CN108038448A - Semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy - Google Patents
Semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy Download PDFInfo
- Publication number
- CN108038448A CN108038448A CN201711323789.4A CN201711323789A CN108038448A CN 108038448 A CN108038448 A CN 108038448A CN 201711323789 A CN201711323789 A CN 201711323789A CN 108038448 A CN108038448 A CN 108038448A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- classification
- remote sensing
- probability
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
Abstract
The present invention relates to a kind of semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy, including basis modeling, the analysis of high-spectrum remote sensing data message, evaluation exports as a result, five steps such as uncertain conversion and mending-leakage calculate.The present invention makes full use of a small amount of training sample for being only capable of obtaining in high-spectrum remote sensing and a large amount of unlabelled samples to carry out the excavation of high spectrum image information, utilize the classification for primarily determining that out atural object by way of decision tree is according to ballot with probability random forest, it is difficult the atural object classification that extracts to filter out meet researcher's demand using weighted entropy algorithm afterwards, the pixel for choosing weighted entropy maximum is added in training sample, it is being iterated processing, untill meeting cut-off condition or when unlabelled sample has run out, stop iteration output category result, it is finally completed the classification of atural object.
Description
Technical field
The present invention relates to the semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy, belongs to remote sensing mapping skill
Art field.
Background technology
With the fast development of space technology, increasing, the traditional remote sensing of the influence of life of the remote sensing image to us
Sorting technique cannot meet the requirement that the image information of present people is excavated.How remote sensing image is sufficiently excavated potential
Information have become we research hot spot.And classification of remote-sensing images technology is the core technology of information excavating, thus it is more next
More scholars is deep among the research to classification of remote-sensing images technology.
High-spectrum remote sensing is used in testing herein.It is a kind of to integrate spectral Dimensions and Spatial Dimension
Imaging remote sensing technology comprising potentially large number of information.There is ambiguity since it includes substantial amounts of information and data, easily by
The influence of Huges noises, therefore traditional classification need being had been unable to meet based on single grader sorting technique in high-spectrum remote-sensing
Ask.In view of the feature of high-spectrum remote sensing, consider to classify using random forests algorithm.
Random forests algorithm is a kind of sorting technique based on CART decision Tree algorithms principles, it is determined by a series of CART
Plan tree is composed, and is voted by decision tree, finally using the most classification final as the atural object classification of number of votes obtained
As a result.Its attribute module used is Gino indexs.
This algorithm is collecting for classification tree, gives the importance of all variables, and this algorithm is faced in shortage of data
It is sane with still comparing when imbalance, it can predict up to thousands of kinds of explanatory variables.Random forest is by producing substantial amounts of point
Class tree establishes the relation of independent variable and dependent variable, this can must successfully calculate variable nonlinear interaction and reciprocation, even if
In the case where interference is bigger, industry is not likely to produce overfitting.Above characteristic based on random forest, it is used for handling
In target in hyperspectral remotely sensed image local extremum and each classification atural object difference is big and the problem of training speed is slower be one very not
Wrong selection.
In view of target in hyperspectral remotely sensed image self information amount is big and it or a kind of new subdivided spectral Imaging remote sensing technology,
It is extremely difficult fully to excavate potential information therein, while it is very difficult to consider that training sample obtains, and cost is very high
Present situation, the method for traditional supervised classification shows and less practical.But although the method for unsupervised classification need not train sample
This, but due to the limitation of its nicety of grading, be seldom used in reality is classified.For these reasons, using semi-supervised point
It can be a good selection that the method for class, which carries out terrain classification,.
Semisupervised classification is a kind of sorting algorithm of Active Learning, in having training process and meanwhile used marker samples and
Unmarked sample.In today of information explosion, classification problem becomes to become increasingly complex, and semisupervised classification algorithm is only obtaining
In the case of sub-fraction classification samples, reach fine and obtained classifying quality, received the favor of people.This sorting algorithm is
Model is created from a small amount of mark example and a large amount of unmarked examples.
Semi-supervised learning (semi-supervised learning) sorting technique used today is to make full use of on a small quantity
Training sample and substantial amounts of unlabelled sample create the self-training machine learning method that model is classified.Facing only
It is very applicable in the case of a small amount of marker samples and a large amount of unlabelled samples.In view of sample labeling it is sufficiently expensive and with into
As the rapid development of spectral technique and the magnanimity of Hyperspectral imaging such as emerge in large numbers at the reason, the sorting technique of semi-supervised learning is studied
It is very significant.This invention can optimize the classification performance of target in hyperspectral remotely sensed image, improve the nicety of grading of high-spectrum remote-sensing
And classification effectiveness.Nowadays the target in hyperspectral remotely sensed image semisupervised classification method applied has many kinds, such as:Based on semi-supervised
It is sparse to differentiate embedded Hyperspectral Remote Sensing Imagery Classification method (Intellectual Property Right Bureau of the RPC's application publication number
CN103593676A) it about subtracts and using stream target in hyperspectral remotely sensed image progress dimension using semi-supervised sparse discriminating embedded mobile GIS
Row study carries out feature extraction and not only only accounts for EO-1 hyperion, and there is complicated nonlinear characteristic also to keep sparse heavy between sample
Structure relation, it takes full advantage of a small amount of marker samples and substantial amounts of unmarked sample is fully excavated and is hidden in high-spectrum remote sensing
In abundant information and low dimensional manifold structure, this algorithm effectively raises the nicety of grading of atural object;Based on figure regularization
Semi-supervised high-spectral remote sensing sorting technique (Intellectual Property Right Bureau of the RPC's application publication number
CN102096825A) it is a kind of sorting technique for getting a good eye value, it belongs to imparametrization and gives full expression to high-spectral data
Manifold characteristic.But the semisupervised classification method of many regularizations is for the scalability and Generalization Capability of high-spectral data
It has been short of.Therefore it is big in order to adapt to high-spectrum remote sensing dimension height, data volume, but the spy that sample extraction is very difficult
Property, it is current problem in urgent need to solve to find a kind of sane high-precision sorting technique.
A kind of semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy proposed by the present invention, is being selected
Classify in the case of taking number of training 10%, it by with probability output based on the random gloomy of CART decision trees
Woods, how much tentative predictions according to votes go out atural object, and the atural object that then recycle weighted entropy algorithm needs to researcher assigns
Certain weight, weight of zero is assigned without reference to the atural object of value for researcher, selects the larger atural object of weighting entropy
It is added to and new training sample is formed in training sample, be then predicted classification again, is carried out according to above step, Zhi Daoman
The condition of sufficient iteration stopping or unlabelled atural object use until exhausted.This sorting technique is economic and practical, and is very suitable for
The characteristic of high-spectrum remote sensing, for the classification of the atural object City Regions of large area, has very high for researcher
Use value.
The content of the invention
In order to solve some shortcomings on existing sorting technique, the present invention provides a kind of semi-supervised random based on weighted entropy
Forest Hyperspectral Remote Sensing Imagery Classification method, solve in target in hyperspectral remotely sensed image information excavating on Spatial Dimension it is insufficient and point
The problems such as class imbalance and larger classification difference, the precision and efficiency of classification are effectively raised, is provided very for researcher
Get well to obtain reference value.
In order to realize effect above-mentioned, it is proposed that the semi-supervised random forest target in hyperspectral remotely sensed image based on weighted entropy
Sorting technique, it comprises the following steps:
The first step, basis modeling, suggests that band probability random data extracts Computation function model first, then that EO-1 hyperion is distant
Sense view data, training sample set data and classification collection data are entered into band probability random data and extract Computation function model respectively
In;
Second step, high-spectrum remote sensing data message analysis, after completing the first step, by training sample set and classification class
Do not collect, the phase judged corresponding to atural object classification that votes are most is calculated using the method for the random forest with probability output
Prestige value, and tentatively judge the probability of happening of the classification in high-spectrum remote sensing representated by each pixel, and to progress of all categories
It is preliminary to judge;
3rd step, evaluation output is as a result, after to second step, the probability of happening output category for the classification that second step computing obtains
As a result and accuracy assessment is carried out, following step is then carried out if first time iteration, otherwise compared with last time output result
Compared with continuing following step if being more than given threshold value if the difference of the two, if being less than given threshold value just final
As a result export;
4th step, uncertainty conversion, meets researcher's demand using the weighted entropy algorithm based on ballot probability and foundation
It is different that different weights is assigned to atural object, the probabilistic assessment result that the 3rd step is evaluated is converted into uncertainty;
5th step, mending-leakage calculate, after completing the 4th step, according to uncertainty unmarked label pixel in Hyperspectral imaging
Mark label pixel is converted to, then new mark label is added in the training set in the first step, and return to the 4th step
In, iteration is run up to untill satisfaction terminates requirement or unmarked training sample uses until exhausted.
Further, in the first step, high-spectrum remote sensing data for satellite high-spectrum remote sensing and nobody
Any one or two kinds in machine high-spectrum remote-sensing figure share, and training sample set is the training sample obtained by priori,
It is that the high-spectrum remote sensing inputted based on step elects training sample;Classification collection is the classification obtained by priori
Collection, i.e., final classification number, is the classification collection that the high-spectrum remote sensing tentative prediction inputted based on step is come out.
Further, in the second step, by training sample set and class categories collection, using with probability output
The method of random forest calculates the probabilistic method of the classification representated by each pixel in high-spectrum remote sensing.Carrying for using is general
The random forest of rate output is the sorting technique based on CART decision Tree algorithms.Since random forest is by a series of CART decision-makings
What tree was composed, and voted by decision tree, its attribute module used is Gino indexs;Gino indexs are specifically public
Formula is as follows:
P in formula (1)iIt is classification CiThe probability occurred in D, Gino desired values are smaller, show " degree of purity " of sample just
It is higher.When only binary divides, D1 and D2 will be divided into for the attribute A in training sample data collection D, then give division D's
Shown in Gini index equation below:
For discrete value attribute, the subset that recursive selection attribute produces minimum Gini indexs in the algorithm is used as it
Division subset;
For Continuous valued attributes, it is necessary to consider all possible split point, its decision-making is similar to the information introduced in ID3 and increases
Beneficial processing method, its formula:
Select given continuous property and produce the point of minimum Gini indexs as the split point for changing attribute.In CART
When structure, no matter whether the node in decision tree is divided, all to corresponding class on each node t marks, using not
Equation is as the differentiation criteria for classifying:
If for all class C in addition to node ijAll set up, then node t is labeled as class;Wherein P (i) represents class Ci
Prior probability, NitBe node t sample in CiThe quantity of class, PC(j/i) represent node i mistake being divided into CjThe cost of class, this
It can be found by searching for decision-making tree matrix;
A certain pixel is represented by the new probability formula x of the semisupervised classification Random Forest model outcome variable with probability output
, DNRepresent classification collection;Shown in c solution formulas such as formula (2):
Estimation coefficient c is solved by the method for maximal possibility estimation.
Further, in the second step, the corresponding classification of each pixel of each pixel class probability output calculated is
Refer to:
Further, output category result and carry out accuracy assessment in the 3rd step, if first time iteration then into
The following step of row, otherwise compared with the last time exports result, continue if given threshold value is more than if the difference of the two into
The following step of row, just exports final result if being less than given threshold value.
Further, in the 4th step, researcher is met using the weighted entropy algorithm based on ballot probability and foundation
Demand difference assigns atural object different weights, and probability is converted into the uncertain probability to each pixel in remote sensing images changes
Refer to for uncertain formula:
The function of weighted entropy algorithm used considers generation of the people to the degree of concern and event of information and people is produced
Raw influence.When calculating the weighting entropy of hyperspectral image data pixel, it may appear that the situation of maximum, in order to ensure to test
As a result accuracy, employs normalized, as shown in formula (7):
Further, in the 5th step, researcher is met using the weighting entropy theory based on ballot probability and foundation
Demand difference assigns atural object different weights, and probability is converted into uncertainty;According to uncertainty in Hyperspectral imaging
Unmarked label pixel is converted to mark label pixel;But it is uncertain it is bigger it includes information it is abundanter, therefore will weighting
The pixel of entropy maximum, which is included in, to be formed new training set and is iterated processing in training set, iteration, which runs up to satisfaction and terminates requirement, is
Only or unmarked training sample uses until exhausted.
Beneficial effects of the present invention:Make full use of a small amount of training sample for being only capable of obtaining in high-spectrum remote sensing and a large amount of
Unlabelled sample carries out the excavation of high spectrum image information, utilizes the side with probability random forest by decision tree according to ballot
Formula primarily determines that out the classification of atural object, and it is difficult to extract afterwards to filter out meet researcher's demand using weighted entropy algorithm
Atural object classification, the pixel for choosing weighted entropy maximum are added in training sample, processing are being iterated, until meeting cut-off condition
Untill or unlabelled sample when have run out, stop iteration output category result, be finally completed the classification of atural object.
Brief description of the drawings
It is next with reference to the accompanying drawings and detailed description that the present invention will be described in detail;
Fig. 1 is the method for the present invention flow chart;
Fig. 2 is flow chart of data processing of the present invention;
Fig. 3 is the corresponding classification sign of high-spectrum remote-sensing original image, object spectrum figure and each atural object that the present invention uses;
Fig. 4 is the final classifying quality figure of each algorithm used in the present invention.
Embodiment
To make the technical means, the creative features, the aims and the efficiencies achieved by the present invention easy to understand, with reference to
Embodiment, the present invention is further explained.
The semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy as described in Fig. 1-4, it includes
Following steps:
The first step, basis modeling, suggests that band probability random data extracts Computation function model first, then that EO-1 hyperion is distant
Sense view data, training sample set data and classification collection data are entered into band probability random data and extract Computation function model respectively
In;
Second step, high-spectrum remote sensing data message analysis, after completing the first step, by training sample set and classification class
Do not collect, the phase judged corresponding to atural object classification that votes are most is calculated using the method for the random forest with probability output
Prestige value, and tentatively judge the probability of happening of the classification in high-spectrum remote sensing representated by each pixel, and to progress of all categories
It is preliminary to judge;
3rd step, evaluation output is as a result, after to second step, the probability of happening output category for the classification that second step computing obtains
As a result and accuracy assessment is carried out, following step is then carried out if first time iteration, otherwise compared with last time output result
Compared with continuing following step if being more than given threshold value if the difference of the two, if being less than given threshold value just final
As a result export;
4th step, uncertainty conversion, meets researcher's demand using the weighted entropy algorithm based on ballot probability and foundation
It is different that different weights is assigned to atural object, the probabilistic assessment result that the 3rd step is evaluated is converted into uncertainty;
5th step, mending-leakage calculate, after completing the 4th step, according to uncertainty unmarked label pixel in Hyperspectral imaging
Mark label pixel is converted to, then new mark label is added in the training set in the first step, and return to the 4th step
In, iteration is run up to untill satisfaction terminates requirement or unmarked training sample uses until exhausted.
In the present embodiment, in the first step, high-spectrum remote sensing data are satellite high-spectrum remote sensing and nothing
Any one or two kinds in man-machine high-spectrum remote-sensing figure share, and training sample set is the training sample obtained by priori
This, is that the high-spectrum remote sensing inputted based on step elects training sample;Classification collection is to be obtained by priori
Classification collection, i.e., final classification number, is the classification collection that the high-spectrum remote sensing tentative prediction inputted based on step is come out.
In the present embodiment, in the second step, by training sample set and class categories collection, using with probability output
The method of random forest calculate the probabilistic method of the classification in high-spectrum remote sensing representated by each pixel.What is used carries
The random forest of probability output is the sorting technique based on CART decision Tree algorithms.Since random forest is determined by a series of CART
Plan tree is composed, and is voted by decision tree, its attribute module used is Gino indexs;Gino indexs are specific
Formula is as follows:
P in formula (1)iIt is classification CiThe probability occurred in D, Gino desired values are smaller, show " degree of purity " of sample just
It is higher.When only binary divides, D1 and D2 will be divided into for the attribute A in training sample data collection D, then give division D's
Shown in Gini index equation below:
For discrete value attribute, the subset that recursive selection attribute produces minimum Gini indexs in the algorithm is used as it
Division subset;
For Continuous valued attributes, it is necessary to consider all possible split point, its decision-making is similar to the information introduced in ID3 and increases
Beneficial processing method, its formula:
Select given continuous property and produce the point of minimum Gini indexs as the split point for changing attribute.In CART
When structure, no matter whether the node in decision tree is divided, all to corresponding class on each node t marks, using not
Equation is as the differentiation criteria for classifying:
If for all class C in addition to node ijAll set up, then node t is labeled as class;Wherein P (i) represents class Ci
Prior probability, NitBe node t sample in CiThe quantity of class, PC(j/i) represent node i mistake being divided into CjThe cost of class, this
It can be found by searching for decision-making tree matrix;
A certain pixel is represented by the new probability formula x of the semisupervised classification Random Forest model outcome variable with probability output
, DNRepresent classification collection;Shown in c solution formulas such as formula (2):
Estimation coefficient c is solved by the method for maximal possibility estimation.
In the present embodiment, in the second step, each pixel class probability calculated exports the corresponding classification of each pixel
Refer to:
Output category result and accuracy assessment is carried out in the present embodiment, in the 3rd step, if first time iteration then
Following step is carried out, otherwise compared with the last time exports result, is continued if given threshold value is more than if the difference of the two
Following step is carried out, just final result is exported if being less than given threshold value.
In the present embodiment, in the 4th step, study using the weighted entropy algorithm based on ballot probability and according to satisfaction
Person's demand difference assigns atural object different weights, and probability is converted into the uncertain probability to each pixel in remote sensing images turns
Uncertain formula is changed to refer to:
The function of weighted entropy algorithm used considers generation of the people to the degree of concern and event of information and people is produced
Raw influence.When calculating the weighting entropy of hyperspectral image data pixel, it may appear that the situation of maximum, in order to ensure to test
As a result accuracy, employs normalized, as shown in formula (7):
In the present embodiment, in the 5th step, study using the weighting entropy theory based on ballot probability and according to satisfaction
Person's demand difference assigns atural object different weights, and probability is converted into uncertainty;According to uncertainty Hyperspectral imaging
In unmarked label pixel be converted to mark label pixel;But it is uncertain it is bigger it includes information it is abundanter, therefore will plus
The pixel of power entropy maximum, which is included in, to be formed new training set and is iterated processing in training set, iteration runs up to satisfaction termination requirement
Untill or unmarked training sample use until exhausted.
Semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy is the same as above-mentioned embodiment
The step of;Hyperion imaging spectrometers are mounted on EO-1 satellite platforms, in November, 2000 by NASA (NASA)
Launch.Its wave-length coverage covered is 400~2500nm, shares 220 wave bands, and spectral resolution is 10nm, space
Resolution ratio is 30 meters, and scan mode is push-broom type.It is domestic that image used herein is located at inner mongolia autonomous region, during acquisition
Between be in August, 2010, since raw video map sheet is larger, high 529 pixels are conveniently determined in order to study, wide 256 pixels
One piece of Experimental Area, after Atmospheric Correction and geometric correction, removes the low wave band of signal-to-noise ratio and selects 139 wave bands to carry out altogether
Analysis.According to image visualization interpretation, the land cover pattern atural object on image mainly includes bare area, building, crops, meadow, sand
Ground, six type of water body.Hyperion data True color synthesis images are as shown in Figure 1, training sample of all categories is distributed such as Fig. 2 institutes
Show, the curve of spectrum of 6 kinds of classification atural objects is as shown in Figure 3.
To prove the Accuracy and high efficiency of this method, by semi-supervised random forest EO-1 hyperion of the present invention based on weighted entropy
Remote Image Classification respectively at the non-supervisory K-means algorithms of classical machine learning classification algorithm, supervised classification most
Maximum-likelihood sorting algorithm and support vector cassification (support vector machine algorithm) algorithm are identical
Comparative result is carried out under sample conditions.Following conclusion is obtained by the result of observation experiment data:K-means algorithms no matter
Run time is still all slightly inferior in terms of nicety of grading, and the phenomenon of wrong point of leakage point is very serious;Maximum likelihood classification algorithm
The Kappa coefficients of nicety of grading and reaction and the actual goodness of fit are all very high, but run time is longer;But support vector machines point
The class algorithm classification precision and goodness of fit is all very high and operational efficiency is higher, the consecutive sort effect applied to large area is very
It is good.But two parameters C and ó of support vector cassification algorithm are to change according to different pieces of information parameter constantly, classification effect
Fruit is not sane enough, it is frequently necessary to adjustment parameter to adapt to the susceptibility to different atural objects, is especially less than tiny atural object,
Classifying quality is not especially good.Set forth herein come method it is insensitive to parameter setting, it gives the importance of all variables,
This algorithm face shortage of data and it is uneven when it is still relatively more sane, can predict up to thousands of kinds of explanatory variables and
The relation of independent variable and dependent variable is established by producing substantial amounts of classification tree, this can must successfully calculate variable nonlinear interaction
And reciprocation, in the case that interference is bigger, it is not easy to produce overfitting.Fig. 3 (a) is K-means algorithms
Result figure, Fig. 3 (b) is the result figure of maximum likelihood classification algorithm.Table 1 lists various different classifications methods in this experiment
Final classification precision.
Set forth herein the algorithm come to make full use of a small amount of mark sample in the case where only obtaining a small amount of training sample
This excavate with a large amount of unmarked Sample Establishing self-training machine learning classification models be hidden in it is big in high-spectrum remote sensing
Useful information is measured, good reference value is provided for the calculating of scientific research and Objects recognition and atural object coverage rate.With other
Sorting algorithm is compared, the time of operation for both having improved the precision of classification or having greatly shortened, while is also solved in assorting process
The classification difference met is big and the unbalanced problem of classification, can also predict the class of atural object well when in face of shortage of data
Not.
Since high-spectrum remote sensing includes substantial amounts of spectral Dimensions information and abundant Spatial Dimension information, more
It must be renovated come more extensive applied to atural object among the fields such as rate prediction, military prospecting, resources observation, spreading network information.But
Due to the limitation of present technology, it is also possible that but for its Spatial Dimension information for the excavation dynamics of its spectrum dimension information
Excavate obvious shortcoming.How key problem that the bulk information in high-spectrum remote sensing hidden be research is fully excavated.But
Classification of hyperspectral remote sensing image technology is the core technology handled high-spectrum remote sensing.The present invention put forward based on
The semi-supervised random forest Hyperspectral RS Classification method of weighted entropy is very suitable for classifying to high-spectrum remote sensing, this
Invent and provide reference value for researcher, be an invention highly significant.
Beneficial effects of the present invention:Make full use of a small amount of training sample for being only capable of obtaining in high-spectrum remote sensing and a large amount of
Unlabelled sample carries out the excavation of high spectrum image information, utilizes the side with probability random forest by decision tree according to ballot
Formula primarily determines that out the classification of atural object, and it is difficult to extract afterwards to filter out meet researcher's demand using weighted entropy algorithm
Atural object classification, the pixel for choosing weighted entropy maximum are added in training sample, processing are being iterated, until meeting cut-off condition
Untill or unlabelled sample when have run out, stop iteration output category result, be finally completed the classification of atural object.
The basic principles, main features and the advantages of the invention have been shown and described above.The technology of the industry
Personnel are it should be appreciated that the present invention is not limited to the above embodiments, and the above embodiments and description only describe this
The principle of invention, without departing from the spirit and scope of the present invention, various changes and modifications of the present invention are possible, these changes
Change and improvement all fall within the protetion scope of the claimed invention.The claimed scope of the invention by appended claims and its
Equivalent thereof.
Claims (7)
1. the semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy, it is characterised in that:It is described based on
The semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method of weighted entropy comprises the following steps:
The first step, basis modeling, suggests that band probability random data extracts Computation function model, then by high-spectrum remote-sensing figure first
Extracted as data, training sample set data and classification collection data are entered into band probability random data respectively in Computation function model;
Second step, high-spectrum remote sensing data message analysis, after completing the first step, by training sample set and class categories
Collection, the expectation judged corresponding to atural object classification that votes are most is calculated using the method for the random forest with probability output
Value, and tentatively judge the probability of happening of the classification in high-spectrum remote sensing representated by each pixel, and at the beginning of progress of all categories
Step judges;
3rd step, evaluation output is as a result, after to second step, the probability of happening output category result for the classification that first step computing obtains
And accuracy assessment is carried out, and following step is then carried out if first time iteration, otherwise compared with the last time exports result, if
The difference of the two is more than given threshold value and then continues following step, if it is just defeated final result to be less than given threshold value
Go out;
4th step, uncertainty conversion, meets researcher's demand difference using the weighted entropy algorithm based on ballot probability and foundation
Different weights is assigned to atural object, the probabilistic assessment result that the 3rd step is evaluated is converted into uncertainty;
5th step, mending-leakage calculate, and after completing the 4th step, unmarked label pixel in Hyperspectral imaging is changed according to uncertain
To mark label pixel, then new mark label is added in the training set in the first step, and is returned in the 4th step, repeatedly
In generation, runs up to untill satisfaction terminates requirement or unmarked training sample uses until exhausted.
2. the semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method according to claim 1 based on weighted entropy, its
It is characterized in that:In the first step, high-spectrum remote sensing data are satellite high-spectrum remote sensing and unmanned plane EO-1 hyperion
Any one or two kinds in remote sensing figure share, and training sample set is the training sample obtained by priori, are based on step
Suddenly the high-spectrum remote sensing inputted elects training sample;Classification collection is the classification collection obtained by priori, i.e., most
Whole classification number, is the classification collection that the high-spectrum remote sensing tentative prediction inputted based on step is come out.
3. the semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method according to claim 1 based on weighted entropy, its
It is characterized in that:In the second step, by training sample set and class categories collection, the random forest with probability output is used
Method calculate the probabilistic method of the classification in high-spectrum remote sensing representated by each pixel.Use with probability output
Random forest is the sorting technique based on CART decision Tree algorithms.Due to random forest be combined by a series of CART decision trees and
Into, and voted by decision tree, its attribute module used is Gino indexs;The specific formula of Gino indexs is as follows:
<mrow>
<mi>G</mi>
<mi>i</mi>
<mi>n</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>D</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mn>1</mn>
<mo>-</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</munderover>
<msup>
<mi>pi</mi>
<mn>2</mn>
</msup>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
P in formula (1)iIt is classification CiThe probability occurred in D, Gino desired values are smaller, show " degree of purity " of sample more
It is high.When only binary divides, D1 and D2 will be divided into for the attribute A in training sample data collection D, then give division D's
Shown in Gini index equation below:
<mrow>
<msub>
<mi>Gini</mi>
<mi>A</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>D</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>|</mo>
<mfrac>
<mrow>
<mi>D</mi>
<mn>1</mn>
</mrow>
<mi>D</mi>
</mfrac>
<mo>|</mo>
<mi>G</mi>
<mi>i</mi>
<mi>n</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>D</mi>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mo>|</mo>
<mfrac>
<mrow>
<mi>D</mi>
<mn>2</mn>
</mrow>
<mi>D</mi>
</mfrac>
<mo>|</mo>
<mi>G</mi>
<mi>i</mi>
<mi>n</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>D</mi>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
</mrow>
For discrete value attribute, the recursive subset for selecting the minimum Gini indexs of attribute generation is as its point in the algorithm
Split subset;
For Continuous valued attributes, it is necessary to consider all possible split point, its decision-making is similar at the information gain introduced in ID3
Reason method, its formula:
<mrow>
<msub>
<mi>Gini</mi>
<mi>A</mi>
</msub>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>v</mi>
</munderover>
<mo>|</mo>
<mfrac>
<mrow>
<mi>D</mi>
<mi>i</mi>
</mrow>
<mi>D</mi>
</mfrac>
<mo>|</mo>
<mo>*</mo>
<mi>G</mi>
<mi>i</mi>
<mi>n</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>D</mi>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
</mrow>
Select given continuous property and produce the point of minimum Gini indexs as the split point for changing attribute.In CART structures
When, no matter whether the node in decision tree is divided, all to corresponding class on each node t marks, made using inequality
To differentiate the criteria for classifying:
<mrow>
<mfrac>
<mrow>
<msub>
<mi>P</mi>
<mi>C</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>|</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<msub>
<mi>Ni</mi>
<mi>t</mi>
</msub>
</mrow>
<mrow>
<msub>
<mi>P</mi>
<mi>C</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>|</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<msub>
<mi>Nj</mi>
<mi>t</mi>
</msub>
</mrow>
</mfrac>
<mo>></mo>
<mfrac>
<msub>
<mi>N</mi>
<mi>i</mi>
</msub>
<msub>
<mi>N</mi>
<mi>j</mi>
</msub>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>4</mn>
<mo>)</mo>
</mrow>
</mrow>
If for all class C in addition to node ijAll set up, then node t is labeled as class;Wherein P (i) represents class CiPriori
Probability, NitBe node t sample in CiThe quantity of class, PC(j/i) represent node i mistake being divided into CjThe cost of class, this can be by looking into
Decision-making tree matrix is looked for find;
A certain pixel, D are represented by the new probability formula x of the semisupervised classification Random Forest model outcome variable with probability outputN
Represent classification collection;Shown in c solution formulas such as formula (2):
<mrow>
<mover>
<mi>c</mi>
<mo>^</mo>
</mover>
<mo>=</mo>
<mi>arg</mi>
<mi> </mi>
<mi>m</mi>
<mi>x</mi>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>c</mi>
<mo>|</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mi>c</mi>
<mo>&Element;</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>,</mo>
<mi>L</mi>
<mo>,</mo>
<msub>
<mi>N</mi>
<mi>C</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>5</mn>
<mo>)</mo>
</mrow>
</mrow>
Estimation coefficient c is solved by the method for maximal possibility estimation.
4. the semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method according to claim 1 based on weighted entropy, its
It is characterized in that:In the second step, the corresponding classification of each pixel of each pixel class probability output calculated refers to:
<mrow>
<mi>C</mi>
<mi>L</mi>
<mi>A</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>arg</mi>
<mi>max</mi>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>c</mi>
</munderover>
<mi>P</mi>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>6</mn>
<mo>)</mo>
</mrow>
<mo>.</mo>
</mrow>
5. the semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method according to claim 1 based on weighted entropy, its
It is characterized in that:Output category result and accuracy assessment is carried out in 3rd step, then carried out if first time iteration following
Step, otherwise compared with the last time exports result, continues following if given threshold value is more than if the difference of the two
Step, just exports final result if being less than given threshold value.
6. the semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method according to claim 1 based on weighted entropy, its
It is characterized in that:In 4th step, meet researcher's demand difference using the weighted entropy algorithm based on ballot probability and foundation
Assign different weight to atural object, by probability be converted into the uncertain probability to each pixel in remote sensing images be converted to it is uncertain
Property formula refers to:
<mrow>
<msub>
<mi>H</mi>
<mi>W</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>X</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>-</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>q</mi>
</munderover>
<msub>
<mi>w</mi>
<mi>i</mi>
</msub>
<msub>
<mi>p</mi>
<mi>i</mi>
</msub>
<msub>
<mi>logp</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>7</mn>
<mo>)</mo>
</mrow>
</mrow>
The function of weighted entropy algorithm used considers what generation of the people to the degree of concern and event of information produced people
Influence.When calculating the weighting entropy of hyperspectral image data pixel, it may appear that the situation of maximum, in order to ensure experimental result
Accuracy, normalized is employed, as shown in formula (7):
<mrow>
<msub>
<mi>R</mi>
<mn>2</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>X</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>-</mo>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mi>l</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<mi>p</mi>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>8</mn>
<mo>)</mo>
</mrow>
<mo>.</mo>
</mrow>
7. the semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method according to claim 1 based on weighted entropy, its
It is characterized in that:In 5th step, meet researcher's demand difference using the weighting entropy theory based on ballot probability and foundation
Different weights is assigned to atural object, probability is converted into uncertainty;According to uncertainty unmarked mark in Hyperspectral imaging
Label pixel is converted to mark label pixel;But it is uncertain it is bigger it includes information it is abundanter, it is therefore that weighted entropy is maximum
Pixel, which is included in, to be formed new training set and is iterated processing in training set, iteration run up to satisfaction terminate require untill or not
Mark training sample uses until exhausted.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711323789.4A CN108038448A (en) | 2017-12-13 | 2017-12-13 | Semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711323789.4A CN108038448A (en) | 2017-12-13 | 2017-12-13 | Semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108038448A true CN108038448A (en) | 2018-05-15 |
Family
ID=62102703
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711323789.4A Pending CN108038448A (en) | 2017-12-13 | 2017-12-13 | Semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108038448A (en) |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108830312A (en) * | 2018-06-01 | 2018-11-16 | 苏州中科天启遥感科技有限公司 | A kind of integrated learning approach adaptively expanded based on sample |
CN109344907A (en) * | 2018-10-30 | 2019-02-15 | 顾海艳 | Based on the method for discrimination for improving judgment criteria sorting algorithm |
CN109726641A (en) * | 2019-01-24 | 2019-05-07 | 常州大学 | A kind of remote sensing image cyclic sort method based on training sample Automatic Optimal |
CN109829425A (en) * | 2019-01-31 | 2019-05-31 | 沈阳农业大学 | A kind of small scale terrain classification method and system of Farmland Landscape |
CN109886340A (en) * | 2019-02-26 | 2019-06-14 | 山东星蓝信息科技有限公司 | A kind of Remote Image Classification |
CN110046674A (en) * | 2019-04-24 | 2019-07-23 | 哈尔滨工业大学 | Classification of hyperspectral remote sensing image method based on feedback arrangement |
CN110070488A (en) * | 2019-04-25 | 2019-07-30 | 北京工业大学 | A kind of multiple-angle thinking image forest height extracting method based on convolutional neural networks |
CN110163101A (en) * | 2019-04-17 | 2019-08-23 | 湖南省中医药研究院 | The difference of Chinese medicine seed and grade quick discrimination method |
CN110443139A (en) * | 2019-07-05 | 2019-11-12 | 中国地质大学(武汉) | A kind of target in hyperspectral remotely sensed image noise wave band detection method of Classification Oriented |
CN110472559A (en) * | 2019-08-13 | 2019-11-19 | 苏州中科天启遥感科技有限公司 | A kind of remote sensing image construction area land use attribute space moving method |
CN110796163A (en) * | 2019-09-19 | 2020-02-14 | 中国地质大学(武汉) | Hyperspectral image space spectrum classification method and device considering spectral importance |
CN111079672A (en) * | 2019-12-20 | 2020-04-28 | 中国科学院遥感与数字地球研究所 | Grape classification extraction method based on maximum entropy features |
CN111091087A (en) * | 2019-12-12 | 2020-05-01 | 河南理工大学 | Land coverage extraction algorithm based on multi-view collaborative canonical correlation forest remote sensing image |
CN111460966A (en) * | 2020-03-27 | 2020-07-28 | 中国地质大学(武汉) | Hyperspectral remote sensing image classification method based on metric learning and neighbor enhancement |
CN111583224A (en) * | 2020-05-08 | 2020-08-25 | 厦门大学 | Medical image augmentation method based on bounding box and area selection clipping |
CN111612086A (en) * | 2020-05-28 | 2020-09-01 | 云南电网有限责任公司电力科学研究院 | Remote sensing image classification method based on ensemble learning algorithm |
CN111666439A (en) * | 2020-05-28 | 2020-09-15 | 重庆渝抗医药科技有限公司 | Working method for rapidly extracting and dividing medical image big data aiming at cloud environment |
CN111898579A (en) * | 2020-08-12 | 2020-11-06 | 中国地质大学(武汉) | Extreme gradient lifting-based unbiased semi-supervised classification model for high-resolution remote sensing images |
CN112085051A (en) * | 2020-07-24 | 2020-12-15 | 西安电子科技大学 | Image classification method and system based on weighted voting and electronic equipment |
CN112198144A (en) * | 2020-09-16 | 2021-01-08 | 安徽泽众安全科技有限公司 | Method and system for rapid sewage tracing |
CN112308151A (en) * | 2020-11-03 | 2021-02-02 | 西安电子科技大学 | Weighting-based classification method for hyperspectral images of rotating forest |
CN113792809A (en) * | 2021-09-18 | 2021-12-14 | 中国空间技术研究院 | Remote sensing picture classification method based on random semi-supervised feature extraction model |
CN113869515A (en) * | 2021-09-17 | 2021-12-31 | 中林信达(北京)科技信息有限责任公司 | Knowledge extraction method fusing genetic algorithm and decision tree algorithm |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101853400A (en) * | 2010-05-20 | 2010-10-06 | 武汉大学 | Multiclass image classification method based on active learning and semi-supervised learning |
CN104376335A (en) * | 2014-11-13 | 2015-02-25 | 河南理工大学 | Semi-supervised hyperspectral remote sensing image classification method based on information entropies |
CN104504393A (en) * | 2014-12-04 | 2015-04-08 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image semi-supervised classification method based on integrated learning |
US20150379426A1 (en) * | 2014-06-30 | 2015-12-31 | Amazon Technologies, Inc. | Optimized decision tree based models |
CN106197424A (en) * | 2016-06-28 | 2016-12-07 | 哈尔滨工业大学 | The unmanned plane during flying state identification method that telemetry drives |
CN106548196A (en) * | 2016-10-20 | 2017-03-29 | 中国科学院深圳先进技术研究院 | A kind of random forest sampling approach and device for non-equilibrium data |
-
2017
- 2017-12-13 CN CN201711323789.4A patent/CN108038448A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101853400A (en) * | 2010-05-20 | 2010-10-06 | 武汉大学 | Multiclass image classification method based on active learning and semi-supervised learning |
US20150379426A1 (en) * | 2014-06-30 | 2015-12-31 | Amazon Technologies, Inc. | Optimized decision tree based models |
CN104376335A (en) * | 2014-11-13 | 2015-02-25 | 河南理工大学 | Semi-supervised hyperspectral remote sensing image classification method based on information entropies |
CN104504393A (en) * | 2014-12-04 | 2015-04-08 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image semi-supervised classification method based on integrated learning |
CN106197424A (en) * | 2016-06-28 | 2016-12-07 | 哈尔滨工业大学 | The unmanned plane during flying state identification method that telemetry drives |
CN106548196A (en) * | 2016-10-20 | 2017-03-29 | 中国科学院深圳先进技术研究院 | A kind of random forest sampling approach and device for non-equilibrium data |
Non-Patent Citations (3)
Title |
---|
S. AMINI等: "SEMI-SUPERVISED CLASSIFICATION OF HYPERSPECTRAL IMAGE USING RANDOM FOREST ALGORITHM", 《IEEE》 * |
盛昭瀚等: "客户流失危机分析的决策树方法", 《管理科学学报》 * |
蔡丽艳著: "《数据挖掘算法及其应用研究》", 28 February 2013 * |
Cited By (37)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108830312A (en) * | 2018-06-01 | 2018-11-16 | 苏州中科天启遥感科技有限公司 | A kind of integrated learning approach adaptively expanded based on sample |
CN108830312B (en) * | 2018-06-01 | 2022-09-09 | 苏州中科天启遥感科技有限公司 | Integrated learning method based on sample adaptive expansion |
CN109344907A (en) * | 2018-10-30 | 2019-02-15 | 顾海艳 | Based on the method for discrimination for improving judgment criteria sorting algorithm |
CN109726641A (en) * | 2019-01-24 | 2019-05-07 | 常州大学 | A kind of remote sensing image cyclic sort method based on training sample Automatic Optimal |
CN109829425A (en) * | 2019-01-31 | 2019-05-31 | 沈阳农业大学 | A kind of small scale terrain classification method and system of Farmland Landscape |
CN109886340A (en) * | 2019-02-26 | 2019-06-14 | 山东星蓝信息科技有限公司 | A kind of Remote Image Classification |
CN110163101A (en) * | 2019-04-17 | 2019-08-23 | 湖南省中医药研究院 | The difference of Chinese medicine seed and grade quick discrimination method |
CN110163101B (en) * | 2019-04-17 | 2022-09-23 | 湖南省中医药研究院 | Method for rapidly distinguishing seeds of traditional Chinese medicinal materials and rapidly judging grades of seeds |
CN110046674A (en) * | 2019-04-24 | 2019-07-23 | 哈尔滨工业大学 | Classification of hyperspectral remote sensing image method based on feedback arrangement |
CN110070488A (en) * | 2019-04-25 | 2019-07-30 | 北京工业大学 | A kind of multiple-angle thinking image forest height extracting method based on convolutional neural networks |
CN110070488B (en) * | 2019-04-25 | 2023-01-03 | 北京工业大学 | Multi-angle remote sensing image forest height extraction method based on convolutional neural network |
CN110443139A (en) * | 2019-07-05 | 2019-11-12 | 中国地质大学(武汉) | A kind of target in hyperspectral remotely sensed image noise wave band detection method of Classification Oriented |
CN110443139B (en) * | 2019-07-05 | 2023-01-03 | 中国地质大学(武汉) | Hyperspectral remote sensing image noise band detection method for classification |
CN110472559A (en) * | 2019-08-13 | 2019-11-19 | 苏州中科天启遥感科技有限公司 | A kind of remote sensing image construction area land use attribute space moving method |
CN110472559B (en) * | 2019-08-13 | 2023-06-23 | 苏州中科天启遥感科技有限公司 | Remote sensing image building area land utilization attribute space migration method |
CN110796163A (en) * | 2019-09-19 | 2020-02-14 | 中国地质大学(武汉) | Hyperspectral image space spectrum classification method and device considering spectral importance |
CN110796163B (en) * | 2019-09-19 | 2021-03-19 | 中国地质大学(武汉) | Hyperspectral image space spectrum classification method and device considering spectral importance |
CN111091087A (en) * | 2019-12-12 | 2020-05-01 | 河南理工大学 | Land coverage extraction algorithm based on multi-view collaborative canonical correlation forest remote sensing image |
CN111079672B (en) * | 2019-12-20 | 2023-10-13 | 中国科学院空天信息创新研究院 | Grape classification extraction method based on maximum entropy characteristics |
CN111079672A (en) * | 2019-12-20 | 2020-04-28 | 中国科学院遥感与数字地球研究所 | Grape classification extraction method based on maximum entropy features |
CN111460966A (en) * | 2020-03-27 | 2020-07-28 | 中国地质大学(武汉) | Hyperspectral remote sensing image classification method based on metric learning and neighbor enhancement |
CN111460966B (en) * | 2020-03-27 | 2024-02-02 | 中国地质大学(武汉) | Hyperspectral remote sensing image classification method based on metric learning and neighbor enhancement |
CN111583224B (en) * | 2020-05-08 | 2023-05-09 | 厦门大学 | Medical image augmentation method based on bounding box and region selection clipping |
CN111583224A (en) * | 2020-05-08 | 2020-08-25 | 厦门大学 | Medical image augmentation method based on bounding box and area selection clipping |
CN111666439B (en) * | 2020-05-28 | 2021-07-13 | 广东唯仁医疗科技有限公司 | Working method for rapidly extracting and dividing medical image big data aiming at cloud environment |
CN111666439A (en) * | 2020-05-28 | 2020-09-15 | 重庆渝抗医药科技有限公司 | Working method for rapidly extracting and dividing medical image big data aiming at cloud environment |
CN111612086B (en) * | 2020-05-28 | 2023-01-20 | 云南电网有限责任公司电力科学研究院 | Remote sensing image classification method based on ensemble learning algorithm |
CN111612086A (en) * | 2020-05-28 | 2020-09-01 | 云南电网有限责任公司电力科学研究院 | Remote sensing image classification method based on ensemble learning algorithm |
CN112085051B (en) * | 2020-07-24 | 2024-02-09 | 西安电子科技大学 | Image classification method and system based on weighted voting and electronic equipment |
CN112085051A (en) * | 2020-07-24 | 2020-12-15 | 西安电子科技大学 | Image classification method and system based on weighted voting and electronic equipment |
CN111898579A (en) * | 2020-08-12 | 2020-11-06 | 中国地质大学(武汉) | Extreme gradient lifting-based unbiased semi-supervised classification model for high-resolution remote sensing images |
CN112198144A (en) * | 2020-09-16 | 2021-01-08 | 安徽泽众安全科技有限公司 | Method and system for rapid sewage tracing |
CN112198144B (en) * | 2020-09-16 | 2023-07-25 | 安徽泽众安全科技有限公司 | Method and system for quickly tracing sewage |
CN112308151A (en) * | 2020-11-03 | 2021-02-02 | 西安电子科技大学 | Weighting-based classification method for hyperspectral images of rotating forest |
CN113869515A (en) * | 2021-09-17 | 2021-12-31 | 中林信达(北京)科技信息有限责任公司 | Knowledge extraction method fusing genetic algorithm and decision tree algorithm |
CN113869515B (en) * | 2021-09-17 | 2024-04-05 | 中林信达(北京)科技信息有限责任公司 | Knowledge extraction method integrating genetic algorithm and decision tree algorithm |
CN113792809A (en) * | 2021-09-18 | 2021-12-14 | 中国空间技术研究院 | Remote sensing picture classification method based on random semi-supervised feature extraction model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108038448A (en) | Semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy | |
Guo et al. | Scene-driven multitask parallel attention network for building extraction in high-resolution remote sensing images | |
CN108038445B (en) | SAR automatic target identification method based on multi-view deep learning framework | |
CN108537192B (en) | Remote sensing image earth surface coverage classification method based on full convolution network | |
Xue et al. | Attention-based second-order pooling network for hyperspectral image classification | |
Tuia et al. | Multiclass feature learning for hyperspectral image classification: Sparse and hierarchical solutions | |
Chang et al. | Change detection of land use and land cover in an urban region with SPOT-5 images and partial Lanczos extreme learning machine | |
Subburayalu et al. | Disaggregation of component soil series on an Ohio County soil survey map using possibilistic decision trees | |
CN111898633B (en) | Marine ship target detection method based on hyperspectral image | |
CN112347970B (en) | Remote sensing image ground object identification method based on graph convolution neural network | |
CN109657610A (en) | A kind of land use change survey detection method of high-resolution multi-source Remote Sensing Images | |
CN101894270A (en) | Method for full-automatic sample selection oriented to classification of remote-sensing images | |
CN111611960B (en) | Large-area ground surface coverage classification method based on multilayer perceptive neural network | |
Ding et al. | An improved ant colony algorithm for optimized band selection of hyperspectral remotely sensed imagery | |
CN115471467A (en) | High-resolution optical remote sensing image building change detection method | |
Ge et al. | Deep residual network-based fusion framework for hyperspectral and LiDAR data | |
Nadiri et al. | Formulating Convolutional Neural Network for mapping total aquifer vulnerability to pollution | |
Diakite et al. | Hyperspectral image classification using 3D 2D CNN | |
Xu et al. | Feature-based constraint deep CNN method for mapping rainfall-induced landslides in remote regions with mountainous terrain: An application to Brazil | |
Xu et al. | The graph attention network and its post-hoc explanation for recognizing mineralization-related geochemical anomalies | |
Hall et al. | Satellite image and machine learning based knowledge extraction in the poverty and welfare domain | |
Zhou et al. | Hyperspectral image classification based on 3-D multihead self-attention spectral–spatial feature fusion network | |
Jing et al. | Time series land cover classification based on semi-supervised convolutional long short-term memory neural networks | |
WO2022271942A1 (en) | Systems and methods for categorizing image pixels | |
Sun et al. | Deep learning in statistical downscaling for deriving high spatial resolution gridded meteorological data: A systematic review |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180515 |