CN109948697A - A method of completed region of the city is extracted using crowd-sourced data auxiliary classification of remote-sensing images - Google Patents
A method of completed region of the city is extracted using crowd-sourced data auxiliary classification of remote-sensing images Download PDFInfo
- Publication number
- CN109948697A CN109948697A CN201910208202.8A CN201910208202A CN109948697A CN 109948697 A CN109948697 A CN 109948697A CN 201910208202 A CN201910208202 A CN 201910208202A CN 109948697 A CN109948697 A CN 109948697A
- Authority
- CN
- China
- Prior art keywords
- training sample
- data
- city
- crowd
- remote
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A30/00—Adapting or protecting infrastructure or their operation
- Y02A30/60—Planning or developing urban green infrastructure
Landscapes
- Image Analysis (AREA)
Abstract
The invention belongs to remote sensing image process fields, disclose a kind of method for extracting completed region of the city using crowd-sourced data auxiliary classification of remote-sensing images, pre-process first to remotely-sensed data;Then the pretreatment operations such as coordinate conversion, redundant data rejecting and filtering are carried out to the crowd-sourced data of acquisition, calculates the frequency of training sample and spectrum similarity in single pixel;Finally completed region of the city is extracted using one-class support vector machines.The present invention realizes the rapidly extracting of completed region of the city, realize the training sample that real-time, intimate zero cost is automatically generated using crowd-sourced data, it has merged crowd-sourced geodata and Methods on Multi-Sensors RS Image extracts completed region of the city, greatly improves the extraction accuracy of completed region of the city.
Description
Technical field
The present invention relates to a kind of methods for extracting completed region of the city using crowd-sourced data auxiliary classification of remote-sensing images, belong to distant
Feel image processing field.
Background technique
Completed region of the city area is to measure the important of urbanization process to want index.Accurate built-up areas boundary of extracting facilitates more
Reasonably planning clay, and the correlative studys such as urban waterlogging, tropical island effect are provided with important reference.The expansion of built-up areas is to area
Domain climate change also has important influence.
Remote sensing technology plays highly important role to the monitoring that large-scale land type changes.Currently, using more
Source remote sensing image, such as DMSP/LOS remote sensing light data, MODIS, Landsat, ENVISAT ASAR, SPOT, build city
A wide range of application has been obtained at the technology that area extracts.So far, it has been developed many for non-supervisory or prison
The method for superintending and directing interpretation of satellite image.Especially supervised classification, since its higher precision performance obtains more a wide range of application.
Although extracting completed region of the city using remote sensing image has many advantages, the acquisition of training sample required for supervised classification
Often influence a key factor of built-up areas extraction accuracy.Supervised classification needs a large amount of practical training sample, to expend
A large amount of human and material resources and financial resources, when especially specification area rises to Global Scale, " three power " to be expended is surprising
, this just proposes urgent demand to the sampling techniques of real-time, inexpensive training sample.Meanwhile the choosing to training sample
It selects and needs operator class has priori knowledge over the ground, artificial selection training sample is time-consuming and laborious, limits completed region of the city extraction
Efficiency.
Summary of the invention
The object of the present invention is to provide a kind of sides that completed region of the city is extracted using crowd-sourced data auxiliary classification of remote-sensing images
Method is automatically generated the training sample of real-time, intimate zero cost using crowd-sourced data, has merged crowd-sourced geodata and multi-source remote sensing
Image data extracts completed region of the city and improves the extraction accuracy of completed region of the city to realize the rapidly extracting of completed region of the city.
To achieve the goals above, the present invention provide it is a kind of using crowd-sourced data auxiliary classification of remote-sensing images extract city build
At the method in area, include the following steps:
(1) data prediction is carried out to the remote sensing image of acquisition;
(2) crowd-sourced data are obtained, select training sample in source data of comforming, are calculated according to the spatial resolution of remote sensing image
The number of training sample in single pixel, and then calculate frequency and spectrum similarity;
(3) impermeable stratum is extracted with one-class support vector machines;
(4) EM cluster is carried out to the impermeable stratum, obtains the completed region of the city to be extracted.
Further, remote sensing image described in step (1) is derived from landsat-8, and the data prediction includes data school
Just, rejecting outliers and cloud exposure mask.
Further, the Data correction is in the case where multispectral data, using the calibration value of offer by landform school
Correction data is converted to radiation value;The abnormality detection is the exceptional value for acquiring inconsistent or calibration error and generating;The cloud exposure mask
Cloud cover is avoided to influence for multispectral image of the selection cloud amount less than 10%.
Further, crowd-sourced data described in step (2) include the two kinds of openings of social media data and Openstreetmap
Data source.
Further, in the step (2), the social media data are derived from Twitter, first calculating Landsat shadow
Tweets quantity as in single pixel, then only takes the pixel of at least one tweets point as training sample;With Ω
={ xi, i=1 ..., l } and indicate training sample set, wherein and x indicates a column vector, and columns is equal to the wave of multispectral image
Number of segment;
Frequency and spectrum similarity to twitter data measure:
1) measure the frequency of Twitter data: the tweet frequency for defining i-th of training sample is Fi:
Wherein l is the Landsat pixel number containing tweets, TiFor tweets number in corresponding pixel, TiValue it is bigger
The probability that expression training sample is located at impermeable stratum is bigger;
2) it measures the spectral similarity of Twitter data: gathering assuming that the training sample derived from from Twitter forms one
Class, distance of the training sample from waterproof class to cluster centre are less than distance of the training sample from permeable class to cluster centre, instruction
The distance for practicing sample to cluster centre carries out quantitative measurment with minimum covariance matrix MCD;When a cloud is symmetrically dispersed in one
When around center, it is assumed that i-th of training sample meets multivariate normal distributions, and probability density function is expressed as
Wherein xiIt is the spectral vector of i-th of training sample, u is the average value of training sample, and Σ is the covariance of mean value
The distance definition of matrix, i-th of training sample and its mean value is spectral similarity Si
WhereinWithIt is sample average and sample covariance matrix, S respectivelyiValue i-th of training sample of smaller expression fall
Probability in waterproof region is bigger;
Based on the frequency and spectral similarity of above-mentioned Twitter data, the weight of i-th of training sample is indicated are as follows:
Further, in the step (2), for being derived from the data of the Openstreetmap, first by impermeable stratum
It is converted into two OSM grating images that spatial resolution is respectively 1m and 30m, i-th of the OSM grating image of 30m resolution ratio
Waterproof pixel will cover the region R that 900 pixels are formed on the OSM grating image that resolution ratio is 1mi;Therefore, i-th it is impermeable
The frequency of water surface pixel is defined as:
WhereinIt is the OSM grid graph region R of 1m resolution ratioiWaterproof pixel number, lOIt is 30m resolution ratio OSM grid
The impermeable stratum pixel number of image;
The spectral similarity S of impermeable surface pixel using OSM as training samplei oIt is expressed as follows:
Wherein yiWithIt is sample average and sample covariance matrix, S respectivelyi oValue i-th of training sample of smaller expression fall
Probability in waterproof region is bigger;The weight w for the training sample that OSM is generatedi oIt is expressed as follows:
Further, in the step (3), classified to remote sensing image using one-class support vector machines OCSVM to mention
Take impermeable stratum information;
OCSVM trains a hypersphere, which has the smallest volume, wherein the maximum training comprising single class
Sample, the distance of sample to boundary are construed to degree of similarity, provide the reference for whether belonging to certain kinds about sample;
Using OCSVM as a constrained convex optimal problem
Wherein a and R is center and the radius of suprasphere respectively;| | | | it is Euler's norm, ξiIt is for controlling loose journey
Slack variable is spent, C is a tradeoff parameter;The dual form of formula (8) is as follows:
Wherein<,>it is inner product, { α1,...,αlIt is Lagrange multiplier;
The weighted value of the training sample from waterproof region is set to be greater than the weighted value of the training sample from permeable region,
By in the weight introduction-type (8) of training sample in formula (4), as follows
Formula (9) accordingly becomes following formula:
With kernel function K < xi,xjInner product<x in>replacement formula (7) and formula (9)i,xj>, it obtains OCSVM and weighting is single
The coring version of class support vector machines (weighted OCSVM, WOCSVM), is determined most suitable by five times of cross-validation methods
The selection of parameter, free parameter is acquired by following formula:
Wherein θ is the entire set of free parameter,It is the optimized parameter collection obtained by five times of cross-validation methods, OA is
Overall accuracy, nSV are the quantity of supporting vector.
Further, EM clustering algorithm described in step (4) is the probability to the impermeable stratum obtained in step (3)
Figure is clustered, and to remove the picture noise of the probability graph, fills hole present in image.
Further, the probability graph is divided into five classes by the EM clustering algorithm, extracted from five class belong to it is impermeable
At least one kind of water layer.
The present invention has merged crowd-sourced data and remote sensing image data, and high-precision city is realized with one-class support vector machines
It extracts city built-up areas: obtaining open and free crowd-sourced geodata first, crowd-sourced data are pre-processed;To remotely-sensed data
It is pre-processed;Finally completed region of the city is extracted using one-class support vector machines, is realized automatic using crowd-sourced data
The training sample for generating real-time, intimate zero cost has merged crowd-sourced geodata and has built with Methods on Multi-Sensors RS Image extraction city
At area, the rapidly extracting of completed region of the city is realized, greatly improves the extraction accuracy of completed region of the city.
The other feature and advantage of the embodiment of the present invention will the following detailed description will be given in the detailed implementation section.
Detailed description of the invention
Attached drawing is to further understand for providing to the embodiment of the present invention, and constitute part of specification, under
The specific embodiment in face is used to explain the present invention embodiment together, but does not constitute the limitation to the embodiment of the present invention.Attached
In figure:
Fig. 1 is the flow chart of one embodiment of the invention.
Fig. 2 is the landsat-8 remotely-sensed data of Tokyo range in one embodiment of the invention.
Fig. 3 is the topography of Fig. 2.
Fig. 4 is distribution map of the social media data twitter as training sample.
Fig. 5 is the result figure that impermeable stratum is extracted using twitter data as training sample.
Fig. 6 is distribution map of the data OSM as training sample of increasing income.
Fig. 7 is the result figure that impermeable stratum is extracted using OSM data as training sample.
Specific embodiment
It is described in detail below in conjunction with specific embodiment of the attached drawing to the embodiment of the present invention.It should be understood that this
Locate described specific embodiment and be merely to illustrate and explain the present invention embodiment, is not intended to restrict the invention embodiment.
In one embodiment of the invention, as shown in Figure 1, being to automatically generate training sample to remote sensing using crowd-sourced data
Image data extracts the method flow diagram of completed region of the city, and specific steps include:
S1. initial data pre-processes.
The remotely-sensed data that the specific embodiment of the invention selects for landsat 8 image.Survey region is Tokyo, is
The radiation difference of different Landsat data sets over time and space used in embodiment is reduced, is used several pre-
Processing step.Specifically include Data correction, abnormality detection and cloud exposure mask.
(a) Data correction: in the case where multispectral data, terrain correction data is converted to using the calibration value of offer
Radiation value.
(b) abnormality detection: acquisition is inconsistent or calibration error generates exceptional value, (the label especially in multispectral data
For 0 or NaN), exceptional value would potentially result in error in classification.Such case mostly occurs on the boundary of image, because sensor is caught
The each wave band grasped has slight delay, leads to the loss of learning of some positions.As previously mentioned, this phenomena impair mostly light
Spectrum data set, and increase the marginal portion that must be excluded from processing.This problem in order to prevent proposes a kind of automatic screen
The method covered.
(c) cloud exposure mask: the presence of cloud layer will affect the precision of classification, select multispectral image thus to avoid cloud cover
It influences.Practical operation selects cloud amount for the image less than 10%.
Fig. 2 is to have chosen spatial discrimination by pretreated Tokyo in October, 2015 Landsat-8OLI image
Rate is 7 wave bands of 30m.The embodiment of the present invention has chosen 2666 (row) * 2728 (column) partial zones from whole scape image
Domain is illustrated (Fig. 3).
S2. training sample is generated from open data.The embodiment of the present invention have chosen respectively social media data (with
Twitter data instance) and two kinds of Openstreetmap (OSM) open data sources.Illustrate crowd-sourced number with both data instances
According to automatically generating training sample strategy.Herein, Twitter is considered as the unique next of the relevant social media in geographical location
Source.The tweets quantity in Landsat image in single pixel is calculated first, then only takes the picture of at least one tweets point
Element is used as training sample.Ω={ xi, i=1 ..., l } indicate training sample set.Wherein, x indicate column vector (compose to
Amount), columns is equal to the wave band number of multispectral image.It should be noted that having the Twitter data of original Geographic Reference
There are two essential characteristics: 1) tweets can repeat to send in 30m pixel;2) tweets can both be sent out on fluid-tight surface
It send, can also be sent on permeable surface.The embodiment of the present invention proposes two kinds of measurement method (i.e. frequencies of twitter data
With spectrum similarity) features of Twitter data described.
1) frequency of Twitter data: Twitter data would generally repeat in the range similar in identical coordinate
(landsat-8 ground space resolution ratio is 30 meters).On this basis, the tweet frequency for defining i-th of training sample is Fi:
Wherein l is the Landsat pixel number containing tweets, TiFor tweets number in corresponding pixel.TiValue it is bigger
Mean that training sample is more likely located in impermeable stratum.For example, being sent out in some pixel on the image of Landsat 8
The number of twitter is 5 times, and has the frequency of maximum hair twitter in some pixel in image, such as maximum frequency
It is 10 times, then the corresponding frequency of that pixel that frequency is 5 is 0.5, i.e. Fi=0.5.
2) spectral similarity of Twitter data: although as described above, have a small amount of tweets from permeable region,
Such as sea/sandy beach, farmland, mountainous region, but major part tweets concentrates on fluid-tight region.It means that coming from
The training sample of Twitter has similar spectral value.Assuming that the training sample derived from from Twitter forms a cluster.Instruction
It is smaller to practice distance of the sample from waterproof class to cluster centre, and distance of the training sample from permeable class to cluster centre is larger.
Training sample to cluster centre distance can with minimum covariance matrix (Minimum Covariance Determinant,
MCD quantitative measurment) is carried out.The core concept of MCD algorithm is to find the smallest h observation of space divergence.The row of covariance matrix
Column is the method for the divergence of a good metric point cloud, when a cloud (being the observation in h bosom in this example)
When being symmetrically dispersed in around a center.Assuming that i-th of training sample meets multivariate normal distributions, probability density function can
To be expressed as
Wherein xiIt is the spectral vector of i-th of training sample, u is the average value of training sample, and Σ is the covariance of mean value
Matrix.I-th of training sample and the distance definition of its mean value are spectral similarity Si
WhereinWithIt is sample average and sample covariance matrix respectively.SiValue smaller mean i-th of training sample
More likely fall in waterproof region.The frequency and spectral similarity for fully considering Twitter data, by i-th of training sample
Weight indicate are as follows:
For example, the frequency when i-th of training sample is 0.5, work as SiWhen=2, weight wi=0.8244;Fig. 4 shows social activity
Distribution map of the media data twitter as training sample.
For OSM, we first by the impermeable stratums such as building, traffic, road be converted into spatial resolution be respectively 1m and
Two grating images of 30m.It is 1m's that i-th of waterproof pixel of the OSM grating image of 30m resolution ratio, which will cover resolution ratio,
900 pixel (region R on OSM grating imagei).Therefore, the frequency of i-th of impermeable surface pixel is defined as:
WhereinIt is the OSM grid graph region R of 1m resolution ratioiWaterproof pixel number, lOIt is 30m resolution ratio OSM grid
The impermeable stratum pixel quantity of image;As region RiIn waterproof pixel number be 300, then the frequency F in the regioni O=
0.3333;
Similar with Twitter, the spectral similarity of the impermeable surface pixel of OSM can also be calculated with same method.
Utilize Fi OReplace Fi, spectral similarity in more new formula (4), the weight for the training sample that available OSM is generated.With OSM
The spectral similarity S of impermeable surface pixel as training samplei oIt is expressed as follows:
Wherein yiWithIt is sample average and sample covariance matrix, S respectivelyi oValue i-th of training sample of smaller expression fall
Probability in waterproof region is bigger;The weight w for the training sample that OSM is generatedi oIt is expressed as follows:
Fig. 6 is distribution map of another open source data (OSM) as training sample.
S3. impermeable stratum is extracted with one-class support vector machines
The embodiment of the present invention classifies to satellite image using one-class support vector machines (OCSVM), extracts impermeable stratum
Information.Compared with finding the standard support vector machines of Optimal Boundary of two classes of isolation, OCSVM trains a hypersphere, should
Hypersphere has the smallest volume, wherein including the maximum training sample of single class.The distance on sample to boundary can be construed to
Degree of similarity, it can provide the reference for whether belonging to certain kinds about sample.OCSVM is a constrained convex optimal problem
Wherein a and R is center and the radius of suprasphere respectively;| | | | it is Euler's norm, ξiIt is slack variable, C is one
A tradeoff parameter, it is used to control loose degree.The dual form of formula (8) is as follows:
Wherein<,>it is inner product, { α1,...,αlIt is Lagrange multiplier.
Twitter data as training sample may be from waterproof and permeable region.In other words, training sample packet
Pixel containing a certain proportion of accidentally label.As step S2 is introduced, we give the training sample one from waterproof region
A biggish weighted value, and give one lesser weighted value of training sample from permeable region.Therefore we assume that a power
The small training sample of weight is weaker to the influence for being fitted hyperspherical parameter (such as center and radius).For this purpose, by training sample in formula (4)
In this weight introduction-type (8), as follows
Formula (8) accordingly becomes following formula:
With kernel function K < xi,xjInner product<x in>replacement formula (9) and formula (11)i,xj>, just obtain OCSVM and
The coring version of WOCSVM.Most suitable parameter is determined by five times of cross-validation methods.In general, when the sample for there was only target label
When available, free parameter is difficult to adjust.In fact, in this case, real rate (sensitivity) can only be calculated, and cannot calculate
Other error respective values (specificity).In this case, the selection of free parameter is acquired by following formula:
Wherein θ is the entire set of free parameter,It is the optimized parameter collection obtained by five times of cross-validation methods, OA is
Overall accuracy, nSV are the quantity of supporting vector.This performance constraints are by keeping lower supporting vector quantity come limited model
Complexity, to realize higher overall accuracy.Cost is calculated in order to reduce large-scale training sample size bring, is used
The LIBSVM packet that GPU accelerates realizes OCSVM classification.
S4. the obtained probability graph of step S3 is clustered.
The impermeable stratum (probability graph) that previous step is obtained using expectation maximization clustering algorithm (EM clustering algorithm) into
Row cluster, the purpose of cluster is to remove picture noise, fills hole present in image.The general step of EM algorithm is divided into:
E step: Qi(z(i)) :=p (z(i)|x(i);θ)
M step:
Wherein Z is hidden variable, and θ is undetermined parameter, and E step is preset parameter θ to ask the expectation of Z, and M step is estimation pole
Maximum-likelihood estimation.Probability graph is fallen into 5 types using EM algorithm, extracts that a kind of or several class for belonging to impermeable stratum, finally
Obtain the completed region of the city to be extracted or impermeable stratum.
Fig. 5, Fig. 7 are to extract the result figure of impermeable stratum using twitter data as training sample and utilize OSM respectively
Data extract the result figure of impermeable stratum as training sample.
The optional embodiment of the embodiment of the present invention is described in detail in conjunction with attached drawing above, still, the embodiment of the present invention is simultaneously
The detail being not limited in above embodiment can be to of the invention real in the range of the technology design of the embodiment of the present invention
The technical solution for applying example carries out a variety of simple variants, these simple variants belong to the protection scope of the embodiment of the present invention.
It is further to note that specific technical features described in the above specific embodiments, in not lance
In the case where shield, it can be combined in any appropriate way.In order to avoid unnecessary repetition, the embodiment of the present invention pair
No further explanation will be given for various combinations of possible ways.
In addition, any combination can also be carried out between a variety of different embodiments of the embodiment of the present invention, as long as it is not
The thought of the embodiment of the present invention is violated, equally should be considered as disclosure of that of the embodiment of the present invention.
Claims (9)
1. a kind of method for extracting completed region of the city using crowd-sourced data auxiliary classification of remote-sensing images, which is characterized in that including such as
Lower step:
(1) data prediction is carried out to the remote sensing image of acquisition;
(2) crowd-sourced data are obtained, select training sample in source data of comforming, are calculated according to the spatial resolution of remote sensing image single
The number of training sample in pixel, and then calculate frequency and spectrum similarity;
(3) impermeable stratum is extracted with one-class support vector machines;
(4) EM cluster is carried out to the impermeable stratum, obtains the completed region of the city to be extracted.
2. the method according to claim 1 for extracting completed region of the city using crowd-sourced data auxiliary classification of remote-sensing images,
It is characterized in that, remote sensing image described in step (1) is derived from landsat-8, and the data prediction includes Data correction, exceptional value
Detection and cloud exposure mask.
3. the method according to claim 2 for extracting completed region of the city using crowd-sourced data auxiliary classification of remote-sensing images,
It is characterized in that, the Data correction is to be turned terrain correction data using the calibration value of offer in the case where multispectral data
It is changed to radiation value;The abnormality detection is the exceptional value for acquiring inconsistent or calibration error and generating;The cloud exposure mask is selection cloud
The multispectral image less than 10% is measured to avoid cloud cover from influencing.
4. the method according to claim 3 for extracting completed region of the city using crowd-sourced data auxiliary classification of remote-sensing images,
It is characterized in that, crowd-sourced data described in step (2) include social media data and two kinds of Openstreetmap open data sources.
5. the method according to claim 4 for extracting completed region of the city using crowd-sourced data auxiliary classification of remote-sensing images,
It is characterized in that, in the step (2), the social media data are derived from Twitter, single in calculating Landsat image first
Then tweets quantity in pixel only takes the pixel of at least one tweets point as training sample;With Ω={ xi, i=
1 ..., l } indicate training sample set, wherein and x indicates a column vector, and columns is equal to the wave band number of multispectral image;
Frequency and spectrum similarity to twitter data measure:
1) measure the frequency of Twitter data: the tweet frequency for defining i-th of training sample is Fi:
Wherein l is the Landsat pixel number containing tweets, TiFor tweets number in corresponding pixel, TiThe bigger expression of value
The probability that training sample is located at impermeable stratum is bigger;
2) spectral similarity of Twitter data is measured: assuming that the training sample derived from from Twitter forms a cluster, instruction
Practice distance of the sample from waterproof class to cluster centre and is less than distance of the training sample from permeable class to cluster centre, training sample
Distance to cluster centre carries out quantitative measurment with minimum covariance matrix MCD;When a cloud is symmetrically dispersed in a center week
When enclosing, it is assumed that i-th of training sample meets multivariate normal distributions, and probability density function is expressed as
Wherein xiIt is the spectral vector of i-th of training sample, u is the average value of training sample, and Σ is the covariance matrix of mean value,
I-th of training sample and the distance definition of its mean value are spectral similarity Si
WhereinWithIt is sample average and sample covariance matrix, S respectivelyiValue i-th of training sample of smaller expression fall in not
The probability in permeable region is bigger;
Based on the frequency and spectral similarity of above-mentioned Twitter data, the weight of i-th of training sample is indicated are as follows:
6. the method according to claim 5 for extracting completed region of the city using crowd-sourced data auxiliary classification of remote-sensing images,
It is characterized in that, in the step (2), for being derived from the data of the Openstreetmap, converts sky for impermeable stratum first
Between resolution ratio be respectively 1m and 30m two OSM grating images, i-th of waterproof picture of the OSM grating image of 30m resolution ratio
Member will cover the region R that 900 pixels are formed on the OSM grating image that resolution ratio is 1mi;Therefore, i-th of impermeable surface picture
The frequency of element is defined as:
WhereinIt is the OSM grid graph region R of 1m resolution ratioiWaterproof pixel number, lOIt is 30m resolution ratio OSM grating image
Impermeable stratum pixel number;
The spectral similarity S of impermeable surface pixel using OSM as training samplei oIt is expressed as follows:
Wherein yiWithIt is sample average and sample covariance matrix, S respectivelyi oValue i-th of training sample of smaller expression fall in not
The probability in permeable region is bigger;The weight w for the training sample that OSM is generatedi oIt is expressed as follows:
7. the method according to claim 1 for extracting completed region of the city using crowd-sourced data auxiliary classification of remote-sensing images,
It is characterized in that, in the step (3), is classified using one-class support vector machines OCSVM to remote sensing image waterproof to extract
Layer information;
OCSVM trains a hypersphere, which has the smallest volume, wherein the maximum training sample comprising single class
This, the distance of sample to boundary is construed to degree of similarity, provides the reference for whether belonging to certain kinds about sample;
Using OCSVM as a constrained convex optimal problem
Wherein a and R is center and the radius of suprasphere respectively;| | | | it is Euler's norm, ξiIt is loose for controlling loose degree
Variable, C are a tradeoff parameters;The dual form of formula (8) is as follows:
Wherein<,>it is inner product, { α1,...,αlIt is Lagrange multiplier;
The weighted value of the training sample from waterproof region is set to be greater than the weighted value of the training sample from permeable region, by formula
(4) in the weight introduction-type (8) of training sample, as follows
Formula (9) accordingly becomes following formula:
With kernel function K < xi,xjInner product<x in>replacement formula (7) and formula (9)i,xj>, it obtains OCSVM and weights single class and support
The coring version of vector machine (weighted OCSVM, WOCSVM), determines most suitable parameter by five times of cross-validation methods, from
It is acquired by the selection of parameter by following formula:
Wherein θ is the entire set of free parameter,It is the optimized parameter collection obtained by five times of cross-validation methods, OA is overall
Precision, nSV are the quantity of supporting vector.
8. the method according to claim 1 for extracting completed region of the city using crowd-sourced data auxiliary classification of remote-sensing images,
It is characterized in that, EM clustering algorithm described in step (4) is gathered to the probability graph of the impermeable stratum obtained in step (3)
Class fills hole present in image to remove the picture noise of the probability graph.
9. the method according to claim 8 for extracting completed region of the city using crowd-sourced data auxiliary classification of remote-sensing images,
Be characterized in that, the probability graph is divided into five classes by the EM clustering algorithm, extracted from five class belong to impermeable stratum to
Few one kind.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910208202.8A CN109948697B (en) | 2019-03-19 | 2019-03-19 | Method for extracting urban built-up area by using multi-source data to assist remote sensing image classification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910208202.8A CN109948697B (en) | 2019-03-19 | 2019-03-19 | Method for extracting urban built-up area by using multi-source data to assist remote sensing image classification |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109948697A true CN109948697A (en) | 2019-06-28 |
CN109948697B CN109948697B (en) | 2022-08-26 |
Family
ID=67008433
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910208202.8A Active CN109948697B (en) | 2019-03-19 | 2019-03-19 | Method for extracting urban built-up area by using multi-source data to assist remote sensing image classification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109948697B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110781948A (en) * | 2019-10-22 | 2020-02-11 | 北京市商汤科技开发有限公司 | Image processing method, device, equipment and storage medium |
CN111125553A (en) * | 2019-11-22 | 2020-05-08 | 中国科学院城市环境研究所 | Intelligent urban built-up area extraction method supporting multi-source data |
CN115270904A (en) * | 2022-04-13 | 2022-11-01 | 广州市城市规划勘测设计研究院 | Method and system for spatialization of proper-age permanent population in compulsory education stage |
CN118501878A (en) * | 2024-07-18 | 2024-08-16 | 北京广安融创房产测绘有限公司 | Real estate mapping method |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160379388A1 (en) * | 2014-07-16 | 2016-12-29 | Digitalglobe, Inc. | System and method for combining geographical and economic data extracted from satellite imagery for use in predictive modeling |
CN109034233A (en) * | 2018-07-18 | 2018-12-18 | 武汉大学 | A kind of high-resolution remote sensing image multi classifier combination classification method of combination OpenStreetMap |
CN109063754A (en) * | 2018-07-18 | 2018-12-21 | 武汉大学 | A kind of remote sensing image multiple features combining classification method based on OpenStreetMap |
CN109299673A (en) * | 2018-09-05 | 2019-02-01 | 中国科学院地理科学与资源研究所 | The green degree spatial extraction method of group of cities and medium |
CN109325085A (en) * | 2018-08-08 | 2019-02-12 | 中南大学 | A kind of urban land identification of function and change detecting method |
-
2019
- 2019-03-19 CN CN201910208202.8A patent/CN109948697B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160379388A1 (en) * | 2014-07-16 | 2016-12-29 | Digitalglobe, Inc. | System and method for combining geographical and economic data extracted from satellite imagery for use in predictive modeling |
CN109034233A (en) * | 2018-07-18 | 2018-12-18 | 武汉大学 | A kind of high-resolution remote sensing image multi classifier combination classification method of combination OpenStreetMap |
CN109063754A (en) * | 2018-07-18 | 2018-12-21 | 武汉大学 | A kind of remote sensing image multiple features combining classification method based on OpenStreetMap |
CN109325085A (en) * | 2018-08-08 | 2019-02-12 | 中南大学 | A kind of urban land identification of function and change detecting method |
CN109299673A (en) * | 2018-09-05 | 2019-02-01 | 中国科学院地理科学与资源研究所 | The green degree spatial extraction method of group of cities and medium |
Non-Patent Citations (2)
Title |
---|
JORDI MŨNOZ-MARÍ等: "Semisupervised One-Class Support Vector Machines for Classification of Remote Sensing Data", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
韩贵金: "加权SVDD算法在人体姿态估计中的研究与应用", 《计算机工程与应用》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110781948A (en) * | 2019-10-22 | 2020-02-11 | 北京市商汤科技开发有限公司 | Image processing method, device, equipment and storage medium |
CN111125553A (en) * | 2019-11-22 | 2020-05-08 | 中国科学院城市环境研究所 | Intelligent urban built-up area extraction method supporting multi-source data |
CN111125553B (en) * | 2019-11-22 | 2022-05-31 | 中国科学院城市环境研究所 | Intelligent urban built-up area extraction method supporting multi-source data |
CN115270904A (en) * | 2022-04-13 | 2022-11-01 | 广州市城市规划勘测设计研究院 | Method and system for spatialization of proper-age permanent population in compulsory education stage |
CN118501878A (en) * | 2024-07-18 | 2024-08-16 | 北京广安融创房产测绘有限公司 | Real estate mapping method |
CN118501878B (en) * | 2024-07-18 | 2024-09-13 | 北京广安融创房产测绘有限公司 | Real estate mapping method |
Also Published As
Publication number | Publication date |
---|---|
CN109948697B (en) | 2022-08-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109948697A (en) | A method of completed region of the city is extracted using crowd-sourced data auxiliary classification of remote-sensing images | |
KR20220000898A (en) | Method to identify shoreline changes based on multi-factor | |
CN103363962B (en) | Remote sensing evaluation method of lake water reserves based on multispectral images | |
CN111539296B (en) | Method and system for identifying illegal building based on remote sensing image change detection | |
Yang et al. | Mapping topsoil electrical conductivity by a mixed geographically weighted regression kriging: A case study in the Heihe River Basin, northwest China | |
CN111695440B (en) | GA-SVR lake level measurement and prediction method based on radar altimeter | |
CN106780586B (en) | A kind of solar energy potential evaluation method based on ground laser point cloud | |
CN111273378B (en) | Typhoon center positioning method based on wind stress disturbance | |
CN109858450A (en) | Ten meter level spatial resolution remote sensing image cities and towns extracting methods of one kind and system | |
CN109299673A (en) | The green degree spatial extraction method of group of cities and medium | |
CN109584284A (en) | A kind of seashore wetland ground object sample extracting method of hierarchical decision making | |
CN109635713A (en) | The shadow region glacier recognition methods of plateau mountainous region | |
CN111222539A (en) | Method for optimizing and expanding supervision classification samples based on multi-source multi-temporal remote sensing image | |
CN109635828A (en) | A kind of typical geographical national conditions elements recognition system and method in ecological protection red line area | |
Liu et al. | Spatial-temporal hidden Markov model for land cover classification using multitemporal satellite images | |
CN105930863A (en) | Determination method for spectral band setting of satellite camera | |
Agarwal et al. | Development of machine learning based approach for computing optimal vegetation index with the use of sentinel-2 and drone data | |
Na et al. | Assessing breeding habitat suitability for the endangered red-crowned crane (Grus japonensis) based on multi-source remote sensing data | |
Jande et al. | Assessment of land use and land cover changes and urban expansion using remote sensing and GIS in Gboko, Benue State, Nigeria | |
CN115690418B (en) | Unsupervised automatic detection method for image waypoints | |
CN115424131B (en) | Cloud detection optimal threshold selection method, cloud detection method and cloud detection system | |
CN110688990B (en) | Rice planting candidate area determination method | |
Meng | Remote Sensing of Urban Green Space | |
CN118115872B (en) | Identification method for north-shift boundary of corn planting zone | |
Jia et al. | ENHANCED U-NET ALGORITHM FOR TYPICAL CROP CLASSIFICATION USING GF-6 WFV REMOTE SENSING IMAGES |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |