CN109829519A - Classifying Method in Remote Sensing Image and system based on adaptive space information - Google Patents
Classifying Method in Remote Sensing Image and system based on adaptive space information Download PDFInfo
- Publication number
- CN109829519A CN109829519A CN201910220213.8A CN201910220213A CN109829519A CN 109829519 A CN109829519 A CN 109829519A CN 201910220213 A CN201910220213 A CN 201910220213A CN 109829519 A CN109829519 A CN 109829519A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- pixel
- current
- indicate
- center pel
- 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
Abstract
The invention discloses a kind of Classifying Method in Remote Sensing Image and system based on adaptive space information.This method comprises: obtaining remote sensing images；Preliminary classification is carried out to remote sensing images using the FCM Algorithms based on Markov random field, obtains initially obscuring subordinated-degree matrix；Using spatial attraction model, the space attraction under current iteration number b in remote sensing images between current center pel and each neighborhood pixel is calculated；Edge detection is carried out to remote sensing images using Sobel operator, obtains spatial structure characteristic；According to spatial structure characteristic, the fringing coefficient of current center pel is calculated using gradient inverse exponential smoothing；According to space attraction and fringing coefficient, the Markov random field of adaptive weighting is constructed；By the Markov random field of adaptive weighting in conjunction with FCM Algorithms, the classification results of remote sensing images are determined.The present invention can be effectively solved the problem of boundary pixel and estimation of spatial information weight coefficient, improve nicety of grading.
Description
Technical field
The present invention relates to technical field of remote sensing image processing, more particularly to a kind of remote sensing based on adaptive space information
Image classification method and system.
Background technique
With the rapid development of remote sensing technology, remote sensing image is widely used in land resources, environmental monitoring, urban planning etc.
Field.In order to meet the application requirement of different field, need to extract a large amount of useful information from remote sensing image, wherein remote sensing shadow
As classification is the key technology of information extraction and the basis of subsequent remote sensing image Objects recognition.Since high spatial resolution is distant
Feel the terrestrial object information comprising complexity in image, the spectrum in similar atural object is heterogeneous obvious, and the spectrum of different atural objects is overlapped,
Prevent it is traditional based on the classification method of spectral signature from obtaining satisfied nicety of grading.In addition, special based on single features
The Remote Image Classification for being only based on spectral signature is easy affected by noise, and obtained classification results are often more scrappy
And the loss of atural object marginal information is larger.Therefore, in order to improve nicety of grading, by the features phase such as spectrum, spatial information and shape
In conjunction with multiple features fusion classification method have become high spatial resolution remote sense image classification hot spot.
The classification method of multiple features fusion is divided into supervised classification method and unsupervised segmentation method.Theoretically, supervised classification
Method is by the training sample using class categories to can obtain better classification results than unsupervised segmentation method, still
Due to the influence in remote sensing image there are many mixed pixels and external condition to remote sensing image, it is difficult to obtain a large amount of reliable
Ground remote sensing image training sample is used for supervised classification, and therefore, the classification results that supervised classification method obtains are difficult satisfactory.And
Unsupervised segmentation method does not need training sample then, it is divided according to the similarity degree of image, so that without prison
It superintends and directs classification method to be widely used in classification of remote-sensing images, such as K-means, fuzzy C-mean algorithm (Fuzzy C-means, FCM), horse
The clustering algorithms such as Er Kefu random field (Markov Random Field, MRF) have been used in unsupervised segmentation.Wherein, FCM is calculated
The classification results of method meet the ambiguity and uncertainty of nature atural object, can retain more image informations, therefore, FCM
Clustering algorithm is widely used in then remote sensing image classification.However, traditional FCM clustering algorithm is not examined in image clustering
The spatial coherence for considering image, causes its noiseproof feature poor, so that there is isolated pixel in image classification result.
In recent years, spatial information is introduced in traditional FCM clustering algorithm to improve clustering precision, has become research people
The common recognition of member.Wherein by MRF be introduced into FCM method be also it is a kind of improve FCM classical way, such as based on markov with
FCM (Markov random fieldbased FCM, the MFCM) algorithm on airport.MRF is a kind of probabilistic model in MFCM algorithm,
It extracts space neighborhood information by the dependence between adjacent picture elements, thus not only have very strong noiseproof feature but also
The spatial neighborhood relationship between image picture elements is fully considered, but the shortcomings that MRF model is its clustering precision by spatial information
The influence of weight, usually experiment obtains repeatedly for the spatial information weight coefficient, and neighborhood pixel centering in traditional MRF
The influence of imago member is identical, therefore is suitable for homogeneous area classification, and bad for fringe region classifying quality, is easily lost side
Edge details.Therefore, the nicety of grading of existing MFCM calculation sorting algorithm is to be improved, and it is this to handle to need more effective way
Boundary pixel and spatial information weight coefficient estimation problem.
Summary of the invention
Based on this, it is necessary to a kind of Classifying Method in Remote Sensing Image and system based on adaptive space information is provided, to have
Effect solves the problems, such as boundary pixel and the estimation of spatial information weight coefficient, improves nicety of grading.
To achieve the above object, the present invention provides following schemes:
Classifying Method in Remote Sensing Image based on adaptive space information, comprising:
Obtain remote sensing images；
Preliminary classification is carried out to the remote sensing images using the FCM Algorithms based on Markov random field, is obtained
Initially fuzzy subordinated-degree matrix；
It is calculated described distant under current iteration number b according to the initially fuzzy subordinated-degree matrix using spatial attraction model
Feel the space attraction in image between current center pel and each neighborhood pixel；The current center pel is the remote sensing
I-th of pixel in image；Each current center pel corresponds to multiple neighborhood pixels；The position of the current center pel
It is adjacent with the position of the neighborhood pixel；
Edge detection is carried out to the remote sensing images using Sobel operator, obtains spatial structure characteristic；
According to the spatial structure characteristic, the edge system of the current center pel is calculated using gradient inverse exponential smoothing
Number；
According to the space attraction and the fringing coefficient, the Markov random field of adaptive weighting is constructed；
By the Markov random field of the adaptive weighting in conjunction with FCM Algorithms, the remote sensing images are determined
Classification results.
Optionally, the Markov random field by the adaptive weighting determines institute in conjunction with FCM Algorithms
The classification results for stating remote sensing images, specifically include:
According to the Markov random field of the adaptive weighting, calculates the current center pel and belong to kth marking class
Prior probability；K ∈ [1, C], C indicate classification number；
According to the prior probability, current cluster centre and current subordinated-degree matrix are calculated；
Judge whether the current cluster centre meets termination condition；The termination condition isWhereinIndicate current cluster centre,Indicate that the cluster centre obtained under last iteration number, ε are to terminate threshold value, ε > 0；
If it is not, then enable b=b+1, and i=i+1, and return described according to the initially fuzzy subordinated-degree matrix, utilizes sky
Between gravity model, calculate under current iteration number b in the remote sensing images between current center pel i and each neighborhood pixel
Space attraction；
If so, determining fuzzy membership matrix according to the preceding cluster centre and the current subordinated-degree matrix；
The classification results of the remote sensing images are determined according to the fuzzy membership matrix.
Optionally, described to calculate current cluster centre and current subordinated-degree matrix according to the prior probability, it is specific to wrap
It includes:
According to the prior probability, objective function is established
Wherein,
Wherein, N indicates pixel total number in the remote sensing images, vkIndicate the cluster centre of kth marking class；uikIndicate the
I pixel belongs to the subordinated-degree matrix of kth marking class, and m indicates constant, xiIndicate the gray value of i-th of pixel, Pk(i) it indicates
I-th of pixel belongs to the prior probability of kth marking class, vtIndicate the cluster centre of t class, t ∈ [1, c], c ∈ [1, C]；
Using lagrange's method of multipliers, the optimal solution of the objective function is calculated；The optimal solution is the objective function
The subordinated-degree matrix of the cluster centre of corresponding kth marking class and kth marking class when minimum；
The cluster centre of kth marking class in the optimal solution is determined as current cluster centre, in the optimal solution
The subordinated-degree matrix of kth marking class is determined as current subordinated-degree matrix.
Optionally, described to calculate current iteration time using spatial attraction model according to the initially fuzzy subordinated-degree matrix
Space attraction under number b in the remote sensing images between current center pel and each neighborhood pixel, specifically includes:
According to the initially fuzzy subordinated-degree matrix, the fuzzy membership of current center pel and the mould of neighborhood pixel are calculated
Paste degree of membership；
According to the fuzzy membership of the current center pel and the fuzzy membership of the current center pel, sky is utilized
Between gravity model calculate the sky under current iteration number b in the remote sensing images between current center pel and each neighborhood pixel
Between attraction
Wherein, j indicates the neighborhood pixel of i-th of pixel, j=0,1 ..., 7, uicIndicate that i-th of pixel belongs to c class
Subordinated-degree matrix, ujcIndicate that neighborhood pixel j belongs to the subordinated-degree matrix of c class, wijIndicate current center pel and neighborhood
The distance between pixel weight coefficient, c ∈ [1, C], C indicate classification number.
Optionally, described according to the spatial structure characteristic, the current middle imago is calculated using gradient inverse exponential smoothing
The fringing coefficient of member, specifically includes:
Calculate the gradient of the current center pel
Wherein,Indicate single order horizontal gradient of i-th of pixel on d-th of wave band,Indicate i-th of pixel
Single order vertical gradient on d-th of wave band；
Gradient according to the current center pel constructs the fringing coefficient of the current center pel
Optionally, the Markov random field of the adaptive weighting, specifically:
P (c)=exp (- α (xi) U (c))/Z,
Wherein, α (xi) indicate that the fringing coefficient of current center pel, Z indicate the normalization constant of cutting function, U (c) table
Show energy function,
Wherein, [1, C] c ∈, C indicate classification number, and f indicates that gesture agglomeration closes, and F indicates gesture agglomeration intersection, If(c) f is indicated
On potential function,
Wherein, c (xi) indicate i-th of pixel classification, c (xj) indicate neighborhood pixel j classification, j indicate i-th of pixel neighbour
Domain pixel, SijIndicate the space attraction between current center pel and neighborhood pixel, xiIndicate the gray value of i-th of pixel, xjTable
Show the gray value of neighborhood pixel j,Indicate the total number of the corresponding neighborhood pixel of i-th of pixel,
The present invention also provides a kind of Classifying System for Remote Sensing based on adaptive space information, comprising:
Image collection module, for obtaining remote sensing images；
Preliminary classification module, for using the FCM Algorithms based on Markov random field to the remote sensing images
Preliminary classification is carried out, obtains initially obscuring subordinated-degree matrix；
Space attraction computing module, for utilizing spatial attraction model, meter according to the initially fuzzy subordinated-degree matrix
Calculate the space attraction under current iteration number b in the remote sensing images between current center pel and each neighborhood pixel；Institute
Stating current center pel is i-th of pixel in the remote sensing images；Each current center pel corresponds to multiple neighborhood pictures
Member；The position of the current center pel is adjacent with the position of the neighborhood pixel；
Edge detection module obtains space structure for carrying out edge detection to the remote sensing images using Sobel operator
Feature；
Fringing coefficient computing module is used for according to the spatial structure characteristic, using described in the calculating of gradient inverse exponential smoothing
The fringing coefficient of current center pel；
Random field constructs module, for constructing adaptive weighting according to the space attraction and the fringing coefficient
Markov random field；
Categorization module, in conjunction with FCM Algorithms, determining the Markov random field of the adaptive weighting
The classification results of the remote sensing images.
Optionally, the categorization module, specifically includes:
First computing unit calculates the current center for the Markov random field according to the adaptive weighting
Pixel belongs to the prior probability of kth marking class；K ∈ [1, C], C indicate classification number；
Second computing unit, for calculating current cluster centre and current subordinated-degree matrix according to the prior probability；
Judging unit, for judging whether the current cluster centre meets termination condition；The termination condition isWhereinIndicate current cluster centre,Indicate that the cluster centre obtained under last iteration number, ε are
Terminate threshold value, ε > 0；If it is not, then enable b=b+1, and i=i+1, and return it is described according to the initially fuzzy subordinated-degree matrix,
Using spatial attraction model, current center pel i and each neighborhood pixel in the remote sensing images are calculated under current iteration number b
Between space attraction；If so, determining fuzzy membership according to the preceding cluster centre and the current subordinated-degree matrix
Matrix, and determine according to the fuzzy membership matrix classification results of the remote sensing images.
Optionally, the space attraction computing module, specifically includes:
Third computing unit, for calculating the fuzzy person in servitude of current center pel according to the initially fuzzy subordinated-degree matrix
The fuzzy membership of category degree and neighborhood pixel；
4th computing unit, for the fuzzy membership and the current center pel according to the current center pel
Fuzzy membership, using current center pel in the remote sensing images under spatial attraction model calculating current iteration number b and often
Space attraction between a neighborhood pixel
Wherein, j indicates the neighborhood pixel of i-th of pixel, j=0,1 ..., 7, uicIndicate that i-th of pixel belongs to c class
Subordinated-degree matrix, ujcIndicate that neighborhood pixel j belongs to the subordinated-degree matrix of c class, wijIndicate current center pel and neighborhood
The distance between pixel weight coefficient, c ∈ [1, C], C indicate classification number.
Optionally, institute's fringing coefficient computing module, specifically includes:
Gradient computing unit, for calculating the gradient of the current center pel
Wherein,Indicate single order horizontal gradient of i-th of pixel on d-th of wave band,Indicate i-th of pixel
Single order vertical gradient on d-th of wave band；
Fringing coefficient computing unit, for constructing the current center pel according to the gradient of the current center pel
Fringing coefficient
Compared with prior art, the beneficial effects of the present invention are:
The invention proposes a kind of Classifying Method in Remote Sensing Image and system based on adaptive space information.This method packet
It includes: obtaining remote sensing images；Preliminary classification is carried out to remote sensing images using the FCM Algorithms based on Markov random field,
It obtains initially obscuring subordinated-degree matrix；Using spatial attraction model, current center in remote sensing images is calculated under current iteration number b
Space attraction between pixel and each neighborhood pixel；Edge detection is carried out to remote sensing images using Sobel operator, obtains sky
Between structure feature；According to spatial structure characteristic, the fringing coefficient of current center pel is calculated using gradient inverse exponential smoothing；Foundation
Space attraction and fringing coefficient construct the Markov random field of adaptive weighting；By the markov of adaptive weighting with
Airport determines the classification results of remote sensing images in conjunction with FCM Algorithms.The present invention is by spatial attraction model, space structure
Feature is combined with traditional Markov random field, constructs the Markov random field of adaptive weighting, and is introduced into mould
It pastes in C means clustering algorithm, influence of noise can not only be overcome also to be able to maintain image edge detailss, and change traditional fixation
The mode of spatial information neighborhood weight, thus the problem of effective solution boundary pixel and spatial information weight coefficient are estimated,
Improve the nicety of grading of remote sensing images.
Detailed description of the invention
It in order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will be to institute in embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without any creative labor, can also be according to these attached drawings
Obtain other attached drawings.
Fig. 1 is the flow chart of Classifying Method in Remote Sensing Image of the embodiment of the present invention based on adaptive space information；
Fig. 2 is 3 × 3 neighborhood system window figures of center pel of the embodiment of the present invention；
Fig. 3 is the distance between center pel of the embodiment of the present invention and neighborhood pixel weight coefficient figure；
Fig. 4 is the original remote sensing images of the embodiment of the present invention；
Fig. 5 is the ground reference image of the embodiment of the present invention；
Fig. 6 is the classification results figure that the present embodiment uses FCM algorithm；
Fig. 7 is the classification results figure that the present embodiment uses MFCM algorithm；
Fig. 8 is the classification results figure that the present embodiment uses AMFCM algorithm；
Fig. 9 is the structural schematic diagram of Classifying System for Remote Sensing of the embodiment of the present invention based on adaptive space information.
Label 1 to 5 indicates five kinds of classifications of atural object in Fig. 4-8, and 1 indicates rose roof, and 2 indicate blue roof, and 3 indicate
Road, 4 indicate shade, and 5 indicate meadow.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing and specific real
Applying mode, the present invention is described in further detail.
Fig. 1 is the flow chart of Classifying Method in Remote Sensing Image of the embodiment of the present invention based on adaptive space information.
Referring to Fig. 1, the Classifying Method in Remote Sensing Image based on adaptive space information of embodiment, comprising:
Step S1: remote sensing images are obtained.
Step S2: FCM Algorithms (the Markov random fieldbased based on Markov random field is used
FCM, MFCM) preliminary classification is carried out to the remote sensing images, it obtains initially obscuring subordinated-degree matrix.
The initially fuzzy subordinated-degree matrix is UC×N, while also obtaining initial category label.
Step S3: current iteration number b is calculated using spatial attraction model according to the initially fuzzy subordinated-degree matrix
Under space attraction in the remote sensing images between current center pel and each neighborhood pixel.
The current center pel is i-th of pixel in the remote sensing images；Each current center pel is corresponding
Multiple neighborhood pixels；The position of the current center pel is adjacent with the position of the neighborhood pixel.Primary iteration is set and counts b
=1.Spatial attraction model is proved to be effectively to characterize the spatial coherence between image picture elements, and spatial attraction model is used for
Markov model, the model include space neighborhood information, further include grayscale information.For two pixel xiAnd xj, they are to
The attraction and their fuzzy membership u of c clustericAnd ujcIt is directly proportional, and with the distance between two pixels at anti-
Than.
The step S3, specifically includes:
1) according to the initially fuzzy subordinated-degree matrix, the fuzzy membership and neighborhood pixel of current center pel are calculated
Fuzzy membership.
2) fuzzy membership according to the fuzzy membership of the current center pel and the current center pel utilizes
Under spatial attraction model calculating current iteration number b in the remote sensing images between current center pel and each neighborhood pixel
Space attraction
Wherein, j indicates the neighborhood pixel of i-th of pixel, j ∈ Ni,j{ j=0,1,2 ..., 7 }, uicIndicate that i-th of pixel is returned
Belong to the subordinated-degree matrix of c class, ujcIndicate that neighborhood pixel j belongs to the subordinated-degree matrix of c class, wijIndicate current center
The distance between pixel and neighborhood pixel weight coefficient, wijFor the inverse of Euclidean distance, neighborhood pixel is closer from center pel,
wijInfluence to center pel is bigger, and c ∈ [1, C], C indicate classification number.
In the present embodiment, spatial attraction exists only in 3 × 3 given neighborhood windows, and other pixels outside window with
Center pel apart from it is too far and cannot be to giving any gravitation on center pel.Fig. 2 be center pel of the embodiment of the present invention 3 ×
3 neighborhood system window figures, Fig. 3 are the distance between center pel of the embodiment of the present invention and neighborhood pixel weight coefficient figure, referring to
Fig. 2 and Fig. 3, space attraction exist only between the center pel in given window and its neighborhood pixel, and outside window
Other pixels and center pel apart from it is too far and cannot be to giving any gravitation on center pel.The main contribution of spatial attraction model
It is: 1) considers the different distance between center pel and its neighborhood pixel；2) come more using the fuzzy membership of image picture elements
Different Effects of the good description neighborhood pixel to center pel.
Step S4: edge detection is carried out to the remote sensing images using Sobel operator, obtains spatial structure characteristic.Edge
It is characterized in one of spatial structure characteristic, and calculating gradient is most common method in edge detection, Sobel operator is examined as edge
Measuring and calculating, is based on first derivative.Because it has used local average operation, which has smoothing effect to noise,
It can be very good to eliminate the influence of noise bring.
Step S5: according to the spatial structure characteristic, the current center pel is calculated using gradient inverse exponential smoothing
Fringing coefficient.
The step S5, specifically includes:
1) gradient of the current center pel is calculated
Wherein,Indicate single order horizontal gradient of i-th of pixel on d-th of wave band,Indicate i-th of pixel
Single order vertical gradient on d-th of wave band.
2) gradient according to the current center pel constructs the fringing coefficient of the current center pel
Noise not only can be effectively reduced in the gradient inverse exponential smoothing that the present embodiment uses, can also retain certain edge and
Detailed information.From above-mentioned formula it can be found that if when pixel is located at smooth region, α (xi) value it is larger, to homogeneity pixel have compared with
Strong smoothing effect；Conversely, when pixel is located at edge or texture area, α (xi) value it is smaller, to the flat of the pixel in the region
Sliding effect is weaker, and lesser region can be retained in image.
Step S6: according to the space attraction and the fringing coefficient, the markov for constructing adaptive weighting is random
?.The Markov random field of the adaptive weighting, specifically:
P (c)=exp (- α (xi) U (c))/Z,
Wherein, α (xi) indicate that the fringing coefficient of current center pel, Z indicate the normalization constant of cutting function, U (c) table
Show energy function,
Wherein, [1, C] c ∈, C indicate classification number, and f indicates that gesture agglomeration closes, and F indicates gesture agglomeration intersection, If(c) f is indicated
On potential function, the I in the present embodimentf(c) what is utilized is the binary potential function Formulas I that Potts model indicates2(c(xi),c
(xj)), I2(c(xi),c(xj)) in only consider is binary potential function on f,
Wherein, c (xi) indicate i-th of pixel classification, c (xj) indicate neighborhood pixel j classification, j indicate i-th of pixel
Neighborhood pixel, SijThe space attraction between current center pel and neighborhood pixel is indicated, for controlling neighborhood pixel centering
The influence of imago member, xiIndicate the gray value of i-th of pixel, xjIndicate the gray value of neighborhood pixel j,Indicate i-th of pixel
The total number of corresponding neighborhood pixel,
In the present embodiment, spatial attraction model and spatial structure characteristic are introduced in Markov random field, is constructed adaptive
The Markov random field (Markov random field of adaptive space information) of weight is answered, adaptive change center pel is carried out
Spatial information weight between neighborhood pixel, while improving the clustering precision to Small object region.
Step S7: it by the Markov random field of the adaptive weighting in conjunction with FCM Algorithms, determines described distant
Feel the classification results of image.
The step S7, specifically includes:
1) Markov random field according to the adaptive weighting calculates the current center pel and belongs to kth label
The prior probability of class；K ∈ [1, C], C indicate classification number.
2) according to the prior probability, current cluster centre and current subordinated-degree matrix is calculated, is specifically included:
21) according to the prior probability, objective function is established
Wherein, objective function JAMFCMObtain conditional extremum necessary condition be
Wherein, N indicates pixel total number in the remote sensing images, vkIndicate the cluster centre of kth marking class；uikIndicate the
I pixel belongs to the subordinated-degree matrix of kth marking class, and m indicates constant, xiIndicate the gray value of i-th of pixel, Pk(i) it indicates
I-th of pixel belongs to the prior probability of kth marking class, vtIndicate the cluster centre of t class, t ∈ [1, c], c ∈ [1, C].
22) lagrange's method of multipliers is used, the optimal solution of the objective function is calculated；The optimal solution is the target letter
The subordinated-degree matrix of number cluster centre of corresponding kth marking class and kth marking class when minimum.
23) cluster centre of the kth marking class in the optimal solution is determined as current cluster centre, in the optimal solution
The subordinated-degree matrix of kth marking class be determined as current subordinated-degree matrix.
3) judge whether the current cluster centre meets termination condition；The termination condition isIts
InIndicate current cluster centre,Indicate that the cluster centre obtained under last iteration number, ε are to terminate threshold value, ε > 0.
If it is not, then enabling b=b+1, and i=i+1, and return to the step S3；If so, according to the preceding cluster centre and
The current subordinated-degree matrix determines fuzzy membership matrix, and determines the remote sensing figure according to the fuzzy membership matrix
The classification results of picture.
Below to the present invention is based on the Classifying Method in Remote Sensing Image of adaptive space information to be verified.
The present invention combines spatial attraction model, spatial structure characteristic with traditional Markov random field, and construction is certainly
The Markov random field of weight is adapted to, and is introduced into Fuzzy C-Means Cluster Algorithm, has obtained believing based on adaptive space
Cease FCM clustering algorithm (the Adaptive Spatial Information MRF-Based FCM of MRF
ClusteringAlgorithm, AMFCM), the influence and preferably reservation of updates neighborhood pixel that can be adaptive to center pel
Image edge detailss information, to improve the nicety of grading of remote sensing images.
In order to verify the validity of proposed algorithm AMFCM, it is compared with FCM, MFCM clustering algorithm, selects high score
Resolution remote sensing image is tested, and carries out precision evaluation to classification results using confusion matrix, is carried out using four indexs
Quantitative assessment, such as overall classification accuracy (OverallAccuracy, OA), Kappa coefficient, producer's precision (Producer '
SAccuracy, PA) and user's precision (User ' s Accuracy, UA), wherein OA is the pixel sum and total picture correctly classified
The ratio of first number.The overall consistency of Kappa coefficient description, is also capable of the consistency of interpretive classification, can be described as
N is reference data pixel sum, xkkIt is the pixel number that all classes in image are correctly classified, xk+It is all classes
Middle reference data pixel number, x+kIt is the pixel number that each class is classified, PA is the picture that the pixel of entire image is correctly divided into A class
First number and A class refer to the ratio of pixel sum.UA is the pixel sum for being correctly divided into A class and classifier by the pixel of entire image
It is divided into the ratio of the pixel sum of A class.
No. two remote sensing images of high score are used to verify the algorithm of proposition.The image capturing time is 2014 9
Month, the multispectral image that the panchromatic image and spatial resolution for being 0.8 meter by spatial resolution are 3.2 meters merge three
The multi-spectrum remote sensing image of wave band, the image are located at Shanghai City.Fig. 4 is the original remote sensing images of the embodiment of the present invention；Fig. 5 is this
The ground reference image of inventive embodiments, it includes 5 classifications: rose that referring to fig. 4 and Fig. 5, research area's size, which are 221 × 192 pixels,
Red roof 1, blue roof 2, road 3, shade 4 and meadow 5.
Fig. 6 is the classification results figure that the present embodiment uses FCM algorithm, and Fig. 7 is the classification that the present embodiment uses MFCM algorithm
Result figure, Fig. 8 are the classification results figure that the present embodiment uses AMFCM algorithm.Show AMFCM most referring to Fig. 6-Fig. 8, Fig. 8
Good classification results.Visually see, due to applying spatial information, the cluster result of MFCM and AMFCM are different from FCM, although
Meadow is attributed to shade class for 3 all results to a certain extent, but the classification results of AMFCM are more preferable.In this reality
In testing, due to having apparent black exposed soil on the bad meadow of growing way, for the light of this part meadow its spectral information and shade
Spectrum information be it is extremely similar, a good classification results cannot be obtained only by the spectral signature of remote sensing images.In Fig. 6
It can be found that containing many shade classes in grassland classification result, this is not consistent with practical situation, and in Fig. 7 and Fig. 8
In, such case is improved, but still the meadow of some is assigned to inside shade class, the reason is that MRF although it is contemplated that
The spatial coherence of pixel is arrived, but the atural object classification closely similar for resolved spectroscopy information, resolution capability are limited
System.Comparison diagram 7 and Fig. 8 can be seen that Fig. 8 and be locally better than Fig. 7, and the classification edge in Fig. 7 is less obvious, and for yin
Small object territorial classification effect as shadow is not very well, the reason is that because MRF is isotropic, so that classification results are lost
The edge detail information of image, and in fig. 8, such case is changed, the reason is that AMFCM is using adaptive space
The Markov random field of information weight, change spatial information weight that not only can be adaptive, but also also introduce edge system
The such spatial structure characteristic of number.
For the above-mentioned algorithm of quantitative comparison, OA the and kappa coefficient of image is listed in table 1, as follows:
Table 1
As shown in Table 1, nicety of grading the ratio FCM high, AMFCM that MFCM and AMFCM is generated obtain optimal general classification
Precision is 79.35%, has increased separately 7.12% and 2.57% than FCM, MFCM.
The Classifying Method in Remote Sensing Image based on adaptive space information in the present embodiment, using being based on, markov is random
The FCM algorithm of field carries out preliminary classification, obtained initial subordinated-degree matrix and initial category label to remote sensing images；Space is drawn
Power model and spatial structure characteristic are introduced into Markov random field, construct the Markov random field of adaptive weighting；
The Markov random field of adaptive weighting is obtained into the FCM algorithm of adaptive space information MRF in conjunction with tradition FCM algorithm；
On the basis of preliminary classification result, the fining point of remote sensing images is carried out by the FCM algorithm of adaptive space information MRF
Class.The problem of this method effective solution boundary pixel and spatial information weight coefficient are estimated improves point of remote sensing images
Class precision.
The present invention also provides a kind of Classifying System for Remote Sensing based on adaptive space information, Fig. 9 is that the present invention is real
Apply the structural schematic diagram of Classifying System for Remote Sensing of the example based on adaptive space information.
The Classifying System for Remote Sensing based on adaptive space information of embodiment, comprising:
Image collection module 901, for obtaining remote sensing images.
Preliminary classification module 902, for using the FCM Algorithms based on Markov random field to the remote sensing figure
As carrying out preliminary classification, obtain initially obscuring subordinated-degree matrix.
Space attraction computing module 903, for utilizing spatial attraction mould according to the initially fuzzy subordinated-degree matrix
Type calculates the space under current iteration number b in the remote sensing images between current center pel and each neighborhood pixel and attracts
Power；The current center pel is i-th of pixel in the remote sensing images；Each current center pel corresponds to multiple neighbours
Domain pixel；The position of the current center pel is adjacent with the position of the neighborhood pixel.
Edge detection module 904 obtains space for carrying out edge detection to the remote sensing images using Sobel operator
Structure feature.
Fringing coefficient computing module 905, for calculating institute using gradient inverse exponential smoothing according to the spatial structure characteristic
State the fringing coefficient of current center pel.
Random field constructs module 906, for constructing adaptive weighting according to the space attraction and the fringing coefficient
Markov random field.The Markov random field of the adaptive weighting, specifically:
P (c)=exp (- α (xi) U (c))/Z,
Wherein, α (xi) indicate that the fringing coefficient of current center pel, Z indicate the normalization constant of cutting function, U (c) table
Show energy function,
Wherein, [1, C] c ∈, C indicate classification number, and f indicates that gesture agglomeration closes, and F indicates gesture agglomeration intersection, If(c) f is indicated
On potential function,
Wherein, c (xi) indicate i-th of pixel classification, c (xj) indicate neighborhood pixel j classification, j indicate i-th of pixel
Neighborhood pixel, SijIndicate the space attraction between current center pel and neighborhood pixel, xiIndicate the gray scale of i-th of pixel
Value, xjIndicate the gray value of neighborhood pixel j,Indicate the total number of the corresponding neighborhood pixel of i-th of pixel,
Categorization module 907, for by the Markov random field of the adaptive weighting in conjunction with FCM Algorithms,
Determine the classification results of the remote sensing images.
As an alternative embodiment, the categorization module 907, specifically includes:
First computing unit calculates the current center for the Markov random field according to the adaptive weighting
Pixel belongs to the prior probability of kth marking class；K ∈ [1, C], C indicate classification number；
Second computing unit, for calculating current cluster centre and current subordinated-degree matrix according to the prior probability；
Judging unit, for judging whether the current cluster centre meets termination condition；The termination condition isWhereinIndicate current cluster centre,Indicate that the cluster centre obtained under last iteration number, ε are
Terminate threshold value, ε > 0；If it is not, then enable b=b+1, and i=i+1, and return it is described according to the initially fuzzy subordinated-degree matrix,
Using spatial attraction model, current center pel i and each neighborhood pixel in the remote sensing images are calculated under current iteration number b
Between space attraction；If so, determining fuzzy membership according to the preceding cluster centre and the current subordinated-degree matrix
Matrix, and determine according to the fuzzy membership matrix classification results of the remote sensing images.
As an alternative embodiment, the space attraction computing module 903, specifically includes:
Third computing unit, for calculating the fuzzy person in servitude of current center pel according to the initially fuzzy subordinated-degree matrix
The fuzzy membership of category degree and neighborhood pixel；
4th computing unit, for the fuzzy membership and the current center pel according to the current center pel
Fuzzy membership, using current center pel in the remote sensing images under spatial attraction model calculating current iteration number b and often
Space attraction between a neighborhood pixel
Wherein, j indicates the neighborhood pixel of i-th of pixel, j=0,1 ..., 7, uicIndicate that i-th of pixel belongs to c class
Subordinated-degree matrix, ujcIndicate that neighborhood pixel j belongs to the subordinated-degree matrix of c class, wijIndicate current center pel and neighborhood
The distance between pixel weight coefficient, c ∈ [1, C], C indicate classification number.
As an alternative embodiment, institute's fringing coefficient computing module 905, specifically includes:
Gradient computing unit, for calculating the gradient of the current center pel
Wherein,Indicate single order horizontal gradient of i-th of pixel on d-th of wave band, ▽vxiD indicates i-th of pixel
Single order vertical gradient on d-th of wave band；
Fringing coefficient computing unit, for constructing the current center pel according to the gradient of the current center pel
Fringing coefficient
As an alternative embodiment, second computing unit, specifically includes:
Objective function establishes subelement, for establishing objective function according to the prior probability
Wherein,
Wherein, N indicates pixel total number in the remote sensing images, vkIndicate the cluster centre of kth marking class；uikIndicate the
I pixel belongs to the subordinated-degree matrix of kth marking class, and m indicates constant, xiIndicate the gray value of i-th of pixel, Pk(i) it indicates
I-th of pixel belongs to the prior probability of kth marking class, vtIndicate the cluster centre of t class, t ∈ [1, c], c ∈ [1, C]；
Subelement is solved, for using lagrange's method of multipliers, calculates the optimal solution of the objective function；The optimal solution
The subordinated-degree matrix of the cluster centre of corresponding kth marking class and kth marking class when for the objective function minimum；
Determine subelement, for the cluster centre of the kth marking class in the optimal solution to be determined as current cluster centre,
The subordinated-degree matrix of kth marking class in the optimal solution is determined as current subordinated-degree matrix.
The Classifying System for Remote Sensing based on adaptive space information of the present embodiment, effective solution boundary pixel and
The problem of spatial information weight coefficient is estimated, improves the nicety of grading of remote sensing images.
For the system disclosed in the embodiment, since it is corresponded to the methods disclosed in the examples, so the ratio of description
Relatively simple, reference may be made to the description of the method.
Used herein a specific example illustrates the principle and implementation of the invention, and above embodiments are said
It is bright to be merely used to help understand method and its core concept of the invention；At the same time, for those skilled in the art, foundation
Thought of the invention, there will be changes in the specific implementation manner and application range.In conclusion the content of the present specification is not
It is interpreted as limitation of the present invention.
Claims (10)
1. the Classifying Method in Remote Sensing Image based on adaptive space information characterized by comprising
Obtain remote sensing images；
Preliminary classification is carried out to the remote sensing images using the FCM Algorithms based on Markov random field, is obtained initial
Fuzzy membership matrix；
The remote sensing figure under current iteration number b is calculated using spatial attraction model according to the initially fuzzy subordinated-degree matrix
Space attraction as between current center pel and each neighborhood pixel；The current center pel is the remote sensing images
In i-th of pixel；Each current center pel corresponds to multiple neighborhood pixels；The position of the current center pel and institute
The position for stating neighborhood pixel is adjacent；
Edge detection is carried out to the remote sensing images using Sobel operator, obtains spatial structure characteristic；
According to the spatial structure characteristic, the fringing coefficient of the current center pel is calculated using gradient inverse exponential smoothing；
According to the space attraction and the fringing coefficient, the Markov random field of adaptive weighting is constructed；
By the Markov random field of the adaptive weighting in conjunction with FCM Algorithms, point of the remote sensing images is determined
Class result.
2. the Classifying Method in Remote Sensing Image according to claim 1 based on adaptive space information, which is characterized in that described
By the Markov random field of the adaptive weighting in conjunction with FCM Algorithms, the classification knot of the remote sensing images is determined
Fruit specifically includes:
According to the Markov random field of the adaptive weighting, the elder generation that the current center pel belongs to kth marking class is calculated
Test probability；K ∈ [1, C], C indicate classification number；
According to the prior probability, current cluster centre and current subordinated-degree matrix are calculated；
Judge whether the current cluster centre meets termination condition；The termination condition isWhereinTable
Show current cluster centre,Indicate that the cluster centre obtained under last iteration number, ε are to terminate threshold value, ε > 0；
If it is not, then enable b=b+1, and i=i+1, and return described according to the initially fuzzy subordinated-degree matrix, is drawn using space
Power model calculates the space under current iteration number b in the remote sensing images between current center pel i and each neighborhood pixel
Attraction；
If so, determining fuzzy membership matrix according to the preceding cluster centre and the current subordinated-degree matrix；
The classification results of the remote sensing images are determined according to the fuzzy membership matrix.
3. the Classifying Method in Remote Sensing Image according to claim 2 based on adaptive space information, which is characterized in that described
According to the prior probabilities, current cluster centre and current subordinated-degree matrix are calculated, is specifically included:
According to the prior probability, objective function is established
Wherein,
Wherein, N indicates pixel total number in the remote sensing images, vkIndicate the cluster centre of kth marking class；uikIt indicates i-th
Pixel belongs to the subordinated-degree matrix of kth marking class, and m indicates constant, xiIndicate the gray value of i-th of pixel, Pk(i) i-th is indicated
A pixel belongs to the prior probability of kth marking class, vtIndicate the cluster centre of t class, t ∈ [1, c], c ∈ [1, C]；
Using lagrange's method of multipliers, the optimal solution of the objective function is calculated；The optimal solution is that the objective function is minimum
When the cluster centre of the corresponding kth marking class and subordinated-degree matrix of kth marking class；
The cluster centre of kth marking class in the optimal solution is determined as current cluster centre, the kth mark in the optimal solution
The subordinated-degree matrix of note class is determined as current subordinated-degree matrix.
4. the Classifying Method in Remote Sensing Image according to claim 1 based on adaptive space information, which is characterized in that described
It is calculated under current iteration number b in the remote sensing images according to the initially fuzzy subordinated-degree matrix using spatial attraction model
Space attraction between current center pel and each neighborhood pixel, specifically includes:
According to the initially fuzzy subordinated-degree matrix, the fuzzy membership of current center pel and the fuzzy person in servitude of neighborhood pixel are calculated
Category degree；
According to the fuzzy membership of the current center pel and the fuzzy membership of the current center pel, drawn using space
Power model calculates the space under current iteration number b in the remote sensing images between current center pel and each neighborhood pixel and inhales
Gravitation
Wherein, j indicates the neighborhood pixel of i-th of pixel, j=0,1 ..., 7, uicIndicate that i-th of pixel belongs to the person in servitude of c class
Category degree matrix, ujcIndicate that neighborhood pixel j belongs to the subordinated-degree matrix of c class, wijIndicate current center pel and neighborhood pixel
The distance between weight coefficient, c ∈ [1, C], C indicate classification number.
5. the Classifying Method in Remote Sensing Image according to claim 1 based on adaptive space information, which is characterized in that described
According to the spatial structure characteristic, the fringing coefficient of the current center pel is calculated using gradient inverse exponential smoothing, it is specific to wrap
It includes:
Calculate the gradient of the current center pel
Wherein,Indicate single order horizontal gradient of i-th of pixel on d-th of wave band,Indicate i-th of pixel in d
Single order vertical gradient on a wave band；
Gradient according to the current center pel constructs the fringing coefficient of the current center pel
6. the Classifying Method in Remote Sensing Image according to claim 1 based on adaptive space information, which is characterized in that described
The Markov random field of adaptive weighting, specifically:
P (c)=exp (- α (xi) U (c))/Z,
Wherein, α (xi) indicate that the fringing coefficient of current center pel, Z indicate that the normalization constant of cutting function, U (c) indicate energy
Flow function,
Wherein, [1, C] c ∈, C indicate classification number, and f indicates that gesture agglomeration closes, and F indicates gesture agglomeration intersection, If(c) gesture on f is indicated
Function,
Wherein, c (xi) indicate i-th of pixel classification, c (xj) indicate neighborhood pixel j classification, j indicate i-th of pixel neighborhood
Pixel, SijIndicate the space attraction between current center pel and neighborhood pixel, xiIndicate the gray value of i-th of pixel, xjTable
Show the gray value of neighborhood pixel j,Indicate the total number of the corresponding neighborhood pixel of i-th of pixel,
7. the Classifying System for Remote Sensing based on adaptive space information characterized by comprising
Image collection module, for obtaining remote sensing images；
Preliminary classification module, for being carried out using the FCM Algorithms based on Markov random field to the remote sensing images
Preliminary classification obtains initially obscuring subordinated-degree matrix；
Space attraction computing module, for according to the initially fuzzy subordinated-degree matrix, using spatial attraction model, calculating to be worked as
Current space attraction between center pel and each neighborhood pixel in the remote sensing images under preceding the number of iterations b；It is described to work as
Preceding center pel is i-th of pixel in the remote sensing images；Each current center pel corresponds to multiple neighborhood pixels；Institute
The position for stating current center pel is adjacent with the position of the neighborhood pixel；
Edge detection module obtains space structure spy for carrying out edge detection to the remote sensing images using Sobel operator
Sign；
Fringing coefficient computing module, for being calculated using gradient inverse exponential smoothing described current according to the spatial structure characteristic
The fringing coefficient of center pel；
Random field constructs module, for constructing the Ma Er of adaptive weighting according to the space attraction and the fringing coefficient
It can husband's random field；
Categorization module, described in conjunction with FCM Algorithms, determining the Markov random field of the adaptive weighting
The classification results of remote sensing images.
8. the Classifying System for Remote Sensing according to claim 7 based on adaptive space information, which is characterized in that described
Categorization module specifically includes:
First computing unit calculates the current center pel for the Markov random field according to the adaptive weighting
Belong to the prior probability of kth marking class；K ∈ [1, C], C indicate classification number；
Second computing unit, for calculating current cluster centre and current subordinated-degree matrix according to the prior probability；
Judging unit, for judging whether the current cluster centre meets termination condition；The termination condition is
WhereinIndicate current cluster centre,Indicate that the cluster centre obtained under last iteration number, ε are to terminate threshold value, ε > 0；
If it is not, then enable b=b+1, and i=i+1, and return described according to the initially fuzzy subordinated-degree matrix, utilizes spatial attraction mould
Type calculates the space under current iteration number b in the remote sensing images between current center pel i and each neighborhood pixel and attracts
Power；If so, determining fuzzy membership matrix, and according to institute according to the preceding cluster centre and the current subordinated-degree matrix
State the classification results that fuzzy membership matrix determines the remote sensing images.
9. the Classifying System for Remote Sensing according to claim 7 based on adaptive space information, which is characterized in that described
Space attraction computing module, specifically includes:
Third computing unit, for calculating the fuzzy membership of current center pel according to the initially fuzzy subordinated-degree matrix
With the fuzzy membership of neighborhood pixel；
4th computing unit, for according to the current center pel fuzzy membership and the current center pel it is fuzzy
Degree of membership utilizes current center pel and each neighbour in the remote sensing images under spatial attraction model calculating current iteration number b
Space attraction between the pixel of domain
Wherein, j indicates the neighborhood pixel of i-th of pixel, j=0,1 ..., 7, uicIndicate that i-th of pixel belongs to the person in servitude of c class
Category degree matrix, ujcIndicate that neighborhood pixel j belongs to the subordinated-degree matrix of c class, wijIndicate current center pel and neighborhood pixel
The distance between weight coefficient, c ∈ [1, C], C indicate classification number.
10. the Classifying System for Remote Sensing according to claim 7 based on adaptive space information, which is characterized in that institute
Fringing coefficient computing module, specifically includes:
Gradient computing unit, for calculating the gradient of the current center pel
Wherein,Indicate single order horizontal gradient of i-th of pixel on d-th of wave band,Indicate i-th of pixel in d
Single order vertical gradient on a wave band；
Fringing coefficient computing unit, for constructing the edge of the current center pel according to the gradient of the current center pel
Coefficient
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910220213.8A CN109829519B (en) | 2019-03-22 | 2019-03-22 | Remote sensing image classification method and system based on self-adaptive spatial information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910220213.8A CN109829519B (en) | 2019-03-22 | 2019-03-22 | Remote sensing image classification method and system based on self-adaptive spatial information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109829519A true CN109829519A (en) | 2019-05-31 |
| CN109829519B CN109829519B (en) | 2020-04-03 |
Family
ID=66870941
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910220213.8A Active CN109829519B (en) | 2019-03-22 | 2019-03-22 | Remote sensing image classification method and system based on self-adaptive spatial information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109829519B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111126467A (en) * | 2019-12-16 | 2020-05-08 | 武汉大学 | Remote sensing image space spectrum clustering method based on multi-target sine and cosine algorithm |
| CN112733888A (en) * | 2020-12-25 | 2021-04-30 | 河南大学 | Deep space exploration image morphology analysis method based on multilayer MRF clustering |
| CN113446992A (en) * | 2021-06-28 | 2021-09-28 | 中国水利水电科学研究院 | Method for optimizing distribution of measuring points in topographic survey |
| CN115578388A (en) * | 2022-12-07 | 2023-01-06 | 滨州华然化纤绳网有限公司 | Data processing method for textile production |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110243417A1 (en) * | 2008-09-03 | 2011-10-06 | Rutgers, The State University Of New Jersey | System and method for accurate and rapid identification of diseased regions on biological images with applications to disease diagnosis and prognosis |
| CN102903102A (en) * | 2012-09-11 | 2013-01-30 | 西安电子科技大学 | Non-local-based triple Markov random field synthetic aperture radar (SAR) image segmentation method |
| CN104123561A (en) * | 2014-07-10 | 2014-10-29 | 中国矿业大学 | Spatial gravity model based fuzzy c-means remote sensing image automatic classification method |
| CN106372612A (en) * | 2016-09-09 | 2017-02-01 | 河海大学 | Multi-temporal remote sensing image change detection method combining FCM with MRF model |
| CN109409438A (en) * | 2018-11-07 | 2019-03-01 | 重庆市勘测院 | The Remote Image Classification inferred based on IFCM cluster with variation |
-
2019
- 2019-03-22 CN CN201910220213.8A patent/CN109829519B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110243417A1 (en) * | 2008-09-03 | 2011-10-06 | Rutgers, The State University Of New Jersey | System and method for accurate and rapid identification of diseased regions on biological images with applications to disease diagnosis and prognosis |
| CN102903102A (en) * | 2012-09-11 | 2013-01-30 | 西安电子科技大学 | Non-local-based triple Markov random field synthetic aperture radar (SAR) image segmentation method |
| CN104123561A (en) * | 2014-07-10 | 2014-10-29 | 中国矿业大学 | Spatial gravity model based fuzzy c-means remote sensing image automatic classification method |
| CN106372612A (en) * | 2016-09-09 | 2017-02-01 | 河海大学 | Multi-temporal remote sensing image change detection method combining FCM with MRF model |
| CN109409438A (en) * | 2018-11-07 | 2019-03-01 | 重庆市勘测院 | The Remote Image Classification inferred based on IFCM cluster with variation |
Non-Patent Citations (6)
| Title |
|---|
| MING HAO等: ""Unsupervised change detection using fuzzy cmeans and MRF from remotely sensed images"", 《REMOTE SENSING LETTERS》 * |
| WEI GU等: ""Change detection method for remote sensing images based on an improved Markov random field"", 《MULTIMEDIA TOOLS AND APPLICATIONS》 * |
| ZHANG HUA等: ""Spatial-attraction-based Markov random field approach for classification of high spatial resolution multispectral imagery"", 《IEEE GEOSCIENCES AND REMOTE SENSING LETTERS》 * |
| 张华等: "《遥感数据可靠性分类方法研究》", 31 March 2016, 测绘出版社 * |
| 王冬雪: ""基于高光谱图像的目标分类技术研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 王冬雪等: ""基于改进马尔科夫分类模型的高光谱图像分类方法"", 《信息通信》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111126467A (en) * | 2019-12-16 | 2020-05-08 | 武汉大学 | Remote sensing image space spectrum clustering method based on multi-target sine and cosine algorithm |
| CN111126467B (en) * | 2019-12-16 | 2022-02-15 | 武汉大学 | Remote sensing image space spectrum clustering method based on multi-target sine and cosine algorithm |
| CN112733888A (en) * | 2020-12-25 | 2021-04-30 | 河南大学 | Deep space exploration image morphology analysis method based on multilayer MRF clustering |
| CN112733888B (en) * | 2020-12-25 | 2022-11-18 | 河南大学 | Deep space exploration image morphology analysis method based on multilayer MRF clustering |
| CN113446992A (en) * | 2021-06-28 | 2021-09-28 | 中国水利水电科学研究院 | Method for optimizing distribution of measuring points in topographic survey |
| CN115578388A (en) * | 2022-12-07 | 2023-01-06 | 滨州华然化纤绳网有限公司 | Data processing method for textile production |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109829519B (en) | 2020-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110443143B (en) | Multi-branch convolutional neural network fused remote sensing image scene classification method | |
| CN108549891B (en) | Multi-scale diffusion well-marked target detection method based on background Yu target priori | |
| CN105488809B (en) | Indoor scene semantic segmentation method based on RGBD descriptors | |
| CN109829519A (en) | Classifying Method in Remote Sensing Image and system based on adaptive space information | |
| Zhang et al. | Adaptive Markov random field approach for classification of hyperspectral imagery | |
| CN106909902B (en) | Remote sensing target detection method based on improved hierarchical significant model | |
| US11854244B2 (en) | Labeling techniques for a modified panoptic labeling neural network | |
| CN109359684A (en) | Fine granularity model recognizing method based on Weakly supervised positioning and subclass similarity measurement | |
| CN103049763B (en) | Context-constraint-based target identification method | |
| CN104123561B (en) | Fuzzy C-mean algorithm remote sensing image automatic classification method based on spatial attraction model | |
| CN102982338B (en) | Classification of Polarimetric SAR Image method based on spectral clustering | |
| CN106611422B (en) | Stochastic gradient Bayes's SAR image segmentation method based on sketch structure | |
| CN109446894B (en) | Multispectral image change detection method based on probability segmentation and Gaussian mixture clustering | |
| CN108960404B (en) | Image-based crowd counting method and device | |
| CN111553397A (en) | Cross-domain target detection method based on regional full convolution network and self-adaption | |
| CN106611423B (en) | SAR image segmentation method based on ridge ripple filter and deconvolution structural model | |
| CN107330875A (en) | Based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images | |
| CN106023257A (en) | Target tracking method based on rotor UAV platform | |
| CN107403434A (en) | SAR image semantic segmentation method based on two-phase analyzing method | |
| CN109948593A (en) | Based on the MCNN people counting method for combining global density feature | |
| Yi et al. | An object-oriented semantic clustering algorithm for high-resolution remote sensing images using the aspect model | |
| CN106651884B (en) | Mean field variation Bayes's SAR image segmentation method based on sketch structure | |
| CN110189292A (en) | A kind of cancer cell detection method based on Faster R-CNN and density estimation | |
| Hao et al. | An advanced superpixel-based Markov random field model for unsupervised change detection | |
| CN112733614B (en) | Pest image detection method with similar size enhanced identification |
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 |