CN107909039A - The ground mulching sorting technique of high-resolution remote sensing image based on parallel algorithm - Google Patents
The ground mulching sorting technique of high-resolution remote sensing image based on parallel algorithm Download PDFInfo
- Publication number
- CN107909039A CN107909039A CN201711138873.9A CN201711138873A CN107909039A CN 107909039 A CN107909039 A CN 107909039A CN 201711138873 A CN201711138873 A CN 201711138873A CN 107909039 A CN107909039 A CN 107909039A
- Authority
- CN
- China
- Prior art keywords
- mtd
- remote sensing
- classification
- sensing image
- msub
- 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
-
- 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
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
- G06T2207/30184—Infrastructure
Abstract
The invention discloses a kind of ground mulching sorting technique of the high-resolution remote sensing image based on parallel algorithm, including:S1 carries out cutting according to the number of computer to high-definition remote sensing image data, obtains the high-resolution remote sensing image block after cutting;S2 is based on the parallel frames of OpenMP and all high-resolution remote sensing image blocks is distributed to m processor, concurrent execution ground mulching classification processing;S3, by all high-resolution remote sensing image block numbers according to merging, obtains final ground mulching classification results according to data segmentation principle.The method of the present invention carries out cutting to data automatically according to size of data and using calculator memory situation, use configuration file tissue typing algorithm flow, realize parallel classification algorithm, so as to adapt to the great scale of data volume, the high-resolution ground mulching that finely divides of object space chart task.
Description
Technical field
The invention belongs to Remote Sensing Image Processing Technology field, the more particularly to high-resolution remote sensing image based on parallel algorithm
Ground mulching sorting technique.
Background technology
Land resource is basis for the survival of mankind, with the hair of urbanization process concerning national society's economic construction
Exhibition, determines the natural quality of reflection soil surface type of ground objects and ground mulching/utilization of social property, for earth system mould
Formula, global environmental change, territory protection, urban development decision-making, hydraulic engineering construction etc. all play an important roll.For this reason, in mistake
Go in decades, the research of international community and each tissue to ground mulching/utilization and its change is paid much attention to, research project
It is commonplace, such as International Geosphere Program (IGBP) and the soil of International Human Dimensions Programme on Global Environmental Change (IHDP)
Changeement project and National 863 plan are utilized using/ground mulching changes the ground mulching of core project, U.S. NASA/
Key project " global seismic cover remote sensing mapping and key technology research " etc., the corresponding earth's surface of achievement in research successively issue are covered
Lid product, according to described in pertinent literature, ground mulching product there are about 21 sets on Global Scale at present, at least 43 on regional scale
Set.Meanwhile in order to which the needs built and developed, country have successively carried out second national land investigation and first time on a large scale
National conditions generally investigate project, and the acquisition of wherein ground mulching/utilize is the vital task of land investigation and geographical national conditions monitoring.To sum up institute
State, extensive global or zonal ground mulching drawing has important research meaning and application value.
The global seismic covering product of highest resolution is the whole world 30 of the China based on Landsat TM data creatings at present
The ground mulching product of rice spatial resolution.The Landsat TM data that 30 meter table covering products use are compared to, in rice
The great feature of high-resolution remote sensing image in the ground mulching drawing application of level spatial resolution is high in order to cover the same area
Resolution remote sense image needs the size of bigger, more number of pixels, thus the great scale of data volume, object space finely draw
Divide and require higher so that the global seismic covering drawing task based on high-resolution remote sensing image becomes extremely difficult.The present invention
Purpose be to be directed to high-resolution remote sensing image, there is provided the full-automatic extensive earth's surface cover classification method based on parallel computation.
The content of the invention
For high-resolution remote sensing image ground mulching drawing application in data volume greatly and it is computationally intensive the problem of, the present invention
Parallel processing mode is calculated using data parallel and algorithm, extracts the spectral signature of remote sensing image, by existing maximum seemingly
So merged with algorithm of support vector machine, reuse connected component labeling algorithm and intermediate result is post-processed, so as to
Quickly to obtain high-resolution ground mulching classification results.
To reach above-mentioned purpose, the technical scheme is that a kind of high-resolution remote sensing image based on parallel algorithm
Ground mulching sorting technique, includes the following steps:
Step 1, carries out cutting to high-definition remote sensing image data according to the number of computer, obtains the height after cutting
Resolution remote sense image block, while the coordinate of each high-resolution remote sensing image block start-stop position is recorded, implementation is as follows,
If h is the length of image, w is the width of the width, then the high-resolution remote sensing image block after cutting of imageLength H=W*r, wherein ratio r=h/w, m are memory number, StFor data type, SsTo be safely
Number;
All high-resolution remote sensing image blocks are distributed to m processor, concurrently by step 2 based on the parallel frames of OpenMP
The classification of execution ground mulching processing;
Step 3, according to the coordinate of each high-resolution remote sensing image block start-stop position, by all high-definition remote sensing shadows
As block number is according to merging, final ground mulching classification results are obtained.
Further, ground mulching classification processing includes the following steps in the step 2:
Step 1, high-resolution remote sensing image block feature extracts, including following sub-step;
Step 1.1, the spectral signature of high-resolution remote sensing image block is extracted;
Step 1.2, feature of the normalized differential vegetation index as extraction vegetation is calculated;
Step 1.3, feature of the normalization water body index as extraction water body is calculated;
Step 1.4, the textural characteristics based on gray scale symbiosis square are calculated as local spatial feature;
Step 1.5, by the spectral signature of extraction, normalized differential vegetation index, normalization water body by way of vector is superimposed
Index and Texture Feature Fusion, the feature as subsequent classification input;
Step 2, the sorting algorithm of structure fusion maximum likelihood and support vector machines, in high-resolution remote sensing image block
Table covering classification is classified, and obtains classification chart picture, including following sub-step:
Step 2.1, the membership probability of ground mulching classification is obtained based on maximum likelihood classification algorithm;
Step 2.2, the atural object class membership probabilities based on acquisition, according to the accuracy of the differentiation of threshold decision atural object, its
Further comprise:
Step 2.2.1, the optimal mark classification of current pixel is determined by maximum a posteriori probability;
Step 2.2.2, differentiates the accuracy of pixel classification, directly using step if classification accuracy is more than given threshold value
The classification results that label corresponding to maximum probability forms in rapid 2.1 maximum likelihood classification algorithm, otherwise, jump into step 2.3;
Step 2.3, using Nonlinear Support Vector Machines sorting algorithm to the picture that is difficult to accurately differentiate in maximum likelihood classification
Member distinguishes in non-linear space;
Step 3, cutting object is produced on classification chart picture based on connected component labeling algorithm, passes through region merging technique strategy pair
Classification results in step 2 are post-processed, and obtain ground mulching classification results in high-resolution remote sensing image block, specific implementation
Mode is as follows,
Step 3.1, on the basis of classification results in step 2, obtained using classical eight neighborhood connected component labeling algorithm
To cutting object O;
Step 3.2, according to the spatial resolution of high-definition remote sensing and atural object characteristic, the cutting object of acquisition is defined
For Oi, wherein i is classification number;Pass through threshold value T1Judge OiWhether belong to noise object, work as OiMore than given threshold value T1It is then direct
Retain, be otherwise merged into neighbouring cutting object.
Further, ground mulching classification processing further includes step 4 in the step 2, according to Kappa coefficients to earth's surface
Cover classification result carries out quality evaluation, and embodiment is as follows:
Step 4.1, confusion matrix is calculated,
Wherein m represents the other number of target class, pijRepresent that the i-th row jth shows pijThe pixel of the i-th class of a actual ownership
Number is predicted to be jth class;Numerical value on matrix leading diagonal is expressed as, certain one kind and corresponding classification are attributed in earth's surface true picture
It is attributed to the number of of a sort pixel, i.e., the number for the pixel correctly classified in figure;
Step 4.2, producer's precision, formula are calculated based on confusion matrixWherein i, j are represented respectively
For the atural object classification shared by the i-th class in confusion matrix and concrete class jth class, AjjRepresent the element on diagonal in confusion matrix
Value;
Calculate overall classification accuracy, formulaWherein N represents authentic specimen sum;
Kappa coefficients are calculated, formula kappa=(d-q)/(N-q), wherein N are authentic specimen sum, and d is confusion matrix
The number of sample on middle diagonal, C be classification number, AijRepresent and i is corresponded in confusion matrix, the element value of j positions, q's
Expression formula is
Step 4.3, when Kappa values are more than α, the uniformity between presentation class result and ground reference information is very big or smart
Degree is very high, and Kappa values represent that uniformity is medium in β-α, and Kappa values then represent that uniformity is very poor less than β.
Further, the calculation formula of normalized differential vegetation index is in step 1.2Wherein ρnirIt is
The reflectance value of the near infrared band of high-resolution multi-spectral remote sensing image, ρredIt is the reflectance value of red band.
Further, water body index calculation formula is normalized in step 1.3 isWherein ρnirIt is
The reflectance value of the near infrared band of high-resolution multi-spectral remote sensing image, ρgreenIt is the reflectance value of image green band.
Further, the textural characteristics in step 1.4 include entropy and homogeney based on gray scale symbiosis square, and implementation is such as
Under,
It is horizontal direction spatial domain to make Kx={ 1,2 ..., Nx }, and Ky={ 1,2 ..., Ny } is vertical direction spatial domain, figure
As gray scale pixel section Kx*Ky=k, wherein entropy can be definedHomogeneyP in formulaijRepresent the element value that the i-th row j is arranged in the gray level co-occurrence matrixes after normalization.
Further, the value of β and α is respectively 0.4 and 0.8 in step 4.3.
Compared with prior art, the invention has the advantages that and beneficial effect:
(1) adaptability is good, can carry out cutting to data automatically according to size of data and using calculator memory situation,
So as to be adapted to different configuration of running environment;
(2) there is computation capability, add the parallel algorithm of adaptability, can be carried out according to the CPU quantity of machine
Parallelization is handled;
(3) there is ease for use, without auxiliary information and manual intervention, calculating speed is fast, can automatic business processing.
(4) there is reusability, the global seismic covering drawing application system based on High Resolution Remote Sensing Satellites uses configuration
File organization sorting algorithm flow, so that different user can repeat to obtain corresponding ground mulching by identical configuration file
Charting results.
The method of the present invention is by way of vector is superimposed by the spectral signature of extraction, normalized differential vegetation index, normalization water
Body index and Texture Feature Fusion, the feature for obtaining fusion as subsequent classification input, and utilize maximum likelihood and supporting vector
The sorting algorithm of machine classifies high-resolution remote sensing image, and locates after being carried out by region merging technique strategy to classification results
Reason, finally calculates Kappa coefficients and carries out quality evaluation to classification chart picture, realize the judgement to nicety of grading.
Brief description of the drawings
Fig. 1 is ground mulching of the present invention classification process chart;
Fig. 2 is hyperplane schematic diagram in Nonlinear Support Vector Machines of the embodiment of the present invention.
Embodiment
Technical solution for a better understanding of the present invention, is the present invention below in conjunction with drawings and examples further in detail
Describe in detail bright.
Step 1, parallel computation frame is established.
Step 1.1, data parallel, since high-definition remote sensing image data amount is big, before ground mulching classification is carried out,
Need high-definition remote sensing image data carrying out cutting.Data parallel is before data operation, according to certain strategy certainly
It is dynamic that cutting is carried out to data, a larger specification data is divided into multiple small data modules, is calculated by multiple core cpus etc.
Resource is calculated, calculate complete after the automatic of data carried out according to data segmentation principles merge, return to requirement format
Data.Data parallel can be calculated with formula below.
Automatically after carrying out cutting to data, the width of each small data moduleRatio r=
H/w, wherein, h is the length of image, and w is the width of image, and m is memory number, StFor data type, for example it is integer, floating-point
Type, SsFor safety coefficient, safety coefficient takes 4 in the present embodiment;The length H=W*r of each small data template;Record decimal at the same time
According to the coordinate of the start-stop position of module.
Step 1.2, algorithm is parallel.Algorithm is to be directed to specific algorithm parallel, in actual calculating process, based on OpenMP
Etc. parallel frame distribute to multiple processors it is concurrent perform calculating.The present invention is realized automatically by the way of flow custom
Table cover classification, user can select extracting method, disaggregated model and the post-classification comparison method of feature by configuration file.
Step 2, feature extraction is carried out for the high-resolution remote sensing image block after cutting, this step further comprises following
Step.
Step 2.1 directly carries out Spectra feature extraction using the multi light spectrum hands of high-resolution remote sensing image block, by general
Thang-kng spectrum band overlapping, which provides, differentiates the basic of atural object.
Step 2.2, feature of the normalized differential vegetation index (NDVI) as extraction vegetation is calculated, calculation formula isWherein ρnirIt is the reflectance value of the near infrared band of high-resolution multi-spectral remote sensing image, ρredIt is
The reflectance value of red band.
Step 2.3, feature of the normalization water body index (NDWI) as extraction water body is calculated, calculation formula isWherein ρnirIt is the reflectance value of the near infrared band of high-resolution multi-spectral remote sensing image, ρgreen
It is the reflectance value of image green band.
Step 2.4, calculate based on the textural characteristics of gray level co-occurrence matrixes (GLCM) as local spatial feature, described with this
Spatial relationship between each pixel between image, it is horizontal direction spatial domain to make Kx={ 1,2 ..., Nx }, Ky=1,2 ...,
Ny } it is vertical direction spatial domain, for given image, Nx, Ny are known quantity, and image can be defined gray scale pixel section
Kx*Ky=k, with entropyAnd homogeneyAs local sky
Between feature, estimate structure textural characteristics Data-Statistics, wherein, p in formulaijRepresent that the i-th row j is arranged in the gray level co-occurrence matrixes after normalization
Element value.Gray level co-occurrence matrixes are pixel distance and the matrix function of angle, it is by calculating certain distance and one in image
The correlation between 2 gray scales in direction is determined, to reflect comprehensive letter of the image on direction, interval, amplitude of variation and speed
Breath, its calculation is the prior art, for details, reference can be made to document [1], and the present invention not writes.
[1] elevation journey, texture feature extraction [J] the computer system applications of Hui Xiaowei based on gray level co-occurrence matrixes,
2010,19(6):195-198.
Step 2.5, by the spectral signature of extraction, normalized differential vegetation index, normalization water body by way of vector is superimposed
Index and Texture Feature Fusion, the feature for obtaining fusion as subsequent classification input.
Step 3, the sorting algorithm of structure fusion maximum likelihood and support vector machines, carries out high-resolution remote sensing image block
Classification, obtains classification chart picture, and the embodiment of this step is as follows:
Step 3.1, the membership probability of ground mulching classification is provided based on maximum likelihood classification algorithm (MLC), to obtain most
Whole ground mulching type provides decision-making foundation.Maximum likelihood classification is by the distribution of remote sensing multi-wavelength data as multidimensional normal state
It is distributed to construct discriminant function.Its basic thought is:The data of all kinds of known pixels form certain in a plane or in space
Point group;As soon as forming a normal distribution on the number axis of oneself per a kind of every one-dimensional data, such multidimensional data forms this
One multiple normal distribution of class, there is all kinds of multiple dimensional distribution models, for the data vector of any one unknown classification,
It can be reversely asked to belong to all kinds of probability;Compare the size of these probability, see that the probability which kind of belongs to is big, just this data
Vector or this pixel are classified as such, are represented by:
In formula, m is wave band number;p(wi) be the i-th class m dimension Density Function of Normal Distribution, by it can be seen that in kth class
There is the probability size of various probable values in middle m n-dimensional random variable ns x.The m dimension data vectors of pixel are represented by:
With miRepresent class ωiThe average of each wave bandThe mean vector formed, such as following formula,
CiRepresent the covariance matrix of class, formula isWherein, nkIt is the pixel number of kth class;WkIt is kth
Mean dispersion error matrix in the class of class, such as following formula,
In formula, ωk11,ωk22,…ωkmmIt is the variance within clusters of kth class;And ωk12,…,ωk1mAnd ωk21,…,ωkm1Deng
It is covariance in the class of kth class, this makes it possible to obtain the covariance matrix of class.
Can be following formula by full scale equation formula abbreviation by abbreviation,
Wherein p (wi) represent class ωiProbability, f represents corresponding class ωiProbability density.Based on maximum likelihood decision function
Value, passes through Pi(l)=exp (Di(f)), l=1 ..., K obtains the membership probability of atural object classification, and wherein K is ground mulching classification
Number.Maximum likelihood classification algorithm (MLC) algorithm is the prior art, for details, reference can be made to document [2].
[2] soup Guoan process in remote sensing digital image processing [M] Science Presses, 2004.
Step 3.2, the atural object class membership probabilities based on acquisition, according to the accuracy of the differentiation of threshold decision atural object, its
Further comprise:
Step 3.2.1, the optimal mark of current pixel, formula are determined by maximum a posteriori probability
Wherein k represents the optimal mark of current pixel;
Step 3.2.2, differentiates the accuracy of pixel classification, directly using step if classification accuracy is more than given threshold value
The classification results that label corresponding to maximum probability forms in rapid 2.1 maximum likelihood classification algorithm, otherwise, jump into step 2.3.Formula
Pi(k) > T, wherein T are the acquiescence accuracy threshold values set, and usual value is 0.8, represent the probably acquirement pair of current pixel
The mark answered.
Step 3.3, using Nonlinear Support Vector Machines (SVM) sorting algorithm to being difficult to accurately differentiate in maximum likelihood classification
Pixel distinguished in non-linear space.Different from maximum likelihood sorting technique based on probability, support vector machines is one
Machine learning algorithm of the kind based on Statistical Learning Theory, using structural risk minimization principle, is minimizing the same of sample error
When reduce model extensive error, so as to improve the generalization ability of model.Formula f (x)=w Φ (x)+b represents hyperplane
Discriminant function, wherein w are weight vector, and b is biasing, and Φ (x) is the vector for being related to parameter x.As shown in fig. 2, there are one two
Dimensional plane, has two kinds of different data in plane, is represented respectively with circle and fork.Since these data are linear separabilities,
This two classes data can be separated with straight line, in two-dimensional space, the straight line is equivalent to hyperplane.Nonlinear Support Vector Machines
Sorting algorithm is the prior art, for details, reference can be made to document [3].
[3] publishing house of Li Hang statistical learning methods [M] Tsinghua University, 2012.
Step 4, cutting object is produced based on connected component labeling algorithm, by region merging technique strategy to point in step 2
Class result is post-processed, and the embodiment of this step is as follows:
Step 4.1, on the basis of classification results in step 2, connected using the classical eight neighborhood of Union-find Sets data structure
Logical zone marker algorithm can obtain cutting object O;
Step 4.2, on the basis of classification results in step 2, according to the spatial resolution of high-resolution remote sensing image
And atural object characteristic, the cutting object of acquisition is defined as Oi, wherein i is classification number.Pass through threshold value T1Judge whether it belongs to
Noise object (T1General value is 0.8) such as larger than given threshold value T1Then directly retain, be otherwise merged into neighbouring cutting object,
In units of object, classification results are obtained by maximum temporal voting strategy, the classification results in step 2 are improved, so as to obtain
Take finer ground mulching classification results.
Step 5, according to the process shown in step 1, the start-stop position coordinates of the recorded small data module of use, inversely
Data merging is carried out, obtains final classification results.
Step 6, quality evaluation is carried out to classification chart picture, the embodiment of this step is as follows:
Step 6.1, confusion matrix is calculated, confusion matrix is by by the position and classification chart of the true pixel of each earth's surface
The classification pixel of relevant position be relatively calculated, the row of confusion matrix represent pixel in classification chart in each classification
In number, row represents the true belonging kinds of data.Confusion matrix is as follows:
Wherein m represents the other number of target class, pijRepresent that the i-th row jth shows pijThe pixel quilt of the i-th class of a actual ownership
It is predicted as jth class.Therefore, the numerical value on matrix leading diagonal is expressed as, certain one kind and corresponding point are attributed in earth's surface true picture
It is attributed to the number of of a sort pixel, i.e., the number for the pixel correctly classified in class figure.By upper, in matrix on leading diagonal
Numerical value is more big, and the pixel number correctly classified is more, and the precision of classification is higher.
Step 6.2, producer's precision (Producer ' s Accuracy, PA), formula are calculated based on confusion matrixWherein i, j are expressed as the i-th class and the atural object classification shown in concrete class jth class in confusion matrix,
AjjRepresent the element value on diagonal in confusion matrix;Calculate overall classification accuracy (Overall Accuracy, OA), formulaWherein N represents the authentic specimen sum being pre-selected.Calculate the injustice between Kappa coefficients reaction classification
Weighing apparatus property, wherein formula kappa=(d-q)/(N-q), N are the number of total authentic specimen, and d is on diagonal in confusion matrix
The number of sample, C be classification number (can be according to User Defined), AijRepresent and i is corresponded in confusion matrix, the member of j positions
Element value, the expression formula of q areKappa values>0.80 is one between classification chart and ground reference information
Cause property is very big or precision is very high, and Kappa values represent that uniformity is medium in 0.40-0.80, and Kappa values are less than 0.40 and represent consistent
Property is very poor.
Specific embodiment described herein is only that spirit of the present invention is illustrated.Technology belonging to the present invention is led
The technical staff in domain can make described specific embodiment various modifications, supplement or replace in a similar way
Generation, but not deviate the spirit or beyond the scope of the appended claims of the present invention.
Claims (7)
1. the ground mulching sorting technique of the high-resolution remote sensing image based on parallel algorithm, it is characterised in that including following step
Suddenly:
Step 1, carries out cutting to high-definition remote sensing image data according to the number of computer, obtains the high-resolution after cutting
Rate remote sensing image block, while the coordinate of each high-resolution remote sensing image block start-stop position is recorded, implementation is as follows,
If h is the length of image, w is the width of the width, then the high-resolution remote sensing image block after cutting of imageLength H=W*r, wherein ratio r=h/w, m are memory number, StFor data type, SsTo be safely
Number;
All high-resolution remote sensing image blocks are distributed to m processor by step 2 based on the parallel frames of OpenMP, and concurrent holds
The classification of row ground mulching is handled;
Step 3, according to the coordinate of each high-resolution remote sensing image block start-stop position, by all high-resolution remote sensing image blocks
Data merge, and obtain final ground mulching classification results.
2. the ground mulching sorting technique of the high-resolution remote sensing image based on parallel algorithm as claimed in claim 1, it is special
Sign is, in the step 2 ground mulching classification processing include the following steps:
Step 1, high-resolution remote sensing image block feature extracts, including following sub-step;
Step 1.1, the spectral signature of high-resolution remote sensing image block is extracted;
Step 1.2, feature of the normalized differential vegetation index as extraction vegetation is calculated;
Step 1.3, feature of the normalization water body index as extraction water body is calculated;
Step 1.4, the textural characteristics based on gray scale symbiosis square are calculated as local spatial feature;
Step 1.5, by the spectral signature of extraction, normalized differential vegetation index, normalization water body index by way of vector is superimposed
And Texture Feature Fusion, the feature as subsequent classification input;
Step 2, the sorting algorithm of structure fusion maximum likelihood and support vector machines, covers earth's surface in high-resolution remote sensing image block
Lid classification is classified, and obtains classification chart picture, including following sub-step:
Step 2.1, the membership probability of ground mulching classification is obtained based on maximum likelihood classification algorithm;
Step 2.2, the atural object class membership probabilities based on acquisition, according to the accuracy of the differentiation of threshold decision atural object, wherein into
One step includes:
Step 2.2.1, the optimal mark classification of current pixel is determined by maximum a posteriori probability;
Step 2.2.2, differentiates the accuracy of pixel classification, and step is directly used if classification accuracy is more than given threshold value
The classification results that label corresponding to maximum probability forms in 2.1 maximum likelihood classification algorithms, otherwise, jump into step 2.3;
Step 2.3, existed using Nonlinear Support Vector Machines sorting algorithm to the pixel for being difficult to accurately differentiate in maximum likelihood classification
Distinguished in non-linear space;
Step 3, cutting object is produced on classification chart picture based on connected component labeling algorithm, by region merging technique strategy to step
Classification results in 2 are post-processed, and obtain ground mulching classification results in high-resolution remote sensing image block, embodiment
It is as follows,
Step 3.1, on the basis of classification results in step 2, divided using classical eight neighborhood connected component labeling algorithm
Cut object O;
Step 3.2, according to the spatial resolution of high-definition remote sensing and atural object characteristic, the cutting object of acquisition is defined as Oi,
Wherein i is classification number;Pass through threshold value T1Judge OiWhether belong to noise object, work as OiMore than given threshold value T1Then directly retain,
Otherwise it is merged into neighbouring cutting object.
3. the ground mulching sorting technique of the high-resolution remote sensing image based on parallel algorithm as claimed in claim 2, it is special
Sign is, in the step 2 ground mulching classification processing further include step 4, according to Kappa coefficients to earth's surface cover classification knot
Fruit carries out quality evaluation, and embodiment is as follows:
Step 4.1, confusion matrix is calculated,
<mrow>
<mi>M</mi>
<mo>=</mo>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>p</mi>
<mn>11</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>p</mi>
<mn>21</mn>
</msub>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<msub>
<mi>p</mi>
<mrow>
<mi>m</mi>
<mn>1</mn>
</mrow>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>p</mi>
<mn>12</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>p</mi>
<mn>22</mn>
</msub>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<msub>
<mi>p</mi>
<mrow>
<mi>m</mi>
<mn>2</mn>
</mrow>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>p</mi>
<mrow>
<mn>1</mn>
<mi>m</mi>
</mrow>
</msub>
</mtd>
<mtd>
<msub>
<mi>p</mi>
<mrow>
<mn>2</mn>
<mi>m</mi>
</mrow>
</msub>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<msub>
<mi>p</mi>
<mrow>
<mi>m</mi>
<mi>m</mi>
</mrow>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
Wherein m represents the other number of target class, pijRepresent that the i-th row jth shows pijThe pixel number of the i-th class of a actual ownership is pre-
Survey as jth class;Numerical value on matrix leading diagonal is expressed as, certain one kind is attributed in earth's surface true picture and is returned with corresponding classification chart
In the number of of a sort pixel, i.e., the number for the pixel correctly classified;
Step 4.2, producer's precision, formula are calculated based on confusion matrixWherein i, j are expressed as mixing
Confuse the i-th class and the atural object classification shared by concrete class jth class, A in matrixjjRepresent the element value on diagonal in confusion matrix;
Calculate overall classification accuracy, formulaWherein N represents authentic specimen sum;
Kappa coefficients are calculated, formula kappa=(d-q)/(N-q), wherein N are authentic specimen sum, and d is right in confusion matrix
The number of sample on linea angulata, C be classification number, AijRepresent and i is corresponded in confusion matrix, the element value of j positions, the expression of q
Formula is
Step 4.3, when Kappa values are more than α, uniformity between presentation class result and ground reference information is very big or precision very
Height, Kappa values represent that uniformity is medium in β-α, and Kappa values then represent that uniformity is very poor less than β.
4. the ground mulching sorting technique of the high-resolution remote sensing image based on parallel algorithm as claimed in claim 3, it is special
Sign is:The calculation formula of normalized differential vegetation index is in step 1.2Wherein ρnirIt is that high-resolution is more
The reflectance value of the near infrared band of spectral remote sensing image, ρredIt is the reflectance value of red band.
5. the ground mulching sorting technique of the high-resolution remote sensing image based on parallel algorithm as claimed in claim 4, it is special
Sign is:Water body index calculation formula is normalized in step 1.3 isWherein ρnirIt is that high-resolution is more
The reflectance value of the near infrared band of spectral remote sensing image, ρgreenIt is the reflectance value of image green band.
6. the ground mulching sorting technique of the high-resolution remote sensing image based on parallel algorithm as claimed in claim 5, it is special
Sign is:Textural characteristics in step 1.4 include entropy and homogeney based on gray scale symbiosis square, and implementation is as follows,
It is horizontal direction spatial domain to make Kx={ 1,2 ..., Nx }, and Ky={ 1,2 ..., Ny } is vertical direction spatial domain, and image can be with
It is defined gray scale pixel section Kx*Ky=k, wherein entropyHomogeney
P in formulaijRepresent the element value that the i-th row j is arranged in the gray level co-occurrence matrixes after normalization.
7. the ground mulching sorting technique of the high-resolution remote sensing image based on parallel algorithm as claimed in claim 6, it is special
Sign is:The value of β and α is respectively 0.4 and 0.8 in step 4.3.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711138873.9A CN107909039B (en) | 2017-11-16 | 2017-11-16 | High-resolution remote sensing image earth surface coverage classification method based on parallel algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711138873.9A CN107909039B (en) | 2017-11-16 | 2017-11-16 | High-resolution remote sensing image earth surface coverage classification method based on parallel algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107909039A true CN107909039A (en) | 2018-04-13 |
CN107909039B CN107909039B (en) | 2020-03-10 |
Family
ID=61845743
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711138873.9A Active CN107909039B (en) | 2017-11-16 | 2017-11-16 | High-resolution remote sensing image earth surface coverage classification method based on parallel algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107909039B (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108827880A (en) * | 2018-04-23 | 2018-11-16 | 吉林大学 | Ground mulching change detecting method based on multispectral image and NDVI time series |
CN108985260A (en) * | 2018-08-06 | 2018-12-11 | 航天恒星科技有限公司 | A kind of remote sensing and meteorological integrated rice yield estimation method |
CN109584284A (en) * | 2018-12-13 | 2019-04-05 | 宁波大学 | A kind of seashore wetland ground object sample extracting method of hierarchical decision making |
CN109657724A (en) * | 2018-12-21 | 2019-04-19 | 浙江中烟工业有限责任公司 | A kind of furrowfilter rod characteristic parameter quick calculation method based on parallel computation |
CN109657598A (en) * | 2018-12-13 | 2019-04-19 | 宁波大学 | Seashore wetland Classification in Remote Sensing Image method based on Stratified Strategy |
CN110033044A (en) * | 2019-04-17 | 2019-07-19 | 北京建筑大学 | Ground mulching product integration or fusion method based on EAGLE matrix |
CN110033499A (en) * | 2019-04-21 | 2019-07-19 | 南京林业大学 | A kind of regional water body drafting method based on Landsat OLI remote sensing image |
CN110796176A (en) * | 2019-10-09 | 2020-02-14 | 武汉大学 | High-resolution image classification method and system based on pixel pair and weighted voting |
CN111199195A (en) * | 2019-12-26 | 2020-05-26 | 中科禾信遥感科技(苏州)有限公司 | Pond state full-automatic monitoring method and device based on remote sensing image |
CN112651465A (en) * | 2021-01-13 | 2021-04-13 | 北京观微科技有限公司 | Equipment target interpretation method based on high-resolution remote sensing image |
CN113222005A (en) * | 2021-05-08 | 2021-08-06 | 兰州交通大学 | Automatic updating method for land coverage |
CN113223022A (en) * | 2021-05-31 | 2021-08-06 | 湖南科技大学 | Multivariate image segmentation method based on multivariate texture image analysis algorithm |
CN113762175A (en) * | 2021-09-10 | 2021-12-07 | 复旦大学 | Two-stage behavior identification fine classification method based on graph convolution network |
CN114283335A (en) * | 2021-12-27 | 2022-04-05 | 河南大学 | Historical period remote sensing identification precision verification preparation method |
CN113762175B (en) * | 2021-09-10 | 2024-04-26 | 复旦大学 | Two-stage behavior recognition fine classification method based on graph convolution network |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020194159A1 (en) * | 2001-06-08 | 2002-12-19 | The Regents Of The University Of California | Parallel object-oriented data mining system |
CN101256677A (en) * | 2008-03-01 | 2008-09-03 | 深圳先进技术研究院 | Automatic monitoring simulation and parallelization process method thereof |
CN102073879A (en) * | 2010-12-02 | 2011-05-25 | 南京大学 | Method for identifying characteristic land categories of ocean remote sensing images of coast on basis of semi-supervised learning |
CN103218783A (en) * | 2013-04-17 | 2013-07-24 | 国家测绘地理信息局卫星测绘应用中心 | Fast geometric correction method for satellite remote sensing image and based on control point image database |
CN103400354A (en) * | 2013-08-14 | 2013-11-20 | 山东大学 | OpenMP-based geometric correcting and parallel processing method for remote-sensing images |
CN104751477A (en) * | 2015-04-17 | 2015-07-01 | 薛笑荣 | Space domain and frequency domain characteristic based parallel SAR (synthetic aperture radar) image classification method |
CN104881867A (en) * | 2015-05-13 | 2015-09-02 | 华中科技大学 | Method for evaluating quality of remote sensing image based on character distribution |
CN107194313A (en) * | 2017-04-19 | 2017-09-22 | 中国国土资源航空物探遥感中心 | A kind of parallel intelligent object-oriented classification method |
-
2017
- 2017-11-16 CN CN201711138873.9A patent/CN107909039B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020194159A1 (en) * | 2001-06-08 | 2002-12-19 | The Regents Of The University Of California | Parallel object-oriented data mining system |
CN101256677A (en) * | 2008-03-01 | 2008-09-03 | 深圳先进技术研究院 | Automatic monitoring simulation and parallelization process method thereof |
CN102073879A (en) * | 2010-12-02 | 2011-05-25 | 南京大学 | Method for identifying characteristic land categories of ocean remote sensing images of coast on basis of semi-supervised learning |
CN103218783A (en) * | 2013-04-17 | 2013-07-24 | 国家测绘地理信息局卫星测绘应用中心 | Fast geometric correction method for satellite remote sensing image and based on control point image database |
CN103400354A (en) * | 2013-08-14 | 2013-11-20 | 山东大学 | OpenMP-based geometric correcting and parallel processing method for remote-sensing images |
CN104751477A (en) * | 2015-04-17 | 2015-07-01 | 薛笑荣 | Space domain and frequency domain characteristic based parallel SAR (synthetic aperture radar) image classification method |
CN104881867A (en) * | 2015-05-13 | 2015-09-02 | 华中科技大学 | Method for evaluating quality of remote sensing image based on character distribution |
CN107194313A (en) * | 2017-04-19 | 2017-09-22 | 中国国土资源航空物探遥感中心 | A kind of parallel intelligent object-oriented classification method |
Non-Patent Citations (4)
Title |
---|
DIPTI PRASAD MUKHERJEE 等: "Ore image segmentation by learning image and shape features", 《PATTERN RECOGNITION LETTERS》 * |
佃袁勇 等: "傅里叶谱纹理和光谱信息结合的高分辨率遥感影像地表覆盖分类", 《武汉大学学报·信息科学版》 * |
匡文慧 等: "亚洲人造地表覆盖遥感精细化分类与分布特征分析", 《中国科学:地球科学》 * |
欧新良 等: "基于动态分界点计算的并行几何校正算法", 《计算机研究与发展》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108827880A (en) * | 2018-04-23 | 2018-11-16 | 吉林大学 | Ground mulching change detecting method based on multispectral image and NDVI time series |
CN108827880B (en) * | 2018-04-23 | 2021-04-16 | 吉林大学 | Earth surface coverage change detection method based on multispectral image and NDVI time sequence |
CN108985260A (en) * | 2018-08-06 | 2018-12-11 | 航天恒星科技有限公司 | A kind of remote sensing and meteorological integrated rice yield estimation method |
CN108985260B (en) * | 2018-08-06 | 2022-03-04 | 航天恒星科技有限公司 | Remote sensing and meteorological integrated rice yield estimation method |
CN109584284A (en) * | 2018-12-13 | 2019-04-05 | 宁波大学 | A kind of seashore wetland ground object sample extracting method of hierarchical decision making |
CN109657598A (en) * | 2018-12-13 | 2019-04-19 | 宁波大学 | Seashore wetland Classification in Remote Sensing Image method based on Stratified Strategy |
CN109657598B (en) * | 2018-12-13 | 2020-05-05 | 宁波大学 | Coastal wetland remote sensing classification method based on layering strategy |
CN109657724A (en) * | 2018-12-21 | 2019-04-19 | 浙江中烟工业有限责任公司 | A kind of furrowfilter rod characteristic parameter quick calculation method based on parallel computation |
CN109657724B (en) * | 2018-12-21 | 2021-08-03 | 浙江中烟工业有限责任公司 | Parallel computing-based method for rapidly computing characteristic parameters of grooved filter sticks |
CN110033044A (en) * | 2019-04-17 | 2019-07-19 | 北京建筑大学 | Ground mulching product integration or fusion method based on EAGLE matrix |
CN110033499A (en) * | 2019-04-21 | 2019-07-19 | 南京林业大学 | A kind of regional water body drafting method based on Landsat OLI remote sensing image |
CN110796176A (en) * | 2019-10-09 | 2020-02-14 | 武汉大学 | High-resolution image classification method and system based on pixel pair and weighted voting |
CN110796176B (en) * | 2019-10-09 | 2022-12-16 | 武汉大学 | High-resolution image classification method and system based on pixel pair and weighted voting |
CN111199195A (en) * | 2019-12-26 | 2020-05-26 | 中科禾信遥感科技(苏州)有限公司 | Pond state full-automatic monitoring method and device based on remote sensing image |
CN111199195B (en) * | 2019-12-26 | 2023-07-18 | 中科禾信遥感科技(苏州)有限公司 | Pond state full-automatic monitoring method and device based on remote sensing image |
CN112651465A (en) * | 2021-01-13 | 2021-04-13 | 北京观微科技有限公司 | Equipment target interpretation method based on high-resolution remote sensing image |
CN112651465B (en) * | 2021-01-13 | 2024-02-02 | 北京观微科技有限公司 | Equipment target interpretation method based on high-resolution remote sensing image |
CN113222005B (en) * | 2021-05-08 | 2022-05-27 | 兰州交通大学 | Automatic updating method for land coverage |
CN113222005A (en) * | 2021-05-08 | 2021-08-06 | 兰州交通大学 | Automatic updating method for land coverage |
CN113223022A (en) * | 2021-05-31 | 2021-08-06 | 湖南科技大学 | Multivariate image segmentation method based on multivariate texture image analysis algorithm |
CN113762175A (en) * | 2021-09-10 | 2021-12-07 | 复旦大学 | Two-stage behavior identification fine classification method based on graph convolution network |
CN113762175B (en) * | 2021-09-10 | 2024-04-26 | 复旦大学 | Two-stage behavior recognition fine classification method based on graph convolution network |
CN114283335A (en) * | 2021-12-27 | 2022-04-05 | 河南大学 | Historical period remote sensing identification precision verification preparation method |
Also Published As
Publication number | Publication date |
---|---|
CN107909039B (en) | 2020-03-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107909039A (en) | The ground mulching sorting technique of high-resolution remote sensing image based on parallel algorithm | |
Wen et al. | A novel automatic change detection method for urban high-resolution remotely sensed imagery based on multiindex scene representation | |
Shihavuddin et al. | Image-based coral reef classification and thematic mapping | |
CN103413151B (en) | Hyperspectral image classification method based on figure canonical low-rank representation Dimensionality Reduction | |
CN110929607B (en) | Remote sensing identification method and system for urban building construction progress | |
CN105046276B (en) | Hyperspectral image band selection method based on low-rank representation | |
CN107358260B (en) | Multispectral image classification method based on surface wave CNN | |
CN107909015A (en) | Hyperspectral image classification method based on convolutional neural networks and empty spectrum information fusion | |
Qin et al. | Object-based land cover change detection for cross-sensor images | |
CN102663391A (en) | Image multifeature extraction and fusion method and system | |
CN104282008B (en) | The method and apparatus that Texture Segmentation is carried out to image | |
CN103793908A (en) | Method for constructing prediction model of multifunctional veins based on brain nuclear magnetic resonance image | |
Yue et al. | Texture extraction for object-oriented classification of high spatial resolution remotely sensed images using a semivariogram | |
Xiao et al. | Segmentation of multispectral high-resolution satellite imagery using log Gabor filters | |
CN106295498A (en) | Remote sensing image target area detection apparatus and method | |
Xu et al. | Generalized morphological component analysis for hyperspectral unmixing | |
Corbane et al. | Mapping natural habitats using remote sensing and sparse partial least square discriminant analysis | |
Liu et al. | MS-CNN: multiscale recognition of building rooftops from high spatial resolution remote sensing imagery | |
Bai et al. | Achieving better category separability for hyperspectral image classification: A spatial–spectral approach | |
Bo et al. | Mean shift‐based clustering analysis of multispectral remote sensing imagery | |
Kishore et al. | Content-Based Image Retrieval System Based on Fusion of Wavelet Transform, Texture and Shape Features. | |
Wang et al. | SiamFuseNet: A pseudo-siamese network for detritus detection from polarized microscopic images of river sands | |
Zou et al. | A spatialized classification approach for land cover mapping using hyperspatial imagery | |
Wang et al. | Simultaneous extracting area and quantity of agricultural greenhouses in large scale with deep learning method and high-resolution remote sensing images | |
CN104050486B (en) | Polarimetric SAR image classification method based on maps and Wishart distance |
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 |