CN102542291A - Hyperspectral remote sensing image classification method based on binary decision tree - Google Patents
Hyperspectral remote sensing image classification method based on binary decision tree Download PDFInfo
- Publication number
- CN102542291A CN102542291A CN2011104404689A CN201110440468A CN102542291A CN 102542291 A CN102542291 A CN 102542291A CN 2011104404689 A CN2011104404689 A CN 2011104404689A CN 201110440468 A CN201110440468 A CN 201110440468A CN 102542291 A CN102542291 A CN 102542291A
- Authority
- CN
- China
- Prior art keywords
- classification
- decision tree
- node
- threshold value
- 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
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
A hyperspectral remote sensing image classification method based on a binary decision tree is used for performing high-efficiency and high-precision classification for hyperspectral remote sensing images. The binary decision tree is used for classifying the hyperspectral remote sensing images, so that the decision tree classification technology is combined with hyperspectral remote sensing image classification, the most useful information in hyperspectral data is sufficiently mined by taking the advantages of the decision tree, and recognition capability and classification precision for ground objects are improved. The hyperspectral remote sensing image classification method in the technical scheme is mainly characterized in that the binary decision tree is used, a tree node data structure is defined, a decision tree algorithm is designed, and selection rules of candidate threshold in the algorithm and selection rules of optimal threshold and wave bands are provided. The hyperspectral remote sensing image classification method can be used for selecting the wave bands of the hyperspectral remote sensing images, generating classification rules, realizing 'integration' and 'automation' of image classification, and achieving high-efficiency and high-precision classification effects for the hyperspectral remote sensing images.
Description
Technical field
The present invention is a kind of sorting technique of target in hyperspectral remotely sensed image, is mainly used in the classification problem that solves Hyperspectral imaging, belongs to the treatment of remote field.
Background technology
Because space flight, satellite, statistics and fast development of computer technology, we can obtain the remote sensing image data of high spectral resolution, and image classification is an important link effectively utilizing these data.Image classification is one of important applied field of remote sensing technology, and it can directly and vivo show the character of silhouette target.Advantages such as at present, the main computer automatic sorting technology that adopts of target in hyperspectral remotely sensed image classification realizes, computer automatic sorting has the precision height, is applicable to quantitative test, rapid speed, reusability are good.How to make computer classification fast, the accurate bulk information of ability and the target in hyperspectral remotely sensed image of deal with data present one of main direction of remote sensing field theory and application research that become that adapts.
The target in hyperspectral remotely sensed image data recording continuous spectrum of ground object target, the information that comprises is horn of plenty more, has possessed the ability of discerning a greater variety of ground object targets and carrying out target classification with higher precision.But its huge data volume causes certain difficulty to information extraction.The method commonly used of target in hyperspectral remotely sensed image classification at present is: earlier image data is carried out dimensionality reduction, like feature selecting, feature extraction etc., and then with different sorting techniques image is classified.Can guarantee to participate in the quantity of information that wave band the comprised maximum of classification like this, the data redundancy between wave band is minimum, thereby reduces the complexity of classification.Do the efficient that has improved classification to a certain extent like this, but before classification, must carry out the dimensionality reduction operation to data separately, make the workload of whole classification increase, do not realize " integrated ", " robotization " of classification fully.To the classification of target in hyperspectral remotely sensed image, a lot of traditional sorting techniques have all demonstrated their deficiency.Research shows with practice: the decision tree classification algorithm is compared with traditional classification of remote-sensing images method, and it is higher to have nicety of grading, advantages such as the classifying rules of generation is directly perceived, easy to understand.
Present decision tree classification method is used wider in other classification of remote-sensing images, but in the target in hyperspectral remotely sensed image classification, uses less.And in the classification of remote-sensing images research of great majority based on decision tree; The foundation of decision tree is not full automatic; General earlier through artificial constructed decision tree; Again by other programs or business software to image classification, very flexible, inefficiency, be unfavorable for making full use of image information and supplementary knowledge is learned on ground.Therefore; System of research and design; Target in hyperspectral remotely sensed image is not being carried out under the situation of dimensionality reduction in advance, letting decision tree select at last image to be classified classifying best wave band and make up classifying rules; Thereby the realization band selection, " integrated " of the generation of classifying rules and image classification and " robotization " very are necessary.
Mainly consider the problem of 4 aspects based on the target in hyperspectral remotely sensed image sorting technique of decision tree: the design of (1) decision tree; (2) decision Tree algorithms design; (3) system of selection of threshold value; (4) definite method of optimal threshold and wave band.The decision tree designing principle mainly contains: select suitable tree construction, rationally arrange node and the branch of tree; Confirm the characteristic that on each nonterminal node, will use; On each nonterminal node, select suitable decision rule.Wherein the binary decision tree structure is directly perceived, is convenient to analyze and understand, and to input data space characteristic and group indication, good elasticity and robustness is arranged, and it is simple, flexible that structure gets up, and has good classifying quality.The design relation of decision Tree algorithms last efficient and the classifying quality that generates decision tree.The definite of threshold value will consider: guarantee to find out all real threshold values of present node; When judging whether an intersection point can be used as threshold value and insert, need to judge this with this intersection point as threshold value whether not effective to region class that present node remains.The confirming of optimal threshold and wave band can guarantee that selected threshold value can distinguish two kinds well, and the advantage of used wave band is greater than its all band.
Summary of the invention
Technical matters:The object of the invention just provides a kind of new target in hyperspectral remotely sensed image sorting technique based on binary decision tree; Solve the classification problem of target in hyperspectral remotely sensed image; This mechanism is a kind of tactic method; Through using this method can realize the target in hyperspectral remotely sensed image band selection, " integrated " of the generation of classifying rules and image classification and " robotization " are for classification provides higher efficient and precision.
Technical scheme:Method of the present invention is a kind of method of tactic; Realize the classification of target in hyperspectral remotely sensed image through a kind of binary decision tree; Thereby the decision tree classification technology is combined with the target in hyperspectral remotely sensed image classification; Utilize the advantage of decision tree, fully excavate Useful Information in the high-spectral data, improved the recognition capability and the nicety of grading of ground object target.
One, architecture
Fig. 1 has provided the target in hyperspectral remotely sensed image categorizing system structural drawing based on binary decision tree, and it mainly comprises five parts: image reads and display module, sample collection module, decision tree generation module, image classification module and classification results precision evaluation module.Decision tree generation module among the figure is the core of native system, has mainly comprised Data Structure Design, decision Tree algorithms design, the system of selection of candidate's threshold value and the optimal threshold and band selection method four parts of decision tree nodes.Generate a binary decision tree that is used for the target in hyperspectral remotely sensed image classification through this module.The present invention has proposed a kind of binary decision tree sorting algorithm in the decision tree generation module, improved the efficient and the precision of target in hyperspectral remotely sensed image classification.
Provide concrete introduction below:
The data structure of decision tree nodes:Data Structure Design is most important, and a good data structure can make algorithm refining more, thereby improves development efficiency.The Data Structure Design of decision tree is primarily aimed at the Data Structure Design of each node that constitutes decision tree, and the tree node structural design is extremely important for making up decision tree, has had enough nodal informations could set up out rational decision tree.Each nodal information of tree all is stored in the self-defining structure.
Decision Tree algorithms:Sample information according to statistics; Begin from the decision tree root node; Be followed successively by each node and select suitable division rule; The atural object classification that present node comprised is divided into two big types, can not be further divided into up to node and ends (be that present node only comprises a kind or do not have suitable division rule according to current fragmentation criterion, the classification that causes present node to comprise can not be segmented) again.
Candidate's threshold value selective rule:At each node place that waits to make a strategic decision, calculate the intersection point between adjacent two classifications on each wave band according to the intersection point calculation formula, if this intersection point satisfies standard, then this intersection point is noted as current candidate's threshold value.
Optimal threshold and band selection rule:Among all candidate's threshold values of present node, the Standard Selection of selecting according to optimal threshold goes out optimal threshold, and this threshold value can be divided current feature space.Can guarantee that so selected threshold value can distinguish two kinds well, the advantage of used wave band is greater than its all band.
Two, method flow
1, decision tree nodes data structure
Each nodal information of tree all is stored in the self-defining structure.Comprised 7 variablees in the decision tree nodes structure, decided classification number to be classified respectively, treat class categories numbering, decision-making characteristic quantity numbering, discrimination threshold, left child node pointer, right child node pointer, father node pointer.Detailed data structure is following:
Struct node // tree node structure
{
Int number; // classification number to be classified
Int category [classNum]; // treat class categories numbering
Int mode; // decision-making characteristic quantity numbering
Float value; // discrimination threshold
Int left; // left child node pointer
Int right; // right child node pointer
Int father; // father node pointer
};
Struct node [2*classNum-1]; The node of // binary decision tree
2, decision Tree algorithms
Obtaining when making up the required priori of decision tree, mainly utilizing the spectral information of target in hyperspectral remotely sensed image.Generally the atural object classification satisfies the characteristic of normal distribution in the one-dimensional characteristic space, just can confirm that according to the average μ and the standard deviation sigma of classification in feature space each classification is interval in the characteristic distribution at each wave band place.Before making up decision tree, through the atural object of each classification is carried out the sampling of gray-scale value, calculate average and the standard deviation of every classification on each wave band respectively, when inserting threshold value, be chosen in selected type of right intersection point place and insert.Under the situation of normal distribution, on each wave band in twos the intersection point between classification be exactly the intersection point of two normal probability paper densimetric curves, its computing formula does
μ wherein
1, σ
1, μ
2, σ
2Sample average and the standard deviation of representing two types of atural objects respectively.Be positioned at the intersection point left side if treat judging point, then it be classified as left side class,, then he is classified as right type if be positioned at the intersection point right side.In normal distribution, the probability that sample characteristics falls into [μ-2 σ, μ+2 σ], [μ-σ, μ+σ] scope is respectively 95.4%, 68.3%.
Separability between two types of atural object can be weighed with the overlapping degree of two types of pairing two normal distribution curves of atural object, and it mainly comprises three kinds of situation: separability is fine between (1) two type of atural object.As shown in Figure 2, the probability density curve of atural object classification is divided into A, B, C, D, five zones of E.The a-quadrant is less than μ-2 σ, and the B zone is [μ-2 σ, μ-σ], and the C zone is [μ-σ, μ+σ], and the D zone is [μ+σ, μ+2 σ], and the E zone is greater than μ+2 σ.If when the intersection point of the probability density curve of two types of atural objects was positioned at A or the E zone of two types of atural object probability density curves, overlapping less between type, the separability of two types of atural objects was fine.Separability is general between (2) two types of atural objects.When the intersection point of the probability density curve of two types of atural objects was positioned at B or the D zone of two types of atural object probability density curves, two types of atural objects existed slight overlapping phenomenon, and separability is general.Separability is relatively poor between (3) two types of atural objects.If the probability density curve intersection point of two types of atural objects is positioned at the C interval, then two types of atural object overlapping phenomenons are serious, and separability is relatively poor.
Make up before the decision tree,, confirm training area, each classification is carried out the sample collection of some, obtain the needed priori of contributing through statistical computation according to the definite atural object classification that exists of existing knowledge and experience; When setting up binary decision tree, begin from root node, each node is judged the separability of its contained classification, confirm to select for use the size of which wave band as decision-making characteristic and decision-making value at this node place; If can divide, then improve the information of this node.The specific algorithm thinking is following:
(1), obtains to wait to distinguish classification and count number and treat other information of region class category [classNum] at the node place of a decision tree;
(2) the sample gray-scale value according to each classification carries out statistical computation, draws gray average and the standard deviation on each wave band of all categories, and the size according to gray average sorts to the atural object classification on each wave band;
(3) calculate the intersection point between adjacent two classifications on each wave band according to formula
; If this intersection point satisfies standard, then this intersection point is noted as current candidate's threshold value;
(4) among all candidate's threshold values of present node, the Standard Selection of selecting according to optimal threshold goes out optimal threshold, and this threshold value can be divided current feature space;
(5) node that is as the criterion when pre-treatment with this threshold value just can generate two node; Left side child node comprises the classification of eigenwert less than threshold value; Right child node comprises the classification of eigenwert greater than threshold value, adds up the number and category [classNum] information of child node simultaneously.
(6) circulation above-mentioned steps successively according to nodal information, is handled all nodes, accomplishes the structure of decision tree.
When making up binary decision tree, at first the standard of foundation is not overlapping between classification, and all leaf nodes of traverse tree mix other leaf node existence if comprise in addition then, then debase the standard, and allow classification slightly overlapping, continue to make up decision tree.The process that decision tree makes up begins from root node exactly, as much as possible each node is divided into two node, in not having appropriate threshold or node, treat sub-category number be one cause dividing again till.After accomplishing the achievement of decision tree, travel through whole tree, if the number of leaf node equals the classification number, then this decision tree can be distinguished all atural object classifications; If the leaf node number is less than the classification number, then this decision tree can not come out all class discriminations.Set up the corresponding relation of classification that leaf node comprises and true atural object classification according to the classification information in the leaf node, set up the discriminant classification rule simultaneously.
The selective rule of candidate's threshold value
When confirming threshold value, will consider two problems: all real threshold values of present node will be guaranteed to find out in (1); (2) when judging whether an intersection point can be used as threshold value and insert, need to judge this with this intersection point as threshold value whether not effective to region class that present node remains.
In the process that makes up tree, when handling a node at every turn, all to remain sub-category gray average and sort, according to dividing standard insertion threshold value to current according to their samples.To judge not only whether this threshold value can will judge also whether this intersection point can be with the sub-category two big class of dividing into of remain with two types of adjacent differentiations.
Stipulate that each classification is only included by a leaf node, promptly when handling some child nodes, the classification of before having been distinguished is just no longer participated in judgement.The degree of depth of control decision tree and the quantity of node make the simple in structure clear of tree, the decision rule easy to understand so effectively.
The selective rule of optimal threshold and wave band
The key issue that in the selection of decision-making characteristic and threshold value, will solve is: because the redundance of high-spectral data is higher; The information that close wave band possibly comprise is very approaching; This will make that the characteristic distribution curve of atural object is also quite similar on these wave bands; If can insert threshold value in these wave bands, select for use which wave band best as the used characteristic of decision tree present node? And in same wave band, the situation of a plurality of threshold values also can appear inserting; Select for use which threshold value classifying quality best actually? The selection of threshold value is except meeting influences the decision rule of present node; Also can the decision rule and the classification division of descendants's node of this node be impacted, select different threshold values can construct different decision trees, classification effectiveness of tree and precision also can differ widely.The invention is intended to select best wave band and best threshold value, the way that addresses these problems is the gauged distance between computation of mean values, and its formula is following:
μ wherein
1, μ
2, σ
1, σ
2Be respectively intersection point corresponding two classifications in the average and the standard deviation of feature space.μ
1, μ
2Gap is big more, and between class distance is big more; σ
1, σ
2More little, then quasi-cohesion is good more, and pairing threshold value is as the judgment rule of present node when getting the d maximum.Can guarantee that so selected threshold value can distinguish two kinds well, the advantage of used wave band is greater than its all band.
Three, beneficial effect
The inventive method has proposed a kind of target in hyperspectral remotely sensed image sorting technique based on binary decision tree; Be mainly used in the classification problem that solves target in hyperspectral remotely sensed image; The method that proposes in the application of the invention can be classified to target in hyperspectral remotely sensed image fast and effectively; Realize the target in hyperspectral remotely sensed image band selection, " integrated " of the generation of classifying rules and image classification and " robotization ".
Provide bright specifically below.
The tree node data structureEach nodal information of tree all is stored in the self-defining structure variable.7 variablees have been comprised in the decision tree nodes structure; Classification number wherein to be classified has been described the quantity of present node place atural object classification to be classified; Treat that class categories numbering described present node place wait the to classify numbering of atural object; Decision-making characteristic quantity numbering has been described the numbering of the decision-making characteristic quantity of present node place employing, and discrimination threshold has been described the discrimination threshold size that the present node place is used, the left child node of left child node pointed present node; The right child node of right child node pointed present node, the father node of father node pointed present node.Above-mentioned Data Structure Design has satisfied needed whole nodal information when decision tree generates in the target in hyperspectral remotely sensed image classification, makes algorithm more terse, has improved development efficiency.
Decision Tree algorithmsConsider from dual precision aspect, at first with not overlapping between classification be standard, obtain at current decision node place and treat class categories number and classification information to be distinguished; Sample gray-scale value according to each classification carries out statistical computation, draws gray average and the standard deviation on each wave band of all categories, and the size according to gray average on each wave band sorts to the atural object classification; Calculate the intersection point of adjacent two classifications on each wave band according to the intersection point calculation formula; Can divide standard if intersection point satisfies, then this intersection point noted as candidate's threshold value, in all candidate's threshold values of present node, select optimal threshold according to the choice criteria of optimal threshold; Make up decision rule with this threshold value and the corresponding wave band of threshold value; Current treating sub-categoryly is divided into two big types, adds up the child node relevant information simultaneously, the circulation above-mentioned steps; Once handle all nodes, accomplish the structure of decision tree according to nodal information.All leaf nodes of traverse tree mix other leaf node existence if comprise in addition then, then debase the standard, and allow classification slightly overlapping, continue to make up decision tree.This algorithm can obtain the higher classification results of precision, simultaneously can be from distinguishing atural object to the full extent.
The selective rule of candidate's threshold valueBe in the achievement process; During node of each processing; All to remain sub-category gray average and sort,, will judge not only whether this threshold value can be with two types of adjacent differentiation according to dividing standard to current according to their samples; Also to judge this intersection point whether can with remain sub-categoryly to divide into two big types, thereby confirm candidate's threshold value.Can guarantee to find out all effective candidate's threshold values of present node like this.
The selective rule of optimal threshold and wave bandSelect for use gauged distance between average as the standard of judging optimal threshold, from candidate's threshold value, select optimal threshold, the wave band that optimal threshold is corresponding is a best band.Best band and optimal threshold constitute the optimal decision rule at present node place jointly, and selected threshold value can be distinguished two kinds well, and the advantage of used wave band has guaranteed the high-class precision greater than its all band.
Description of drawings
Fig. 1 is based on the target in hyperspectral remotely sensed image categorizing system structural drawing of binary decision tree.Mainly comprise: image reads and display module, sample collection module, decision tree generation module, sort module, classification results precision evaluation module.The core of decision tree generation module comprises the selective rule of selective rule, optimal threshold and the best band of decision tree nodes data structure, decision Tree algorithms, candidate's threshold value.
Fig. 2 is a separability situation synoptic diagram between two types of atural objects.
Fig. 3 is the core process synoptic diagram of the inventive method.
Embodiment
Describe for ease, we have following application example at hypothesis:
Certain has the target in hyperspectral remotely sensed image of N wave band, for further research and utilization, need classify to it.Choose and show carrying out image, after the sample collection, utilize the decision tree generation module to generate the classification binary decision tree, extract the classifying rules of decision tree image is classified, last, the classifying quality of image is carried out precision evaluation.
Its concrete embodiment is:
(1) uses image to choose and open the target in hyperspectral remotely sensed image that needs classification with display module;
(2) classification of using the specimen sample module to be respectively required differentiation is chosen the sample point of some, and system notes sampled result;
(3) in the decision tree generation module, through the sample training decision tree classification device of sampling.At the root node place, obtain classification to be distinguished and count number and classification information category [classNum] to be distinguished;
(4) the sample gray-scale value according to each classification carries out statistical computation, draws gray average and the standard deviation on each wave band of all categories, and the size according to gray average sorts to the atural object classification on each wave band;
(5) calculate the intersection point between adjacent two classifications on each wave band according to formula
; If this intersection point satisfies standard, then this intersection point is noted as current candidate's threshold value;
(6) among all candidate's threshold values of present node; Standard
according to optimal threshold is selected is selected optimal threshold, and this threshold value can be divided current feature space;
(7) node that is as the criterion when pre-treatment with this threshold value just can generate two node; Left side child node comprises the classification of eigenwert less than threshold value; Right child node comprises the classification of eigenwert greater than threshold value, adds up the number and category [classNum] information of child node simultaneously;
(8) circulation above-mentioned steps successively according to nodal information, is handled all nodes, accomplishes the structure of decision tree;
(9) from the binary decision tree that is generated, extract classifying rules, image is classified through sort module;
(10) utilize classification results precision evaluation module, classification results is carried out precision evaluation.
Claims (3)
1. based on the target in hyperspectral remotely sensed image sorting technique of binary decision tree, it is characterized in that, may further comprise the steps:
(1), obtains to wait to distinguish classification and count number and treat other information of region class category [classNum] at the node place of a decision tree;
(2) the sample gray-scale value according to each classification carries out statistical computation, draws gray average and the standard deviation on each wave band of all categories, and the size according to gray average sorts to the atural object classification on each wave band;
(3) calculate the intersection point between adjacent two classifications on each wave band according to formula
; If this intersection point satisfies standard, then this intersection point is noted as current candidate's threshold value;
(4) among all candidate's threshold values of present node, the Standard Selection of selecting according to optimal threshold goes out optimal threshold, and this threshold value can be divided current feature space;
(5) node that is as the criterion when pre-treatment with this threshold value just can generate two node; Left side child node comprises the classification of eigenwert less than threshold value; Right child node comprises the classification of eigenwert greater than threshold value, adds up the number and category [classNum] information of child node simultaneously;
(6) circulation above-mentioned steps successively according to nodal information, is handled all nodes, accomplishes the structure of decision tree;
When making up binary decision tree, at first the standard of foundation is not overlapping between classification, and all leaf nodes of traverse tree mix other leaf node existence if comprise in addition then, then debase the standard, and allow classification slightly overlapping, continue to make up decision tree.
2. the target in hyperspectral remotely sensed image sorting technique based on binary decision tree according to claim 1 is characterized in that:
In the process that makes up tree; During node of each processing; All to remain sub-category gray average and sort current according to their samples; According to dividing standard to insert threshold value, to judge not only whether this threshold value can will judge also whether this intersection point can be with the sub-category two big class of dividing into of remain with two types of adjacent differentiations; Stipulate that each classification is only included by a leaf node, promptly when handling some child nodes, the classification of before having been distinguished is just no longer participated in judgement.
3. the target in hyperspectral remotely sensed image sorting technique based on binary decision tree according to claim 1 is characterized in that:
When selecting the threshold value of best wave band and the best, the gauged distance between computation of mean values, its formula is:
μ wherein
1, μ
2, σ
1, σ
2Be respectively intersection point corresponding two classifications in the average and the standard deviation of feature space; μ
1, μ
2Gap is big more, and between class distance is big more; σ
1, σ
2More little, then quasi-cohesion is good more, get d when maximum pairing threshold value guarantee that as the judgment rule of present node selected threshold value can distinguish two kinds well, the advantage of used wave band is greater than its all band.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011104404689A CN102542291A (en) | 2011-12-23 | 2011-12-23 | Hyperspectral remote sensing image classification method based on binary decision tree |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011104404689A CN102542291A (en) | 2011-12-23 | 2011-12-23 | Hyperspectral remote sensing image classification method based on binary decision tree |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102542291A true CN102542291A (en) | 2012-07-04 |
Family
ID=46349141
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011104404689A Pending CN102542291A (en) | 2011-12-23 | 2011-12-23 | Hyperspectral remote sensing image classification method based on binary decision tree |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102542291A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104820839A (en) * | 2015-04-24 | 2015-08-05 | 深圳信息职业技术学院 | Respective positive and negative example correct rate setting-based controllable confidence machine algorithm |
| CN105894041A (en) * | 2016-04-26 | 2016-08-24 | 国网山东省电力公司经济技术研究院 | Method of extracting substation information in power distribution system based on hyperspectral remote sensing images |
| CN106975617A (en) * | 2017-04-12 | 2017-07-25 | 北京理工大学 | A kind of Classification of materials method based on color selector |
| CN108388916A (en) * | 2018-02-26 | 2018-08-10 | 武汉大学 | A kind of water body floating material automatic identifying method and system based on artificial intelligence |
| CN109034250A (en) * | 2018-07-31 | 2018-12-18 | 佛山科学技术学院 | A kind of information representation system of spacecraft monitoring big data |
| CN110495879A (en) * | 2019-07-30 | 2019-11-26 | 福建亿能达信息技术股份有限公司 | Brain wave patterns time-frequency characteristics extracting method based on information gain |
| CN111898681A (en) * | 2020-07-31 | 2020-11-06 | 北京林业大学 | Burned area fine extraction method based on remote sensing cloud platform and decision tree |
| CN112183426A (en) * | 2020-10-10 | 2021-01-05 | 中国科学院空天信息创新研究院 | Method for determining target area in hyperspectral image and related device |
-
2011
- 2011-12-23 CN CN2011104404689A patent/CN102542291A/en active Pending
Non-Patent Citations (2)
| Title |
|---|
| 巴桑等: "决策树在遥感影像分类中的应用", 《高原山地气象研究》 * |
| 闫培洁等: "基于遥感影像光谱信息的二叉决策分类树自动生成方法研究", 《测绘科学》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104820839A (en) * | 2015-04-24 | 2015-08-05 | 深圳信息职业技术学院 | Respective positive and negative example correct rate setting-based controllable confidence machine algorithm |
| CN105894041A (en) * | 2016-04-26 | 2016-08-24 | 国网山东省电力公司经济技术研究院 | Method of extracting substation information in power distribution system based on hyperspectral remote sensing images |
| CN106975617A (en) * | 2017-04-12 | 2017-07-25 | 北京理工大学 | A kind of Classification of materials method based on color selector |
| CN106975617B (en) * | 2017-04-12 | 2018-10-23 | 北京理工大学 | A kind of Classification of materials method based on color selector |
| CN108388916A (en) * | 2018-02-26 | 2018-08-10 | 武汉大学 | A kind of water body floating material automatic identifying method and system based on artificial intelligence |
| CN109034250A (en) * | 2018-07-31 | 2018-12-18 | 佛山科学技术学院 | A kind of information representation system of spacecraft monitoring big data |
| CN110495879A (en) * | 2019-07-30 | 2019-11-26 | 福建亿能达信息技术股份有限公司 | Brain wave patterns time-frequency characteristics extracting method based on information gain |
| CN110495879B (en) * | 2019-07-30 | 2021-11-09 | 福建亿能达信息技术股份有限公司 | Brain waveform time-frequency feature extraction method based on information gain |
| CN111898681A (en) * | 2020-07-31 | 2020-11-06 | 北京林业大学 | Burned area fine extraction method based on remote sensing cloud platform and decision tree |
| CN111898681B (en) * | 2020-07-31 | 2023-09-05 | 北京林业大学 | Fire trace fine extraction method based on remote sensing cloud platform and decision tree |
| CN112183426A (en) * | 2020-10-10 | 2021-01-05 | 中国科学院空天信息创新研究院 | Method for determining target area in hyperspectral image and related device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102542291A (en) | Hyperspectral remote sensing image classification method based on binary decision tree | |
| CN101604322B (en) | Decision level text automatic classified fusion method | |
| CN109902953B (en) | Power consumer classification method based on self-adaptive particle swarm clustering | |
| CN108363810A (en) | A kind of file classification method and device | |
| CN102208037B (en) | Hyper-spectral image classification method based on Gaussian process classifier collaborative training algorithm | |
| CN111914090B (en) | Method and device for enterprise industry classification identification and characteristic pollutant identification | |
| CN103116762A (en) | Image classification method based on self-modulated dictionary learning | |
| Wan et al. | UAV swarm based radar signal sorting via multi-source data fusion: A deep transfer learning framework | |
| CN102176698A (en) | Method for detecting abnormal behaviors of user based on transfer learning | |
| CN107103332A (en) | A kind of Method Using Relevance Vector Machine sorting technique towards large-scale dataset | |
| CN109934354A (en) | Abnormal deviation data examination method based on Active Learning | |
| CN106228389A (en) | Network potential usage mining method and system based on random forests algorithm | |
| CN104834940A (en) | Medical image inspection disease classification method based on support vector machine (SVM) | |
| WO2016177069A1 (en) | Management method, device, spam short message monitoring system and computer storage medium | |
| CN111062425B (en) | Unbalanced data set processing method based on C-K-SMOTE algorithm | |
| CN111881909A (en) | Coal and gangue identification method and device, electronic equipment and storage medium | |
| CN106778603A (en) | A kind of pedestrian recognition method that SVM classifier is cascaded based on gradient type | |
| CN106021578A (en) | Improved text classification algorithm based on integration of cluster and membership degree | |
| CN103092931A (en) | Multi-strategy combined document automatic classification method | |
| CN110751121A (en) | Unsupervised radar signal sorting method based on clustering and SOFM | |
| CN112733936A (en) | Recyclable garbage classification method based on image recognition | |
| Wang et al. | An improved k NN text classification method | |
| CN103780588A (en) | User abnormal behavior detection method in digital home network | |
| CN100416599C (en) | Not supervised classification process of artificial immunity in remote sensing images | |
| Chu et al. | Co-training based on semi-supervised ensemble classification approach for multi-label data stream |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20120704 |