CN107909062A - A kind of Algorithm for Discretization of Remote Sensing Image Attribute and system based on comentropy - Google Patents
A kind of Algorithm for Discretization of Remote Sensing Image Attribute and system based on comentropy Download PDFInfo
- Publication number
- CN107909062A CN107909062A CN201711311000.3A CN201711311000A CN107909062A CN 107909062 A CN107909062 A CN 107909062A CN 201711311000 A CN201711311000 A CN 201711311000A CN 107909062 A CN107909062 A CN 107909062A
- Authority
- CN
- China
- Prior art keywords
- discretization
- comentropy
- section
- image information
- attribute
- 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
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
Abstract
The present invention relates to a kind of Algorithm for Discretization of Remote Sensing Image Attribute and system based on comentropy, the method comprising the steps of：Establish the image information decision table based on pixel value and generic；Section based on comentropy separates model computation interval comentropy, and interal separation is carried out based on comentropy；Discretization is carried out to the image information decision table, and calculates the Indiscernible relation diversity factor of the forward and backward image information table of discretization respectively；The result of discretization is assessed, selects optimal discretization scheme.The method of the present invention and system can reduce discretization section number, avoid the redundancy of block information, alleviate discretization data distortion problem.
Description
Technical field
The present invention relates to characteristics of remote sensing image extractive technique field, more particularly to a kind of remote sensing image based on comentropy is special
Levy discretization method and system.
Background technology
Environmental monitoring is the important component of marine informatization construction, is the environmental protection and management of ocean, and resource is opened
Send out and utilize and provide strong scientific basis.Seasat effectively compensate for the deficiency of Conventional marine observation method, based on more
Kind of the remote sensor continuously observation to ocean, makes the mankind greatly deepen the understanding to ocean, in preventing and reducing natural disasters for Oceanic disasters,
Development of resources, ocean right-safeguarding, the numerous areas such as marine ecology and environmental protection play an important role.Remote sensing that will be to collecting
Image data is analyzed and processed it is necessary to first carry out feature extraction to it, these image features are often continuous data.However, work as
Preceding most knowledge extraction (Knowledge Extraction) and data mining (Data Mining) algorithm can only handle from
Type data are dissipated, although some algorithms can also handle continuous data, poor-performing.Therefore, it is necessary to be converted into continuous feature
Discrete features reduce time and the space expense of algorithm, improve cluster energy of the system to sample to adapt to these intellectualized algorithms
Power, strengthening system anti-noise ability, and the study precision of algorithm is improved, expand application range.
A kind of data reducti techniques of the discretization (Discretization) as data preprocessing phase, were subject in recent years
Extensive concern and research, and achieve plentiful and substantial achievement in research.The essence of discretization is exactly to determine that selection is more briefly
Few cut-point and definite cut-point position.Whether the method for discretization is very much, can be it comprising classification information according to data
Be divided into and have supervision discretization and unsupervised discretization, there is supervision discretization to need to consider classification information, and unsupervised discretization
Then need not.
At this stage, the unsupervised method of comparative maturity is also fewer, in the case of no category information, to be got well
Discretization results are relatively difficult, and the result of discretization is also difficult to weigh.And in having supervision discretization algorithm, single argument has
Supervise discretization algorithm and do not have the relation of interdependence considered between attribute, simply from unsupervised discretization mode development to orphan
On the spot consider some attribute to the unidirectional dependence or attribute of classification and the relation of interdependence of classification.When it is such from
When dispersion algorithm is applied to assorting process, the missing of substantial amounts of redundancy breakpoint and necessary breakpoint will cause overall process between attribute
Failure.And existing multivariable has supervision discretization algorithm (Multi Variable Supervised Discretization)
Although compensate for above-mentioned single argument to a certain extent has the deficiency of supervision discretization algorithm, also fail to reach and rejecting
The Indiscernible relation of decision table is kept while redundancy breakpoint.It is when selecting breakpoint, not in view of between attribute and belonging to
Property internal breakpoints alternative, this obviously be not suitable for processing with multiple features remote sensing image data.
The content of the invention
It is an object of the invention to improve the deficiency in the presence of the prior art, there is provided a kind of remote sensing shadow based on comentropy
As feature discretization method and system.
In order to realize foregoing invention purpose, an embodiment of the present invention provides following technical scheme：
A kind of Algorithm for Discretization of Remote Sensing Image Attribute based on comentropy, comprises the following steps：
List and pixel value and generic of the sample in each wave band are selected in remote sensing image, and establish and be based on pixel value
With the image information decision table of generic；
Section based on comentropy separates model computation interval comentropy, and interal separation is carried out based on comentropy；
To the image information decision table carry out discretization, and respectively calculate the forward and backward image information table of discretization can not
Resolution relation diversity factor；
The result of discretization is assessed according to the Indiscernible relation diversity factor, selects optimal discretization side
Case.
On the other hand, the embodiment of the present invention additionally provides a kind of characteristics of remote sensing image discretized system based on comentropy,
Including：
Image information decision table establishes unit, and pixel value of the sample in each wave band is selected in remote sensing image for listing
And generic, and establish the image information decision table based on pixel value and generic；
Comentropy computing unit, separates model computation interval comentropy for the section based on comentropy, and is based on information
Entropy carries out interal separation；
Discretization unit, for carrying out discretization to the image information decision table, and calculates the forward and backward shadow of discretization respectively
As the Indiscernible relation diversity factor of information table；
Assessment unit, for being assessed according to the Indiscernible relation diversity factor the result of discretization, selection is most
Excellent discretization scheme.
Compared with prior art, beneficial effects of the present invention：
Firstly, since the present invention is the interal separation modelling technique based on comentropy, the division of discrete segment is built upon
Carried out on the basis of the information gain of attribute independent, and existing other methods be substantially foundation in the generic attribute degree of correlation and
Carried out on the basis of similarity between intervals, by experimental data below it can be confirmed that the method for the present invention can reduce discretization
Section number；Secondly, when being selected to Candidate point, calculated by section entropy, avoid the redundancy of block information
Property, improve the calculating accuracy rate that subsequent classification prediction is handled；Finally, due to it is the calculating based on Indiscernible relation, therefore
The data distortion problem after discretization is alleviated to a certain extent, and this method is concentrated use in more long, acquisition in data
Discretization scheme it is more excellent.
Brief description of the drawings
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be to needed in the embodiment attached
Figure is briefly described, it will be appreciated that the following drawings illustrate only certain embodiments of the present invention, therefore be not construed as pair
The restriction of scope, for those of ordinary skill in the art, without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings.
Fig. 1 is the flow chart of the Algorithm for Discretization of Remote Sensing Image Attribute of the invention based on comentropy.
Fig. 2 divides schematic diagram for approximate set.
Fig. 3 is the comprehensive schematic diagram of approximate set.
Fig. 4 is the performance comparison figure of several algorithms.
Fig. 5 is the influence schematic diagram of Indiscernible relation diversity factor.
Fig. 6 analyzes schematic diagram for data consistency.
Fig. 7 is the composition frame chart of the characteristics of remote sensing image discretized system of the invention based on comentropy.
Embodiment
Below in conjunction with attached drawing in the embodiment of the present invention, the technical solution in the embodiment of the present invention is carried out clear, complete
Ground describes, it is clear that described embodiment is only part of the embodiment of the present invention, instead of all the embodiments.Usually exist
The component of the embodiment of the present invention described and illustrated in attached drawing can be arranged and designed with a variety of configurations herein.Cause
This, the detailed description of the embodiment of the present invention to providing in the accompanying drawings is not intended to limit claimed invention below
Scope, but it is merely representative of the selected embodiment of the present invention.Based on the embodiment of the present invention, those skilled in the art are not doing
Go out all other embodiments obtained on the premise of creative work, belong to the scope of protection of the invention.
Referring to Fig. 1, the Algorithm for Discretization of Remote Sensing Image Attribute based on comentropy provided in the present embodiment, including with
Lower step：
S101, by carrying out radiation calibration, atmospheric correction, sample selection processing to remote sensing image, is listed in remote sensing image
Pixel value and generic of the sample in each wave band are selected, and establishes the image information based on pixel value and generic and determines
Plan table.
In the present embodiment, specifically, this step can be realized in the following ways：
Training dataset of the significant region of several block features as discretization is selected from remote sensing image, is denoted as TDS；Together
When, wave band attribute is isolated in TDS, a newly-increased row are used as category attribute, and respectively every piece of region is labeled.It is right herein
Remote sensing images carry out information extraction, isolate the method for the various information such as geographic coordinate information, wave band attribute can use it is existing
Technology realizes that this discrete method does not improve this, therefore to simplify description, this is not stated carefully herein.
Multiple samples are chosen by image sampling, and obtain the pixel value of each selected sample, and at each
Corresponding Land cover types are marked in sample, all samples form sample set.For example, according to area marking in region
Sample assignment in category attribute row；Then, the wave band attribute of each sample is selected in the much information obtained after TDS separation
With category attribute (other information give up should not), merge wave band attribute and category attribute, to build decision table matrix D S；Most
Afterwards, the value of category attribute column is ranked up sample, sample set of the generation with pixel value and land type label.
According to the sample set with pixel value and land type label, establish using wave band as conditional attribute, with type of ground objects
For the decision table matrix of decision attribute.Matrix pattern is as follows：
Each of which row represents 1 sample item, sample set U={ u1,u2,...,un, i.e., sample is concentrated with n sample, with
Wave band is the attribute column C={ c of conditional attribute1,c2,...,cm, represent pixel value of the sample in m wave band, last row is
Attribute column D using type of ground objects as decision attribute, identifies the classification information of sample.Each sample item is by sample sequence number, wave band
Attribute and category attribute composition.The value range of wave band attribute is 0≤cij≤ 1, wherein cijIt is i-th of sample in j-th of wave band
Pixel value；Category attribute value is represented with nature numerical value, and value range is determined according to the classification number of definition, such as：Assuming that definition
Classification number be 5, then value range is D={ 1,2,3,4,5 }.
S102, the section based on comentropy separate model computation interval comentropy, and interal separation is carried out based on comentropy.
In the present embodiment, specifically, this step can be accomplished by the following way：
A) according to the status information in current all sections, (status information refers to maximum, minimum value in the entropy table of section
And entropy) establish or improve section entropy table；
In order to be quickly found the subinterval of entropy maximum, it is necessary to establish a table to record the entropy in current all sections
Value, i.e. section entropy table (IET).For IET altogether comprising 3 row, the 1st row preserve the minimum value in section, and the 2nd row preserve the maximum in section,
3rd row are corresponding entropy obtained by calculation, as shown in table 1.
Table 1 section entropy table (IET) structure
Every a line in table corresponds to a subinterval, originally, a line is comprised only in IET, i.e., whole connection attribute section, with
The operation of algorithm, starts to divide (b below) step), by increasing new a line and renewal when predivision section position
The minimum and maximum value in two sections, realizes preservations of the IET to all sections.
B) current all corresponding comentropies in section are calculated, judge whether the comentropy in each section is more than given threshold,
If it is the section is continued to split, until the entropy in the section is less than given threshold, updates section entropy table.
Sample size according to contained by section, calculates the comentropy in section in above-mentioned IET tables successively, and the calculating of comentropy is public
Formula is as follows：
Wherein, S is object set；K is classification number；CiExpression classification in object set S is the example number of i；P
(Ci, S) and represent the probability that classification i occurs in object set S, i.e. and the example number of class i accounts for the percentage of whole object set S
Than.
Try to achieve in IET tables after the comentropy in all sections, arranged from small to large according to entropy.Every time since last
Search entropy it is maximum can by stages, can by stages refer to including at least two breakpoints (upper bound in section and lower bound are unequal)
And entropy is more than the section of given threshold value.
A), b) process can be merged understanding, that is, that completes section entropy table establishes process, if the entropy of current interval
More than given threshold, then current interval is split, obtain two sections, and update the maximum in two sections, minimum
Value, threshold value, then the entropy in each section is calculated, and progress interal separation is determined whether according to entropy, so constantly divided
Cut, until the entropy in all sections is respectively less than or equal to given threshold.
C) the segmentation section of each wave band is represented with Candidate point successively, and preserved with a matrix type.
Section can be represented using the lower bound in section as break value, the Candidate point of all wave bands is protected with a matrix type
Deposit, as shown in table 2.
The Candidate point matrix of 2 wave band of table
Wherein, BandjRepresent j-th of wave band, breakpointijRepresent i-th of Candidate point value in j-th of wave band,
totalkRepresent the Candidate point number that k-th of wave band includes.
S103, discretization is carried out to the raw video information decision table obtained in step S101, and calculates discretization respectively
The Indiscernible relation diversity factor of forward and backward image information table.
Specifically, in this embodiment, this step can be accomplished by the following way：
1) discretization is carried out to raw video information decision table using the Candidate point of each wave band successively.
For example, for wave band j, if the pixel value bandvalue of i-th of sampleiK-th of section of this wave band is fallen into,
That is, Lk≤bandvaluei≤Uk, then, and bandvaluei=Lk=breakpointkj.Wherein, breakpointkjIt is in wave band j
K-th of Candidate point being sorted in ascending order, LkAnd UkIt is lower bound and the upper bound in the segmentation section that this Candidate point represents respectively.Will
All samples, by aforesaid operations, complete the sliding-model control to raw information decision table in each wave band.
2) lower and upper approximations on decision attribute of the forward and backward image information decision table of discretization are solved respectively.
Given decision table S=(U, R, V, f), wherein, U is limited object set, i.e. domain, and R is attribute set, bag
Collection containing conditional attribute C and decision kind set D.Binary crelation can not be differentiated：For each attribute setDefinition can not
Binary crelation IND (A) is differentiated, i.e.,：
Equivalence class in domain U on attribute set A：
To each subsetAnd the equivalence class in domain U on attribute set A, the lower and upper approximations of X respectively by
It is defined as follows：
Here A=C, X ∈ U are taken | IND (d), d are the decision attribute of decision kind set D, before being calculated discretization respectively,
The lower aprons collection C of each decision attribute values is corresponded in the decision attribute d of image information decision table afterwards-(dX), C-(dX) ' and it is upper
Approximate set C-(dX), C-(dX) ', C-(dX) be discretization before lower aprons collection, C-(dX) ' be discretization after lower aprons collection, C-
(dX) be discretization before upper approximate set, C-(dX) ' be discretization after upper approximate set.
3) the Indiscernible relation diversity factor of the forward and backward image information decision table of discretization is calculated.
, can be in the hope of the difference number N of the forward and backward lower and upper approximations of discretization according to above-mentioned definitiond=Nl+Nu, wherein, Nl
For the difference number of lower aprons collection, NuFor the difference number of upper approximate set, Nd, NlAnd Nu0 is initialized as respectively.Discretization is forward and backward
The difference number of lower and upper approximations is the Indiscernible relation diversity factor of the forward and backward image information decision table of discretization.NdCalculating
Step is as follows：
Step 1：Element d is chosen from the decision attribute d of raw video information decision tablei(i=1,2 ..., n), its
In, n is the number of different values of the decision attribute d on domain U, i.e. classification number；
Step 2：Calculate diLower aprons collection C-(di) and upper approximate set C-(di), if also having element not in decision attribute d
Calculated, then return to Step 1 and continue to execute, otherwise, into next step Step 3；
Step 3：With the IET ultimately generated to raw video information decision table discretization, new image information decision-making is obtained
Table SE=(U, R, VE,fE)；
Step 4：Element d is chosen from the decision attribute d of new image information decision tablei(i=1,2 ..., n), calculate
diLower aprons collection C-(di) ' and upper approximate set C- (di)'；
Step 5：The forward and backward d of discretization is judged respectivelyiLower and upper approximations it is whether equal, if C-(di)'≠C-(di),
Then Nl=Nl+ 1, if C- (di)'≠C-(di), then Nu=Nu+1；If also having element not calculated in decision attribute d, return
Step 4 is returned to continue to execute, otherwise, Nd=Nl+Nu, terminate.
S104, assesses the result of discretization according to the Indiscernible relation diversity factor, selects optimal discrete
Change scheme.
Wherein, this step assessment is carried out to the result of discretization can be in the following way：
The deliberated index of the result of discretization is essentially from nicety of grading, the uniformity of section number and sample, and classifies
Precision and the uniformity of sample depend on the size that Indiscernible relation changes degree again, section number then by the threshold value that sets and
Iterations controls.Based on the Indiscernible relation diversity factor being calculated in step 103, different threshold values and iterations are set
The discretization results fixed score, and formula is as follows：
Wherein, NdFor the Indiscernible relation diversity factor of the forward and backward image information decision table of discretization, | d | it is the class of sample
Other number, each classification correspond to a pair of of lower and upper approximations；It is for the Candidate point sum of all wave bands, i.e., discrete
The section number obtained after change,For the breakpoint sum of raw video information decision table.
Select optimal discretization scheme to recommend targeted customer described in this step, can by the different threshold values of setting and
Discretization assessment result under iterations is ranked up, and selects the highest discretization scheme of score value to recommend targeted customer.
The step of Algorithm for Discretization of Remote Sensing Image Attribute of the present invention based on comentropy is described above, discretization of the present invention
Method has the redundancy that can be reduced discretization section number, avoid block information, alleviates discretization data distortion problem.Under
Face compared with the method for the present invention, will be verified with several discretization methods of current mainstream in terms of theoretical and test data two
The superiority of the method for the present invention.
1) generic information calculates
There is the discretization algorithm of supervision due to the calculating based on generic information, have and make full use of class label and target
The superiority of attribute information, can be more easily found suitable breakpoint location than unsupervised discretization algorithm.Have supervision from
Dispersion method has very much, wherein more typical generic information calculation formula mainly has following three kinds.
The a generic degrees of correlation
Wherein, CAIR is class-attribute relevant redundancy degree, and I (C, D | F) represents class-attribute of generic attribute C and discrete division D
Shared mutual information, H (C, D | F) represent the combination entropy between generic attribute C and discrete division D.
B similarity between intervals
Wherein, χ2Represent the departure degree between observed value and theoretical value, AijRepresent the sample for belonging to class j in the i-th section
This number, EijRepresent the expecterd frequency to the class j in the i-th section.
C information gain degrees
Wherein, Gain (S) is the information gain amount before and after interval division, and Ent (S) represents the information content of section S, S1And S2
Represent respectively by two sections of breakpoint division generation；| S | represent the example number that section S is included.
The traditional generic information computational methods of above three have more complete theoretical foundation.Calculation based on the generic degree of correlation
Method mainly has CADD, CAIM and CACC etc..CADD algorithms need user to specify interval number in initialization, and CAIM need not be pre-
First set interval number, and make certain improvements in the calculating of degree of correlation standard.Although CAIM compensate for some of CADD
Deficiency, but since warp is frequently with interval number approximate target attribute value number, and for a section, the discrete differentiation provided
Formula only considers that the section includes the generic attribute of most samples, have ignored the distribution situation of other generic attributes, eventually produces excessive
Section and cause over training, cause overfitting.CACC is similar with CAIM, and simply discretization evaluation criterion is repaiied
Change, determine whether section needs to split by the coefficient of association cacc of computation interval, and replace in CAIR standards using log (n)
N, so as to compensate for the deficiency of CAIM.Although CACC can accelerate the process of discretization, overfitting is prevented, to single category
Performance obtains the preferable breakpoint of quality, but since CACC accepted standards are to maximize the generic degree of correlation, only to single attribute
Realize that Interval Discrete result is optimal, lack the description to data Global Information, also do not account for the uniformity of data before and after the processing,
It will necessarily cause the loss of initial data important information amount.
ChiMerge, Chi2 and Extended Chi2 scheduling algorithms employ the generic information based on similarity between intervals and calculate
Method carries out differentiation merging to adjacent interval.These methods apply the Pearson came statistic in statistics to differentiate current breakpoint
Whether should be removed, i.e., adjacent with the breakpoint two interval whether the merging.Can by the Pearson came theorem in mathematical statistics
Know, statistic χ2Progressive distribution be the free degree be k-1 χ2Distribution, as given level of signifiance α, it may be determined that corresponding critical
Value χα.ChiMerge is using fixed level of signifiance α, and Chi2 is then constantly to decline level of signifiance α, so that critical value
χαConstantly increase, passes through mergingMaximum Candidate point, improves the utilization rate of computational efficiency and information.But
It is that Chi2 uses total classification number that fixed free degree v=k-1, wherein k are systems.In fact, the selection of the free degree should
The classification number that the foundation section is associated, total classification number without that should be system, therefore, ExtendedChi2 algorithms are in Chi2
On improved, useInstead of in Chi2 algorithmsCalculated.However, Extended Chi2 are calculated
Used in methodAlthough the discrimination standard as breakpoint importance is than usingIt is rational more, but be a lack of
Corresponding theoretical foundation, at the same it is computationally also inaccurate.
The most common discretization method calculated using information gain generic information, is based on most short description length principle
(MDLP) method.It is after connection attribute according to value size sequence, and the border between different target class is set to candidate's division points,
Then being found out in candidate's division points makes the division border T of information gain degree maximum as two points of divergent boundaries, successively iteration time
Definite potential division points (Potential cut point) are gone through until meeting minimum description length principle.The algorithm is due to using
The measurement standard of information gain, for more above-mentioned algorithm, can largely ensure the uniformity of section internal specimen generic,
The situation that suitable Target Attribute values are evenly distributed.But in remote sensing image processing, the sample taken often be subject to noise and
The influence of other impurity, Target Attribute values be distributed after sequence it is still more scattered, to each attribute need carry out N-1 times
Information gain degree calculates, and wherein N is the number of samples with different attribute value, therefore, when data scale is very big, can be caused non-
Often big time overhead.
2) approximately equivalent model up and down
Information system is four-tuple (U, Q, V, f).Wherein, U is object set, also referred to as domain；Q is attribute set, bag
Collection containing conditional attribute C and decision kind set D；V is the codomain of attribute；F is mapping function, represents that the attribute of object is mapped to attribute
Some value of codomain.Knowledge is referred to as according to the ability that the characteristic difference of things is classified.It can not be differentiated each other in domain
Object composition set be form knowledge particle (granule).Knowledge has granularity, and the smaller explanation of granularity can accurate table
The concept reached is more, and granularity uses Indiscernible relation, i.e. equivalence class is described, and is the least unit for representing knowledge.
In assorting process, the individual subject being not much different is attributed to same class, and the relation formed between them is exactly not
Distinguishable relation, then be defined as below：
If IND (P)={ (x, y) ∈ U × U:F (x, a)=f (y, a), a ∈ P }, be on any one attribute set P can not
Resolution relation.Judge whether an object a belongs to set X according to the Indiscernible relation cluster in knowledge base, three kinds of situations can be divided：
A necessarily belongs to set X；A is necessarily not belonging to set X；A, which may belong to, may also be not belonging to set X.So, set X is on can not
Differentiate the lower aprons (Lower approximation) of relations IIt is exactly according to I by those
Judgement belongs to the maximum set that the object of X is formed certainly, also referred to as the positive domain (positive region) of X, is denoted as POS
(X)；Upper approximate (Upper approximations) of the set X on Indiscernible relation IIt is by the union of all equivalence class I (X) for intersecting non-NULL with X, is that those may belong to
In the minimal set that the object of X forms.If approximation set equality up and down, X is a precise set, is otherwise one coarse
Set, wherein lower aprons are known as the positive region of the concept, up and down it is approximate it is poor be referred to as border, it is upper it is approximate beyond region be known as
Negative region (Negative region), is denoted as NEG (x), and the model thus established is known as approximately equivalent model up and down.
It is assumed that information system S=(U, Q, V, f), there is two attribute Q1And Q2, attribute Q1There are 5 values, attribute Q2There are 6 values.
Given approximation set X, uses attribute Q1And Q2Divided, situation is as shown in Figure 2.
Such as Fig. 2, attribute Q1Domain U 5 row, attribute Q have been divided into2Domain U 6 rows, Q have been divided into1×Q2Constitute one
A two dimensional character, is divided into 30 lattices by domain U, that is, generates 30 knowledge particles altogether.Circle represents approximate set X,
The region of lattice composition disjoint with circle is the negative region on X, its expression formula is：
NEG (X)=U-I-(X) (11)
Lower aprons I-(X) and upper approximation I-(X) extracted respectively from the knowledge particles of division, as shown in Figure 3.Can
To see, the region with five-pointed star mark is the positive region on X, and the region with No. # mark is the frontier district of approximate set X
Domain, its expression formula are：
BND (X)=I-(X)-I-(X) (12)
The concept such as lower aprons, upper approximate and borderline region carves the approximation properties for having changed an ambiguous set in border.Coarse journey
Degree is calculated as follows：
In formula | I- (X) | and | I- (X) | the radix or gesture of lower aprons collection and upper approximate set are represented respectively, that is, are wrapped in gathering
The element number contained.Obvious 0≤α1(X)≤1, if α1(X)=1, then set X is claimed relative to Indiscernible relation I to be clear
, if α1(X) ＜ 1, then title X is coarse relative to I.α1(X) it is regarded as approaching set X's under Indiscernible relation I
Precision.
Ambiguous set can not judge whether some elements belong to this without clearly border according to existing knowledge in domain
Set.In rough set, uncertain concept is for element is under the jurisdiction of the degree of set.By using pass can not be differentiated
System, it is as follows to the coarse membership function of set X, formula under equivalence relation I can to define element x：
Indiscernible relation changes the importance for the conditional attribute set C that degree is actually reflected under decision attribute D
Change, i.e. decision attribute set D is formulated as follows the degree of dependence of conditional attribute set C：
Wherein, POSc(D) it is positive regions of the attribute set C in U/IND (D).As can be seen from the above equation, degree of dependence is just
It is ratio of the lower aprons set element with whole domain radix of the corresponding equivalence class of all categories value in decision attribute D.
Since each approximation set X correspond to a lower aprons collection I of given equivalence relation I-(X) and one is gone up approximate set I-
(X), the change of degree of dependence means that approximate set is also changed, and can replace asking by calculating the variable number of approximate set
The ratio of positive domain and domain radix carrys out the intensity of variation of reaction information system.In fact, the change of information system is exactly by knowledge
Caused by the equivalence relation in storehouse changes, i.e., directly portray the I of set X-(X) changed with I- (X).And γC(D)
Simply reflect the ratio of element number, identical γC(D) different I may be correspond to-(X) and I-(X) change number, because
This, directly uses I-(X) and I-(X) change number reflects that the intensity of variation of information system is relatively reasonable.According to above-mentioned analysis,
Construction solves the function f of Indiscernible relation diversity factor:
I. the input of function is two decision tables before and after discretization, is represented respectively with DT and NDT；
Ii. all approximate sets, and being compared up and down of DT and NDT on decision attribute are obtained respectively；
Iii. the difference number of lower aprons collection and the difference number of upper approximate set that the output of function is DT and NDT.
Diversity factor function f reflects the intensity of variation of knowledge system, and lower aprons set and the upper approximate number gathered are by sample
The value number of this class number, i.e. decision attribute determines.
The principle of the present invention is that the change for dividing front and rear information gain according to section decides whether to more than threshold value
Section continues to divide, and entropy, the linear module as information uncertainty can reflect the steady of section generic well
Determine degree.The knowledge base of information system is made of internal equivalence relation at the same time, so, closed using the equivalence of decision table
System's change is to reflect the otherness between information system, and utilizing the relation diversity factor calculated, simultaneously combining information gain-splitting is calculated
Method, the final discretization scheme for obtaining image feature.The method of the present invention introduces in the section fission process larger to entropy
The equivalence model of rough set is compared calculating to the decision table before and after discretization, by the difference number of obtained approximate set up and down
Mesh knows the change degree of Indiscernible relation, so as to obtaining optimal discretization interval number.
When being verified by experiment to the method for the present invention, it is South Sea inshore region to test the test data set used
A width GF-2 satellite images, which includes 4 wave bands, the atural object on image is divided into settlement place in experiment, forest land is naked
Ground, arable land and five major class of water body.Local sampling is carried out to imagery zone, selects the significant imagery zone of several block features as discrete
The training dataset of change, the sample set with pixel value and class label finally obtained include 8379 samples, 4 wave bands altogether
Initial breakpoint number be 6328,6596,6720 and 6611 respectively, sum be 26255.
The visualization of this experiment, programs, emulation, and test and numerical computations processing are in MATLAB (R2016a versions) environment
Lower realization；The radiation calibration of image, air calibration, and comparison of classification is carried out in ENVI5.3 rings to the result before and after discretization
Completed under border.First by carrying out radiation calibration to remote sensing image, atmospheric correction, sample selection processing, lists in image and is chosen
Sample and establishes the information model of image, i.e. image information decision table in the pixel value and generic of each wave band.Image information
Decision table is altogether comprising 5 row, and the 1st is classified as the numbering of sample, and the 2nd to the 5th is classified as wave band attribute, i.e. conditional attribute, and the 6th is classified as classification
Attribute, i.e. decision attribute.This input data of image information decision table as discretization algorithm performs, every time with one-dimension array
Form is exported as a result, last, and all results are preserved in the form of two-dimensional matrix, matrix totally 5 row, the 1st to 4 row respectively from
The breakpoint number of each wave band after dispersion, the 5th is classified as Indiscernible relation difference number.
In order to verify the validity for proposing algorithm, experiment is accurate by section number, Indiscernible relation diversity factor, data
Rate and run time this several respect analyze implementing result.And there is supervision with the former algorithm based on entropy and current main-stream
Discretization algorithm 1R, ChiMerge are compared.
Figure 4, it is seen that the run time of four kinds of algorithms is similar.Since classification number is set as 5 major classes, so shadow
As information decision table on the lower and upper approximations of category attribute sum be 10.Although former algorithm and side of the present invention based on entropy
Method obtains almost consistent breakpoint number, but since Indiscernible relation difference number is 8, the number obtained than the method for the present invention
6 is high, and therefore, the accuracy 78.5% classified is similar 10 percentage points low compared with the method for the present invention 90.8%.
The Indiscernible relation difference number of ChiMerge algorithms is consistent with the method for the present invention, but is obtained more than the method for the present invention
Breakpoint number, and the accuracy classified also is slightly less than the method for the present invention.1R algorithms are relatively other due to the process in splitting section
Algorithm is simple, and the breakpoint number of minimum is achieved in 4 wave bands, but the difference number of Indiscernible relation is maximum, reaches 10,
Mean that the distortion level of decision table maximizes, it is clear that the accuracy rate of classification has also reached minimum 53.6% in four algorithms.
Performance Evaluation is carried out to four kinds of algorithms using DME formula, the assessed value for trying to achieve four kinds of algorithms is respectively：0.2896,0.1441,
0.1828 and 0.Where it can be seen that the overall performance of the method for the present invention is more preferably.
By the analysis to Fig. 4, while accuracy rate generation of the Indiscernible relation diversity factor to classification of drawing a conclusion is important
Influence, i.e. classification accuracy under different relation diversity factoies is different, and is reduced with the increase of difference number.
From figure 5 it can be seen that by varying the discretization flow iterations of the method for the present invention threshold different with setting
Value, can obtain different relation diversity factoies, the value in [0,2,4,6,8,10] scope.Number of breakpoints is with difference number
Reduce and increase, when heteromerism mesh of being on duty is 0, the number of breakpoints and original breakpoint sum that discretization obtains are basically identical, occur
Over-fitting.Calculated with crossing, obtain the DME values under this 6 diversity factoies, be respectively 0.0251,0.0631,0.1253,
0.2896,0.1629 and 0.As can be seen that when diversity factor is 6, the DME values of maximum are obtained.Draw a conclusion, although diversity factor reduces
The compatibility of system after discretization can be kept, but also brings breakpoint enormous amount, over-fitting occurs, obtained DME values
It is relatively low.Therefore, the method for the present invention can therefrom obtain optimum discretization scheme under balance number of breakpoints and relation diversity factor.
Mainly the uniformity of data after each wave band discretization is analyzed in figure 6.Here uniformity refers to
The accuracy rate of sample, i.e., after discretization, the sample in same discrete segment has identical category attribute.As can be seen that
As difference number constantly increases, the accuracy rate of sample gradually reduces in each wave band.Wave band 1 difference number in [0,2,4,
6,8,10] during value, the accuracy rate of sample is respectively [0.985 0.965 0.913 0.895 0.767 0.665]；Wave band 2
Sample accuracy rate is [0.963 0.957 0.935 0.925 0.793 0.632]；The sample accuracy rate of wave band 3 is [0.956
0.924 0.906 0.894 0.802 0.763]；The sample accuracy rate of wave band 4 is [0.972 0.943 0.883 0.856
0.757 0.602].This illustrates that the change of decision table Indiscernible relation can produce large effect to data consistency.And
Experiment simulation and comprehensive analysis more than demonstrate the validity of the method for the present invention, and more other algorithms can obtain less area
Between the classification accuracy of number and higher, while also alleviate the data distortion problem of system to a certain extent, improve discrete
Change scheme recommends efficiency and the scalability of algorithm.
Referring to Fig. 7, the present invention gives the characteristics of remote sensing image discretized system based on comentropy at the same time, in the system
Including：
Image information decision table establishes unit, and pixel value of the sample in each wave band is selected in remote sensing image for listing
And generic, and establish the image information decision table based on pixel value and generic；
Comentropy computing unit, separates model computation interval comentropy for the section based on comentropy, and is based on information
Entropy carries out interal separation；
Discretization unit, for carrying out discretization to the image information decision table, and calculates the forward and backward shadow of discretization respectively
As the Indiscernible relation diversity factor of information table；
Assessment unit, for being assessed according to the Indiscernible relation diversity factor the result of discretization, selection is most
Excellent discretization scheme.
Wherein, in the present embodiment, image information decision table is established unit and is specifically included：Sample choose module, for from
Multiple samples are chosen in remote sensing image；Sample set establishes module, for obtaining each selected sample by image sampling
Pixel value, and corresponding Land cover types are marked in each sample, all samples form sample set；Decision table square
Battle array establishes module, for according to the sample set with pixel value and land type label, establishing using wave band as conditional attribute, with ground
Species type is the decision table matrix of decision attribute.
Comentropy computing unit specifically includes：Section entropy table establishes module, for the state letter according to current all sections
Breath establishes section entropy table；Comentropy computing module, for calculating current all corresponding comentropies in section, is obtained based on comentropy
The position in next section to be split；Preserving module, for successively representing in the segmentation section of each wave band simultaneously with Candidate point
Preserve with a matrix type.
Discretization unit specifically includes：Descretization module, for utilizing the Candidate point of each wave band successively to the shadow
As information decision table carries out discretization；Approximate set asks for module, for image information table before solving discretization respectively on decision-making
Upper approximate set, the lower aprons collection of attribute, and upper approximate set, lower aprons collection of the image information table on decision attribute after discretization；
Diversity factor computing module, for calculating the Indiscernible relation diversity factor of the forward and backward image information table of discretization respectively.
Only the characteristics of remote sensing image discretized system based on comentropy is briefly described herein, is described in more detail
It may refer to the associated description in the above-mentioned Algorithm for Discretization of Remote Sensing Image Attribute based on comentropy.
The above description is merely a specific embodiment, but protection scope of the present invention is not limited thereto, any
Those familiar with the art the invention discloses technical scope in, change or replacement can be readily occurred in, should all be contained
Cover within protection scope of the present invention.
Claims (10)
1. a kind of Algorithm for Discretization of Remote Sensing Image Attribute based on comentropy, it is characterised in that comprise the following steps：
List and pixel value and generic of the sample in each wave band are selected in remote sensing image, and establish and be based on pixel value and institute
Belong to the image information decision table of classification；
Section based on comentropy separates model computation interval comentropy, and interal separation is carried out based on comentropy；
Discretization is carried out to the image information decision table, and calculates can not differentiating for the forward and backward image information table of discretization respectively
Relation diversity factor；
The result of discretization is assessed according to the Indiscernible relation diversity factor, selects optimal discretization scheme.
2. according to the method described in claim 1, it is characterized in that, described list is selected sample in each ripple in remote sensing image
The pixel value and generic of section, and the image information decision table based on pixel value and generic is established, including step：
Multiple samples are chosen from remote sensing image；
The pixel value of each selected sample is obtained by image sampling, and corresponding soil is marked in each sample
Cover type, all samples form sample set；
According to the sample set with pixel value and land type label, establish using wave band as conditional attribute, using type of ground objects as certainly
The decision table matrix of plan attribute.
3. according to the method described in claim 1, it is characterized in that, the section based on comentropy separates model computation interval
Comentropy, interal separation, including step are carried out based on comentropy：
Section entropy table is established according to the status information in current all sections；
Current all corresponding comentropies in section are calculated, judge whether the comentropy in each section is more than given threshold, if
Then the section is continued to split, until the entropy in the section is less than given threshold, update section entropy table；
The segmentation section of each wave band is represented with Candidate point successively and is preserved with a matrix type.
4. according to the method described in claim 3, it is characterized in that, it is described to the image information decision table carry out discretization,
And the Indiscernible relation diversity factor of the forward and backward image information table of discretization is calculated respectively, including step：
Discretization is carried out to the image information decision table using the Candidate point of each wave band successively；
Upper approximate set, lower aprons collection of the image information table on decision attribute before solution discretization respectively, and image after discretization
Upper approximate set, lower aprons collection of the information table on decision attribute；
The Indiscernible relation diversity factor of the forward and backward image information table of discretization is calculated respectively.
5. according to the method described in claim 1, it is characterized in that, it is described according to the Indiscernible relation diversity factor to discrete
The result of change is assessed, and selects optimal discretization scheme, including step：
Scored using equation below the discrete results under different set threshold value and iterations：Wherein, NdFor the forward and backward image information table of discretization
Indiscernible relation diversity factor, | d | be the classification number of sample, each classification correspond to a pair of of lower and upper approximations；It is total for the Candidate point of all wave bands,Breakpoint for raw video information table is total
Number, m is wave band number, and i, j, m is integer；
Choose the highest discretization scheme of score value.
A kind of 6. characteristics of remote sensing image discretized system based on comentropy, it is characterised in that including：
Image information decision table establishes unit, and pixel value and institute of the sample in each wave band are selected in remote sensing image for listing
Belong to classification, and establish the image information decision table based on pixel value and generic；
Comentropy computing unit, for based on comentropy section separate model computation interval comentropy, and based on comentropy into
Row interal separation；
Discretization unit, for carrying out discretization to the image information decision table, and calculates the forward and backward image letter of discretization respectively
Cease the Indiscernible relation diversity factor of table；
Assessment unit, for being assessed according to the Indiscernible relation diversity factor the result of discretization, selects optimal
Discretization scheme.
7. system according to claim 6, it is characterised in that image information decision table, which establishes unit, to be included：
Sample chooses module, for choosing multiple samples from remote sensing image；
Sample set establishes module, for obtaining the pixel value of each selected sample by image sampling, and at each
Corresponding Land cover types are marked in sample, all samples form sample set；
Decision table matrix establishes module, for according to the sample set with pixel value and land type label, establish using wave band as
Conditional attribute, the decision table matrix using type of ground objects as decision attribute.
8. system according to claim 6, it is characterised in that comentropy computing unit includes：
Section entropy table establishes module, for establishing section entropy table according to the status information in current all sections；
Comentropy computing module, for calculating current all corresponding comentropies in section, judge each section comentropy whether
More than given threshold, if it is the section is continued to split, until the entropy in the section is less than given threshold, update section
Entropy table；
Preserving module, for being represented the segmentation section of each wave band with Candidate point and being preserved with a matrix type successively.
9. system according to claim 8, it is characterised in that discretization unit includes：
Descretization module, discretization is carried out for the Candidate point successively using each wave band to the image information decision table；
Approximate set asks for module, for solving the upper approximate set, lower near of image information table before discretization on decision attribute respectively
Like upper approximate set of the image information table after collection, and discretization on decision attribute, lower aprons collection；
Diversity factor computing module, for calculating the Indiscernible relation diversity factor of the forward and backward image information table of discretization respectively.
10. system according to claim 6, it is characterised in that assessment unit is specifically used for：Using equation below to difference
Discrete results under given threshold and iterations score, and choose the highest discretization scheme of score value：
Wherein, NdFor the forward and backward image letter of discretization
The Indiscernible relation diversity factor of table is ceased, | d | it is the classification number of sample, each classification correspond to a pair of of lower and upper approximations；It is total for the Candidate point of all wave bands,Breakpoint for raw video information table is total
Number, m is wave band number, and i, j, m is integer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711311000.3A CN107909062B (en) | 2017-12-11 | 2017-12-11 | Information entropy-based remote sensing image feature discretization method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711311000.3A CN107909062B (en) | 2017-12-11 | 2017-12-11 | Information entropy-based remote sensing image feature discretization method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107909062A true CN107909062A (en) | 2018-04-13 |
| CN107909062B CN107909062B (en) | 2021-05-07 |
Family
ID=61865137
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711311000.3A Active CN107909062B (en) | 2017-12-11 | 2017-12-11 | Information entropy-based remote sensing image feature discretization method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107909062B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109076761A (en) * | 2018-08-08 | 2018-12-25 | 安徽省现代农业装备产业技术研究院有限公司 | One kind being based on remote sensing images variable fertilization method |
| CN112580483A (en) * | 2020-12-15 | 2021-03-30 | 海南大学 | Remote sensing image characteristic discretization method based on rough fuzzy model |
| CN113743464A (en) * | 2021-08-02 | 2021-12-03 | 昆明理工大学 | Continuous characteristic discretization loss information compensation method and application thereof |
| CN117075884A (en) * | 2023-10-13 | 2023-11-17 | 南京飓风引擎信息技术有限公司 | Digital processing system and method based on visual script |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1652146A1 (en) * | 2003-07-21 | 2006-05-03 | Tubitak-Bilten ( Turkiye Bilimsel Ve Teknik Arastirma Kurumu-Bilgi Teknolojileri Ve Elektronik Arastirma Enstitusu) | Implementation of the jpeg2000 compression algorithm in hardware |
| CN101702172A (en) * | 2009-11-10 | 2010-05-05 | 大连理工大学 | Data discretization method based on category-attribute relation dependency |
| CN102879677A (en) * | 2012-09-24 | 2013-01-16 | 西北工业大学 | Intelligent fault diagnosis method based on rough Bayesian network classifier |
| CN102930275A (en) * | 2012-10-29 | 2013-02-13 | 福州大学 | Remote sensing image feature selection method based on Cramer's V index |
| CN104573050A (en) * | 2015-01-20 | 2015-04-29 | 安徽科力信息产业有限责任公司 | Continuous attribute discretization method based on Canopy clustering and BIRCH hierarchical clustering |
| CN105139021A (en) * | 2015-07-08 | 2015-12-09 | Tcl集团股份有限公司 | Method and system for realizing television user rapid classification based on rough set theory |
| CN106971204A (en) * | 2017-03-31 | 2017-07-21 | 上海海洋大学 | A kind of dimension reduction method of the new high-spectrum remote sensing data based on rough set |
| TWI599896B (en) * | 2016-10-21 | 2017-09-21 | 嶺東科技大學 | Multiple decision attribute selection and data discretization classification method |
-
2017
- 2017-12-11 CN CN201711311000.3A patent/CN107909062B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1652146A1 (en) * | 2003-07-21 | 2006-05-03 | Tubitak-Bilten ( Turkiye Bilimsel Ve Teknik Arastirma Kurumu-Bilgi Teknolojileri Ve Elektronik Arastirma Enstitusu) | Implementation of the jpeg2000 compression algorithm in hardware |
| CN101702172A (en) * | 2009-11-10 | 2010-05-05 | 大连理工大学 | Data discretization method based on category-attribute relation dependency |
| CN102879677A (en) * | 2012-09-24 | 2013-01-16 | 西北工业大学 | Intelligent fault diagnosis method based on rough Bayesian network classifier |
| CN102930275A (en) * | 2012-10-29 | 2013-02-13 | 福州大学 | Remote sensing image feature selection method based on Cramer's V index |
| CN104573050A (en) * | 2015-01-20 | 2015-04-29 | 安徽科力信息产业有限责任公司 | Continuous attribute discretization method based on Canopy clustering and BIRCH hierarchical clustering |
| CN105139021A (en) * | 2015-07-08 | 2015-12-09 | Tcl集团股份有限公司 | Method and system for realizing television user rapid classification based on rough set theory |
| TWI599896B (en) * | 2016-10-21 | 2017-09-21 | 嶺東科技大學 | Multiple decision attribute selection and data discretization classification method |
| CN106971204A (en) * | 2017-03-31 | 2017-07-21 | 上海海洋大学 | A kind of dimension reduction method of the new high-spectrum remote sensing data based on rough set |
Non-Patent Citations (4)
| Title |
|---|
| FENGJIANG ET AL: "A novel approach for discretization of continuous attributes in rough set theory", 《KNOWLEDGE-BASED SYSTEMS》 * |
| STEPHEN D. BAY ET AL: "Multivariate discretization of continuous variables for set mining", 《KDD "00: PROCEEDINGS OF THE SIXTH ACM SIGKDD INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING》 * |
| 梁静等: "基于信息熵的遥感影像特征离散化方法", 《地理空间信息》 * |
| 王举范等: "基于信息熵的粗糙集连续属性多变量离散化算法", 《青岛科技大学学报（自然科学版）》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109076761A (en) * | 2018-08-08 | 2018-12-25 | 安徽省现代农业装备产业技术研究院有限公司 | One kind being based on remote sensing images variable fertilization method |
| CN109076761B (en) * | 2018-08-08 | 2021-01-08 | 安徽省现代农业装备产业技术研究院有限公司 | Variable fertilization method based on remote sensing image |
| CN112580483A (en) * | 2020-12-15 | 2021-03-30 | 海南大学 | Remote sensing image characteristic discretization method based on rough fuzzy model |
| WO2022127075A1 (en) * | 2020-12-15 | 2022-06-23 | 海南大学 | Feature discretization method for remote sensing image on the basis of rough fuzzy model |
| CN112580483B (en) * | 2020-12-15 | 2023-12-01 | 海南大学 | Remote sensing image feature discretization method based on rough fuzzy model |
| CN113743464A (en) * | 2021-08-02 | 2021-12-03 | 昆明理工大学 | Continuous characteristic discretization loss information compensation method and application thereof |
| CN113743464B (en) * | 2021-08-02 | 2023-09-05 | 昆明理工大学 | Continuous characteristic discretization loss information compensation method and application thereof |
| CN117075884A (en) * | 2023-10-13 | 2023-11-17 | 南京飓风引擎信息技术有限公司 | Digital processing system and method based on visual script |
| CN117075884B (en) * | 2023-10-13 | 2023-12-15 | 南京飓风引擎信息技术有限公司 | Digital processing system and method based on visual script |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107909062B (en) | 2021-05-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2019149147A1 (en) | Method for dividing ecological and geological environment types based on coal resource development | |
| CN107909062A (en) | A kind of Algorithm for Discretization of Remote Sensing Image Attribute and system based on comentropy | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| CN102930275B (en) | Based on the characteristics of remote sensing image system of selection of Cramer ' s V index | |
| CN103336801A (en) | Multi-feature locality sensitive hashing (LSH) indexing combination-based remote sensing image retrieval method | |
| CN108764527B (en) | Screening method for soil organic carbon library time-space dynamic prediction optimal environment variables | |
| CN111611960B (en) | Large-area ground surface coverage classification method based on multilayer perceptive neural network | |
| CN112087316B (en) | Network anomaly root cause positioning method based on anomaly data analysis | |
| CN107392887A (en) | A kind of heterogeneous method for detecting change of remote sensing image based on the conversion of homogeneity pixel | |
| Lunga et al. | Resflow: A remote sensing imagery data-flow for improved model generalization | |
| CN116129262A (en) | Cultivated land suitability evaluation method and system for suitable mechanized transformation | |
| Cheng et al. | Multi-view graph convolutional network with spectral component decompose for remote sensing images classification | |
| Zhang et al. | Geochemically constrained prospectivity mapping aided by unsupervised cluster analysis | |
| Nong et al. | Airborne LiDAR point cloud classification using PointNet++ network with full neighborhood features | |
| Wu et al. | Geo-parcel-based geographical thematic mapping using C5. 0 decision tree: A case study of evaluating sugarcane planting suitability | |
| CN117093890A (en) | Comprehensive evaluation method for ecological environment of energy resource development area | |
| Nguyen | Land cover change detection in northwestern Vietnam using Landsat images and Google Earth Engine | |
| Du et al. | An ensemble learning approach for land use/land cover classification of arid regions for climate simulation: A case study of Xinjiang, Northwest China | |
| CN115719453B (en) | Deep learning-based remote sensing extraction method for rice planting structure | |
| Roudier et al. | Soilscapes of New Zealand: Pedologic diversity as organised along environmental gradients | |
| Maurya et al. | Estimation of major agricultural crop with effective yield prediction using data mining | |
| Ji et al. | Multicascaded feature fusion-based deep learning network for local climate zone classification based on the So2Sat LCZ42 benchmark dataset | |
| Jiang et al. | Remote sensing recognition and classification of forest vegetation based on image feature depth learning | |
| Zhang et al. | A hierarchical growth method for extracting 3D phenotypic trait of apple tree branch in edge computing | |
| Li et al. | Feature-guided dynamic graph convolutional network for wetland hyperspectral image classification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |