CN105975973A - Forest biomass-based remote sensing image feature selection method and apparatus - Google Patents
Forest biomass-based remote sensing image feature selection method and apparatus Download PDFInfo
- Publication number
- CN105975973A CN105975973A CN201610282178.9A CN201610282178A CN105975973A CN 105975973 A CN105975973 A CN 105975973A CN 201610282178 A CN201610282178 A CN 201610282178A CN 105975973 A CN105975973 A CN 105975973A
- Authority
- CN
- China
- Prior art keywords
- feature
- feature set
- eigenvalue
- forest
- remote sensing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
- G06F18/211—Selection of the most significant subset of features
- G06F18/2113—Selection of the most significant subset of features by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a forest biomass-based remote sensing image feature selection method and a forest biomass-based remote sensing image feature selection apparatus. The method includes the following steps that: feature values are extracted from a forest remote sensing image, the feature values are preprocessed through an SR (stepwise regression) algorithm, and feature values corresponding to multicollinearity are removed from the preprocessed feature values, so that a feature set can be generated, the initial set of the feature set is a full set; the feature set is updated repeatedly according to the following processes: an SVM (support vector machine) algorithm is trained according to the initialization feature set, so that the weights of feature values in the initialization feature set are determined, an SVM-REF (support vector machine-recursive feature elimination) algorithm and the weights are adopted to construct the feature sequencing coefficient of the feature values in the feature set, and the feature values in the feature set are sequenced according to the feature sequencing coefficient, and the feature set is updated according to the sequence of the feature set; and update operation is carried out continuously until the number of feature values in the current feature set is equal to a preset number of feature values, and the current feature set is determined as the optimal feature set used for forest biomass. With the method and apparatus of the invention adopted, the effect of remote sensing image feature selection can be optimized.
Description
Technical field
Remote sensing image technical field of the present invention, in particular to a kind of characteristics of remote sensing image for forest biomass
System of selection and device.
Background technology
Characteristics of remote sensing image be image unique have, for being different from this qualitative attribution of other image.And shadow
As feature complexity of a great variety, both included landform, vegetation, the such physical feature of the hydrology, included again that house and road were such
The upper atural object of people, and the relation each other between these features is also complicated, thus in the analysis to remote sensing image, distant
Sense image feature selecting technology plays key player wherein.
Forest biomass accounts for Global land vegetation biomass 90%, is not only the important symbol of forest carbon sequestration capacity, also
It it is the important parameter of assessment Forest Carbon revenue and expenditure.Thus research forest biomass has important meaning for the understanding of forest ecosystem
Justice.Nowadays people are by using remote sensing image technology to obtain forest biomass correlated characteristic, to carry out correlational study.So
And, the characteristics of remote sensing image broad categories of forest biomass, such as single band feature, vegetation index, textural characteristics, terrain factor
Deng, every category feature is subdivided into various features, up to hundreds of；Scientific research personnel is to selecting needed for forest biomass Optimized model
Characteristics of remote sensing image time highly desirable disclosure satisfy that following effect: the characteristics of remote sensing image selected by (1) can make model result
Compared with true forest biomass, error is less；(2) select the time used by the process of characteristics of remote sensing image shorter；(3) select
Characteristics of remote sensing image negligible amounts but higher with forest biomass dependency.But the method the most generally used can spend relatively
The long time selects more feature, and the resultant error of the forest biomass Optimized model derived is relatively big, causes comprehensive
Poor effect.
For the problem of the characteristics of remote sensing image poor effect chosen in said method, effective solution is the most not yet proposed
Scheme.
Summary of the invention
In view of this, the purpose of the embodiment of the present invention is to provide a kind of characteristics of remote sensing image for forest biomass to select
Selection method and device, to optimize the effect that characteristics of remote sensing image is chosen.
First aspect, embodiments provides a kind of characteristics of remote sensing image system of selection for forest biomass,
Including:
Eigenvalue is extracted from forest remote sensing image；
By SR (Stepwise Regression, stepwise regression analysis) algorithm, eigenvalue is carried out pretreatment, from pre-
Eigenvalue after reason is rejected multicollinearity characteristic of correspondence value, generates feature set；Wherein, the initial set of feature set is characterized
The full collection of value；
Repeat as steps described below to update feature set: according to initialization feature collection training SVM (Support Vector
Machine, support vector machine) algorithm, determine that initialization feature concentrates the weight of each eigenvalue；Use SVM-REF
(Support Vector Machine-Recursive Feature Elimination, support vector machine-recursive feature disappears
Removing) algorithm and weight structural feature concentrate the feature ordering coefficient of each eigenvalue, according to feature ordering coefficient in feature set
Eigenvalue be ranked up；Ranking replacement feature set according to feature set；
Until the number of eigenvalue is equal to the eigenvalue number preset in current feature set, current feature set is determined
For the optimal characteristics collection for forest biomass.
In conjunction with first aspect, embodiments provide the first possible embodiment of first aspect, wherein, adopt
The feature ordering coefficient of each eigenvalue in current feature set is constructed, according to feature ordering system by SVM-REF algorithm and weight
Several it is ranked up including to the eigenvalue in feature set: structural feature ordering rule；Set up according to SVM-REF algorithm and weight
Scoring functions；The feature ordering coefficient that in feature set, each eigenvalue is relevant to weight is calculated by scoring functions；According to above-mentioned
Eigenvalue in feature set is ranked up by feature ordering rule and feature ordering coefficient.
In conjunction with first aspect, embodiments provide the embodiment that the second of first aspect is possible, wherein, on
Stating scoring functions is:
Wherein, J is characterized sequence coefficient, and ω is weight.
In conjunction with first aspect, embodiments provide the third possible embodiment of first aspect, wherein, root
Include according to the ranking replacement feature set of features described above collection: according to the sequence of feature set, determine minimum sequence coefficient characteristic of correspondence
Value；Minimum sequence coefficient characteristic of correspondence value is removed, as the feature set after updating in feature set.
In conjunction with first aspect, embodiments provide the 4th kind of possible embodiment of first aspect, wherein, also
Including: utilize optimal characteristics collection to build forest biomass Optimized model；Application forest biomass Optimized model is to forest remote sensing shadow
As being predicted, obtain the forest biomass of prediction；Corresponding with forest remote sensing image by the forest biomass of comparison prediction
Actual forest biomass checking optimal characteristics collection.
Second aspect, embodiments provides a kind of characteristics of remote sensing image for forest biomass and selects device,
Including:
Characteristics extraction module, for extracting eigenvalue from forest remote sensing image；
Feature set generation module, for carrying out pretreatment by SR algorithm to eigenvalue, from pretreated eigenvalue
Reject multicollinearity characteristic of correspondence value, generate feature set；Wherein, the initial set of feature set is characterized the full collection of value；
Feature set more new module, for repeating to update feature set according to following function: according to initialization feature collection training SVM
Algorithm, determines that initialization feature concentrates the weight of each eigenvalue；SVM-REF algorithm and weight structural feature is used to concentrate each
The feature ordering coefficient of individual eigenvalue, is ranked up the eigenvalue in feature set according to feature ordering coefficient；According to feature set
Ranking replacement feature set；
Optimal characteristics collection determines module, for the eigenvalue being equal to preset until the number of eigenvalue in current feature set
Number, is defined as the optimal characteristics collection for forest biomass by current feature set.
In conjunction with second aspect, embodiments provide the first possible embodiment of second aspect, wherein, on
State feature set more new module to include: feature ordering rule construct unit, for structural feature ordering rule；Scoring functions is set up single
Unit, for setting up scoring functions according to SVM-REF algorithm and weight；Feature ordering coefficient calculation unit, for by marking letter
Number calculates the feature ordering coefficient that in feature set, each eigenvalue is relevant to weight；Eigenvalue sequencing unit, for according to feature
Eigenvalue in feature set is ranked up by ordering rule and feature ordering coefficient.
In conjunction with second aspect, embodiments provide the embodiment that the second of second aspect is possible, wherein, on
Stating the scoring functions that scoring functions sets up in unit is:
Wherein, J is characterized sequence coefficient, and ω is weight.
In conjunction with second aspect, embodiments provide the third possible embodiment of second aspect, wherein, on
State feature set more new module also to include: eigenvalue determines unit, for the sequence according to feature set, determine minimum sequence coefficient pair
The eigenvalue answered；Feature set updating block, for removing minimum sequence coefficient characteristic of correspondence value, as renewal in feature set
After feature set.
In conjunction with second aspect, embodiments provide the 4th kind of possible embodiment of second aspect, wherein, on
State device also to include: Biomass Optimized model construction unit, be used for utilizing optimal characteristics collection to build forest biomass Optimized model；
Biomass predicting unit, is used for applying forest biomass Optimized model to be predicted forest remote sensing image, obtains the gloomy of prediction
Woods Biomass；Optimal characteristics collection authentication unit, for corresponding with forest remote sensing image by the forest biomass of comparison prediction
Actual forest biomass checking optimal characteristics collection.
The characteristics of remote sensing image system of selection for forest biomass of embodiment of the present invention offer and device, calculated by SR
Method carries out pretreatment to the eigenvalue extracted from forest remote sensing image, and rejects multicollinearity eigenvalue, generates feature set；
Repeat to use SVM algorithm and the feature ordering coefficient of SVM-REF each eigenvalue of algorithm construction, be characterized the eigenvalue of concentration
Sort and according to ranking replacement feature set；During until the eigenvalue number in feature set is equal to the eigenvalue number preset, determine
Optimal characteristics collection for forest biomass.Compared with the mode choosing characteristics of remote sensing image of the prior art, the present invention can
To select the characteristics of remote sensing image higher with forest biomass dependency in the short period of time, and then make to utilize optimal characteristics
The forest biomass that collection is derived is less with actual forest biomass phase ratio error, optimizes the effect that characteristics of remote sensing image is chosen
Really.
For making the above-mentioned purpose of the present invention, feature and advantage to become apparent, preferred embodiment cited below particularly, and coordinate
Appended accompanying drawing, is described in detail below.
Accompanying drawing explanation
In order to be illustrated more clearly that the technical scheme of the embodiment of the present invention, below by embodiment required use attached
Figure is briefly described, it will be appreciated that the following drawings illustrate only certain embodiments of the present invention, and it is right to be therefore not construed as
The restriction of scope, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to according to this
A little accompanying drawings obtain other relevant accompanying drawings.
Fig. 1 shows a kind of characteristics of remote sensing image system of selection for forest biomass that the embodiment of the present invention is provided
Flow chart；
Fig. 2 shows that a kind of characteristics of remote sensing image for forest biomass that the embodiment of the present invention is provided selects device
Structured flowchart.
Detailed description of the invention
For making the purpose of the embodiment of the present invention, technical scheme and advantage clearer, below in conjunction with the embodiment of the present invention
Middle accompanying drawing, is clearly and completely described the technical scheme in the embodiment of the present invention, it is clear that described embodiment is only
It is a part of embodiment of the present invention rather than whole embodiments.Generally real with the present invention illustrated described in accompanying drawing herein
The assembly executing example can be arranged with various different configurations and design.Therefore, below to the present invention's provided in the accompanying drawings
The detailed description of embodiment is not intended to limit the scope of claimed invention, but is merely representative of the selected reality of the present invention
Execute example.Based on embodiments of the invention, the institute that those skilled in the art are obtained on the premise of not making creative work
There are other embodiments, broadly fall into the scope of protection of the invention.
The longer time can be spent to select more feature in view of method used in the prior art, and derive
The resultant error of forest biomass Optimized model relatively big, cause the problem that resultant effect is the best, embodiments provide
A kind of characteristics of remote sensing image system of selection for forest biomass and device, be described below by embodiment.
Embodiment 1
The flow chart of a kind of characteristics of remote sensing image system of selection for forest biomass shown in Figure 1, the method
Comprise the steps:
Step S102, extracts eigenvalue from forest remote sensing image；Wherein, forest remote sensing image is record forest electromagnetic wave
The satellite image of characteristic or aerial image, reflect each feature and attribute that forest is comprised.Eigenvalue can be divided into again list
The characteristic types such as wave band feature, vegetation index, textural characteristics, terrain factor, each characteristic type can be subdivided into again multiple feature.Example
As vegetation index characteristic type specifically includes that difference vegetation index, normalized differential vegetation index, ratio vegetation index, environment vegetation
Index, soil vegetative cover index, perpendicular vegetation index, the brightness index through K-T Transformation, green degree index, humidity index, Yi Jijing
The features such as the first principal component of main composition conversion, Second principal component, the 3rd main constituent；Texture index feature classification specifically includes that
The features such as average, variance, homogeneity, contrast, diversity, comentropy and dependency.Thus each characteristic type is refined
After, from forest remote sensing image, extractible eigenvalue can items the most up to a hundred.
Step S104, carries out pretreatment by SR algorithm to eigenvalue, rejects multiple common from pretreated eigenvalue
Linear characteristic of correspondence value, generates feature set；Wherein, the initial set of feature set is characterized the full collection of value；
Concrete, the process of above-mentioned pretreatment is: feature introduces model one by one, will carry out after often introducing a feature
F checks, and the feature being selected into carries out T inspection one by one, becomes due to the introducing of feature below when the original feature introduced
Time no longer notable, then it is deleted.To guarantee before introducing new feature, regression equation only to comprise significant characteristics every time.
Step S106, repeats to update feature set: according to initialization feature collection training SVM algorithm, determine as steps described below
Initialization feature concentrates the weight of each eigenvalue；SVM-REF algorithm and weight structural feature is used to concentrate each eigenvalue
Feature ordering coefficient, is ranked up the eigenvalue in feature set according to feature ordering coefficient；Ranking replacement according to feature set
Feature set；
Wherein, on the basis of SVM algorithm is built upon Statistical Learning Theory, according to limited sample information answering at model
Between polygamy (i.e. the study precision to specific training sample) and learning capacity (identifying the ability of arbitrary sample the most error-free)
Seek optimal compromise, use SVM algorithm can objectively determine that initialization feature concentrates the weight of each eigenvalue.Additionally,
SVM-REF algorithm is a kind of based on the backward sequence reduction algorithm of largest interval principle in SVM algorithm, and concrete steps can refer to
Existing method, does not repeats them here.
Step S108, until in current feature set, the number of eigenvalue is equal to the eigenvalue number preset, by current
Feature set is defined as the optimal characteristics collection for forest biomass.
In the said method of the present embodiment, by SR algorithm, the eigenvalue extracted from forest remote sensing image is carried out from pre-
Reason, and reject multicollinearity eigenvalue, generate feature set；Repeat to use SVM algorithm and each spy of SVM-REF algorithm construction
The feature ordering coefficient of value indicative, is characterized the eigenvalue sequence of concentration and according to ranking replacement feature set；Until in feature set
When eigenvalue number is equal to the eigenvalue number preset, determine the optimal characteristics collection for forest biomass.The embodiment of the present invention
The characteristics of remote sensing image higher with forest biomass dependency can be selected in the short period of time, and then make utilization optimum special
The forest biomass that collection is derived is less with actual forest biomass phase ratio error, optimizes the effect that characteristics of remote sensing image is chosen
Really.
In order to weigh each eigenvalue and the dependency of forest biomass in feature set, the embodiment of the present invention is implementing
Time, use SVM-REF algorithm and weight to construct the feature ordering coefficient of each eigenvalue in current feature set, according to feature
The step that eigenvalue in feature set is ranked up by sequence coefficient may include that
(1) structural feature ordering rule；Wherein, this feature ordering rule can be incremented by sequence, it is also possible to is the row that successively decreases
Sequence.
(2) scoring functions is set up according to SVM-REF algorithm and weight；
(3) calculate, by scoring functions, the feature ordering coefficient that in feature set, each eigenvalue is relevant to weight；
(4) according to feature ordering rule and feature ordering coefficient, the eigenvalue in feature set is ranked up.
By the way, each eigenvalue in feature set can be carried out objective and accurate judgement；The sequence drawn
Result represents each eigenvalue and the dependency of forest biomass in feature set more intuitively, provides visitor for selection eigenvalue
The foundation seen.
Wherein, above-mentioned scoring functions can be:
J is characterized sequence coefficient, and ω is weight.
It should be noted that above-mentioned scoring functions only realizes a kind of mode of the embodiment of the present invention, when implementing,
Other scoring functions can also be used.
Additionally, during the implementing of the present embodiment, the step according to the ranking replacement feature set of feature set is permissible
Including:
(1) according to the sequence of feature set, minimum sequence coefficient characteristic of correspondence value is determined；
(2) in feature set, minimum sequence coefficient characteristic of correspondence value is removed, as the feature set after updating.
By the way of above-mentioned renewal feature set, can effectively get rid of in feature set little with forest biomass dependency
The eigenvalue of (sequence coefficient is low), leaves the eigenvalue of higher with forest biomass dependency (sequence coefficient is high).
Further, in order to verify the Selection effect of optimal characteristics collection, the embodiment of the present invention based on the above method, is gone back
May include that
(1) optimal characteristics collection is utilized to build forest biomass Optimized model；
(2) forest remote sensing image is predicted by application forest biomass Optimized model, obtains the forest biomass of prediction；
(3) optimum is verified by the actual forest biomass that the forest biomass of comparison prediction is corresponding with forest remote sensing image
Feature set.
Aforesaid way by will the prediction forest biomass that be derived from by the remote sensing image optimal characteristics collection chosen with
Actual forest biomass compares, it may be verified that the effect of the optimal characteristics collection selected by method that employing the present embodiment is provided,
Make result more objective.
A kind of characteristics of remote sensing image system of selection for forest biomass that the present embodiment is provided, mainly utilizes SR to calculate
Method and SVM-REF algorithm effectively choose the characteristics of remote sensing image higher with forest biomass degree of association, and then can improve gloomy
Woods Biomass Optimized model.Wherein, the main thought of SR algorithm is by introducing independent variable one by one, and introduce affects result every time
The most significant independent variable, and the original variable in equation is tested one by one, becoming inapparent variable one by one from equation
Middle rejecting, neither misses in final regression equation and result is affected significant variable, does not the most comprise result impact the most notable
Variable.The main thought of SVM-REF algorithm is to carry out structural feature sequence according to SVM weight vector w of generation when training to be
Number, each iteration is removed a minimum feature of sequence coefficient, is finally given the sort descending of all characteristic attributes.By combining
Above-mentioned algorithm, can make the characteristics of remote sensing image selected reach following effect: to select the process of optimized remote sensing image feature simultaneously
Used time is shorter；The characteristics of remote sensing image negligible amounts that selects but higher with forest biomass dependency, by characteristics of remote sensing image
Predicting the outcome that the forest biomass model derived generates is less with actual forest biomass phase ratio error.
Embodiment 2
The method provided corresponding to above-described embodiment, the embodiment of the present invention additionally provides a kind of distant for forest biomass
Sense image feature selects device, sees Fig. 2, and this device includes with lower module:
Characteristics extraction module 202, for extracting eigenvalue from forest remote sensing image；
Feature set generation module 204, for carrying out pretreatment by SR algorithm to eigenvalue, from pretreated eigenvalue
Middle rejecting multicollinearity characteristic of correspondence value, generates feature set；Wherein, the initial set of feature set is characterized the full collection of value；
Feature set more new module 206, for repeating to update feature set according to following function: according to SVM algorithm, determines initial
Change the weight of each eigenvalue in feature set；SVM-REF algorithm and weight structural feature is used to concentrate the feature of each eigenvalue
Sequence coefficient, is ranked up the eigenvalue in feature set according to feature ordering coefficient；Ranking replacement feature according to feature set
Collection；
Optimal characteristics collection determines module 208, for the spy being equal to preset until the number of eigenvalue in current feature set
Value indicative number, is defined as the optimal characteristics collection for forest biomass by current feature set.
In the said apparatus of the present embodiment, by SR algorithm, the eigenvalue extracted from forest remote sensing image is carried out from pre-
Reason, and reject multicollinearity eigenvalue, generate feature set；Repeat to use SVM algorithm and each spy of SVM-REF algorithm construction
The feature ordering coefficient of value indicative, is characterized the eigenvalue sequence of concentration and according to ranking replacement feature set；Until in feature set
When eigenvalue number is equal to the eigenvalue number preset, determine the optimal characteristics collection for forest biomass.The present invention can be
Select the characteristics of remote sensing image higher with forest biomass dependency in relatively short period of time, and then make to utilize optimal characteristics collection to push away
The forest biomass derived is less with actual forest biomass phase ratio error, optimizes the effect that characteristics of remote sensing image is chosen.
In order to weigh each eigenvalue and the dependency of forest biomass in feature set, the embodiment of the present invention is implementing
Time, feature set more new module can include with lower unit:
Feature ordering rule construct unit, for structural feature ordering rule；
Scoring functions sets up unit, for setting up scoring functions according to SVM-REF algorithm and weight；
Wherein, above-mentioned scoring functions can be:
J is characterized sequence coefficient, and ω is weight.
Feature ordering coefficient calculation unit is relevant to weight for calculating each eigenvalue in feature set by scoring functions
Feature ordering coefficient；
Eigenvalue sequencing unit, for entering the eigenvalue in feature set according to feature ordering rule and feature ordering coefficient
Row sequence.
Each eigenvalue in feature set can be carried out objective and accurate by the said units included by feature set more new module
Judgement；The ranking results drawn represents each eigenvalue and the dependency of forest biomass in feature set more intuitively, for
Eigenvalue is selected to provide objective foundation.
Additionally, in order to effectively get rid of eigenvalue little with forest biomass dependency in feature set, stay and forest
The eigenvalue that Biomass dependency is higher, during the implementing of the present embodiment, features described above collection more new module is all right
Including: eigenvalue determines unit, for the sequence according to feature set, determines minimum sequence coefficient characteristic of correspondence value；Feature set
Updating block, for removing minimum sequence coefficient characteristic of correspondence value, as the feature set after updating in feature set.
Said units can effectively be screened for the high eigenvalue of forest biomass dependency, it is ensured that stays in feature set
Eigenvalue be current optimal characteristics.
Further, in order to verify the Selection effect of optimal characteristics collection, the embodiment of the present invention is on the basis of said apparatus, also
Can include with lower unit: Biomass Optimized model construction unit, be used for utilizing optimal characteristics collection to build forest biomass optimization
Model；Biomass predicting unit, is used for applying forest biomass Optimized model to be predicted forest remote sensing image, is predicted
Forest biomass；Optimal characteristics collection authentication unit, for by the forest biomass of comparison prediction and forest remote sensing image pair
The actual forest biomass checking optimal characteristics collection answered.
By said units, make the prediction forest biomass that is derived from by the remote sensing image optimal characteristics collection chosen with
Actual forest biomass compares, it may be verified that the effect of the optimal characteristics collection selected by device that employing the present embodiment is provided,
Make result more objective.
The device that the embodiment of the present invention is provided, it realizes principle and the technique effect of generation and preceding method embodiment phase
With, for briefly describing, the not mentioned part of device embodiment part, refer to corresponding contents in preceding method embodiment.
In the said apparatus of the present embodiment, mainly by utilizing SR algorithm and SVM-REF algorithm effectively to choose and forest
The characteristics of remote sensing image that Biomass degree of association is higher, can reach following effect: the forest derived by remote sensing image can be made raw
Predicting the outcome of thing amount model generation is less with actual forest biomass phase ratio error；Select the process of optimized remote sensing image feature
Used time is shorter；The characteristics of remote sensing image negligible amounts that selects but higher with forest biomass dependency.
For above-mentioned a kind of characteristics of remote sensing image system of selection for forest biomass and device, the embodiment of the present invention is also
Provide following method:
1, choose have regional representativeness and research importance Mount Taishan Scenic Spot for investigation target, consider Regional Distribution,
The factors such as land occupation condition, age group structure choose the sample ground having adequate representation, and have chosen 48 groups altogether such as the present embodiment has abundant generation
The sample ground of table；
2, the remote sensing image that above-mentioned sample ground is corresponding is obtained；
3, from above-mentioned remote sensing image, extract eigenvalue, extract 132 stack features values, wherein every stack features value such as this test altogether
It is 48 dimensional vectors；
4, with SR algorithm, above-mentioned 132 stack features values are carried out pretreatment, eliminate the eigenvalue with multicollinearity, set up
Feature set；
5, repeat to update feature set as steps described below:
5.1, utilize feature set to train SVM algorithm, determine the weights omega of each eigenvalue in feature set；
5.2, it is expressed as according to SVM-REF algorithm and ω foundationScoring functions, calculate each eigenvalue
Feature ordering coefficient J；
5.3, it is characterized each eigenvalue sequence of concentration, and determines that minimum sequence coefficient characteristic of correspondence value is (i.e. with gloomy
The eigenvalue that woods Biomass degree of association is little)；
5.4, remove in feature set and minimum sequence coefficient characteristic of correspondence collection, as the feature set after updating；
6, repeat step 5, until feature set only includes 6 stack features values (6 be eigenvalue number set in advance), then this 6
Stack features value is the optimal characteristics the highest with forest biomass degree of association, and this feature integrates as optimal characteristics collection；
7, utilize above-mentioned optimal characteristics to set up forest biomass Optimized model, drawn by this forest biomass Optimized model
The predictive value of forest biomass；
8, according to the true measurement data on institute's sampling ground, the actual value of forest biomass is calculated；
9, predictive value and the actual value of forest biomass, the characteristics of remote sensing image effect that checking is chosen are compared.
Result shows as follows: when choosing 6 characteristics of remote sensing images, 1.864 seconds used times, and squared correlation coefficient is 0.356, institute
The prediction forest biomass lowest mean square root error compared with actual forest biomass drawn is 36.589 tons/hectare；
And use other method selecting remote sensing features, and such as only with SVM-RFE algorithm, then when choosing 6, the used time
4.099 seconds, squared correlation coefficient was 0.274, and its lowest mean square root error is 38.641 tons/hectare；And if minimizing all
Square error is 36.589 tons/hectare, needs to choose 8 characteristics of remote sensing images only with SVM-RFE algorithm.
Change the characteristics of remote sensing image value preset, all can draw and compare similar result with above-mentioned data, no longer arrange at this
Go out.
Synthesis result shows compared with other characteristics of remote sensing image choosing method, by above-mentioned distant for forest biomass
Sense image feature system of selection and device can reach following effect:
(1) the process used time selecting optimized remote sensing image feature is shorter；
(2) the characteristics of remote sensing image negligible amounts that selects but higher with forest biomass dependency；
(3) what the forest biomass model derived by remote sensing image generated predicts the outcome compared with actual forest biomass
Error is less.
It is last it is noted that the detailed description of the invention of above example, the only present invention, in order to the skill of the present invention to be described
Art scheme, is not intended to limit, and protection scope of the present invention is not limited thereto, although entering the present invention with reference to previous embodiment
Go detailed description, it will be understood by those within the art that: any those familiar with the art is at this
In the technical scope that invention discloses, the technical scheme described in previous embodiment still can be modified or can think easily by it
To change, or wherein portion of techniques feature is carried out equivalent；And these are revised, change or replace, do not make corresponding
The essence of technical scheme departs from the spirit and scope of embodiment of the present invention technical scheme.All should contain in protection scope of the present invention
Within.Therefore, protection scope of the present invention should be as the criterion with scope of the claims.
Claims (10)
1. the characteristics of remote sensing image system of selection for forest biomass, it is characterised in that including:
Eigenvalue is extracted from forest remote sensing image；
By stepwise regression analysis SR algorithm, described eigenvalue is carried out pretreatment, reject from pretreated described eigenvalue
Multicollinearity characteristic of correspondence value, generates feature set；Wherein, the initial set of described feature set is the full collection of described eigenvalue；
Repeat to update described feature set as steps described below: according to described initialization feature collection training support vector machines algorithm,
Determine that described initialization feature concentrates the weight of each eigenvalue；Use support vector machine-recursive feature to eliminate SVM-REF to calculate
Method and described weight construct the feature ordering coefficient of each eigenvalue in described feature set, according to described feature ordering coefficient to institute
The eigenvalue stated in feature set is ranked up；Feature set described in ranking replacement according to described feature set；
Until the number of eigenvalue is equal to the eigenvalue number preset in current feature set, described current feature set is determined
For the optimal characteristics collection for described forest biomass.
Method the most according to claim 1, it is characterised in that use support vector machine-recursive feature to eliminate SVM-REF
Algorithm and described weight construct the feature ordering coefficient of each eigenvalue in current described feature set, according to described feature ordering
Eigenvalue in described feature set is ranked up including by coefficient:
Structural feature ordering rule；
Scoring functions is set up according to SVM-REF algorithm and described weight；
The feature ordering coefficient that in described feature set, each eigenvalue is relevant to weight is calculated by described scoring functions；
According to the regular and described feature ordering coefficient of described feature ordering, the eigenvalue in described feature set is ranked up.
Method the most according to claim 2, it is characterised in that described scoring functions is:
Wherein, J is characterized sequence coefficient, and ω is weight.
Method the most according to claim 1, it is characterised in that according to feature set bag described in the ranking replacement of described feature set
Include:
According to the sequence of described feature set, determine minimum sequence coefficient characteristic of correspondence value；
Described minimum sequence coefficient characteristic of correspondence value is removed, as the described feature set after updating in described feature set.
Method the most according to claim 1, it is characterised in that described method also includes:
Described optimal characteristics collection is utilized to build forest biomass Optimized model；
Apply described forest biomass Optimized model that forest remote sensing image is predicted, obtain the forest biomass of prediction；
The actual forest biomass corresponding with described forest remote sensing image by the forest biomass of relatively described prediction verifies institute
State optimal characteristics collection.
6. the characteristics of remote sensing image for forest biomass selects device, it is characterised in that including:
Characteristics extraction module, for extracting eigenvalue from forest remote sensing image；
Feature set generation module, for carrying out pretreatment to described eigenvalue, from pretreatment by stepwise regression analysis SR algorithm
After described eigenvalue in reject multicollinearity characteristic of correspondence value, generate feature set；Wherein, the initial set of described feature set
Full collection for described eigenvalue；
Feature set more new module, for repeating to update described feature set according to following function: assemble for training according to described initialization feature
Practice support vector machines algorithm, determine that described initialization feature concentrates the weight of each eigenvalue；Employing support vector machine-pass
Return feature to eliminate SVM-REF algorithm and described weight constructs the feature ordering coefficient of each eigenvalue in described feature set, according to
Eigenvalue in described feature set is ranked up by described feature ordering coefficient；Spy described in ranking replacement according to described feature set
Collection；
Optimal characteristics collection determines module, for individual equal to the eigenvalue preset until the number of eigenvalue in current feature set
Number, is defined as the optimal characteristics collection for described forest biomass by described current feature set.
Device the most according to claim 6, it is characterised in that described feature set more new module includes:
Feature ordering rule construct unit, for structural feature ordering rule；
Scoring functions sets up unit, for setting up scoring functions according to SVM-REF algorithm and described weight；
Feature ordering coefficient calculation unit, for calculating each eigenvalue and weight in described feature set by described scoring functions
Relevant feature ordering coefficient；
Eigenvalue sequencing unit, is used for according to the regular and described feature ordering coefficient of described feature ordering in described feature set
Eigenvalue is ranked up.
Device the most according to claim 6, it is characterised in that described scoring functions is set up the scoring functions in unit and is:
Wherein, J is characterized sequence coefficient, and ω is weight.
Device the most according to claim 6, it is characterised in that described feature set more new module also includes:
Eigenvalue determines unit, for the sequence according to described feature set, determines minimum sequence coefficient characteristic of correspondence value；
Feature set updating block, for removing described minimum sequence coefficient characteristic of correspondence value, as more in described feature set
Described feature set after Xin.
Device the most according to claim 6, it is characterised in that described device also includes:
Biomass Optimized model construction unit, is used for utilizing described optimal characteristics collection to build forest biomass Optimized model；
Biomass predicting unit, is used for applying described forest biomass Optimized model to be predicted forest remote sensing image, obtains
The forest biomass of prediction；
Optimal characteristics collection authentication unit, for corresponding with described forest remote sensing image by the forest biomass of relatively described prediction
Actual forest biomass verify described optimal characteristics collection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610282178.9A CN105975973A (en) | 2016-04-29 | 2016-04-29 | Forest biomass-based remote sensing image feature selection method and apparatus |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610282178.9A CN105975973A (en) | 2016-04-29 | 2016-04-29 | Forest biomass-based remote sensing image feature selection method and apparatus |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105975973A true CN105975973A (en) | 2016-09-28 |
Family
ID=56993435
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610282178.9A Pending CN105975973A (en) | 2016-04-29 | 2016-04-29 | Forest biomass-based remote sensing image feature selection method and apparatus |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105975973A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107085705A (en) * | 2017-03-28 | 2017-08-22 | 中国林业科学研究院资源信息研究所 | A kind of forest parameters remote sensing estimation method of efficient feature selection |
| CN109919990A (en) * | 2019-02-19 | 2019-06-21 | 北京工业大学 | Forest Height Prediction method is carried out using depth perception network and parallax remote sensing image |
| CN110956101A (en) * | 2019-11-19 | 2020-04-03 | 广东省城乡规划设计研究院 | Remote sensing image yellow river ice detection method based on random forest algorithm |
| CN111861704A (en) * | 2020-07-10 | 2020-10-30 | 深圳无域科技技术有限公司 | Wind control feature generation method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101278304A (en) * | 2005-09-02 | 2008-10-01 | 皇家创新有限公司 | Feature selection |
| CN104361592A (en) * | 2014-11-14 | 2015-02-18 | 中国林业科学研究院资源信息研究所 | Method and device for estimating forest above-ground biomass |
| CN104778451A (en) * | 2015-03-31 | 2015-07-15 | 中国科学院上海技术物理研究所 | Grassland biomass remote sensing inversion method considering grassland height factor |
-
2016
- 2016-04-29 CN CN201610282178.9A patent/CN105975973A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101278304A (en) * | 2005-09-02 | 2008-10-01 | 皇家创新有限公司 | Feature selection |
| CN104361592A (en) * | 2014-11-14 | 2015-02-18 | 中国林业科学研究院资源信息研究所 | Method and device for estimating forest above-ground biomass |
| CN104778451A (en) * | 2015-03-31 | 2015-07-15 | 中国科学院上海技术物理研究所 | Grassland biomass remote sensing inversion method considering grassland height factor |
Non-Patent Citations (7)
| Title |
|---|
| ISABELLE GUYON ET AL.: "Gene Selection for Cancer Classification using Support Vector Machines", 《MACHINE LEARNING》 * |
| PIOTR S. GROMSKI ET AL.: "A comparative investigation of modern feature selection and classification approaches for the analysis of mass spectrometry data", 《ANALYTICA CHIMICA ACTA》 * |
| 卢运梅: "SVM_RFE算法在数据分析中的应用", 《中国优秀硕士学位论文全文数据库 信息科技辑（月刊）》 * |
| 周南等: "云冷杉林生物量遥感估测模型研究", 《林业调查规划》 * |
| 杨帆等: "面向局部特征的支持向量机递归特征消除", 《计算机工程与应用》 * |
| 穆喜云: "森林地上生物量遥感估测方法研究", 《中国博士学位论文全文数据库 农业科技辑（月刊）》 * |
| 薛巍等: "基于遥感技术的森林生物量估算应用", 《水土保持研究》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107085705A (en) * | 2017-03-28 | 2017-08-22 | 中国林业科学研究院资源信息研究所 | A kind of forest parameters remote sensing estimation method of efficient feature selection |
| CN107085705B (en) * | 2017-03-28 | 2020-08-04 | 中国林业科学研究院资源信息研究所 | Forest parameter remote sensing estimation method based on feature selection |
| CN109919990A (en) * | 2019-02-19 | 2019-06-21 | 北京工业大学 | Forest Height Prediction method is carried out using depth perception network and parallax remote sensing image |
| CN110956101A (en) * | 2019-11-19 | 2020-04-03 | 广东省城乡规划设计研究院 | Remote sensing image yellow river ice detection method based on random forest algorithm |
| CN110956101B (en) * | 2019-11-19 | 2020-08-07 | 广东省城乡规划设计研究院 | Remote sensing image yellow river ice detection method based on random forest algorithm |
| CN111861704A (en) * | 2020-07-10 | 2020-10-30 | 深圳无域科技技术有限公司 | Wind control feature generation method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108090510A (en) | A kind of integrated learning approach and device based on interval optimization | |
| CN105184316B (en) | A kind of support vector machines electrical network business classification method based on feature power study | |
| CN103702416B (en) | Semi-supervised learning indoor positioning method based on support vector machine | |
| Irawan | Implementation Of Data Mining For Determining Majors Using K-Means Algorithm In Students Of SMA Negeri 1 Pangkalan Kerinci | |
| CN107274020B (en) | Learner subject total measured result prediction system and method based on collaborative filtering thought | |
| CN106779087A (en) | A kind of general-purpose machinery learning data analysis platform | |
| CN108846259A (en) | A kind of gene sorting method and system based on cluster and random forests algorithm | |
| CN105975973A (en) | Forest biomass-based remote sensing image feature selection method and apparatus | |
| CN104933428B (en) | A kind of face identification method and device based on tensor description | |
| CN116108758B (en) | Landslide susceptibility evaluation method | |
| CN108288014A (en) | Intelligent road extracting method and device, extraction model construction method and hybrid navigation system | |
| CN106651519A (en) | Method and system for personalized recommendation based on label information | |
| CN107944986A (en) | A kind of O2O Method of Commodity Recommendation, system and equipment | |
| CN103810299A (en) | Image retrieval method on basis of multi-feature fusion | |
| CN109118119A (en) | Air control model generating method and device | |
| CN105976070A (en) | Key-element-based matrix decomposition and fine tuning method | |
| CN108549907A (en) | A kind of data verification method based on multi-source transfer learning | |
| CN104239496A (en) | Collaborative filtering method based on integration of fuzzy weight similarity measurement and clustering | |
| CN106202377A (en) | A kind of online collaborative sort method based on stochastic gradient descent | |
| CN112381644A (en) | Credit scene risk user assessment method based on space variable reasoning | |
| Yan | Student performance prediction using XGBoost method from a macro perspective | |
| CN106780258A (en) | A kind of method for building up and device of minor crime decision tree | |
| CN105550711A (en) | Firefly algorithm based selective ensemble learning method | |
| CN107341558A (en) | Multiply the computational methods in path and the computing device and record media using the method altogether | |
| CN113191359A (en) | Small sample target detection method and system based on support and query samples |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160928 |
|
| RJ01 | Rejection of invention patent application after publication |