CN109829519B - Remote sensing image classification method and system based on self-adaptive spatial information - Google Patents
Remote sensing image classification method and system based on self-adaptive spatial information Download PDFInfo
- Publication number
- CN109829519B CN109829519B CN201910220213.8A CN201910220213A CN109829519B CN 109829519 B CN109829519 B CN 109829519B CN 201910220213 A CN201910220213 A CN 201910220213A CN 109829519 B CN109829519 B CN 109829519B
- Authority
- CN
- China
- Prior art keywords
- pixel
- current
- remote sensing
- representing
- neighborhood
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 239000011159 matrix material Substances 0.000 claims abstract description 83
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 44
- 238000003708 edge detection Methods 0.000 claims abstract description 13
- 238000009499 grossing Methods 0.000 claims abstract description 13
- 230000003044 adaptive effect Effects 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 18
- 238000010276 construction Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 11
- 230000000694 effects Effects 0.000 description 5
- 230000003595 spectral effect Effects 0.000 description 5
- 244000025254 Cannabis sativa Species 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 230000005484 gravity Effects 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000000747 cardiac effect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 241000220317 Rosa Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011158 quantitative evaluation Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 230000004304 visual acuity Effects 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a remote sensing image classification method and system based on self-adaptive spatial information. The method comprises the following steps: acquiring a remote sensing image; carrying out initial classification on the remote sensing images by adopting a fuzzy C-means algorithm based on a Markov random field to obtain an initial fuzzy membership matrix; calculating the space attraction between the current center pixel and each neighborhood pixel in the remote sensing image under the current iteration number b by using a space attraction model; carrying out edge detection on the remote sensing image by adopting a Sobel operator to obtain spatial structure characteristics; calculating the edge coefficient of the current central pixel by adopting a gradient reciprocal smoothing method according to the spatial structure characteristics; constructing a Markov random field of self-adaptive weight according to the space attraction and the edge coefficient; and combining the Markov random field of the self-adaptive weight with a fuzzy C-means algorithm to determine a classification result of the remote sensing image. The invention can effectively solve the problem of the weight coefficient estimation of the boundary pixels and the spatial information and improve the classification precision.
Description
Technical Field
The invention relates to the technical field of remote sensing image processing, in particular to a remote sensing image classification method and system based on self-adaptive spatial information.
Background
With the rapid development of remote sensing technology, remote sensing images are widely applied to the fields of homeland resources, environmental monitoring, urban planning and the like. In order to meet application requirements of different fields, a large amount of useful information needs to be extracted from the remote sensing images, wherein the classification of the remote sensing images is a key technology for information extraction and is also a basis for subsequent remote sensing image ground feature identification. Because the high-spatial-resolution remote sensing image contains complex ground feature information, the spectrum heterogeneity of the same ground feature is obvious, and the spectra of different ground features are mutually overlapped, so that the traditional classification method based on the spectral characteristics cannot obtain satisfactory classification accuracy. In addition, the remote sensing image classification method based on single characteristics, especially only based on spectral characteristics, is susceptible to noise, and the obtained classification result is often fragmented and the loss of feature edge information is large. Therefore, in order to improve the classification accuracy, a classification method of multi-feature fusion combining features such as spectrum, spatial information and shape has become a hot spot for classifying high-spatial-resolution remote sensing images.
The classification method of multi-feature fusion is classified into a supervised classification method and an unsupervised classification method. Theoretically, the supervised classification method can obtain a better classification result than the unsupervised classification method by using the training samples of the classification classes, but because a plurality of mixed pixels exist in the remote sensing image and the influence of external conditions on the remote sensing image, a large number of reliable remote sensing image training samples are difficult to obtain for supervised classification, so that the classification result obtained by the supervised classification method is difficult to satisfy. The unsupervised classification method does not need training samples and is divided according to the similarity of images, so that the unsupervised classification method is widely used for remote sensing image classification, and clustering algorithms such as K-means, Fuzzy C-means (FCM), Markov Random Field (MRF) and the like are used for unsupervised classification. The classification result of the FCM algorithm conforms to the ambiguity and uncertainty of the natural ground objects, and more image information can be reserved, so that the FCM clustering algorithm is widely applied to remote sensing image classification. However, the traditional FCM clustering algorithm does not consider the spatial correlation of images during image clustering, resulting in poor anti-noise performance, so that isolated pixels exist in the image classification result.
In recent years, it has become common knowledge of researchers to introduce spatial information into a conventional FCM clustering algorithm to improve clustering accuracy. The introduction of MRF into the FCM method is also a classical method to improve FCM, such as Markov random field based FCM (MFCM) algorithm. MRF in the MFCM algorithm is a probability model, and extracts spatial neighborhood information through the dependency relationship between adjacent pixels, so that the MRF has strong anti-noise performance and fully considers the spatial neighborhood relationship between image pixels, but the MRF model has the defect that the clustering precision is influenced by the spatial information weight, the spatial information weight coefficient is usually obtained through repeated experiments, and the influence of the neighborhood pixels on the central pixel in the traditional MRF is the same, so that the MRF is suitable for uniform region classification, and the MRF is poor in edge region classification effect and easy to lose edge details. Therefore, the classification accuracy of the existing MFCM classification algorithm needs to be improved, and a more effective method is needed to solve the problem of boundary pixel and spatial information weight coefficient estimation.
Disclosure of Invention
Therefore, a remote sensing image classification method and system based on adaptive spatial information are needed to effectively solve the problem of estimation of boundary pixels and spatial information weight coefficients and improve classification accuracy.
In order to achieve the purpose, the invention provides the following scheme:
the remote sensing image classification method based on the self-adaptive spatial information comprises the following steps:
acquiring a remote sensing image;
carrying out initial classification on the remote sensing images by adopting a fuzzy C-means algorithm based on a Markov random field to obtain an initial fuzzy membership matrix;
calculating the space attraction between the current central pixel and each neighborhood pixel in the remote sensing image under the current iteration number b by utilizing a space attraction model according to the initial fuzzy membership matrix; the current central pixel is the ith pixel in the remote sensing image; each current center pixel corresponds to a plurality of neighborhood pixels; the position of the current center pixel is adjacent to the position of the neighborhood pixel;
performing edge detection on the remote sensing image by using a Sobel operator to obtain spatial structure characteristics;
calculating the edge coefficient of the current central pixel by adopting a gradient reciprocal smoothing method according to the spatial structure characteristics;
constructing a Markov random field of self-adaptive weight according to the space attraction and the edge coefficient;
and combining the Markov random field of the self-adaptive weight with a fuzzy C-means algorithm to determine a classification result of the remote sensing image.
Optionally, the determining the classification result of the remote sensing image by combining the markov random field with the adaptive weight with the fuzzy C-means algorithm specifically includes:
calculating the prior probability of the current central pixel belonging to the kth mark class according to the Markov random field of the self-adaptive weight; k belongs to [1, C ], and C represents the number of categories;
calculating a current clustering center and a current membership matrix according to the prior probability;
judging whether the current clustering center meets a termination condition;the termination condition is
Wherein
The center of the current cluster is represented,
representing the clustering center obtained under the last iteration times, wherein epsilon is a termination threshold value and is more than 0;
if not, making b equal to b +1 and i equal to i +1, and returning to the initial fuzzy membership matrix, and calculating the spatial attraction between the current central pixel i and each neighborhood pixel in the remote sensing image under the current iteration number b by using a spatial attraction model;
if so, determining a fuzzy membership matrix according to the previous clustering center and the current membership matrix;
and determining the classification result of the remote sensing image according to the fuzzy membership matrix.
Optionally, the calculating a current clustering center and a current membership matrix according to the prior probability specifically includes:
establishing an objective function according to the prior probability
Wherein,
wherein N represents the total number of pixels in the remote sensing image, vkA cluster center representing a kth label class; u. ofikA membership matrix for representing the ith pixel belonging to the kth mark class, m representsConstant, xiRepresenting the gray value, P, of the ith pixelk(i) Representing the prior probability, v, that the ith pixel belongs to the kth label classtRepresenting the clustering center of the t-th class, t ∈ [1, c ]]，c∈[1,C]；
Calculating the optimal solution of the target function by adopting a Lagrange multiplier method; the optimal solution is a clustering center of a kth mark class and a membership matrix of the kth mark class corresponding to the minimum objective function;
and determining the clustering center of the kth mark class in the optimal solution as a current clustering center, and determining the membership matrix of the kth mark class in the optimal solution as a current membership matrix.
Optionally, the calculating, according to the initial fuzzy membership matrix, a spatial attraction model by using a spatial attraction model, a spatial attraction between a current center pixel and each neighborhood pixel in the remote sensing image under the current iteration number b specifically includes:
calculating the fuzzy membership degree of the current center pixel and the fuzzy membership degree of the neighborhood pixels according to the initial fuzzy membership degree matrix;
according to the fuzzy membership degree of the current center pixel and the fuzzy membership degree of the current center pixel, calculating the space attraction between the current center pixel and each neighborhood pixel in the remote sensing image under the current iteration number b by using a space attraction model
Wherein j represents a neighborhood pixel of the ith pixel element, and j is 0,1icA membership matrix u representing the i-th pixel belonging to the c-th classjcMembership matrix, w, representing the membership of a neighborhood pixel j to class cijRepresenting the distance weight coefficient between the current center pixel and the neighborhood pixel, C belongs to [1, C ∈]And C represents the number of categories.
Optionally, the calculating, according to the spatial structure characteristic, an edge coefficient of the current center pixel by using a gradient reciprocal smoothing method specifically includes:
calculating the gradient of the current center pixel
Wherein,
representing a first order horizontal gradient of the ith pixel over the d-th band,
representing a first-order vertical gradient of the ith pixel in the d-th band;
constructing the edge coefficient of the current center pixel according to the gradient of the current center pixel
Optionally, the self-adaptive weighted markov random field specifically includes:
P(c)＝exp(-α(xi)U(c))/Z，
of these, α (x)i) Representing the edge coefficients of the current center pel, Z representing the normalized constant of the slicing function, u (c) representing the energy function,
wherein C is [1, C ]]C represents the number of categories, F represents a potential group set, F represents a potential group set, If(c) The potential function on the f is represented,
wherein, c (x)i) Indicates the class of the i-th pixel, c (x)j) Watch (A)Indicating the class of the neighborhood pixel j, j indicating the neighborhood pixel of the ith pixel, SijRepresenting the spatial attraction, x, between the current center pixel and the neighborhood pixelsiRepresenting the gray value, x, of the ith pixeljRepresenting the gray value of the neighborhood pixel j,
representing the total number of neighborhood pixels corresponding to the ith pixel,
the invention also provides a remote sensing image classification system based on the self-adaptive spatial information, which comprises the following steps:
the image acquisition module is used for acquiring a remote sensing image;
the initial classification module is used for carrying out initial classification on the remote sensing images by adopting a fuzzy C mean algorithm based on a Markov random field to obtain an initial fuzzy membership matrix;
the spatial attraction calculation module is used for calculating the spatial attraction between the current center pixel and each neighborhood pixel in the remote sensing image under the current iteration number b by utilizing a spatial attraction model according to the initial fuzzy membership matrix; the current central pixel is the ith pixel in the remote sensing image; each current center pixel corresponds to a plurality of neighborhood pixels; the position of the current center pixel is adjacent to the position of the neighborhood pixel;
the edge detection module is used for carrying out edge detection on the remote sensing image by adopting a Sobel operator to obtain a spatial structure characteristic;
the edge coefficient calculation module is used for calculating the edge coefficient of the current central pixel by adopting a gradient reciprocal smoothing method according to the spatial structure characteristics;
the random field construction module is used for constructing a Markov random field of self-adaptive weight according to the space attraction and the edge coefficient;
and the classification module is used for combining the Markov random field of the self-adaptive weight with a fuzzy C mean algorithm to determine a classification result of the remote sensing image.
Optionally, the classification module specifically includes:
the first calculating unit is used for calculating the prior probability that the current central pixel belongs to the kth mark class according to the Markov random field of the self-adaptive weight; k belongs to [1, C ], and C represents the number of categories;
the second calculation unit is used for calculating the current clustering center and the current membership matrix according to the prior probability;
the judging unit is used for judging whether the current clustering center meets a termination condition; the termination condition is
Wherein
The center of the current cluster is represented,
representing the clustering center obtained under the last iteration times, wherein epsilon is a termination threshold value and is more than 0; if not, making b equal to b +1 and i equal to i +1, and returning to the initial fuzzy membership matrix, and calculating the spatial attraction between the current central pixel i and each neighborhood pixel in the remote sensing image under the current iteration number b by using a spatial attraction model; and if so, determining a fuzzy membership matrix according to the previous clustering center and the current membership matrix, and determining a classification result of the remote sensing image according to the fuzzy membership matrix.
Optionally, the spatial attraction force calculation module specifically includes:
the third calculating unit is used for calculating the fuzzy membership degree of the current central pixel and the fuzzy membership degree of the neighborhood pixels according to the initial fuzzy membership degree matrix;
a fourth calculating unit, configured to calculate, according to the fuzzy membership of the current center pixel and the fuzzy membership of the current center pixel, a spatial attraction model based on the spatial attraction model, where the spatial attraction between the current center pixel and each neighborhood pixel in the remote sensing image is calculated according to the current iteration number b
Wherein j represents a neighborhood pixel of the ith pixel element, and j is 0,1icA membership matrix u representing the i-th pixel belonging to the c-th classjcMembership matrix, w, representing the membership of a neighborhood pixel j to class cijRepresenting the distance weight coefficient between the current center pixel and the neighborhood pixel, C belongs to [1, C ∈]And C represents the number of categories.
Optionally, the edge coefficient calculating module specifically includes:
a gradient calculation unit for calculating the gradient of the current center pixel
Wherein,
representing a first order horizontal gradient of the ith pixel over the d-th band,
representing a first-order vertical gradient of the ith pixel in the d-th band;
an edge coefficient calculation unit for constructing the edge coefficient of the current center pixel according to the gradient of the current center pixel
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a remote sensing image classification method and system based on self-adaptive spatial information. The method comprises the following steps: acquiring a remote sensing image; carrying out initial classification on the remote sensing images by adopting a fuzzy C-means algorithm based on a Markov random field to obtain an initial fuzzy membership matrix; calculating the space attraction between the current center pixel and each neighborhood pixel in the remote sensing image under the current iteration number b by using a space attraction model; carrying out edge detection on the remote sensing image by adopting a Sobel operator to obtain spatial structure characteristics; calculating the edge coefficient of the current central pixel by adopting a gradient reciprocal smoothing method according to the spatial structure characteristics; constructing a Markov random field of self-adaptive weight according to the space attraction and the edge coefficient; and combining the Markov random field of the self-adaptive weight with a fuzzy C-means algorithm to determine a classification result of the remote sensing image. The invention combines the spatial gravitation model, the spatial structure characteristics and the traditional Markov random field to construct the Markov random field with self-adaptive weight, and introduces the Markov random field into the fuzzy C-means clustering algorithm, thereby not only overcoming the noise influence, but also keeping the edge details of the image, and changing the traditional mode of fixing the neighborhood weight of the spatial information, thereby effectively solving the problem of the estimation of the boundary pixel and the spatial information weight coefficient, and improving the classification precision of the remote sensing image.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a remote sensing image classification method based on adaptive spatial information according to an embodiment of the present invention;
FIG. 2 is a diagram of a 3 × 3 neighborhood system window of a cardiac pixel in an embodiment of the present invention;
FIG. 3 is a graph of distance weighting coefficients between a cardiac pel and a neighborhood pel in an embodiment of the invention;
FIG. 4 is an original remote sensing image of an embodiment of the present invention;
FIG. 5 is a ground reference image of an embodiment of the present invention;
FIG. 6 is a diagram illustrating the classification result of the FCM algorithm according to this embodiment;
FIG. 7 is a diagram illustrating the classification result of the MFCM algorithm according to the present embodiment;
FIG. 8 is a diagram illustrating the classification result of the AMFCM algorithm according to the present embodiment;
fig. 9 is a schematic structural diagram of a remote sensing image classification system based on adaptive spatial information according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
FIG. 1 is a flowchart of a method for classifying remote sensing images based on adaptive spatial information according to an embodiment of the present invention.
Referring to fig. 1, the remote sensing image classification method based on adaptive spatial information according to the embodiment includes:
step S1: and acquiring a remote sensing image.
Step S2: and carrying out initial classification on the remote sensing image by adopting a Markov random field based fuzzy C-means algorithm (MFCM) to obtain an initial fuzzy membership matrix.
The initial fuzzy membership matrix is UC×NAnd meanwhile, obtaining an initial category label.
Step S3: and calculating the space attraction between the current central pixel and each neighborhood pixel in the remote sensing image under the current iteration number b by utilizing a space attraction model according to the initial fuzzy membership matrix.
The current central pixel is the ith pixel in the remote sensing image; each current center pixel corresponds to a plurality of neighborhood pixels; and the position of the current center pixel is adjacent to the position of the neighborhood pixel. The initial iteration count b is set to 1. The spatial gravitation model is proved to be effective for representing the spatial correlation among image pixels, and is used for a Markov model, and the model contains spatial neighborhood information and gray information. For two picture elements xiAnd xjTheir attractiveness to the c-th cluster and their fuzzy membership uicAnd ujcProportional and inversely proportional to the distance between two picture elements.
The step S3 specifically includes:
1) and calculating the fuzzy membership of the current center pixel and the fuzzy membership of the neighborhood pixels according to the initial fuzzy membership matrix.
2) According to the fuzzy membership degree of the current center pixel and the fuzzy membership degree of the current center pixel, calculating the space attraction between the current center pixel and each neighborhood pixel in the remote sensing image under the current iteration number b by using a space attraction model
Wherein j represents the neighborhood pixel of the ith pixel, and j belongs to Ni,j{j＝0,1,2,…,7}，uicA membership matrix u representing the i-th pixel belonging to the c-th classjcMembership matrix, w, representing the membership of a neighborhood pixel j to class cijRepresents the distance weight coefficient, w, between the current center pixel and the neighborhood pixelijIs the reciprocal of Euclidean distance, the closer the neighborhood pixel is to the center pixel, the wijThe larger the influence on the central pixel element is, the C belongs to [1, C ∈]And C represents the number of categories.
In this embodiment, spatial gravity exists only in a given 3 × 3 neighborhood window, and other pixels outside the window are too far from the center pixel to impart any gravity on the center pixel. Fig. 2 is a 3 × 3 neighborhood system window diagram of a heart pixel in the embodiment of the present invention, fig. 3 is a distance weight coefficient diagram between the heart pixel and a neighborhood pixel in the embodiment of the present invention, referring to fig. 2 and fig. 3, a spatial attraction exists only between a center pixel and its neighborhood pixel in a given window, and other pixels outside the window are too far away from the center pixel to give any attraction to the center pixel. The main contributions of the spatial gravity model are: 1) different distances between the central pixel and the neighborhood pixels are considered; 2) the fuzzy membership degree of the image pixels is used for better describing different influences of the neighborhood pixels on the central pixels.
Step S4: and carrying out edge detection on the remote sensing image by adopting a Sobel operator to obtain the spatial structure characteristics. The edge feature is one of the spatial structure features, while calculating the gradient is the most common method in edge detection, and the Sobel operator is used as an edge detection operator and is based on the first derivative. Because the local average operation is applied, the operator has a smoothing effect on noise, and the influence caused by the noise can be well eliminated.
Step S5: and calculating the edge coefficient of the current central pixel by adopting a gradient reciprocal smoothing method according to the spatial structure characteristics.
The step S5 specifically includes:
1) calculating the gradient of the current center pixel
Wherein,
representing a first order horizontal gradient of the ith pixel over the d-th band,
representing the first order vertical gradient of the ith pixel over the d-th band.
2) Constructing the edge coefficient of the current center pixel according to the gradient of the current center pixel
From the above formula, it can be found that α (x) is obtained when the image element is in the smooth regioni) Has a larger value and has a stronger smoothing effect on homogeneous pixels, whereas when the pixel is located in an edge or texture region, α (x)i) Smaller values of (a) have a weaker smoothing effect on the picture elements in the area, and smaller areas in the image can be retained.
Step S6: and constructing a Markov random field of self-adaptive weight according to the space attraction and the edge coefficient. The self-adaptive weight Markov random field specifically comprises the following steps:
P(c)＝exp(-α(xi)U(c))/Z，
of these, α (x)i) Representing the edge coefficients of the current center pel, Z representing the normalized constant of the slicing function, u (c) representing the energy function,
wherein C is [1, C ]]C represents the number of categories, F represents a potential group set, F represents a potential group set, If(c) Denotes the potential function over f, I in this examplef(c) Uses a binary potential function formula I represented by a Potts model2(c(xi),c(xj))，I2(c(xi),c(xj) Only the binary potential function at f is considered,
wherein, c (x)i) Indicates the class of the i-th pixel, c (x)j) Representing the class of a neighborhood pixel j, j representing the neighborhood pixel of the ith pixel, SijRepresenting the space attraction between the current central pixel and the neighborhood pixels, for controlling the influence of the neighborhood pixels on the central pixel, xiRepresenting the gray value, x, of the ith pixeljRepresenting the gray value of the neighborhood pixel j,
representing the total number of neighborhood pixels corresponding to the ith pixel,
in the embodiment, a spatial gravitation model and spatial structure characteristics are introduced into the Markov random field, a Markov random field with self-adaptive weight (self-adaptive spatial information Markov random field) is constructed, the spatial information weight between a center pixel and a neighborhood pixel is adaptively changed, and the clustering precision of a small target area is improved.
Step S7: and combining the Markov random field of the self-adaptive weight with a fuzzy C-means algorithm to determine a classification result of the remote sensing image.
The step S7 specifically includes:
1) calculating the prior probability of the current central pixel belonging to the kth mark class according to the Markov random field of the self-adaptive weight; k is equal to [1, C ], and C represents the number of categories.
2) According to the prior probability, calculating a current clustering center and a current membership matrix, and specifically comprising the following steps:
21) establishing an objective function according to the prior probability
Wherein the objective function JAMFCMThe condition necessary for obtaining the condition extremum is
Wherein N represents the total number of pixels in the remote sensing image, vkA cluster center representing a kth label class; u. ofikA membership matrix representing that the ith pixel belongs to the kth mark class, m represents a constant, xiRepresenting the gray value, P, of the ith pixelk(i) Representing the prior probability, v, that the ith pixel belongs to the kth label classtRepresenting the clustering center of the t-th class, t ∈ [1, c ]]，c∈[1,C]。
22) Calculating the optimal solution of the target function by adopting a Lagrange multiplier method; and the optimal solution is the clustering center of the k mark class and the membership matrix of the k mark class corresponding to the minimum objective function.
23) And determining the clustering center of the kth mark class in the optimal solution as a current clustering center, and determining the membership matrix of the kth mark class in the optimal solution as a current membership matrix.
3) Judging whether the current clustering center meets a termination condition; the termination condition is
Wherein
The center of the current cluster is represented,
representing the clustering center obtained under the last iteration number, wherein epsilon is a termination threshold value and is larger than 0.
If not, let b be b +1 and i be i +1, and return to the step S3; and if so, determining a fuzzy membership matrix according to the previous clustering center and the current membership matrix, and determining a classification result of the remote sensing image according to the fuzzy membership matrix.
The remote sensing image classification method based on the self-adaptive spatial information is verified below.
The invention combines a Spatial gravitation model, Spatial structure characteristics and a traditional Markov random field to construct a Markov random field with self-Adaptive weight, and introduces the Markov random field into a fuzzy C-means clustering algorithm to obtain an FCM (Adaptive Spatial Information MRF-Based FCMdifferential Algorithm, AMFCM) clustering algorithm Based on self-Adaptive Spatial Information MRF, which can self-adaptively update the influence of neighborhood pixels on center pixels and better retain image edge detail Information, thereby improving the classification precision of remote sensing images.
In order to verify the effectiveness of the AMFCM algorithm, the AMFCM algorithm is compared with FCM and MFCM clustering algorithms, a high-resolution remote sensing image is selected for carrying out experiments, a confusion matrix is used for carrying out precision evaluation on a classification result, and four indexes are adopted for carrying out quantitative evaluation, such as overall classification precision (OA), Kappa coefficient, Producer's Precision (PA) and User's precision (UA), wherein the OA is the ratio of the total number of correctly classified pixels to the total number of pixels. The Kappa coefficient describes the consistency of the population and also the consistency of the classification, and can be described as
N is the total number of reference data pixels, xkkIs the number of correctly classified pixels, x, of all classes in the imagek+Is the number of reference data pixels, x, in all classes+kIs the number of pixels classified in each class, and PA is the ratio of the number of pixels correctly classified as a class to the total number of a reference pixels. UA is the ratio of the total number of pixels correctly classified as class A to the total number of pixels classified as class A by the classifier.
The high-score second remote sensing image is used for verifying the proposed algorithm. The image acquisition time is 9 months in 2014, and a three-band multispectral remote sensing image is obtained by fusing a panchromatic image with the spatial resolution of 0.8 m and a multispectral image with the spatial resolution of 3.2 m and is located in Shanghai city. FIG. 4 is an original remote sensing image of an embodiment of the present invention; fig. 5 is a ground reference image according to an embodiment of the present invention, and referring to fig. 4 and 5, the size of the research area is 221 × 192 pixels, which includes 5 categories: rose roof 1, blue roof 2, road 3, shade 4 and grass 5.
Fig. 6 is a diagram of a classification result of the FCM algorithm according to the present embodiment, fig. 7 is a diagram of a classification result of the MFCM algorithm according to the present embodiment, and fig. 8 is a diagram of a classification result of the AMFCM algorithm according to the present embodiment. Referring to fig. 6-8, fig. 8 shows the best classification results for AMFCM. Visually, the clustering results of MFCM and AMFCM are different from FCM due to the application of spatial information, and although grassland is classified to some extent to shadow classes for all 3 results, the classification results of AMFCM are better. In the experiment, due to the fact that the grassland with poor growth has obvious black bare soil, the spectral information of the grassland is extremely similar to that of the shadow, and a good classification result cannot be obtained only through the spectral characteristics of the remote sensing image. In fig. 6 it can be seen that the grass classification results contain many shadow classes, which does not correspond to the actual situation, while in fig. 7 and 8 the situation is improved, but still a part of the grass is classified into shadow classes, because the MRF, although taking into account the spatial correlation of the image elements, has a limited resolving power for resolving terrain classes with very similar spectral information. Comparing fig. 7 and 8, it can be seen that fig. 8 is locally better than fig. 7, the class edges in fig. 7 are less obvious, and the classification effect is not very good for small target regions such as shadows because MRF is isotropic, so that the classification result loses the edge detail information of the image, while in fig. 8, the situation is changed because AMFCM adopts a markov random field of adaptive spatial information weights, which can adaptively change the spatial information weights, and also introduces the spatial structure features such as edge coefficients.
For quantitative comparison of the above algorithms, the OA and kappa coefficients of the images are listed in Table 1, as follows:
TABLE 1
As can be seen from table 1, the classification accuracy produced by MFCM and AMFCM is higher than FCM, and AMFCM achieves the best overall classification accuracy of 79.35%, which is 7.12% and 2.57% higher than FCM and MFCM, respectively.
In the remote sensing image classification method based on the self-adaptive spatial information, the FCM algorithm based on the Markov random field is used for carrying out initial classification on the remote sensing images to obtain an initial membership matrix and an initial category label; introducing a spatial gravitation model and spatial structure characteristics into a Markov random field, and constructing a Markov random field with self-adaptive weight; combining the Markov random field with the self-adaptive weight with the traditional FCM algorithm to obtain an FCM algorithm of self-adaptive spatial information MRF; and on the basis of the initial classification result, carrying out refined classification on the remote sensing image through an FCM algorithm of the self-adaptive spatial information MRF. The method effectively solves the problem of boundary pixel and spatial information weight coefficient estimation, and improves the classification precision of the remote sensing image.
The invention also provides a remote sensing image classification system based on the adaptive spatial information, and FIG. 9 is a schematic structural diagram of the remote sensing image classification system based on the adaptive spatial information according to the embodiment of the invention.
The remote sensing image classification system based on the self-adaptive spatial information comprises the following components:
an image obtaining module 901, configured to obtain a remote sensing image.
And the initial classification module 902 is configured to perform initial classification on the remote sensing image by using a fuzzy C-means algorithm based on a markov random field to obtain an initial fuzzy membership matrix.
A spatial attraction calculation module 903, configured to calculate, according to the initial fuzzy membership matrix, a spatial attraction between a current center pixel and each neighborhood pixel in the remote sensing image for the current iteration number b by using a spatial attraction model; the current central pixel is the ith pixel in the remote sensing image; each current center pixel corresponds to a plurality of neighborhood pixels; and the position of the current center pixel is adjacent to the position of the neighborhood pixel.
And the edge detection module 904 is configured to perform edge detection on the remote sensing image by using a Sobel operator to obtain a spatial structure characteristic.
And the edge coefficient calculating module 905 is configured to calculate an edge coefficient of the current center pixel by using a gradient reciprocal smoothing method according to the spatial structure characteristic.
A random field constructing module 906, configured to construct a self-adaptive weighted markov random field according to the spatial attraction force and the edge coefficient. The self-adaptive weight Markov random field specifically comprises the following steps:
P(c)＝exp(-α(xi)U(c))/Z，
of these, α (x)i) Representing the edge coefficients of the current center pel, Z representing the normalized constant of the slicing function, u (c) representing the energy function,
wherein C is [1, C ]]C represents the number of categories, F represents a potential group set, F represents a potential group set, If(c) The potential function on the f is represented,
wherein, c (x)i) Indicates the class of the i-th pixel, c (x)j) Representing the class of a neighborhood pixel j, j representing the neighborhood pixel of the ith pixel, SijRepresenting the spatial attraction, x, between the current center pixel and the neighborhood pixelsiRepresenting the gray value, x, of the ith pixeljRepresenting the gray value of the neighborhood pixel j,
representing the total number of neighborhood pixels corresponding to the ith pixel,
and the classification module 907 is used for combining the markov random field with the adaptive weight with a fuzzy C-means algorithm to determine a classification result of the remote sensing image.
As an optional implementation manner, the classification module 907 specifically includes:
the first calculating unit is used for calculating the prior probability that the current central pixel belongs to the kth mark class according to the Markov random field of the self-adaptive weight; k belongs to [1, C ], and C represents the number of categories;
the second calculation unit is used for calculating the current clustering center and the current membership matrix according to the prior probability;
the judging unit is used for judging whether the current clustering center meets a termination condition; the termination condition is
Wherein
The center of the current cluster is represented,
representing the clustering center obtained under the last iteration times, wherein epsilon is a termination threshold value and is more than 0; if not, making b equal to b +1 and i equal to i +1, and returning to the initial fuzzy membership matrix, and calculating the spatial attraction between the current central pixel i and each neighborhood pixel in the remote sensing image under the current iteration number b by using a spatial attraction model; and if so, determining a fuzzy membership matrix according to the previous clustering center and the current membership matrix, and determining a classification result of the remote sensing image according to the fuzzy membership matrix.
As an optional implementation manner, the spatial attraction force calculation module 903 specifically includes:
the third calculating unit is used for calculating the fuzzy membership degree of the current central pixel and the fuzzy membership degree of the neighborhood pixels according to the initial fuzzy membership degree matrix;
a fourth calculating unit, configured to calculate, according to the fuzzy membership of the current center pixel and the fuzzy membership of the current center pixel, a spatial attraction model based on the spatial attraction model, where the spatial attraction between the current center pixel and each neighborhood pixel in the remote sensing image is calculated according to the current iteration number b
Wherein j represents a neighborhood pixel of the ith pixel element, and j is 0,1icA membership matrix u representing the i-th pixel belonging to the c-th classjcMembership matrix, w, representing the membership of a neighborhood pixel j to class cijRepresenting the distance weight coefficient between the current center pixel and the neighborhood pixel, C belongs to [1, C ∈]And C represents the number of categories.
As an optional implementation manner, the edge coefficient calculating module 905 specifically includes:
a gradient calculation unit for calculating the gradient of the current center pixel
Wherein,
representing the first order horizontal gradient of the ith pixel over the d-th band, ▽vxid represents the first-order vertical gradient of the ith pixel on the d wave band;
an edge coefficient calculation unit for constructing the edge coefficient of the current center pixel according to the gradient of the current center pixel
As an optional implementation manner, the second computing unit specifically includes:
an objective function establishing subunit, configured to establish an objective function according to the prior probability
Wherein,
wherein N represents the total number of pixels in the remote sensing image, vkA cluster center representing a kth label class; u. ofikA membership matrix representing that the ith pixel belongs to the kth mark class, m represents a constant, xiRepresenting the gray value, P, of the ith pixelk(i) Representing the prior probability, v, that the ith pixel belongs to the kth label classtRepresenting the clustering center of the t-th class, t ∈ [1, c ]]，c∈[1,C]；
The solving subunit is used for calculating the optimal solution of the objective function by adopting a Lagrange multiplier method; the optimal solution is a clustering center of a kth mark class and a membership matrix of the kth mark class corresponding to the minimum objective function;
and the determining subunit is used for determining the clustering center of the kth mark class in the optimal solution as the current clustering center, and determining the membership matrix of the kth mark class in the optimal solution as the current membership matrix.
The remote sensing image classification system based on the self-adaptive spatial information effectively solves the problem of boundary pixel and spatial information weight coefficient estimation, and improves the classification precision of the remote sensing image.
For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (9)
1. The remote sensing image classification method based on the self-adaptive spatial information is characterized by comprising the following steps:
acquiring a remote sensing image;
carrying out initial classification on the remote sensing images by adopting a fuzzy C-means algorithm based on a Markov random field to obtain an initial fuzzy membership matrix;
calculating the space attraction between the current central pixel and each neighborhood pixel in the remote sensing image under the current iteration number b by utilizing a space attraction model according to the initial fuzzy membership matrix; the current central pixel is the ith pixel in the remote sensing image; each current center pixel corresponds to a plurality of neighborhood pixels; the position of the current center pixel is adjacent to the position of the neighborhood pixel;
performing edge detection on the remote sensing image by using a Sobel operator to obtain spatial structure characteristics;
calculating the edge coefficient of the current central pixel by adopting a gradient reciprocal smoothing method according to the spatial structure characteristics;
constructing a Markov random field of self-adaptive weight according to the space attraction and the edge coefficient;
combining the Markov random field of the self-adaptive weight with a fuzzy C-means algorithm to determine a classification result of the remote sensing image;
the self-adaptive weight Markov random field specifically comprises the following steps:
P(c)＝exp(-α(xi)U(c))/Z，
of these, α (x)i) Representing the edge coefficients of the current center pel, Z representing the normalized constant of the slicing function, u (c) representing the energy function,
wherein C is [1, C ]]C represents the number of categories, F represents a potential group set, F represents a potential group set, If(c) The potential function on the f is represented,
wherein, c (x)i) Indicates the class of the i-th pixel, c (x)j) Representing the class of a neighborhood pixel j, j representing the neighborhood pixel of the ith pixel, SijRepresenting the spatial attraction, x, between the current center pixel and the neighborhood pixelsiRepresenting the gray value, x, of the ith pixeljRepresenting the gray value, N, of a neighborhood pixel ji,jRepresents the total number of neighborhood pixels corresponding to the ith pixel, Ni,j∈{0,1,2,...,7}。
2. The method for classifying remote sensing images based on adaptive spatial information according to claim 1, wherein the step of determining the classification result of the remote sensing images by combining the markov random field with adaptive weight with a fuzzy C-means algorithm specifically comprises the steps of:
calculating the prior probability of the current central pixel belonging to the kth mark class according to the Markov random field of the self-adaptive weight; k belongs to [1, C ], and C represents the number of categories;
calculating a current clustering center and a current membership matrix according to the prior probability;
judging whether the current clustering center meets a termination condition; the termination condition is
Wherein
The center of the current cluster is represented,
representing the clustering center obtained under the last iteration times, wherein epsilon is a termination threshold value and is more than 0;
if not, making b equal to b +1 and i equal to i +1, and returning to the initial fuzzy membership matrix, and calculating the spatial attraction between the current central pixel i and each neighborhood pixel in the remote sensing image under the current iteration number b by using a spatial attraction model;
if so, determining a fuzzy membership matrix according to the previous clustering center and the current membership matrix;
and determining the classification result of the remote sensing image according to the fuzzy membership matrix.
3. The method for classifying remote sensing images based on adaptive spatial information according to claim 2, wherein the calculating a current cluster center and a current membership matrix according to the prior probability specifically comprises:
establishing an objective function according to the prior probability
Wherein,
wherein N represents the total number of pixels in the remote sensing image, vkA cluster center representing a kth label class; u. ofikMembership moment representing that ith pixel belongs to kth mark classArray, m denotes a constant, xiRepresenting the gray value, P, of the ith pixelk(i) Representing the prior probability, v, that the ith pixel belongs to the kth label classtRepresenting the clustering center of the tth labeled class, te [1, c]，c∈[1,C]；
Calculating the optimal solution of the target function by adopting a Lagrange multiplier method; the optimal solution is a clustering center of a kth mark class and a membership matrix of the kth mark class corresponding to the minimum objective function;
and determining the clustering center of the kth mark class in the optimal solution as a current clustering center, and determining the membership matrix of the kth mark class in the optimal solution as a current membership matrix.
4. The method for classifying remote sensing images based on adaptive spatial information according to claim 1, wherein the calculating the spatial attraction between the current central pixel and each neighborhood pixel in the remote sensing image at the current iteration number b by using a spatial attraction model according to the initial fuzzy membership matrix specifically comprises:
calculating the fuzzy membership degree of the current center pixel and the fuzzy membership degree of the neighborhood pixels according to the initial fuzzy membership degree matrix;
according to the fuzzy membership degree of the current center pixel and the fuzzy membership degree of the neighborhood pixels, calculating the space attraction between the current center pixel and each neighborhood pixel in the remote sensing image under the current iteration number b by using a space attraction model
Wherein j represents a neighborhood pixel of the ith pixel element, and j is 0,1icA membership matrix u representing the i-th pixel belonging to the c-th classjcMembership matrix, w, representing the membership of a neighborhood pixel j to class cijRepresenting the distance weight coefficient between the current center pixel and the neighborhood pixel, C belongs to [1, C ∈]And C represents the number of categories.
5. The method for classifying remote sensing images based on adaptive spatial information according to claim 1, wherein the calculating the edge coefficient of the current center pixel by using a gradient reciprocal smoothing method according to the spatial structure characteristics specifically comprises:
calculating the gradient of the current center pixel
Wherein,
representing a first order horizontal gradient of the ith pixel over the d-th band,
representing a first-order vertical gradient of the ith pixel in the d-th band;
constructing the edge coefficient of the current center pixel according to the gradient of the current center pixel
6. The remote sensing image classification system based on self-adaptive spatial information is characterized by comprising the following steps:
the image acquisition module is used for acquiring a remote sensing image;
the initial classification module is used for carrying out initial classification on the remote sensing images by adopting a fuzzy C mean algorithm based on a Markov random field to obtain an initial fuzzy membership matrix;
the spatial attraction calculation module is used for calculating the spatial attraction between the current center pixel and each neighborhood pixel in the remote sensing image under the current iteration number b by utilizing a spatial attraction model according to the initial fuzzy membership matrix; the current central pixel is the ith pixel in the remote sensing image; each current center pixel corresponds to a plurality of neighborhood pixels; the position of the current center pixel is adjacent to the position of the neighborhood pixel;
the edge detection module is used for carrying out edge detection on the remote sensing image by adopting a Sobel operator to obtain a spatial structure characteristic;
the edge coefficient calculation module is used for calculating the edge coefficient of the current central pixel by adopting a gradient reciprocal smoothing method according to the spatial structure characteristics;
the random field construction module is used for constructing a Markov random field of self-adaptive weight according to the space attraction and the edge coefficient;
the classification module is used for combining the Markov random field of the self-adaptive weight with a fuzzy C mean algorithm to determine a classification result of the remote sensing image;
the self-adaptive weight Markov random field specifically comprises the following steps:
P(c)＝exp(-α(xi)U(c))/Z，
of these, α (x)i) Representing the edge coefficients of the current center pel, Z representing the normalized constant of the slicing function, u (c) representing the energy function,
wherein C is [1, C ]]C represents the number of categories, F represents a potential group set, F represents a potential group set, If(c) The potential function on the f is represented,
wherein, c (x)i) Indicates the class of the i-th pixel, c (x)j) Representing the class of a neighborhood pixel j, j representing the neighborhood pixel of the ith pixel, SijRepresenting the space between the current center pixel and the neighborhood pixelsAttractive force, xiRepresenting the gray value, x, of the ith pixeljRepresenting the gray value, N, of a neighborhood pixel ji,jRepresents the total number of neighborhood pixels corresponding to the ith pixel, Ni,j∈{0,1,2,...,7}。
7. The system for classifying remote sensing images based on adaptive spatial information according to claim 6, wherein the classification module specifically comprises:
the first calculating unit is used for calculating the prior probability that the current central pixel belongs to the kth mark class according to the Markov random field of the self-adaptive weight; k belongs to [1, C ], and C represents the number of categories;
the second calculation unit is used for calculating the current clustering center and the current membership matrix according to the prior probability;
the judging unit is used for judging whether the current clustering center meets a termination condition; the termination condition is
Wherein
The center of the current cluster is represented,
representing the clustering center obtained under the last iteration times, wherein epsilon is a termination threshold value and is more than 0; if not, making b equal to b +1 and i equal to i +1, and returning to the initial fuzzy membership matrix, and calculating the spatial attraction between the current central pixel i and each neighborhood pixel in the remote sensing image under the current iteration number b by using a spatial attraction model; and if so, determining a fuzzy membership matrix according to the previous clustering center and the current membership matrix, and determining a classification result of the remote sensing image according to the fuzzy membership matrix.
8. The system for classifying remote sensing images based on adaptive spatial information according to claim 6, wherein the spatial attraction calculation module specifically comprises:
the third calculating unit is used for calculating the fuzzy membership degree of the current central pixel and the fuzzy membership degree of the neighborhood pixels according to the initial fuzzy membership degree matrix;
a fourth calculating unit, configured to calculate, according to the fuzzy membership of the current center pixel and the fuzzy membership of the neighborhood pixels, a spatial attraction model based on the spatial attraction model, where the spatial attraction between the current center pixel and each of the neighborhood pixels in the remote sensing image is calculated according to the current iteration number b
Wherein j represents a neighborhood pixel of the ith pixel element, and j is 0,1icA membership matrix u representing the i-th pixel belonging to the c-th classjcMembership matrix, w, representing the membership of a neighborhood pixel j to class cijRepresenting the distance weight coefficient between the current center pixel and the neighborhood pixel, C belongs to [1, C ∈]And C represents the number of categories.
9. The system for classifying remote sensing images based on adaptive spatial information according to claim 6, wherein the edge coefficient calculation module specifically comprises:
a gradient calculation unit for calculating the gradient of the current center pixel
Wherein,
representing a first order horizontal gradient of the ith pixel over the d-th band,
representing a first-order vertical gradient of the ith pixel in the d-th band;
an edge coefficient calculation unit for constructing the edge coefficient of the current center pixel according to the gradient of the current center pixel
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 CN109829519A (en) | 2019-05-31 |
| CN109829519B true 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) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN112733888B (en) * | 2020-12-25 | 2022-11-18 | 河南大学 | Deep space exploration image morphology analysis method based on multilayer MRF clustering |
| CN113446992B (en) * | 2021-06-28 | 2023-06-16 | 中国水利水电科学研究院 | Method for optimizing distribution of topographic survey points in topographic survey |
| CN114419561A (en) * | 2022-02-25 | 2022-04-29 | 广东粤港澳大湾区硬科技创新研究院 | Method and device for detecting sprinkled objects |
| CN115578388B (en) * | 2022-12-07 | 2023-04-21 | 滨州华然化纤绳网有限公司 | Data processing method for textile production |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010027476A1 (en) * | 2008-09-03 | 2010-03-11 | 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 |
| 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 |
| CN109409438B (en) * | 2018-11-07 | 2021-09-07 | 重庆市勘测院 | Remote sensing image classification method based on IFCM clustering and variational inference |
-
2019
- 2019-03-22 CN CN201910220213.8A patent/CN109829519B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
Non-Patent Citations (5)
| Title |
|---|
| "Change detection method for remote sensing images based on an improved Markov random field";Wei Gu等;《Multimedia Tools and Applications》;20150926;第17719-17734页 * |
| "Spatial-attraction-based Markov random field approach for classification of high spatial resolution multispectral imagery";Zhang Hua等;《IEEE Geosciences and Remote Sensing Letters》;20130904;第489-493页 * |
| "Unsupervised change detection using fuzzy cmeans and MRF from remotely sensed images";Ming Hao等;《Remote Sensing Letters》;20131120;第4卷(第12期);第1185-1194页 * |
| "基于改进马尔科夫分类模型的高光谱图像分类方法";王冬雪等;《信息通信》;20160131(第1期);第16-17页 * |
| "基于高光谱图像的目标分类技术研究";王冬雪;《中国优秀硕士学位论文全文数据库 信息科技辑》;20170315;第2017年卷(第3期);I140-1727 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109829519A (en) | 2019-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829519B (en) | Remote sensing image classification method and system based on self-adaptive spatial information | |
| CN110443143B (en) | Multi-branch convolutional neural network fused remote sensing image scene classification method | |
| Zhang et al. | VPRS-based regional decision fusion of CNN and MRF classifications for very fine resolution remotely sensed images | |
| 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 | |
| CN104123561B (en) | Fuzzy C-mean algorithm remote sensing image automatic classification method based on spatial attraction model | |
| Du et al. | Learning discriminative sparse representations for hyperspectral image classification | |
| CN102982338B (en) | Classification of Polarimetric SAR Image method based on spectral clustering | |
| CN106096655B (en) | A kind of remote sensing image airplane detection method based on convolutional neural networks | |
| US9449395B2 (en) | Methods and systems for image matting and foreground estimation based on hierarchical graphs | |
| CN106611422B (en) | Stochastic gradient Bayes's SAR image segmentation method based on sketch structure | |
| CN113609889A (en) | High-resolution remote sensing image vegetation extraction method based on sensitive feature focusing perception | |
| CN106373146A (en) | Target tracking method based on fuzzy learning | |
| CN106056165B (en) | A kind of conspicuousness detection method based on super-pixel relevance enhancing Adaboost classification learning | |
| CN104408731B (en) | Region graph and statistic similarity coding-based SAR (synthetic aperture radar) image segmentation method | |
| CN111274964B (en) | Detection method for analyzing water surface pollutants based on visual saliency of unmanned aerial vehicle | |
| CN102169631A (en) | Manifold-learning-based traffic jam event cooperative detecting method | |
| CN113255727A (en) | Multi-sensor remote sensing image fusion classification method capable of layering dense fusion network | |
| CN113723449B (en) | Agent-driven multi-target evolutionary fuzzy clustering method based on preference information | |
| CN112767376B (en) | Multi-scale fusion image change detection method for gravitation model optimization conditional random field | |
| Chen et al. | A semisupervised context-sensitive change detection technique via gaussian process | |
| CN106157330A (en) | A kind of visual tracking method based on target associating display model | |
| CN112329818B (en) | Hyperspectral image non-supervision classification method based on graph convolution network embedded characterization | |
| CN106874862A (en) | People counting method based on submodule technology and semi-supervised learning | |
| CN110717531A (en) | Method for detecting classified change type based on uncertainty analysis and Bayesian fusion |
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 |