperion is distant
Sense image classification Quality Down.
Summary of the invention
The present invention is to solve the problem that prior art needs substantial amounts of training sample and nicety of grading to reduce, and proposes
The grader construction method of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery.
The grader construction method of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery realizes according to the following steps:
Step one: input a target in hyperspectral remotely sensed image to be sorted;
Step 2: classification number M of input Hyperspectral Remote Sensing Imagery Classification, input has the training sample set of tag along sort
LC, randomly selects in a target in hyperspectral remotely sensed image and obtains the training sample set LUC without tag along sort, integrates and has contingency table
The training sample set signed and the training sample set without tag along sort construct sample set L;
Step 3: target in hyperspectral remotely sensed image comprises Band Set Band, constructs sparse description vector, structure to each wave band
Become sparse description set of vectors V:
V=(V1,V2,…,Vi,…,VBN)
V1For Band1Description vector, V2For Band2Description vector, ViFor BandiDescription vector, VBNFor BandBN
Description vector;
Band=(Band1,Band2,…,Bandi,…,BandBN)
Band1 is the 1st wave band, and Band2 is the 2nd wave band, BandiFor i-th wave band, BandBNIt is the BN wave band;
Step 4: according to sparse description set of vectors V, all wave bands of target in hyperspectral remotely sensed image are divided into M isomery group,
Constitute isomery Groups List GL;
Step 5: in isomery Groups List GL, first in each group of taking-up group and last wave band, constitute and wait to instruct
Practice wave band list TBList;
Step 6: construct training sample subset with the training sample set LS having tag along sort according to TBList, by propping up
Hold vector machine Algorithm Learning and obtain the grader of target in hyperspectral remotely sensed image.
Invention effect:
The present invention provides the grader construction method of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery, utilizes this
Method can utilize less sample to choose one group of wave band having the target in hyperspectral remotely sensed image of enough diversityes, and utilizes these
Wave band builds a disaggregated model.The Hyperspectral Remote Sensing Imagery Classification of degree of precision can be realized by the method.Can be extensive
Be applied to the fields such as Surface classification, agricultural monitoring, environmental management, especially for training sample obtain difficulty, type of ground objects
Mix and be difficult to determine that the area of sample has preferable using value.
Accompanying drawing explanation
Fig. 1 is flow chart of the present invention;
Fig. 2 is the flow chart of structure sample set L;
Fig. 3 is the flow chart obtaining sparse description set of vectors V;
Fig. 4 is the flow chart obtaining isomery Groups List GL;
Fig. 5 is the flow chart obtaining wave band list TBList to be trained;
Fig. 6 is the flow chart of the grader obtaining target in hyperspectral remotely sensed image.
Detailed description of the invention
Detailed description of the invention one: as it is shown in figure 1, the grader of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery
Construction method comprises the following steps:
Step one: input a target in hyperspectral remotely sensed image to be sorted;
Step 2: as in figure 2 it is shown, classification number M of input Hyperspectral Remote Sensing Imagery Classification, input has the instruction of tag along sort
Practice sample set LC, randomly select in a target in hyperspectral remotely sensed image and obtain the training sample set LUC without tag along sort, whole
Close and have the training sample set of tag along sort and the training sample set without tag along sort to construct sample set L;
Step 3: target in hyperspectral remotely sensed image comprises Band Set Band, constructs sparse description vector, structure to each wave band
Become sparse description set of vectors V:
V=(V1,V2,…,Vi,…,VBN)
V1For Band1Description vector, V2For Band2Description vector, ViFor BandiDescription vector, VBNFor BandBN
Description vector;
Band=(Band1,Band2,…,Bandi,…,BandBN)
Band1 is the 1st wave band, and Band2 is the 2nd wave band, BandiFor i-th wave band, BandBNIt is the BN wave band;
Step 4: according to sparse description set of vectors V, all wave bands of target in hyperspectral remotely sensed image are divided into M isomery group,
Constitute isomery Groups List GL;
Step 5: in isomery Groups List GL, first in each group of taking-up group and last wave band, constitute and wait to instruct
Practice wave band list TBList;
Step 6: construct training sample subset with the training sample set LS having tag along sort according to TBList, by propping up
Hold vector machine Algorithm Learning and obtain the grader of target in hyperspectral remotely sensed image.
Detailed description of the invention two: present embodiment is unlike detailed description of the invention one: the Gao Guang in described step one
The width of spectrum remote sensing image is Width, and height is Height, comprises BN wave band, the set constituting wave band of all wave bands
Band。
Other step and parameter are identical with detailed description of the invention one.
Detailed description of the invention three: present embodiment is unlike detailed description of the invention one or two: LC in described step 2
It it is a set comprising N number of sample;
LC=(LC1,LC2,…LCI,…LCN)
LC in LC1It is the 1st sample, LC2It is the 2nd sample, LCIIt is i-th sample, 1≤I≤N, LCNIt it is n-th
Sample;
For each sample LCI=(X, Y, C), wherein, X is this sample X-coordinate in target in hyperspectral remotely sensed image, and Y is
This sample Y coordinate in target in hyperspectral remotely sensed image;C is classification numbering, and span is 1 to M.
Other step and parameter are identical with detailed description of the invention one or two.
Detailed description of the invention four: present embodiment is unlike one of detailed description of the invention one to three: described step 2
Middle LUC comprises U sample;
LUC=(LUC1,LUC2,…LUCJ,…LUCU)
LUC in LUC1It is the 1st sample, LUC2It is the 2nd sample, LUCJIt is j-th classification, 1≤J≤N, LUCUIt is
The U classification;LUCJ=(UX, UY, UC);
Wherein, UX=Random (1~Width) is the random integers between 1 to Width, distant at EO-1 hyperion for this sample
X-coordinate in sense image;UY=Random (1~Height) is the random integers between 1 to Height, for this sample at Gao Guang
Y coordinate in spectrum remote sensing image;UC=-1 represents that corresponding sample generic is unknown.
Other step and parameter are identical with one of detailed description of the invention one to three.
Detailed description of the invention five: present embodiment is unlike one of detailed description of the invention one to four: described step 2
Middle structure sample set L particularly as follows:
L=LC ∪ LUC
Sample set L be tag along sort training sample set LC with without the training sample set of tag along sort and the union of LUC,
L is expressed as:
L=(L1,L2,…,LU+N)
U+N sample, L is comprised altogether in L1It is the 1st sample, L2It is the 2nd sample, LU+NIt is the U+N sample;
Wherein any one element Lk=(SX, SY, SC), SX are its abscissa, and SY is its vertical coordinate, and SC is its corresponding class
Mesh;If this sample is from LC, then SX=X, SY=Y, SC=C;If this sample comes from LUC, then SX=UX, SY=UY, SC=
UC。
Other step and parameter are identical with one of detailed description of the invention one to four.
Detailed description of the invention six: present embodiment is unlike one of detailed description of the invention one to five: as it is shown on figure 3,
Band in described step 3iDescription vector ViConstruction process be:
Step 3 one: initialize BandiDescription vector Vi;
Vi=zeros (M)
Wherein, zeros (M) represents the vector generating a M element, and this vector all elements value is 0;
Step 3 two: calculate BandiIn maximum Max and minimum M in;
Step 3 three: describe vector enumerator Counter=1, vector enumerator Counter will be described and be set to 1;
Step 3 four: count initialized vector C1 and counting vector C2;
C1=zeros (10), wherein zeros (10) represents one 10 element of initialization, and value is all the vector of 0;
C2=zeros (10), wherein zeros (10) represents one 10 element of initialization, and value is all the vector of 0;
Step 3 five: fragment counter SCounter=1, will be set to 1 by fragment counter SCounter;
Step 3 six: structure codomain is interval [Lower, Upper];Wherein, [Lower, Upper] passes through equation below meter
Calculate:
Lower=Min+ (SCounter-1) × (Max-Min)/10
Upper=Min+SCounter × (Max-Min)/10
Step Radix Notoginseng: for sample set L, each sample takes out, according to itself SX with SY coordinate, the value that i-th wave band is corresponding
Value, statistical magnitude falls into the number of samples in interval [Lower, Upper] and counts among C1 and C2;
C1 [SCounter]=and in sample set L, the value of sample is in interval [Lower, Upper], and generic
The SC number of samples equal to Counter;
C2 [SCounter]=and in sample set L, the value of sample is in interval [Lower, Upper], and generic
The SC number of samples equal to-1;
Step 3 eight: SCounter=SCounter+1, the numerical value of fragment counter SCounter increases by 1;
Step three nine-day periods after the winter solstice: if SCounter > 10, forward step 3 ten to, otherwise go to step 3 six;
Step 3 ten: calculate according to C1 and C2 and have the marker samples sparse degree for Counter classification;Calculate public affairs
Formula is:
Wherein C1 [t] is the t element in C1, and C2 [t] is the t element in C2;
Step 3 11: by ViThe Counter element be entered as Sparse;
Vi[Counter]=Sparse
Step 3 12: Counter=Counter+1, the numerical value describing vector enumerator Counter increases by 1;
Step 3 13: if Counter > M, then forward step 3 14 to, otherwise forward step 3 four to;
Step 3 14: calculate ViProcess terminate.
Other step and parameter are identical with one of detailed description of the invention one to five.
Detailed description of the invention seven: present embodiment is unlike one of detailed description of the invention one to six: described step 4
Middle GL particularly as follows:
GL=(GL1,GL2,…,GLi,..,GLM)
Wherein, GL1It is the 1st packet that classification is corresponding, GL2It is the 2nd packet that classification is corresponding, GLiFor i-th classification
Corresponding packet, 1≤i≤M, GLMFor the packet that m-th classification is corresponding.
Other step and parameter are identical with one of detailed description of the invention one to six.
Detailed description of the invention eight: present embodiment is unlike one of detailed description of the invention one to seven: as shown in Figure 4,
Described GLiConstruction process be:
Step a: initialize GLi, its content is empty, GLi=();
Step b: describe vector list V, the numerical value of i-th element describing vector according to each, finds numerical value maximum
Vector position j, i.e. jth wave band;
Step c: by jth wave band BandjIt is incorporated to GLiIn, GLi=GLi ∪ Bandj;
Step d: by VjAll elements be set to-1;
Step e: if GLiIn element number < (BN/M) then forwards step b to, otherwise forwards step f to;BN/M be BN divided by
The business of M.
Step f: structure GLiProcess terminate.
Other step and parameter are identical with one of detailed description of the invention one to seven.
Detailed description of the invention nine: present embodiment is unlike one of detailed description of the invention one to eight: as it is shown in figure 5,
The detailed process constituting wave band list TBList to be trained in described step 5 is:
On step May Day: initialize TBList, its content be sky, TBList=();
Step 5 two: GCounter=1, is set to 1 by group counter GCounter;
Step 5 three: CGroup=GL [GCounter], the GCounter the element taken out in GL puts into current group
In variable CGroup;
Step the May 4th: TBList=TBList ∪ CGroup [1], adds first element of current group variable CGroup
Enter in TBList;
Step 5 five: TBList=TBList ∪ CGroup [last], by last of current group variable CGroup
Element joins in TBList, and wherein last represents last element of CGroup position in set;
Step 5 six: GCounter=GCounter+1, the numerical value of group counter GCounter increases by 1;
Step 5 seven: if GCounter < M, then go to step 5 three, otherwise go to step 5 eight;
Step 5 eight: structure TBList process terminates.
Other step and parameter are identical with one of detailed description of the invention one to eight.
Detailed description of the invention ten: present embodiment is unlike one of detailed description of the invention one to nine: as shown in Figure 6,
The detailed process of the grader obtaining target in hyperspectral remotely sensed image in described step 6 is:
Step 6 one: SubCounter=1, is set to 1 by sample architecture enumerator SubCounter;
Step 6 two: SubSamples=(), is set to null set by training sample subset SubSamples;
Step 6 three: [X, Y, C]=LC [SubCounter], takes out the X of SubCounter sample, Y coordinate in LC
And its generic C;
Step 6 four: take out all wave bands of TBList and put at X, the numerical value of Y location and generic C one sample of composition
In sample variable Sample;
Step 6 five: SubSamples=SubSamples ∪ Sample, the content that sample variable Sample is stored
Join in training sample subset SubSamples;
Step 6 six: SubCounter=SubCounter+1, the numerical value of sample architecture enumerator SubCounter increases
1;
Step 6 seven: if SubCounter≤N, then forward step 6 three to, otherwise go to step 6 eight;
Step 6 eight: utilize all samples in algorithm of support vector machine study SubSamples to obtain a grader;
Step 6 nine: output category device, target in hyperspectral remotely sensed image is classified by grader automatically.
Other step and parameter are identical with one of detailed description of the invention one to nine.
Claims (10)
1. the grader construction method of the target in hyperspectral remotely sensed image being grouped based on sparse isomery, it is characterised in that described point
Class device construction method comprises the following steps:
Step one: input a target in hyperspectral remotely sensed image to be sorted;
Step 2: classification number M of input Hyperspectral Remote Sensing Imagery Classification, input has the training sample set LC of tag along sort,
A target in hyperspectral remotely sensed image randomly selects and obtains the training sample set LUC without tag along sort, integrate and have tag along sort
Training sample set and without tag along sort training sample set construct sample set L;
Step 3: target in hyperspectral remotely sensed image comprises Band Set Band, constructs sparse description vector to each wave band, constitutes dilute
Dredge and describe set of vectors V:
V=(V1,V2,…,Vi,…,VBN)
V1For Band1Description vector, V2For Band2Description vector, ViFor BandiDescription vector, VBNFor BandBNDescription
Vector;
Band=(Band1,Band2,…,Bandi,…,BandBN)
Band1 is the 1st wave band, and Band2 is the 2nd wave band, BandiFor i-th wave band, BandBNIt is the BN wave band;
All wave bands of target in hyperspectral remotely sensed image are divided into M isomery group by step 4: according to sparse description set of vectors V, constitute
Isomery Groups List GL;
Step 5: in isomery Groups List GL, first in each group of taking-up group and last wave band, constitute ripple to be trained
Duan Liebiao TBList;
Step 6: according to TBList and have tag along sort training sample set LS construct training sample subset, by support to
Amount machine Algorithm Learning obtains the grader of target in hyperspectral remotely sensed image.
The grader structure side of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery the most according to claim 1
Method, it is characterised in that the width of the target in hyperspectral remotely sensed image in described step one is Width, height is Height, comprises BN
Wave band, the set Band constituting wave band of all wave bands.
The grader structure side of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery the most according to claim 2
Method, it is characterised in that in described step 2, LC is a set comprising N number of sample;
LC=(LC1,LC2,…LCI,…LCN)
LC in LC1It is the 1st sample, LC2It is the 2nd sample, LCIIt is i-th sample, 1≤I≤N, LCNIt it is n-th sample;
For each sample LCI=(X, Y, C), wherein, X is this sample X-coordinate in target in hyperspectral remotely sensed image, and Y is this sample
Y coordinate in target in hyperspectral remotely sensed image;C is classification numbering, and span is 1 to M.
The grader structure side of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery the most according to claim 3
Method, it is characterised in that in described step 2, LUC comprises U sample;
LUC=(LUC1,LUC2,…LUCJ,…LUCU)
LUC in LUC1It is the 1st sample, LUC2It is the 2nd sample, LUCJIt is j-th classification, 1≤J≤N, LUCUIt it is U
Classification;LUCJ=(UX, UY, UC);
Wherein, UX=Random (1~Width) is the random integers between 1 to Width, for this sample at high-spectrum remote-sensing shadow
X-coordinate in Xiang;UY=Random (1~Height) is the random integers between 1 to Height, distant at EO-1 hyperion for this sample
Y coordinate in sense image;UC=-1 represents that corresponding sample generic is unknown.
The grader structure side of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery the most according to claim 4
Method, it is characterised in that in described step 2 construct sample set L particularly as follows:
L=LC ∪ LUC
Sample set L be tag along sort training sample set LC with without the training sample set of tag along sort and the union of LUC, L table
It is shown as:
L=(L1,L2,…,LU+N)
U+N sample, L is comprised altogether in L1It is the 1st sample, L2It is the 2nd sample, LU+NIt is the U+N sample;
Wherein any one element Lk=(SX, SY, SC), SX are its abscissa, and SY is its vertical coordinate, and SC is its corresponding classification;If
This sample from LC, then SX=X, SY=Y, SC=C;If this sample comes from LUC, then SX=UX, SY=UY, SC=UC.
The grader structure side of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery the most according to claim 5
Method, it is characterised in that Band in described step 3iDescription vector ViConstruction process be:
Step 3 one: initialize BandiDescription vector Vi;
Vi=zeros (M)
Wherein, zeros (M) represents the vector generating a M element, and this vector all elements value is 0;
Step 3 two: calculate BandiIn maximum Max and minimum M in;
Step 3 three: description vector enumerator Counter is set to 1;
Step 3 four: count initialized vector C1 and counting vector C2;
C1=zeros (10), wherein zeros (10) represents one 10 element of initialization, and value is all the vector of 0;
C2=zeros (10), wherein zeros (10) represents one 10 element of initialization, and value is all the vector of 0;
Step 3 five: fragment counter SCounter is set to 1;
Step 3 six: structure codomain is interval [Lower, Upper];Wherein, [Lower, Upper] is calculated by equation below:
Lower=Min+ (SCounter-1) × (Max-Min)/10
Upper=Min+SCounter × (Max-Min)/10
Step Radix Notoginseng: for sample set L, each sample takes out, according to itself SX with SY coordinate, the value that i-th wave band is corresponding
Value, statistical magnitude falls into the number of samples in interval [Lower, Upper] and counts among C1 and C2;
C1 [SCounter]=and in sample set L, the value of sample is interior at interval [Lower, Upper], and generic SC etc.
Number of samples in Counter;
C2 [SCounter]=and in sample set L, the value of sample is interior at interval [Lower, Upper], and generic SC etc.
In the number of samples of-1;
Step 3 eight: the numerical value of fragment counter SCounter increases by 1;
Step three nine-day periods after the winter solstice: if SCounter > 10, forward step 3 ten to, otherwise go to step 3 six;
Step 3 ten: calculate according to C1 and C2 and have the marker samples sparse degree for Counter classification;Computing formula
For:
Wherein C1 [t] is the t element in C1, and C2 [t] is the t element in C2;
Step 3 11: by ViThe Counter element be entered as Sparse;
Vi[Counter]=Sparse
Step 3 12: the numerical value describing vector enumerator Counter increases by 1;
Step 3 13: if Counter > M, then forward step 3 14 to, otherwise forward step 3 four to;
Step 3 14: calculate ViProcess terminate.
The grader structure side of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery the most according to claim 6
Method, it is characterised in that GL in described step 4 particularly as follows:
GL=(GL1,GL2,…,GLi,..,GLM)
Wherein, GL1It is the 1st packet that classification is corresponding, GL2It is the 2nd packet that classification is corresponding, GLiCorresponding for i-th classification
Packet, 1≤i≤M, GLMFor the packet that m-th classification is corresponding.
The grader structure side of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery the most according to claim 7
Method, it is characterised in that described GLiConstruction process be:
Step a: initialize GLi, its content is empty, GLi=();
Step b: describe vector list V, the numerical value of i-th element describing vector according to each, finds the arrow that numerical value is maximum
Amount position j, i.e. jth wave band;
Step c: by jth wave band BandjIt is incorporated to GLiIn;
Step d: by VjAll elements be set to-1;
Step e: if GLiIn element number < (BN/M) then forwards step b to, otherwise forwards step f to;
Step f: structure GLiProcess terminate.
The grader structure side of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery the most according to claim 8
Method, it is characterised in that the detailed process constituting wave band list TBList to be trained in described step 5 is:
On step May Day: initialize TBList, its content be sky, TBList=();
Step 5 two: group counter GCounter is set to 1;
Step 5 three: the GCounter the element taken out in GL is put in current group variable CGroup;
Step the May 4th: first element of current group variable CGroup is joined in TBList;
Step 5 five: last element of current group variable CGroup is joined in TBList;
Step 5 six: the numerical value of group counter GCounter increases by 1;
Step 5 seven: if GCounter < M, then go to step 5 three, otherwise go to step 5 eight;
Step 5 eight: structure TBList process terminates.
The grader structure side of a kind of target in hyperspectral remotely sensed image being grouped based on sparse isomery the most according to claim 9
Method, it is characterised in that the detailed process of the grader obtaining target in hyperspectral remotely sensed image in described step 6 is:
Step 6 one: sample architecture enumerator SubCounter is set to 1;
Step 6 two: training sample subset SubSamples is set to null set;
Step 6 three: take out the X of SubCounter sample, Y coordinate and its generic C in LC;
Step 6 four: take out all wave bands of TBList and put into sample at X, the numerical value of Y location and generic C one sample of composition
In variable Sample;
Step 6 five: the content that sample variable Sample is stored is joined in training sample subset SubSamples;
Step 6 six: the numerical value of sample architecture enumerator SubCounter increases by 1;
Step 6 seven: if SubCounter≤N, then forward step 6 three to, otherwise go to step 6 eight;
Step 6 eight: utilize all samples in algorithm of support vector machine study SubSamples to obtain a grader;
Step 6 nine: output category device, target in hyperspectral remotely sensed image is classified by grader automatically.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610472840.7A CN106096663B (en) | 2016-06-24 | 2016-06-24 | A kind of classifier construction method of the target in hyperspectral remotely sensed image based on the grouping of sparse isomery |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610472840.7A CN106096663B (en) | 2016-06-24 | 2016-06-24 | A kind of classifier construction method of the target in hyperspectral remotely sensed image based on the grouping of sparse isomery |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106096663A true CN106096663A (en) | 2016-11-09 |
CN106096663B CN106096663B (en) | 2019-03-26 |
Family
ID=57252670
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610472840.7A Active CN106096663B (en) | 2016-06-24 | 2016-06-24 | A kind of classifier construction method of the target in hyperspectral remotely sensed image based on the grouping of sparse isomery |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106096663B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107071858A (en) * | 2017-03-16 | 2017-08-18 | 许昌学院 | A kind of subdivision remote sensing image method for parallel processing under Hadoop |
CN111666313A (en) * | 2020-05-25 | 2020-09-15 | 中科星图股份有限公司 | Correlation construction and multi-user data matching method based on multi-source heterogeneous remote sensing data |
CN112257603A (en) * | 2020-10-23 | 2021-01-22 | 深圳大学 | Hyperspectral image classification method and related equipment |
CN115346009A (en) * | 2022-05-18 | 2022-11-15 | 上海航遥信息技术有限公司 | Geographic entity semantic modeling method based on hyperspectral data and inclined three-dimensional data |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090092299A1 (en) * | 2007-10-03 | 2009-04-09 | Siemens Medical Solutions Usa, Inc. | System and Method for Joint Classification Using Feature Space Cluster Labels |
CN102096825A (en) * | 2011-03-23 | 2011-06-15 | 西安电子科技大学 | Graph-based semi-supervised high-spectral remote sensing image classification method |
CN104751191A (en) * | 2015-04-23 | 2015-07-01 | 重庆大学 | Sparse self-adaptive semi-supervised manifold learning hyperspectral image classification method |
-
2016
- 2016-06-24 CN CN201610472840.7A patent/CN106096663B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090092299A1 (en) * | 2007-10-03 | 2009-04-09 | Siemens Medical Solutions Usa, Inc. | System and Method for Joint Classification Using Feature Space Cluster Labels |
CN102096825A (en) * | 2011-03-23 | 2011-06-15 | 西安电子科技大学 | Graph-based semi-supervised high-spectral remote sensing image classification method |
CN104751191A (en) * | 2015-04-23 | 2015-07-01 | 重庆大学 | Sparse self-adaptive semi-supervised manifold learning hyperspectral image classification method |
Non-Patent Citations (2)
Title |
---|
CE ZHANG ET.AL: "A novel multi-parameter support vector machine for image classification", 《INTERNATIONAL JOURNAL OF REMOTE SENSING》 * |
王立国 等: "引入负相似的高光谱图像半监督分类", 《信号处理》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107071858A (en) * | 2017-03-16 | 2017-08-18 | 许昌学院 | A kind of subdivision remote sensing image method for parallel processing under Hadoop |
CN111666313A (en) * | 2020-05-25 | 2020-09-15 | 中科星图股份有限公司 | Correlation construction and multi-user data matching method based on multi-source heterogeneous remote sensing data |
CN111666313B (en) * | 2020-05-25 | 2023-02-07 | 中科星图股份有限公司 | Correlation construction and multi-user data matching method based on multi-source heterogeneous remote sensing data |
CN112257603A (en) * | 2020-10-23 | 2021-01-22 | 深圳大学 | Hyperspectral image classification method and related equipment |
CN112257603B (en) * | 2020-10-23 | 2022-06-17 | 深圳大学 | Hyperspectral image classification method and related equipment |
CN115346009A (en) * | 2022-05-18 | 2022-11-15 | 上海航遥信息技术有限公司 | Geographic entity semantic modeling method based on hyperspectral data and inclined three-dimensional data |
CN115346009B (en) * | 2022-05-18 | 2023-04-28 | 上海航遥信息技术有限公司 | Geographic entity semantic modeling method based on hyperspectral data and oblique three-dimensional data |
Also Published As
Publication number | Publication date |
---|---|
CN106096663B (en) | 2019-03-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104102929B (en) | Hyperspectral remote sensing data classification method based on deep learning | |
CN110321963A (en) | Based on the hyperspectral image classification method for merging multiple dimensioned multidimensional sky spectrum signature | |
CN109800736A (en) | A kind of method for extracting roads based on remote sensing image and deep learning | |
CN106503739A (en) | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics | |
Wang et al. | Identification of apple leaf diseases by improved deep convolutional neural networks with an attention mechanism | |
CN106096663A (en) | A kind of grader construction method of the target in hyperspectral remotely sensed image being grouped based on sparse isomery | |
CN109871830A (en) | Spatial-spectral fusion hyperspectral image classification method based on three-dimensional depth residual error network | |
CN104182985B (en) | Remote sensing image change detection method | |
CN107909015A (en) | Hyperspectral image classification method based on convolutional neural networks and empty spectrum information fusion | |
CN112070078B (en) | Deep learning-based land utilization classification method and system | |
CN107358142A (en) | Polarimetric SAR Image semisupervised classification method based on random forest composition | |
CN106529508A (en) | Local and non-local multi-feature semantics-based hyperspectral image classification method | |
CN111160127B (en) | Remote sensing image processing and detecting method based on deep convolutional neural network model | |
CN101894270A (en) | Method for full-automatic sample selection oriented to classification of remote-sensing images | |
Aleksandrowicz et al. | Change detection algorithm for the production of land cover change maps over the European Union countries | |
CN108898096B (en) | High-resolution image-oriented information rapid and accurate extraction method | |
CN105160351B (en) | Semi-supervised hyperspectral classification method based on anchor point sparse graph | |
Zhang et al. | EMMCNN: An ETPS-based multi-scale and multi-feature method using CNN for high spatial resolution image land-cover classification | |
CN107895136A (en) | A kind of colliery area recognizing method and system | |
Dalal et al. | Optimized deep learning with learning without forgetting (LwF) for weather classification for sustainable transportation and traffic safety | |
CN104134069A (en) | Automatic identification method for shale microsections | |
Xu et al. | Multi-modal deep learning for weeds detection in wheat field based on RGB-D images | |
Jilge et al. | Detecting unknown artificial urban surface materials based on spectral dissimilarity analysis | |
Guo et al. | Automatic and accurate calculation of rice seed setting rate based on image segmentation and deep learning | |
Luo et al. | Test and Analysis of Vegetation Coverage in Open-Pit Phosphate Mining Area around Dianchi Lake Using UAV–VDVI |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |