CN109657616A - A kind of remote sensing image land cover pattern automatic classification method - Google Patents

A kind of remote sensing image land cover pattern automatic classification method Download PDF

Info

Publication number
CN109657616A
CN109657616A CN201811557247.8A CN201811557247A CN109657616A CN 109657616 A CN109657616 A CN 109657616A CN 201811557247 A CN201811557247 A CN 201811557247A CN 109657616 A CN109657616 A CN 109657616A
Authority
CN
China
Prior art keywords
sample
land cover
classification
sensing image
primitive
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
Application number
CN201811557247.8A
Other languages
Chinese (zh)
Inventor
谭力
程熙
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sichuan Liwei Space Information Technology Co Ltd
Original Assignee
Sichuan Liwei Space Information Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Sichuan Liwei Space Information Technology Co Ltd filed Critical Sichuan Liwei Space Information Technology Co Ltd
Priority to CN201811557247.8A priority Critical patent/CN109657616A/en
Publication of CN109657616A publication Critical patent/CN109657616A/en
Pending legal-status Critical Current

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
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Molecular Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Astronomy & Astrophysics (AREA)
  • Remote Sensing (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a kind of remote sensing image land cover pattern automatic classification method, including Image Segmentation, feature extraction, sample automatically selects and the several steps of land cover classification.The present invention carries out land cover pattern to remote sensing image using supervised learning algorithm and classifies automatically, by system flow repetitive exercise classifier, classification results can be made to reach higher precision level, it is opposite simultaneously with existing land cover classification method, it can automatically select different classes of land cover pattern sample, and operand is smaller.Furthermore, the present invention is before selecting and classifying to sample, Image Segmentation is carried out to satellite-remote-sensing image using mean shift algorithm first, priori knowledge is required few, training data is fully relied on to be estimated, the estimation that can be used for arbitrary shape density function has good adaptability and robustness for the data of different structure.

Description

A kind of remote sensing image land cover pattern automatic classification method
Technical field
The invention belongs to Land Cover Remote Sensing Monitoring technical fields, and in particular to a kind of remote sensing image land cover pattern divides automatically The design of class method.
Background technique
Land cover pattern refers to the various biologies of earth land surface or the cover type of physics, the vegetation (day including earth's surface So or manually), construction land (building, road), lake, glacier, naked rock and desert etc., the main natural category for describing earth surface Property.Land cover pattern plays key player in reflecting the people of the mankind and natural relation interactive system, affects earth system Various aspects in the process, and be the important parameter of the models such as the weather of earth system, biochemistry.In addition, being covered in soil Lid application aspect, every profession and trade department, country requires ground mulching information at present, as land administration department needs to understand fully that soil covers It should also be understood that the relationship of land cover pattern and settlement place, hydraulic department need to know the earth's surface near national water conservancy projects except lid Coverage condition, environment department it should be understood that major rivers periphery factory scenario, agricultural sector it should be understood that arable land distribution feelings Condition etc., all these specialized departments require comprehensive ground mulching data.Therefore, accurate, in time, comprehensively obtain soil Coverage information has important scientific meaning.
Currently, the remote sensing monitoring of land cover pattern is mostly monitored by Land cover types, area and situation of change, To know that the variation of land cover pattern area necessarily be unable to do without land cover classification, relative to single type of ground objects extraction and Speech, the algorithm that land cover pattern classification is automatically extracted from remote sensing image is relatively fewer, at present between single band threshold method, multiband spectrum The land cover classifications methods such as relations act, unsupervised classification method, supervised classification are applied in succession, however the above method is equal There is a problem of that operand is big and classification results precision is lower.
Summary of the invention
The purpose of the present invention is to solve existing land cover classification methods for artificial collecting sample heavy workload And the problem that classification results precision is unstable, propose a kind of remote sensing image land cover pattern automatic classification method.
The technical solution of the present invention is as follows: a kind of remote sensing image land cover pattern automatic classification method, comprising the following steps:
S1, Image Segmentation is carried out to satellite-remote-sensing image using mean shift algorithm, and segmentation result is polygon with vector In the unity of form storage to generalized information system database of shape.
S2, using each vector polygon as an independent primitive, extract the feature of each primitive and with vector attribute Form be stored in database.
S3, sample is automatically selected according to the feature of each primitive.
S4, whether judge the sample selected for Land cover types sample, if then entering step S5, otherwise return step S3。
S5, classified using supervised classification algorithm to Land cover types sample, and export land cover classification result.
Further, the mean shift algorithm in step S1 specifically:
A1, any one data point x, and the point centered on data point x are chosen in satellite-remote-sensing image, calculating is being set Radius is all data point x in the circular space of RiWith the offset mean value m (x) of central point x:
Wherein wiIt is i-th of data point xiWeight coefficient, meet constraint conditionN is circle The data point number in space, K () they are kernel function, and:
Wherein xsIndicate the space segment of central point x characteristic vector, xrIndicate the color part of central point x characteristic vector, p For satellite-remote-sensing image dimension, hs,hrRespectively space nucleus band is wide and color nucleus band is wide, and C is normaliztion constant, | | | | it indicates Two norms, k () be in space and color gamut all use identical core, and:
A2, central point is moved to offset position mean value m (x), iteration step A1, until central point and offset The distance of mean value is less than the distance threshold of setting, and all data points in the secondary iteration circular space are clustered, and realizes image Segmentation.
Further, the method for the feature of each primitive is extracted in step S2 specifically: mention using linear algebraic transformation method The spectral signature for taking each primitive extracts the space characteristics of each primitive using semivariogram method, using fourier spectrum point Analysis method extracts the features of terrain of each primitive.
Further, step S3 include it is following step by step:
S31, similarity between non-similar primitive is reduced to improve similarity between similar primitive according to the feature of each primitive For learning objective, exercises supervision study in conjunction with the land use survey data in generalized information system database, obtain transformation matrix.
Similitude between S32, the more unknown primitive of transformation matrix obtained according to supervised learning, is screened according to Spreading requirements To initial sample.
S33, the initial of spectral signature exception is rejected with index characteristic in conjunction with the ground-object spectrum data in generalized information system database Sample.
S34, Land cover types sample is converted by the land use pattern sample in initial sample, obtains complete sample This.
Further, the supervised classification algorithm in step S5 is artificial neural network algorithm, SVM algorithm or C5.0 decision tree Algorithm.
Further, artificial neural network algorithm specifically:
Four layers that neuron building includes an input layer, two hidden layers and an output layer are chosen in the library FANN Neural network, and Land cover types sample is inputted into neural network, obtain land cover classification result.
The neuron number that input layer includes is identical as image wave band number, the neuron number that output layer includes and sample text The classification number of part setting is identical.
The threshold cell that neuron is multi input, singly exports, the relationship between input and output are as follows:
Wherein xiIndicate i-th of input of neuron, i=1,2 ..., N, N are the input number of neuron, WiFor xi's Weight coefficient, θiIndicating the threshold value of i-th of neuron, y indicates the output of neuron, and f () is transforming function transformation function, and:
Further, SVM algorithm specifically:
There are a largest interval hyperplane, expression formula by training sample foundation in higher-dimension mathematical space are as follows:
D (x)=wx+w0 (6)
Wherein D () indicates that largest interval hyperplane function, x are that the vector of training sample indicates, w, w0Respectively indicate power Weight coefficient and biasing coefficient, calculation formula are as follows:
Wherein xiIndicate the vector representation of i-th of training sample, i=1,2 ..., I, I are training sample sum, yi Indicate the classification marker of i-th of training sample, and:
yi=sgn [D (xi)] (8)
Wherein sgn [] indicates sign function.
Land cover types sample is mapped in the form of vectors in higher-dimension mathematical space, is calculated according to formula (8) The classification to Land cover types sample is completed in the classification marker of each Land cover types sample.
Further, C5.0 decision Tree algorithms specifically:
B1, the comentropy for calculating Land cover types sample set:
Wherein Info (S) indicates the comentropy of Land cover types sample set S, freq (Ci, S) and it indicates to belong to CiClass Sample number, i=1,2 ..., k, k are classification sum, | S | indicate the sample number of sample set S.
B2, the conditional entropy introduced after attribute variable T in sample set S is calculated:
Wherein Info (T) indicates to introduce the conditional entropy after attribute variable T, T in sample set SjIndicate the of attribute variable T J classification, j=1,2 ..., n, n are the classification number of attribute variable T, Info (Tj) indicate TjConditional entropy.
B3, after introducing attribute variable T according to the comentropy of Land cover types sample set S and in sample set S The information gain Gain (T) of conditional entropy computation attribute variable T:
Gain (T)=Info (S)-Info (T) (11)
B4, the maximum attribute variable of information gain is chosen as grouping variable, generate the branch of decision tree.
B5, decision tree is successively trimmed upwards from the leaf node of decision tree, child node error is greater than father node error Node trims, node error calculation formula are as follows:
Wherein e is node error, and f is prediction error rate, i.e. the ratio of prediction error number and prediction sum, and z is critical Value, N are prediction sum.
B6, Land cover types sample is inputted to the decision tree trimmed, obtains land cover classification result.
The beneficial effects of the present invention are: the present invention classifies to land cover pattern sample using supervised learning algorithm, pass through System flow ground repetitive exercise classifier can make classification results reach higher precision level, at the same it is opposite with it is existing Land cover classification method, operand are smaller.In addition, the present invention uses first before sample is selected and classified Mean shift algorithm carries out Image Segmentation to satellite-remote-sensing image, requires less priori knowledge, fully relies on training data progress Estimation, can be used for the estimation of arbitrary shape density function, have good adaptability and steady for the data of different structure Property.
Detailed description of the invention
Fig. 1 show a kind of remote sensing image land cover pattern automatic classification method flow chart provided in an embodiment of the present invention.
Fig. 2 show classification results schematic diagram provided in an embodiment of the present invention.
Specific embodiment
Carry out detailed description of the present invention illustrative embodiments with reference to the drawings.It should be appreciated that shown in attached drawing and The embodiment of description is only exemplary, it is intended that is illustrated the principle and spirit of the invention, and is not limited model of the invention It encloses.
The embodiment of the invention provides a kind of remote sensing image land cover pattern automatic classification methods, as shown in Figure 1, including following Step S1~S5:
S1, Image Segmentation is carried out to satellite-remote-sensing image using mean shift algorithm, and segmentation result is polygon with vector In the unity of form storage to generalized information system database of shape.
Mean shift algorithm is a kind of method of non-parametric estmation density function, requires less, to fully rely on to priori knowledge Training data is estimated, can be used for the estimation of arbitrary shape density function, has for the data of different structure good Adaptability and robustness do not need that classification number is determined in advance, each in feature space can be made to put by effectively counting iteration " drift " arrives the Local modulus maxima of density function.Meanwhile merging rule by formulating the different of mean shift algorithm, it is easy to It realizes based on the multiple dimensioned merging process on the basis of mean filter, so that the primitive under realizing different scale merges, reaches more rulers Spend the purpose of segmentation.
In the embodiment of the present invention, specific step is as follows for mean shift algorithm:
A1, any one data point x, and the point centered on data point x are chosen in satellite-remote-sensing image, calculating is being set Radius is all data point x in the circular space of RiWith the offset mean value m (x) of central point x:
Wherein wi is i-th of data point xiWeight coefficient, meet constraint conditionN is circle The data point number in space, K () they are kernel function, and:
Wherein xsIndicate the space segment of central point x characteristic vector, xrIndicate the color part of central point x characteristic vector, p For satellite-remote-sensing image dimension, hs,hrRespectively space nucleus band is wide and color nucleus band is wide, and C is normaliztion constant, | | | | it indicates Two norms, k () be in space and color gamut all use identical core, and:
A2, central point is moved to offset position mean value m (x), iteration step A1, until central point and offset The distance of mean value is less than the distance threshold of setting, and all data points in the secondary iteration circular space are clustered, and realizes image Segmentation.
S2, using each vector polygon as an independent primitive, extract the feature of each primitive and with vector attribute Form be stored in database.
In the embodiment of the present invention, the spectral signature of each primitive is extracted using linear algebraic transformation method, using semivariogram Method extracts the space characteristics of each primitive, and the features of terrain of each primitive is extracted using Fourier analysis method.
Linear algebraic transformation method includes main composition transformation (Karhunen-Loeve transformation), K-T Transformation (K-T transformation), minimal noise separation Convert (MNF transformation) etc..
Since spatial texture feature is actually related with space scale, suitable calculating textural characteristics are selected Neighborhood window size, it is extremely important for effectively quantitative expression textural characteristics, semivariogram can be used (semivariogram) etc. space statistics method estimates to extract the optimal window size of texture information, appropriate by selecting Parametrization semivariogram statistical model (such as exponential model, logarithmic model) semivariogram is fitted, and to be used Semivariogram model derived from parameter as measure texture information standard.
Fourier analysis method carries out the image in each subregion by dividing an image into several sub-regions Fourier transform.In this way, the spectrum energy in open grain region is concentrated mainly on low frequency domain, the spectrum energy in close grain region It is concentrated mainly on high-frequency domain.Using polar coordinates with different radiuses circle domain in power spectral integral, obtained integral Value can reflect the fineness of texture；And it utilizes and texture can reflect to power spectral integral in the fan-shaped region of different directions Directionality.It is noted that textural characteristics are generally to participate in classification as supplemental characteristic to improve point in image classification Class precision, sort operation is not participated in individually.
In the real-time example of the present invention, the Partial Feature of extraction is as shown in table 1.
Table 1
S3, sample is automatically selected according to the feature of each primitive.
Step S3 includes following S31~S34 step by step:
S31, similarity between non-similar primitive is reduced to improve similarity between similar primitive according to the feature of each primitive For learning objective, exercises supervision study in conjunction with the land use survey data in generalized information system database, obtain transformation matrix.
Similitude between S32, the more unknown primitive of transformation matrix obtained according to supervised learning, is screened according to Spreading requirements To initial sample.
S33, the initial of spectral signature exception is rejected with index characteristic in conjunction with the ground-object spectrum data in generalized information system database Sample (there may be mistakes for land use data).
S34, Land cover types sample is converted by the land use pattern sample in initial sample, obtains complete sample This.
Land cover types sample is based on level-one class, including forest land, meadow, building site (city), waters, agricultural use Ground (arable land), unused land (bare area) etc..
S4, whether judge the sample selected for Land cover types sample, if then entering step S5, otherwise return step S3。
S5, classified using supervised classification algorithm to Land cover types sample, and export land cover classification result.
In the embodiment of the present invention, supervised classification algorithm can use artificial neural network algorithm, SVM algorithm or C5.0 decision Tree algorithm, in which:
(1) artificial neural network algorithm specifically:
Four layers that neuron building includes an input layer, two hidden layers and an output layer are chosen in the library FANN Neural network, and Land cover types sample is inputted into neural network, obtain land cover classification result.
The neuron number that input layer includes is identical as image wave band number, the neuron number that output layer includes and sample text The classification number of part setting is identical.
The threshold cell that neuron is multi input, singly exports, the relationship between input and output are as follows:
Wherein xiIndicate i-th of input of neuron, i=1,2 ..., N, N are the input number of neuron, WiFor xi's Weight coefficient, θiIndicating the threshold value of i-th of neuron, y indicates the output of neuron, and f () is transforming function transformation function, and:
(2) SVM algorithm specifically:
There are a largest interval hyperplane, expression formula by training sample foundation in higher-dimension mathematical space are as follows:
D (x)=wx+w0 (6)
Wherein D () indicates that largest interval hyperplane function, x are that the vector of training sample indicates, w, w0Respectively indicate power Weight coefficient and biasing coefficient, calculation formula are as follows:
Wherein xiIndicate the vector representation of i-th of training sample, i=1,2 ..., I, I are training sample sum, yi Indicate the classification marker of i-th of training sample, and:
yi=sgn [D (xi)] (8)
Wherein sgn [] indicates sign function.
Land cover types sample is mapped in the form of vectors in higher-dimension mathematical space, is calculated according to formula (8) The classification to Land cover types sample is completed in the classification marker of each Land cover types sample.
It should be noted that sign function sgn [] is only capable of for sample being divided into two classes, and needed in the embodiment of the present invention pair The classification that Land cover types sample is classified is likely larger than two classes, it is therefore desirable to which successive ignition uses SVM algorithm.Such as such as Fruit needs for Land cover types sample to be divided into five class of A, B, C, D, E, then is classified as A class and non-A using SVM algorithm for the first time The sample of non-A class is divided into B class and non-B class using SVM algorithm again for the second time by class, and so on, until all sample classifications are complete At.
(3) C5.0 decision Tree algorithms specifically:
B1, the comentropy for calculating Land cover types sample set:
Wherein Info (S) indicates the comentropy of Land cover types sample set S, freq (Ci, S) and it indicates to belong to CiClass Sample number, i=1,2 ..., k, k are classification sum, | S | indicate the sample number of sample set S.
B2, the conditional entropy introduced after attribute variable T in sample set S is calculated:
Wherein Info (T) indicates to introduce the conditional entropy after attribute variable T, T in sample set SjIndicate the of attribute variable T J classification, j=1,2 ..., n, n are the classification number of attribute variable T, Info (Tj) indicate TjConditional entropy.
B3, after introducing attribute variable T according to the comentropy of Land cover types sample set S and in sample set S The information gain Gain (T) of conditional entropy computation attribute variable T:
Gain (T)=Info (S)-Info (T) (11)
B4, the maximum attribute variable of information gain is chosen as grouping variable, generate the branch of decision tree.
B5, decision tree is successively trimmed upwards from the leaf node of decision tree, child node error is greater than father node error Node trims, node error calculation formula are as follows:
Wherein e is node error, and f is prediction error rate, i.e. the ratio of prediction error number and prediction sum, and z is critical Value, N are prediction sum.
B6, Land cover types sample is inputted to the decision tree trimmed, obtains land cover classification result.
Classifying quality of the invention is verified with specific test example below.
Choosing the groups of cities regions such as Dongguan, Shenzhen, Hong Kong of the Delta of the Pearl River is test area, to cover trial zone Same rail No. three images of 3 scape resource as main experimental data, other auxiliary datas include the land use of partial region Survey data, GDEM altitude data (30m), impermeable aqua index, trial zone typical feature spectral data etc., these data are by several What, radiant correction and mutually registration after be managed collectively under GIS platform.
Final classification result as shown in Figure 2 (since the 2nd scape is main region, when display is covered the 1st, 3 scapes), from mesh From the point of view of visual effect fruit, the classifications such as city, waters, forest land, arable land are substantially consistent with image, due to distributed more widely due to classification chart Information ways of presentation, or even rural residential area more indistinguishable in image can clearly be shown, and ploughed, meadow, bare area Order of accuarcy is relatively difficult to determine.In order to quantify to verify the precision of classification results, uniform piecemeal is generally carried out to survey region Random acquisition test sample, for having the region of the field investigation data then piecemeal no longer stochastical sampling, remaining stochastical sampling point Block area typicalness and randomness are taken into account generally with the artificial interpretation result supplement of the high-resolution data of identical phase.To take into account The demand of object oriented classification and statistical accuracy pixel-by-pixel, uses following precision test method: final statistic sampling number is 561 A, each sample corresponds to 3 × 3 homogeneity pixel region in image.It is big that sample areas accuracy is corresponded in result in precision evaluation It with correct counting of classifying when 7/9, is otherwise counted with classification error, finally to three scape image Individual accuracy statistical result such as tables 2 It is shown.
Table 2
From table 2 it can be seen that the classification results of single scape image have all reached higher precision level, this is not merely single The effect of a algorithm, but the result of whole system procedure iterative calculation.
Those of ordinary skill in the art will understand that the embodiments described herein, which is to help reader, understands this hair Bright principle, it should be understood that protection scope of the present invention is not limited to such specific embodiments and embodiments.This field Those of ordinary skill disclosed the technical disclosures can make according to the present invention and various not depart from the other each of essence of the invention The specific variations and combinations of kind, these variations and combinations are still within the scope of the present invention.

Claims (8)

1. a kind of remote sensing image land cover pattern automatic classification method, which comprises the following steps:
S1, Image Segmentation is carried out to satellite-remote-sensing image using mean shift algorithm, and by segmentation result with vector polygon In unity of form storage to generalized information system database；
S2, using each vector polygon as an independent primitive, extract the feature of each primitive and with the shape of vector attribute Formula is stored in database；
S3, sample is automatically selected according to the feature of each primitive；
S4, whether judge the sample selected for Land cover types sample, if then entering step S5, otherwise return step S3；
S5, classified using supervised classification algorithm to Land cover types sample, and export land cover classification result.
2. remote sensing image land cover pattern automatic classification method according to claim 1, which is characterized in that in the step S1 Mean shift algorithm specifically:
A1, any one data point x, and the point centered on data point x are chosen in satellite-remote-sensing image, calculate in setting radius For data point x all in the circular space of RiWith the offset mean value m (x) of central point x:
Wherein wiIt is i-th of data point xiWeight coefficient, meet constraint conditionN is circular space Data point number, K () be kernel function, and:
Wherein xsIndicate the space segment of central point x characteristic vector, xrIndicate that the color part of central point x characteristic vector, p are distant Feel satellite image dimension, hs,hrRespectively space nucleus band is wide and color nucleus band is wide, and C is normaliztion constant, | | | | indicate two models Number, k () be in space and color gamut all use identical core, and:
A2, central point is moved to offset position mean value m (x), iteration step A1, until central point and offset mean value Distance be less than setting distance threshold, all data points in the secondary iteration circular space are clustered, realize Image Segmentation.
3. remote sensing image land cover pattern automatic classification method according to claim 1, which is characterized in that in the step S2 The method for extracting the feature of each primitive specifically: extract the spectral signature of each primitive using linear algebraic transformation method, use Semivariogram method extracts the space characteristics of each primitive, and the landform for extracting each primitive using Fourier analysis method is special Sign.
4. remote sensing image land cover pattern automatic classification method according to claim 1, which is characterized in that the step S3 packet Include it is following step by step:
S31, according to the feature of each primitive, to improve similarity between similar primitive, similarity is to learn between reducing non-similar primitive Target is practised, exercises supervision study in conjunction with the land use survey data in generalized information system database, obtains transformation matrix；
Similitude between S32, the more unknown primitive of transformation matrix obtained according to supervised learning screens to obtain just according to Spreading requirements Beginning sample；
S33, the initial sample for rejecting spectral signature exception with index characteristic in conjunction with the ground-object spectrum data in generalized information system database This；
S34, Land cover types sample is converted by the land use pattern sample in initial sample, obtains full sample.
5. remote sensing image land cover pattern automatic classification method according to claim 1, which is characterized in that in the step S5 Supervised classification algorithm be artificial neural network algorithm, SVM algorithm or C5.0 decision Tree algorithms.
6. remote sensing image land cover pattern automatic classification method according to claim 5, which is characterized in that the artificial neuron Network algorithm specifically:
Four layers of nerve that neuron building includes an input layer, two hidden layers and an output layer are chosen in the library FANN Network, and Land cover types sample is inputted into neural network, obtain land cover classification result；
The neuron number that the input layer includes is identical as image wave band number, the neuron number and sample that the output layer includes The classification number of this document setting is identical；
The threshold cell that the neuron is multi input, singly exports, the relationship between input and output are as follows:
Wherein xiIndicate i-th of input of neuron, i=1,2 ..., N, N are the input number of neuron, WiFor xiWeight Coefficient, θiIndicating the threshold value of i-th of neuron, y indicates the output of neuron, and f () is transforming function transformation function, and:
7. remote sensing image land cover pattern automatic classification method according to claim 5, which is characterized in that the SVM algorithm Specifically:
There are a largest interval hyperplane, expression formula by training sample foundation in higher-dimension mathematical space are as follows:
D (x)=wx+w0 (6)
Wherein D () indicates that largest interval hyperplane function, x are that the vector of training sample indicates, w, w0Respectively indicate weight coefficient With biasing coefficient, calculation formula are as follows:
Wherein xiIndicate the vector representation of i-th of training sample, i=1,2 ..., I, I are training sample sum, yiIt indicates The classification marker of i-th of training sample, and:
yi=sgn [D (xi)] (8)
Wherein sgn [] indicates sign function；
Land cover types sample is mapped in the form of vectors in higher-dimension mathematical space, is calculated according to formula (8) each The classification to Land cover types sample is completed in the classification marker of Land cover types sample.
8. remote sensing image land cover pattern automatic classification method according to claim 5, which is characterized in that the C5.0 decision Tree algorithm specifically:
B1, the comentropy for calculating Land cover types sample set:
Wherein Info (S) indicates the comentropy of Land cover types sample set S, freq (Ci, S) and it indicates to belong to CiThe sample of class Number, i=1,2 ..., k, k are classification sum, | S | indicate the sample number of sample set S；
B2, the conditional entropy introduced after attribute variable T in sample set S is calculated:
Wherein Info (T) indicates to introduce the conditional entropy after attribute variable T, T in sample set SjIndicate j-th point of attribute variable T Class, j=1,2 ..., n, n are the classification number of attribute variable T, Info (Tj) indicate TjConditional entropy；
B3, the condition after attribute variable T is introduced according to the comentropy of Land cover types sample set S and in sample set S The information gain Gain (T) of entropy computation attribute variable T:
Gain (T)=Info (S)-Info (T) (11)
B4, the maximum attribute variable of information gain is chosen as grouping variable, generate the branch of decision tree；
B5, decision tree is successively trimmed upwards from the leaf node of decision tree, child node error is greater than to the node of father node error It trims, node error calculation formula are as follows:
Wherein e is node error, and f is prediction error rate, i.e. the ratio of prediction error number and prediction sum, and z is critical value, N For prediction sum；
B6, Land cover types sample is inputted to the decision tree trimmed, obtains land cover classification result.
CN201811557247.8A 2018-12-19 2018-12-19 A kind of remote sensing image land cover pattern automatic classification method Pending CN109657616A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811557247.8A CN109657616A (en) 2018-12-19 2018-12-19 A kind of remote sensing image land cover pattern automatic classification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811557247.8A CN109657616A (en) 2018-12-19 2018-12-19 A kind of remote sensing image land cover pattern automatic classification method

Publications (1)

Publication Number Publication Date
CN109657616A true CN109657616A (en) 2019-04-19

Family

ID=66115001

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811557247.8A Pending CN109657616A (en) 2018-12-19 2018-12-19 A kind of remote sensing image land cover pattern automatic classification method

Country Status (1)

Country Link
CN (1) CN109657616A (en)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110929572A (en) * 2019-10-18 2020-03-27 天博电子信息科技有限公司 Forest fire identification method and system
CN111160127A (en) * 2019-12-11 2020-05-15 中国资源卫星应用中心 Remote sensing image processing and detecting method based on deep convolutional neural network model
CN111222539A (en) * 2019-11-22 2020-06-02 国际竹藤中心 Method for optimizing and expanding supervision classification samples based on multi-source multi-temporal remote sensing image
CN111460943A (en) * 2020-03-24 2020-07-28 山西大学 Remote sensing image ground object classification method and system
CN111783904A (en) * 2020-09-04 2020-10-16 平安国际智慧城市科技股份有限公司 Data anomaly analysis method, device, equipment and medium based on environmental data
CN112070078A (en) * 2020-11-16 2020-12-11 武汉思众空间信息科技有限公司 Deep learning-based land utilization classification method and system
CN112101159A (en) * 2020-09-04 2020-12-18 国家林业和草原局中南调查规划设计院 Multi-temporal forest remote sensing image change monitoring method
CN112966657A (en) * 2021-03-25 2021-06-15 中国科学院空天信息创新研究院 Remote sensing automatic classification method for large-scale water body coverage
CN113570153A (en) * 2021-08-06 2021-10-29 四川省水利科学研究院 Rainfall station network point number optimization method
CN115100541A (en) * 2022-07-21 2022-09-23 文娟 Satellite remote sensing data processing method and system and cloud platform
CN110929572B (en) * 2019-10-18 2023-11-10 天博电子信息科技有限公司 Forest fire identification method and system

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103208011A (en) * 2013-05-05 2013-07-17 西安电子科技大学 Hyperspectral image space-spectral domain classification method based on mean value drifting and group sparse coding
CN105844287A (en) * 2016-03-15 2016-08-10 民政部国家减灾中心 Domain self-adaptive method and system for remote sensing image classification
CN107392926A (en) * 2017-09-18 2017-11-24 河海大学 Characteristics of remote sensing image system of selection based on soil thematic map early stage

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103208011A (en) * 2013-05-05 2013-07-17 西安电子科技大学 Hyperspectral image space-spectral domain classification method based on mean value drifting and group sparse coding
CN105844287A (en) * 2016-03-15 2016-08-10 民政部国家减灾中心 Domain self-adaptive method and system for remote sensing image classification
CN107392926A (en) * 2017-09-18 2017-11-24 河海大学 Characteristics of remote sensing image system of selection based on soil thematic map early stage

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
林卉，等.: "基于特征向量的遥感影像自动分类研究", 《计算机工程与应用》 *
沈佳洁，等.: "基于均值漂移算法的高分辨率遥感图像分割方法研究", 《全国测绘科技信息网中南分网第二十八次学术信息交流会》 *
王怀警，等.: "C5.0决策树Hyperion影像森林类型精细分类方法", 《浙江农林大学学报》 *
邵惠鹤.: "《工业过程高级控制》", 30 November 1997 *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110929572A (en) * 2019-10-18 2020-03-27 天博电子信息科技有限公司 Forest fire identification method and system
CN110929572B (en) * 2019-10-18 2023-11-10 天博电子信息科技有限公司 Forest fire identification method and system
CN111222539A (en) * 2019-11-22 2020-06-02 国际竹藤中心 Method for optimizing and expanding supervision classification samples based on multi-source multi-temporal remote sensing image
CN111160127A (en) * 2019-12-11 2020-05-15 中国资源卫星应用中心 Remote sensing image processing and detecting method based on deep convolutional neural network model
CN111460943A (en) * 2020-03-24 2020-07-28 山西大学 Remote sensing image ground object classification method and system
CN111783904A (en) * 2020-09-04 2020-10-16 平安国际智慧城市科技股份有限公司 Data anomaly analysis method, device, equipment and medium based on environmental data
CN112101159A (en) * 2020-09-04 2020-12-18 国家林业和草原局中南调查规划设计院 Multi-temporal forest remote sensing image change monitoring method
CN112070078A (en) * 2020-11-16 2020-12-11 武汉思众空间信息科技有限公司 Deep learning-based land utilization classification method and system
CN112966657A (en) * 2021-03-25 2021-06-15 中国科学院空天信息创新研究院 Remote sensing automatic classification method for large-scale water body coverage
CN113570153A (en) * 2021-08-06 2021-10-29 四川省水利科学研究院 Rainfall station network point number optimization method
CN113570153B (en) * 2021-08-06 2024-04-30 四川省水利科学研究院 Rainfall station network point number optimization method
CN115100541A (en) * 2022-07-21 2022-09-23 文娟 Satellite remote sensing data processing method and system and cloud platform

Similar Documents

Publication Publication Date Title
CN109657616A (en) A kind of remote sensing image land cover pattern automatic classification method
CN111666918B (en) Coastline change identification method based on multiple factors
CN110458048A (en) Take population distribution Spatio-temporal Evolution and the cognition of town pattern feature into account
CN102646200B (en) Image classifying method and system for self-adaption weight fusion of multiple classifiers
CN108830870B (en) Satellite image high-precision farmland boundary extraction method based on multi-scale structure learning
CN110414577A (en) A kind of laser radar point cloud multiple target Objects recognition method based on deep learning
CN105260729B (en) A kind of satellite remote-sensing image cloud amount computational methods based on random forest
Quan et al. GIS-based landslide susceptibility mapping using analytic hierarchy process and artificial neural network in Jeju (Korea)
CN105608474B (en) Region adaptivity plant extraction method based on high resolution image
CN104732244B (en) The Classifying Method in Remote Sensing Image integrated based on wavelet transformation, how tactful PSO and SVM
CN108734171A (en) A kind of SAR remote sensing image ocean floating raft recognition methods of depth collaboration sparse coding network
CN110264484A (en) A kind of improvement island water front segmenting system and dividing method towards remotely-sensed data
CN105005789B (en) A kind of remote sensing images terrain classification method of view-based access control model vocabulary
CN109446894A (en) The multispectral image change detecting method clustered based on probabilistic segmentation and Gaussian Mixture
CN105184297B (en) Classification of Polarimetric SAR Image method based on the sparse self-encoding encoder of tensor sum
Chen et al. Agricultural remote sensing image cultivated land extraction technology based on deep learning
CN106228130A (en) Remote sensing image cloud detection method of optic based on fuzzy autoencoder network
CN112949738A (en) Multi-class unbalanced hyperspectral image classification method based on EECNN algorithm
Nourani et al. Classification of groundwater level data using SOM to develop ANN-based forecasting model
CN107121681B (en) Residential area extraction system based on high score satellite remote sensing date
Ma et al. A multisource remotely sensed data oriented method for “ghost city” phenomenon identification
CN104573662B (en) A kind of cloud sentences method and system
Zheng et al. Accuracy comparison and driving factor analysis of LULC changes using multi-source time-series remote sensing data in a coastal area
CN111611960A (en) Large-area ground surface coverage classification method based on multilayer perceptive neural network
CN117611993B (en) Method for estimating vegetation classification based on remote sensing actual evapotranspiration

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: 20190419