CN102831440A - Method and device for decision tree based wide-area remote sensing image classification - Google Patents
Method and device for decision tree based wide-area remote sensing image classification Download PDFInfo
- Publication number
- CN102831440A CN102831440A CN2012102974644A CN201210297464A CN102831440A CN 102831440 A CN102831440 A CN 102831440A CN 2012102974644 A CN2012102974644 A CN 2012102974644A CN 201210297464 A CN201210297464 A CN 201210297464A CN 102831440 A CN102831440 A CN 102831440A
- Authority
- CN
- China
- Prior art keywords
- image
- decision tree
- classification
- remote sensing
- wide area
- 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
Images
Landscapes
- Image Processing (AREA)
Abstract
An embodiment of the invention provides a method and a device for decision tree based wide-area remote sensing image classification. The method includes acquiring an image set to be classified; dividing the image set to be classified into multiple groups according to time phase characteristics and enabling each group of images to have the same spectrum characteristics; respectively sampling the same groups of images uniformly to acquire sample data, extracting characteristics of the same groups of images, and then carrying out band combination so as to acquire multi-characteristic band-combination images; and inputting the sample data and the multi-characteristic band-combination images of the same groups of images to a decision tree classifier to acquire classified images. The method and the device have the technical advantages that a wide-area land cover classification decision according to decision tree based sensing image classification is provided, and wide-area remote sensing image classification precision is improved.
Description
Technical field
The present invention relates to the classification of remote sensing image, relate in particular to a kind of wide area remote sensing image decision tree sorting technique and device.
Background technology
Remote sensing image be utilize spaceborne or the reaction earth that airborne sensor obtains outwardly object space distribute and the image document of spectral information; It has wide coverage; Characteristics such as imaging cycle is short; Development along with the classification of remote-sensing images technology makes the classification of remote-sensing images technology in the cover classification of the wide area face of land, obtain increasing application.Existing classification of remote-sensing images method and system has: supervision such as minimum distance method, parallel six methods, maximum likelihood method and ISODATA (iteration self-organization data analysis technique), K-Means (K-mean cluster method), unsupervised classification method, and emerging fuzzy clustering method, neural network method, decision tree, SVMs and object-oriented classification.Wherein the decision tree classification method mainly has the following advantages:
(1) the categorised decision tree has clear in structure, and easy to understand realizes that simply travelling speed is fast, characteristics such as accuracy height.Can also can be input in the expert system for analysis expert, judgement and correction.
(2) the decision tree classification method need not supposed prior probability distribution; The characteristics of this imparametrization make it have better dirigibility and robustness; Therefore; When the space distribution of remote sensing image data characteristic is very complicated, when perhaps multi-source data has different statistic distribution and yardstick, can obtain desirable classification results with the decision tree classification method.
(3) decision tree can effectively be handled a large amount of high dimensional datas and nonlinear relationship.
(4) traditional decision-tree can suppress training sample decision attribute disappearance problem effectively, therefore can solve because training sample (possibly mixed by sensor noise, scanning leakage, signal, reason such as various cause) to make the problem that nicety of grading reduces.
The decision tree classification algorithm is a kind of forecast model in the data mining technology, and it infers the decision tree form of expression through out of order, random sample data collection, and is used for the classification of target data set.In classification of remote-sensing images, lack corresponding classification policy and algorithm, to improve wide area classification of remote-sensing images precision.
Summary of the invention
The embodiment of the invention provides a kind of wide area remote sensing image decision tree sorting technique and device, so that a kind of wide area face of land cover classification strategy based on the decision tree classification of remote-sensing images to be provided, and improves wide area classification of remote-sensing images precision.
On the one hand, the embodiment of the invention provides a kind of wide area remote sensing image decision tree sorting technique, and said wide area remote sensing image decision tree sorting technique comprises:
Obtain image set to be classified;
According to the time phase character image set to be classified is divided into some groups, every group image has identical wave spectrum characteristic;
Respectively same group image is unified sampling to obtain sample data, reach respectively to carrying out band combination after the same group image feature extraction to obtain many characteristic wave bands combination images;
Sample data and many characteristic wave bands combination image of said same group image are input in the decision tree classification device, to obtain the classification image.
Optional, in an embodiment of the present invention, said feature extraction comprises: characteristic exponent is extracted, normalized differential vegetation index NDVI extracts, texture transformation.
Optional; In an embodiment of the present invention; Sample data and many characteristic wave bands combination image of said same group image are input in the decision tree classification device,, comprise: with sample data and many characteristic wave bands combination image input decision tree classification device of said same group image to obtain the classification image; And take the mode of ten folding cross validations to test, to obtain training set.
Optional; In an embodiment of the present invention; Sample data and many characteristic wave bands combination image of said same group image are input in the decision tree classification device; To obtain the classification image, comprising: the training set of said generation is input in the decision tree of the binary tree structure of classification of remote-sensing images learns, to obtain the rule set that is used for classification of remote-sensing images.
Optional; In an embodiment of the present invention, the sample data of said same group image and many characteristic wave bands combination image are input in the decision tree classification device, to obtain the classification image; Comprise: through bringing in constant renewal in the weight of each sample; Iterative technique is applied in the decision tree of above-mentioned binary tree structure, makes up the iteration decision tree, to improve the nicety of grading of rule set.
On the other hand, the embodiment of the invention provides a kind of wide area remote sensing image decision tree sorter, and said wide area remote sensing image decision tree sorter comprises:
Acquiring unit is used to obtain image set to be classified;
Grouped element, be used for according to the time phase character image set to be classified is divided into some groups, every group image has identical wave spectrum characteristic;
Unified sampling unit is used for respectively same group image being unified sampling to obtain sample data;
The band combination unit is used for respectively to carrying out band combination after the same group image feature extraction to obtain many characteristic wave bands combination images;
Taxon is used for sample data and many characteristic wave bands combination image of said same group image are input to the decision tree classification device, to obtain the classification image.
Optional, in an embodiment of the present invention, said feature extraction comprises: characteristic exponent is extracted, normalized differential vegetation index NDVI extracts, texture transformation.
Optional, in an embodiment of the present invention, said taxon is further used for sample data and many characteristic wave bands combination image input decision tree classification device with said same group image, and takes the mode of ten folding cross validations to test, to obtain training set.
Optional, in an embodiment of the present invention, said taxon is further used for training set with said generation and is input in the decision tree of the binary tree structure of classification of remote-sensing images and learns, to obtain the rule set that is used for classification of remote-sensing images.
Optional, in an embodiment of the present invention, said taxon is further used for iterative technique being applied in the decision tree of above-mentioned binary tree structure through bringing in constant renewal in the weight of each sample, makes up the iteration decision tree, to improve the nicety of grading of rule set.
Technique scheme has following beneficial effect: obtain image set to be classified because adopt; According to the time phase character image set to be classified is divided into some groups, every group image has identical wave spectrum characteristic; Respectively same group image is unified sampling to obtain sample data, reach respectively to carrying out band combination after the same group image feature extraction to obtain many characteristic wave bands combination images; Sample data and many characteristic wave bands combination image of said same group image are input in the decision tree classification device; To obtain the technological means of classification image; So reached following technique effect: a kind of wide area face of land cover classification strategy based on the decision tree classification of remote-sensing images is provided, has improved wide area classification of remote-sensing images precision.
Description of drawings
In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art; To do to introduce simply to the accompanying drawing of required use in embodiment or the description of the Prior Art below; Obviously, the accompanying drawing in describing below only is some embodiments of the present invention, for those of ordinary skills; Under the prerequisite of not paying creative work, can also obtain other accompanying drawing according to these accompanying drawings.
Fig. 1 is a kind of wide area remote sensing image decision tree of embodiment of the invention sorting technique process flow diagram;
Fig. 2 is a kind of wide area remote sensing image decision tree of embodiment of the invention sorter structural representation;
Fig. 3 uses the precision evaluation synoptic diagram of wide area remote sensing image decision tree classification for application example of the present invention;
Fig. 4 is the concrete classification policy synoptic diagram of single scape image in the application example of the present invention;
Fig. 5 implements synoptic diagram for the unified sampling of application example grouping of the present invention;
Fig. 6 is an application example decision tree classification of remote-sensing images practical implementation process flow diagram of the present invention.
Embodiment
To combine the accompanying drawing in the embodiment of the invention below, the technical scheme in the embodiment of the invention is carried out clear, intactly description, obviously, described embodiment only is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills are not making the every other embodiment that is obtained under the creative work prerequisite, all belong to the scope of the present invention's protection.
As shown in Figure 1, be a kind of wide area remote sensing image decision tree of embodiment of the invention sorting technique process flow diagram, said wide area remote sensing image decision tree sorting technique comprises:
101, obtain image set to be classified;
102, according to the time phase character image set to be classified is divided into some groups, every group image has identical wave spectrum characteristic;
103, respectively same group image is unified sampling to obtain sample data, reach respectively to carrying out band combination after the same group image feature extraction to obtain many characteristic wave bands combination images;
104, sample data and the many characteristic wave bands combination image with said same group image is input in the decision tree classification device, to obtain the classification image.
Optional, said feature extraction comprises: characteristic exponent is extracted, normalized differential vegetation index NDVI extracts, texture transformation.
Optional; Sample data and many characteristic wave bands combination image of said same group image are input in the decision tree classification device; To obtain the classification image; Comprise: with sample data and many characteristic wave bands combination image input decision tree classification device of said same group image, and take the mode of ten folding cross validations to test, to obtain training set.
Optional; Sample data and many characteristic wave bands combination image of said same group image are input in the decision tree classification device; To obtain the classification image; Comprise: the training set of said generation is input in the decision tree of the binary tree structure of classification of remote-sensing images learns, to obtain the rule set that is used for classification of remote-sensing images.
Optional; Sample data and many characteristic wave bands combination image of said same group image are input in the decision tree classification device; To obtain the classification image, comprising:, iterative technique is applied in the decision tree of above-mentioned binary tree structure through bringing in constant renewal in the weight of each sample; Make up the iteration decision tree, to improve the nicety of grading of rule set.
Corresponding to method embodiment, as shown in Figure 2, be a kind of wide area remote sensing image decision tree of embodiment of the invention sorter structural representation, said wide area remote sensing image decision tree sorter comprises:
Acquiring unit 21 is used to obtain image set to be classified;
Optional, said feature extraction comprises: characteristic exponent is extracted, normalized differential vegetation index NDVI extracts, texture transformation.
Optional, said taxon 25 is further used for sample data and many characteristic wave bands combination image input decision tree classification device with said same group image, and takes the mode of ten folding cross validations to test, to obtain training set.
Optional, said taxon 25 is further used for training set with said generation and is input in the decision tree of the binary tree structure of classification of remote-sensing images and learns, to obtain the rule set that is used for classification of remote-sensing images.
Optional, said taxon 25 is further used for iterative technique being applied in the decision tree of above-mentioned binary tree structure through bringing in constant renewal in the weight of each sample, makes up the iteration decision tree, to improve the nicety of grading of rule set.
Embodiment of the invention said method or device technique scheme have following beneficial effect: obtain image set to be classified because adopt; According to the time phase character image set to be classified is divided into some groups, every group image has identical wave spectrum characteristic; Respectively same group image is unified sampling to obtain sample data, reach respectively to carrying out band combination after the same group image feature extraction to obtain many characteristic wave bands combination images; Sample data and many characteristic wave bands combination image of said same group image are input in the decision tree classification device; To obtain the technological means of classification image; So reached following technique effect: a kind of wide area face of land cover classification strategy based on the decision tree classification of remote-sensing images is provided, has improved wide area classification of remote-sensing images precision.
Below lifting application example is elaborated:
As shown in Figure 3; Use the precision evaluation synoptic diagram of wide area remote sensing image decision tree classification for application example of the present invention; Before classification, application example of the present invention at first is divided into some groups according to the time phase character of raw video with image set to be classified, and every group image has identical wave spectrum characteristic; Then every group image is sampled; Simultaneously raw video being carried out characteristic exponent extracts (like KT conversion (kautlr-thomas transformation; The red-tasselled official hat conversion), NDVI (Normalized Difference Vegetation Index, normalized differential vegetation index) index extraction, texture transformation etc.); Then many characteristic wave bands combination images and sample data are input in the decision tree classification device, obtain final classification image through generating training set-create-rule collection-image classification-sequence of operations such as human-edited.
With TM image (the multiband scan-image that 4～No. 5 thematic mappers of U.S.'s Landsat (thematic mapper) are obtained) is example, as shown in Figure 4, is the concrete classification policy synoptic diagram of single scape image in the application example of the present invention; As follows: in this strategy; Application example of the present invention has at first carried out feature extraction to the TM image, promptly do index characteristic extract (, the NDVI index extracts, texture transformation, simultaneously with DEM (the Digital Elevation Model of correspondence; Digital elevation model) resamples; Then The above results is carried out band combination, import sample data and generate the training set that has 15 characteristic variables, it is imported in the decision tree classification device; Utilize the rule set and the many characteristic wave bands combination image that generate to classify, obtain classification results.
Sample collection:
As shown in Figure 5; For synoptic diagram is implemented in the unified sampling of application example grouping of the present invention; As follows: in order to improve the efficient of whole assorting process; Application example of the present invention is implemented phase packet samples on time according to last figure with image to be classified, and each classification all has the corresponding sample collection in the categorizing system that in the practical implementation process, will guarantee to formulate in advance, and the ratio of all kinds of samples is roughly identical with image face of land covering classification ratio.When sample is tested; Need use ten folding cross validation functions of decision tree classification device; Promptly earlier with sample data and the wave band input decision tree classification device that combines; Generate training set, take the mode of ten folding cross validations to test to training set then, so-called ten folding cross validations (English name is called 10-fold cross-validation) are the method for testings of using always.Data set is divided into very, in turn will be wherein 9 parts as training data, 1 part as test data, makes an experiment.Each test all can draw corresponding accuracy (or error rate).The mean value of 10 times result's accuracy (or error rate) generally also need carry out repeatedly 10 folding cross validations (for example 10 times 10 folding cross validations) as the estimation to arithmetic accuracy, asks its average again, as the estimation to the algorithm accuracy.Why selecting data set is divided into 10 parts, is because a large amount of tests through utilizing the mass data collection, using different learning arts to carry out show that 10 foldings are the appropriate selections that obtain best estimation of error.
The decision tree classification device:
As shown in Figure 6, for application example decision tree classification of remote-sensing images practical implementation process flow diagram of the present invention, can combine Fig. 3 to understand decision tree classification implement body implementing procedure in the application example of the present invention.
Single GLC (Globle Land Cover) tree:
In the application example of the present invention, single GLC tree algorithm primary structure is as shown in table 1 below:
Single GLC tree algorithm of table 1 primary structure table
The GLC tree construction:
In numerous decision Tree algorithms; The structure of its tree is divided into two kinds of binary tree and multiway trees, and application example of the present invention is through evidence, in the remote sensing image decision tree classification; Binary tree structure is convenient to handle continuity data (need not do the discretize of connection attribute), precision is high; And the rule set that generates is described simple, no matter be in the generation of rule set or in the last image classification, all more to have superiority than multiway tree like this, so the GLC tree has been adopted binary tree structure in the application example of the present invention.
Wave band selection of threshold index:
GLC tree wave band selection of threshold index has adopted the information gain ratio in the application example of the present invention.Correlation formula and being described below:
The information gain ratio is defined as:
Wherein, information gain Gain (A) is defined as poor between original information requirement (promptly only based on this type of ratio) and the new demand (obtaining after promptly A being divided),
Gain(A)≡Info(D)-Info
A(D)
And quantity of information Info (D) is defined as:
Wherein, p
iBe any tuple type of belonging to C among the D
iProbability.Hypothesis application example of the present invention is divided the tuple among the D according to wave band A now, after the division, also need how much information in order to obtain accurate classification application example of the present invention? This amount is measured by following formula:
Division information SplitInf is similar to Info (D), defines as follows:
This value representative is through being divided into training dataset D v information of dividing generation corresponding to v output of wave band A test.
Stop condition:
Because the GLC tree has been adopted binary tree structure in the application example of the present invention, so formulated the stop condition of following GLC tree growth:
All samples all belong to same type in the sample set of node. and at this moment, this node is set at leaf node.
The wave band value of all samples is identical in the sample set of node, but affiliated classification is different. and indicate this node with most classes in the sample this moment, and be set and be leaf node.
GLC height of tree degree arrives the threshold values that the user is provided with, and indicate this node with most classes in the sample set of node this moment, and is set and is leaf node.
Pruning method:
When GLC tree is created because noise and outlier in the training set, many branches reflections be unusual in the training data.Need handle this undue fitting data problem through pruning method, in application example of the present invention, adopt pessimistic wrong pruning method).Roughly thinking is following:
Suppose that Tt is for being the subtree of root with internal node t.N (t) is for arriving all numbers of samples of node t.E (t) is for arriving node t but do not belong to the number of samples of the classification that node t identified.R (t) is the error sample rate of node t.See that from the angle of probability error sample rate r (t) can regard in the inferior experiment of n (t) certain incident as and the inferior probability of e (t) take place and can obtain a fiducial interval [L about the error sample rate
CF, U
CF], establish the confidence level of CF, L for this fiducial interval is set
CFAnd U
CFBe respectively the lower bound and the upper bound of this fiducial interval.The value of CF can be used for controlling the degree of beta pruning: the high more beta pruning of value is few more, and the low more beta pruning of value is many more. and default value is 0.25 in the application example of the present invention, and the wrong sample rate that divides of hypothesis is obeyed binomial distribution.
Output:
In classification of remote-sensing images, the GLC tree is stored and uses with the structure of tree all is inconvenient, so application example of the present invention converts the GLC tree to rule set when final output, so not only is beneficial to storage, and is beneficial to the enforcement of classification of remote-sensing images.Every rule is represented a leaf node in the GLC tree in application example of the present invention, and all decision conditions on having write down from root node to this leaf node path, and the prediction effect of every paths is that every rule has been given weight in generating according to the GLC tree simultaneously:
Divide position accumulative total hits to be illustrated in the GLC tree growth course, for the correct sample number of this rule prediction of training set.
Divide the position sample number to be illustrated in the GLC tree growth course, use the sample number of this rule for training set.
Total hits are illustrated in the GLC tree growth course, for the correct sample number of training set whole GLC tree prediction.
Total sample number is illustrated in the GLC tree growth course, for total sample number of training set.
Single GLC tree growth course:
If training set D is each wave band value and category label in the corresponding decision tree classification of remote-sensing images strategy of sample point, then the growth course of single GLC tree can be described as:
Output:
(1) creates a node N;
(2) if the tuple among the D all is same type of C, return N as leaf node, with class C mark;
(3) if all wave band values of the tuple among the D are all identical or the height of tree reaches user-defined height, return N, be labeled as most types among the D as leaf node;
(4) calculate candidate's threshold value of each wave band: the intermediate value of per two values;
(5) the information gain ratio under each candidate's threshold value two tunnel is divided in all wave bands among the calculating D is found out best wave band and threshold value;
(6) the two tunnel divide tuple D, and each is divided generation subtree D1, D2;
(7) respectively D1 and D2 are repeated aforesaid operations, until whole arrival leaf nodes;
(8) carry out pessimistic wrong beta pruning.
Iteration GLC tree:
In order to improve the precision of prediction of GLC tree, application example of the present invention is incorporated into iterative technique in the use of GLC tree, and in iteration, each sample all is endowed a weight, shows that it is selected into the probability of training set by certain GLC tree.If certain sample point is by classification exactly, in the next training set of structure, its selected probability just is lowered so; On the contrary, if certain sample point not by correct classification, its weight just obtains improving so.Through such mode, can " focus on " those the difficulty (richer information) sample on.On concrete the realization, make the weight of each sample all equate at first.For the k time iterative operation, application example of the present invention is chosen sample point according to these weights, and then training GLC tree C
kAccording to this GLC tree, improve by the weight of its wrong those sample that divide then, and reduce can be by the sample weights of correct classification, the sample set that upgraded of weight is used to train next GLC to set C then
K+1Whole training process so goes on.The judgement of last overall classification can use each GLC tree weighted mean to obtain.
For the ease of arthmetic statement, application example of the present invention is established total n the sample of training sample set S, and T promptly carries out T time sample training altogether for the number (iterations) that the GLC that makes up sets.The decision-tree model that has the t time training to produce is designated as C
tThe compound decision tree that is finally combined to obtain by this T GLC tree-model is designated as C
* For the weight of sample i in the t time GLC tree building process (i=1,2 ..., n; T=1,2 ..., T),
For
Normalized factor, β
tThe adjustment factor for weighted value.
Define a 0-1 function:
The step of sample training is following:
(1) sets initial weight: the number T (being 10 generally speaking) that sets the GLC tree that makes up.Make t=1,
(2) calculate
; (making
).
(3) give normalized weight for each sample
On the basis of this probability distribution, make up C
t
(4) calculate the error rate
of t GLC tree to sample
(5) if ε
t>0.5, then finish whole training process, make T=T-1; If ε
t=0, finish whole training process, make t=T; If 0<ε
t≤0.5, then continue step 6).
(6) calculate β
t=ε
t/ (1-ε
t).
(7) according to the error rate weighted value of new samples more:
(8) if t=T, training process finishes.Otherwise, make t=t+1, go to step 2) and carry out training next time.
The purpose of application example of the present invention is to provide a kind of wide area face of land to cover iteration decision tree classification method and system.Establishment is applied to single decision tree of classification of remote-sensing images---the GLC tree; Iterative technique is incorporated in the GLC tree; Formulating wide area face of land covering decision tree classification strategy is three innovation point of application example of the present invention: 1, classification policy: application example of the present invention is according to remote sensing image wave spectrum characteristic; Remote sensing image to pending divides into groups on time mutually; The generation of Sample selection, training set and rule set is carried out in unification, utilizes the unified rule set that generates to carrying out batch processing with group image at last.Formulated the concrete classification policy of TM image simultaneously according to the characteristics of TM image.2, the introducing of GLC tree establishment scheme and iterative technique: the establishment scheme of GLC tree and the introducing of iterative technique are the cores of application example of the present invention; Application example of the present invention is first according to the characteristics of classification of remote-sensing images; Formulated the decision Tree algorithms GLC tree that is specifically designed to classification of remote-sensing images; And first iterative technique is incorporated in the decision tree classification of remote-sensing images, design has realized remote sensing image iteration decision tree classification system.3, the one-piece construction of remote sensing image iteration decision tree classification system: the support system that application example of the present invention designed has not only satisfied whole demands of application example scheme of the present invention, and has realized some subsidiary classifications and simple statistical classification function.Its one-piece construction has been reacted the techniqueflow of application example scheme of the present invention; Simultaneously application example of the present invention has also designed some internal file forms, and like rule set: application example of the present invention converts the GLC tree that generates storage and the use of rule set to make things convenient for DECISION KNOWLEDGE to.
The characteristics that application example of the present invention is at first classified according to the wide area remote sensing image decision tree; Formulated classification policy, created the decision tree growth algorithm that is applicable to classification of remote-sensing images then---the GLC tree, in order to improve nicety of grading; Application example of the present invention is incorporated into iterative technique in the application of GLC tree; Formed the iteration decision tree, and realized remote sensing image iteration decision tree classification system, for the wide area classification of remote-sensing images strategy of formulating provides technical support according to this iteration decision tree design.The beneficial effect that application example technical scheme of the present invention is brought:
1, the overall technology flow process of having formulated big regional remote sensing image face of land cover classification has made full use of the advantage based on the decision Tree algorithms classification of remote-sensing images; Making becomes possibility to the face of land cover classification of carrying out mass with the remote sensing image under the phase condition for the moment; Improved the big regional face of land greatly and covered classification of remote-sensing images speed, strong technical support is provided for the big regional face of land covers classification of remote-sensing images.
2, created the decision Tree algorithms GLC tree that is applicable to classification of remote-sensing images, and iterative technique is incorporated in the use of GLC tree, improved the nicety of grading of GLC tree greatly, design has simultaneously realized remote sensing image iteration decision tree classification system.
3, formulate the classification policy of TM image; In classification, not only utilized the wave spectrum characteristic of himself; And utilized that its TC conversion, NDVI index extract, result and DEM resampling result behind the texture transformation; Improve the nicety of grading of TM image greatly, and in support system, realized corresponding function.
Those skilled in the art can also recognize the various illustrative components, blocks (illustrative logical block) that the embodiment of the invention is listed, and unit and step can be passed through electronic hardware, computer software, or both combinations realize.Be the clear replaceability (interchangeability) of showing hardware and software, above-mentioned various illustrative components (illustrative components), unit and step have been described their function generally.Such function is to realize depending on the designing requirement of certain applications and total system through hardware or software.Those skilled in the art can be for every kind of certain applications, and can make ins all sorts of ways realizes described function, but this realization should not be understood that to exceed the scope of embodiment of the invention protection.
Various illustrative logical block described in the embodiment of the invention; Or the unit can pass through general processor, digital signal processor, special IC (ASIC); Field programmable gate array (FPGA) or other programmable logic device; Discrete gate or transistor logic, discrete hardware components, or the design of above-mentioned any combination realizes or operates described function.General processor can be microprocessor, and alternatively, this general processor also can be any traditional processor, controller, microcontroller or state machine.Processor also can realize through the combination of calculation element, for example digital signal processor and microprocessor, a plurality of microprocessors, Digital Signal Processor Core of one or more microprocessors associatings, or any other similarly configuration realize.
The method described in the embodiment of the invention or the step of algorithm can directly embed hardware, the software module of processor execution or the two combination.Software module can be stored in the storage medium of other arbitrary form in RAM storer, flash memory, ROM storer, eprom memory, eeprom memory, register, hard disk, moveable magnetic disc, CD-ROM or this area.Exemplarily, storage medium can be connected with processor, so that processor can read information from storage medium, and can deposit write information to storage medium.Alternatively, storage medium can also be integrated in the processor.Processor and storage medium can be arranged among the ASIC, and ASIC can be arranged in the user terminal.Alternatively, processor and storage medium also can be arranged in the various parts in the user terminal.
In one or more exemplary designs, the described above-mentioned functions of the embodiment of the invention can realize in hardware, software, firmware or this three's combination in any.If in software, realize, these functions can be stored on the media with computer-readable, or are transmitted on the media of computer-readable with one or more instructions or code form.The computer-readable media comprises that the computer storage medium lets computer program transfer to other local telecommunication media from a place with being convenient to make.Storage medium can be the useable medium that any general or special computer can access.For example; Such computer readable media can include but not limited to RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage device, or other any can be used to carry or store with instruction or data structure and other can be read the media of the program code of form by general or special computer or general or special processor.In addition; Any connection can suitably be defined as the computer-readable media; For example, if software is through a concentric cable, fiber optic cables, twisted-pair feeder, Digital Subscriber Line (DSL) or also being comprised in the defined computer-readable media with wireless mode transmission such as for example infrared, wireless and microwaves from a web-site, server or other remote resource.Described video disc (disk) and disk (disc) comprise Zip disk, radium-shine dish, CD, DVD, floppy disk and Blu-ray Disc, and disk is usually with the magnetic duplication data, and video disc carries out the optical reproduction data with laser usually.Above-mentioned combination also can be included in the computer-readable media.
Above-described embodiment; The object of the invention, technical scheme and beneficial effect have been carried out further explain, and institute it should be understood that the above is merely embodiment of the present invention; And be not used in qualification protection scope of the present invention; All within spirit of the present invention and principle, any modification of being made, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (10)
1. a wide area remote sensing image decision tree sorting technique is characterized in that, said wide area remote sensing image decision tree sorting technique comprises:
Obtain image set to be classified;
According to the time phase character image set to be classified is divided into some groups, every group image has identical wave spectrum characteristic;
Respectively same group image is unified sampling to obtain sample data, reach respectively to carrying out band combination after the same group image feature extraction to obtain many characteristic wave bands combination images;
Sample data and many characteristic wave bands combination image of said same group image are input in the decision tree classification device, to obtain the classification image.
2. wide area remote sensing image decision tree sorting technique according to claim 1, it is characterized in that said feature extraction comprises: characteristic exponent is extracted, normalized differential vegetation index NDVI extracts, texture transformation.
3. wide area remote sensing image decision tree sorting technique according to claim 1 is characterized in that, sample data and many characteristic wave bands combination image of said same group image is input in the decision tree classification device, to obtain the classification image, comprising:
With sample data and many characteristic wave bands combination image input decision tree classification device of said same group image, and take the mode of ten folding cross validations to test, to obtain training set.
4. wide area remote sensing image decision tree sorting technique according to claim 1 is characterized in that, sample data and many characteristic wave bands combination image of said same group image is input to the decision tree classification device, to obtain the classification image, comprising:
The training set of said generation is input in the decision tree of the binary tree structure of classification of remote-sensing images learns, to obtain the rule set that is used for classification of remote-sensing images.
5. like the said wide area remote sensing image decision tree of claim 4 sorting technique, it is characterized in that, the sample data of said same group image and many characteristic wave bands combination image be input in the decision tree classification device,, comprising to obtain the classification image:
Through bringing in constant renewal in the weight of each sample, iterative technique is applied in the decision tree of said binary tree structure, make up the iteration decision tree, to improve the nicety of grading of rule set.
6. a wide area remote sensing image decision tree sorter is characterized in that, said wide area remote sensing image decision tree sorter comprises:
Acquiring unit is used to obtain image set to be classified;
Grouped element, be used for according to the time phase character image set to be classified is divided into some groups, every group image has identical wave spectrum characteristic;
Unified sampling unit is used for respectively same group image being unified sampling to obtain sample data;
The band combination unit is used for respectively to carrying out band combination after the same group image feature extraction to obtain many characteristic wave bands combination images;
Taxon is used for sample data and many characteristic wave bands combination image of said same group image are input to the decision tree classification device, to obtain the classification image.
7. like the said wide area remote sensing image decision tree of claim 6 sorter, it is characterized in that said feature extraction comprises: characteristic exponent is extracted, normalized differential vegetation index NDVI extracts, texture transformation.
8. like the said wide area remote sensing image decision tree of claim 6 sorter, it is characterized in that,
Said taxon is further imported the decision tree classification device with sample data and many characteristic wave bands combination image of said same group image, and is taked the mode of ten folding cross validations to test, to obtain training set.
9. like the said wide area remote sensing image decision tree of claim 6 sorter, it is characterized in that,
Said taxon further is input to the training set of said generation in the decision tree of the binary tree structure of classification of remote-sensing images and learns, to obtain the rule set that is used for classification of remote-sensing images.
10. like the said wide area remote sensing image decision tree of claim 9 sorter, it is characterized in that,
Said taxon further through bringing in constant renewal in the weight of each sample, is applied to iterative technique in the decision tree of said binary tree structure, makes up the iteration decision tree, to improve the nicety of grading of rule set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210297464.4A CN102831440B (en) | 2012-08-20 | 2012-08-20 | Method and device for decision tree based wide-area remote sensing image classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210297464.4A CN102831440B (en) | 2012-08-20 | 2012-08-20 | Method and device for decision tree based wide-area remote sensing image classification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102831440A true CN102831440A (en) | 2012-12-19 |
| CN102831440B CN102831440B (en) | 2015-03-18 |
Family
ID=47334568
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210297464.4A Active CN102831440B (en) | 2012-08-20 | 2012-08-20 | Method and device for decision tree based wide-area remote sensing image classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102831440B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103500344A (en) * | 2013-09-02 | 2014-01-08 | 中国测绘科学研究院 | Method and module for extracting and interpreting information of remote-sensing image |
| CN103902591A (en) * | 2012-12-27 | 2014-07-02 | 中国科学院深圳先进技术研究院 | Decision tree classifier establishing method and device |
| CN105893970A (en) * | 2016-03-31 | 2016-08-24 | 杭州电子科技大学 | Nighttime road vehicle detection method based on luminance variance characteristics |
| CN106874339A (en) * | 2016-12-20 | 2017-06-20 | 北京华宇信息技术有限公司 | A kind of methods of exhibiting of circulant Digraph and its application |
| CN107194464A (en) * | 2017-04-25 | 2017-09-22 | 北京小米移动软件有限公司 | The training method and device of convolutional neural networks model |
| CN107967696A (en) * | 2017-11-23 | 2018-04-27 | 湖南文理学院 | A kind of water surface ship radar remote sensing detection method, electronic equipment |
| CN108197225A (en) * | 2017-12-28 | 2018-06-22 | 广东欧珀移动通信有限公司 | Sorting technique, device, storage medium and the electronic equipment of image |
| CN109472304A (en) * | 2018-10-30 | 2019-03-15 | 厦门理工学院 | Tree species classification method, device and equipment based on SAR Yu optical remote sensing time series data |
| CN110837875A (en) * | 2019-11-18 | 2020-02-25 | 国家基础地理信息中心 | Method and device for judging quality abnormity of earth surface coverage data |
| CN111401467A (en) * | 2020-03-26 | 2020-07-10 | 遥相科技发展(北京)有限公司 | Remote sensing image classification method, image processing equipment and computer readable storage device |
| CN113592770A (en) * | 2021-06-23 | 2021-11-02 | 中国科学院南京地理与湖泊研究所 | Algal bloom remote sensing identification method for removing influence of aquatic weeds |
| CN115248876A (en) * | 2022-08-18 | 2022-10-28 | 北京数慧时空信息技术有限公司 | Remote sensing image overall planning recommendation method based on content understanding |
| CN117786544A (en) * | 2024-02-28 | 2024-03-29 | 浪潮通信信息***有限公司 | User satisfaction obtaining method and device, electronic equipment and storage medium |
-
2012
- 2012-08-20 CN CN201210297464.4A patent/CN102831440B/en active Active
Non-Patent Citations (3)
| Title |
|---|
| 姜丽华，杨晓蓉: "《基于决策树分类技术的遥感影像分类方法研究》", 《农业网络信息》 * |
| 潘琛 等: "《一种基于植被指数的遥感影像决策树分类方法》", 《计算机应用》 * |
| 罗来平 等: "《遥感图像决策树分类器研究与实现》", 《遥感信息》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103902591A (en) * | 2012-12-27 | 2014-07-02 | 中国科学院深圳先进技术研究院 | Decision tree classifier establishing method and device |
| CN103902591B (en) * | 2012-12-27 | 2019-04-23 | 中国科学院深圳先进技术研究院 | Construct the method and device of decision tree classifier |
| CN103500344B (en) * | 2013-09-02 | 2017-02-08 | 中国测绘科学研究院 | Method and module for extracting and interpreting information of remote-sensing image |
| CN103500344A (en) * | 2013-09-02 | 2014-01-08 | 中国测绘科学研究院 | Method and module for extracting and interpreting information of remote-sensing image |
| CN105893970A (en) * | 2016-03-31 | 2016-08-24 | 杭州电子科技大学 | Nighttime road vehicle detection method based on luminance variance characteristics |
| CN106874339A (en) * | 2016-12-20 | 2017-06-20 | 北京华宇信息技术有限公司 | A kind of methods of exhibiting of circulant Digraph and its application |
| CN107194464A (en) * | 2017-04-25 | 2017-09-22 | 北京小米移动软件有限公司 | The training method and device of convolutional neural networks model |
| CN107194464B (en) * | 2017-04-25 | 2021-06-01 | 北京小米移动软件有限公司 | Training method and device of convolutional neural network model |
| CN107967696A (en) * | 2017-11-23 | 2018-04-27 | 湖南文理学院 | A kind of water surface ship radar remote sensing detection method, electronic equipment |
| CN108197225A (en) * | 2017-12-28 | 2018-06-22 | 广东欧珀移动通信有限公司 | Sorting technique, device, storage medium and the electronic equipment of image |
| CN109472304B (en) * | 2018-10-30 | 2021-04-16 | 厦门理工学院 | Tree species classification method, device and equipment based on SAR and optical remote sensing time sequence data |
| CN109472304A (en) * | 2018-10-30 | 2019-03-15 | 厦门理工学院 | Tree species classification method, device and equipment based on SAR Yu optical remote sensing time series data |
| CN110837875A (en) * | 2019-11-18 | 2020-02-25 | 国家基础地理信息中心 | Method and device for judging quality abnormity of earth surface coverage data |
| CN111401467A (en) * | 2020-03-26 | 2020-07-10 | 遥相科技发展(北京)有限公司 | Remote sensing image classification method, image processing equipment and computer readable storage device |
| CN111401467B (en) * | 2020-03-26 | 2023-12-19 | 苏州征图智能科技有限公司 | Remote sensing image classification method, image processing device and computer readable storage device |
| CN113592770A (en) * | 2021-06-23 | 2021-11-02 | 中国科学院南京地理与湖泊研究所 | Algal bloom remote sensing identification method for removing influence of aquatic weeds |
| CN113592770B (en) * | 2021-06-23 | 2024-02-23 | 中国科学院南京地理与湖泊研究所 | Algal bloom remote sensing identification method for removing influence of aquatic weeds |
| CN115248876A (en) * | 2022-08-18 | 2022-10-28 | 北京数慧时空信息技术有限公司 | Remote sensing image overall planning recommendation method based on content understanding |
| CN115248876B (en) * | 2022-08-18 | 2024-05-10 | 北京数慧时空信息技术有限公司 | Remote sensing image overall recommendation method based on content understanding |
| CN117786544A (en) * | 2024-02-28 | 2024-03-29 | 浪潮通信信息***有限公司 | User satisfaction obtaining method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102831440B (en) | 2015-03-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102831440A (en) | Method and device for decision tree based wide-area remote sensing image classification | |
| CN105184307B (en) | A kind of generation method of medical domain image, semantic similarity matrix | |
| CN110363049B (en) | Method and device for detecting, identifying and determining categories of graphic elements | |
| CN110414577A (en) | A kind of laser radar point cloud multiple target Objects recognition method based on deep learning | |
| CN110378331A (en) | A kind of end-to-end Vehicle License Plate Recognition System and its method based on deep learning | |
| WO2022134794A1 (en) | Method and apparatus for processing public opinions about news event, storage medium, and computer device | |
| CN102208037B (en) | Hyper-spectral image classification method based on Gaussian process classifier collaborative training algorithm | |
| Fu et al. | Comparison of optimized object-based RF-DT algorithm and SegNet algorithm for classifying Karst wetland vegetation communities using ultra-high spatial resolution UAV data | |
| AU2021101918A4 (en) | Method and system of fusion classification of rock spectral signature | |
| CN104834693A (en) | Depth-search-based visual image searching method and system thereof | |
| CN107194498B (en) | Hydrologic monitoring network optimization method | |
| CN102543074B (en) | Agricultural product data acquisition system as well as voice recognition system and method of mobile equipment | |
| CN104462063B (en) | Positional information structuring extracting method based on semantic locations model and system | |
| CN104462301A (en) | Network data processing method and device | |
| CN116681176B (en) | Traffic flow prediction method based on clustering and heterogeneous graph neural network | |
| CN104035996A (en) | Domain concept extraction method based on Deep Learning | |
| CN104376335A (en) | Semi-supervised hyperspectral remote sensing image classification method based on information entropies | |
| CN117235647B (en) | Mineral resource investigation business HSE data management method based on edge calculation | |
| CN106951455A (en) | A kind of similar track analysis system and its analysis method | |
| Rousi et al. | Semantically enriched crop type classification and linked earth observation data to support the common agricultural policy monitoring | |
| CN109448366A (en) | A kind of space domain sector degree of crowding prediction technique based on random forest | |
| CN115205481A (en) | Spectrum map construction method and system based on graph neural network | |
| CN104778480A (en) | Hierarchical spectral clustering method based on local density and geodesic distance | |
| CN109063752A (en) | The method for sorting of the multiple dimensioned real-time stream of multi-source higher-dimension neural network based | |
| CN102208027B (en) | Method for evaluating land utilization spatial pattern based on clearance degree dimension |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |