CN112507937B - Satellite image unsupervised classification method and device fusing multi-source data - Google Patents

Satellite image unsupervised classification method and device fusing multi-source data Download PDF

Info

Publication number
CN112507937B
CN112507937B CN202011492176.5A CN202011492176A CN112507937B CN 112507937 B CN112507937 B CN 112507937B CN 202011492176 A CN202011492176 A CN 202011492176A CN 112507937 B CN112507937 B CN 112507937B
Authority
CN
China
Prior art keywords
data
clustering
loss
representing
data source
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
Application number
CN202011492176.5A
Other languages
Chinese (zh)
Other versions
CN112507937A (en
Inventor
马千里
郑佳炜
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN202011492176.5A priority Critical patent/CN112507937B/en
Publication of CN112507937A publication Critical patent/CN112507937A/en
Application granted granted Critical
Publication of CN112507937B publication Critical patent/CN112507937B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Biomedical Technology (AREA)
  • Multimedia (AREA)
  • Remote Sensing (AREA)
  • Astronomy & Astrophysics (AREA)
  • Image Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a satellite image unsupervised classification method and device fusing multi-source data, which comprises the following steps: acquiring satellite image data sets of a plurality of data sources, and separating sample information and category information of the data; constructing a codec model, and designing a codec suitable for the data type and the data dimension of each data source for the data of each data source; introducing K-means clustering loss into each codec, fusing the extracted features with clustering targets, and adding constraints among different data sources; inputting satellite images of a plurality of data sources into a depth clustering model, training through a back propagation algorithm, and guiding the generation of hidden layer features; and clustering the hidden layer characteristics of each data source generated in the training process by using a K-means algorithm to obtain a plurality of clustering results, and finally obtaining a final category distribution result in a voting mode. The invention fuses satellite images of a plurality of data sources to realize an effective unsupervised classification target.

Description

Satellite image unsupervised classification method and device fusing multi-source data
Technical Field
The invention relates to the technical field of multi-source data analysis, in particular to a satellite image unsupervised classification method and device fusing multi-source data.
Background
The satellite image is analyzed, and the method has important practical significance for improving monitoring of dangerous and disastrous weather and improving weather forecast. The key work of satellite image processing is remote sensing image classification, which is a very effective method for extracting and distinguishing geographic information elements. In the remote sensing satellite image classification processing, the classification method can be roughly divided into two types: the method mainly comprises the steps of supervised classification and unsupervised classification, wherein the most important difference between the supervised classification and the unsupervised classification is whether validated remote sensing satellite data samples are required to be subjected to rule training before classification processing.
At present, unsupervised classification research on satellite images is mainly focused on a scene with a single data source, but because meteorological satellite data formats and imaging principles of images of the meteorological satellite data formats are different, any single image cannot comprehensively reflect the characteristics of an observation target, and therefore, various satellite images with different characteristics need to be combined, so that information complementation is realized, the advantages and the disadvantages are made up, the characteristics can be more comprehensively extracted, and the result is more accurate and reliable.
Disclosure of Invention
The invention aims to solve the defects in the prior art and provides a satellite image unsupervised classification method and device fusing multi-source data.
The first purpose of the invention can be achieved by adopting the following technical scheme:
a satellite image unsupervised classification method fusing multi-source data comprises the following steps:
s1, acquiring satellite image data sets of a plurality of data sources, preprocessing the data and separating sample information and category information of the data;
and S2, constructing a codec model, and designing a codec suitable for the data type and the data dimension of each data source for the data of each data source. The design is favorable for fully considering the data characteristics and better extracting the data characteristics;
s3, inputting satellite image data of a plurality of data sources into an encoding and decoding model, extracting key features of the images by the encoding and decoding model, introducing K-means clustering loss into each encoder and decoder, fusing the key features of the extracted images with a clustering target, and adding constraints among different data sources to enable feature learning of a certain specific data source to obtain help and guidance of other data sources, so that information among the features is complementary and consistent;
s4, inputting the sample information of the data into a deep clustering model, training through a back propagation algorithm, and guiding the generation of hidden layer characteristics, so that the original sample can be reconstructed, and a cluster-shaped structure can be formed, thereby better clustering operation can be performed, and the clustering precision is improved;
and S5, clustering the hidden layer characteristics of each data source generated in the training process in the step S4 by using a K-means algorithm to obtain a plurality of clustering results, and finally obtaining a final category distribution result in a voting mode.
Further, the encoding and decoding process of the encoding and decoding model is as follows:
Figure BDA0002841030730000021
Figure BDA0002841030730000022
wherein the content of the first and second substances,
Figure BDA0002841030730000023
the ith sample representing the v-th data source,
Figure BDA0002841030730000024
which represents the v-th encoder, decoder,
Figure BDA0002841030730000025
represents the parameters of the v-th encoder and decoder,
Figure BDA0002841030730000026
hidden layer features representing the ith sample of the vth data source,
Figure BDA0002841030730000027
an ith sample representing the reconstructed vth data source;
reconstructing sample data through encoding and decoding operations of the encoding and decoding model, wherein in the process, the reconstruction loss of the v-th encoding and decoding model is as follows:
Figure BDA0002841030730000031
wherein the content of the first and second substances,
Figure BDA0002841030730000032
representing the reconstruction loss of the v-th codec model,
Figure BDA0002841030730000033
the ith sample representing the v-th data source,
Figure BDA0002841030730000034
an ith sample representing the reconstructed vth data source, n representing the number of samples of the data set,
Figure BDA0002841030730000035
representing the two-norm of the vector.
Further, the process of introducing K-means clustering loss into each codec in step S3 is as follows:
define the K-means clustering penalty as:
Figure BDA0002841030730000036
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0002841030730000037
k-means clustering loss representing the vth data source, tr represents the trace of the matrix, H v Hidden layer representation of the v-th data source, F v Is a pseudo tag matrix for the v-th data source, and F v Is an orthonormal matrix, I represents an identity matrix;
the constraints between different data sources are established as follows:
Figure BDA0002841030730000038
wherein the content of the first and second substances,
Figure BDA0002841030730000039
denotes the loss of consistency of the v-th data source, F v A pseudo tag matrix representing the v-th data source, Y representing a consistency class matrix, I representing an identity matrix,
Figure BDA00028410307300000310
frobsen representing a matrixThe us norm.
Further, the step S4 process is as follows
Constructing an overall loss function according to the reconstruction loss, the K-means clustering loss and the consistency loss, and determining an overall training target of the depth clustering model through the overall loss function, wherein the overall loss function is as follows:
Figure BDA00028410307300000311
in the formula (I), the compound is shown in the specification,
Figure BDA00028410307300000312
for the reconstruction loss of the v-th codec model,
Figure BDA00028410307300000313
indicating a loss of consistency for the vth data source,
Figure BDA0002841030730000041
and K-means clustering loss of the v-th data source is represented, m represents the number of the data sources, and lambda represents the weight of the K-means clustering loss.
And minimizing the total loss function by using a BP algorithm so as to achieve the aim of optimizing network parameters.
Further, the step S5 process is as follows:
defining a loss function according to the original K-means algorithm, expressed as follows:
Figure BDA0002841030730000042
wherein, J v Representing the raw K-means loss for the vth data source,
Figure BDA0002841030730000043
i sample, C, representing the v data source i Indicates the class cluster to which the ith sample belongs,
Figure BDA0002841030730000044
respectively represent the total number of samples and
Figure BDA0002841030730000045
the center of the closest cluster is the center of the cluster,
Figure BDA0002841030730000046
a two-norm representation of a vector;
and obtaining a clustering class by taking a loss function of the minimized K-means algorithm as a target, and finally obtaining a final class result in a voting mode through different data sources.
The other purpose of the invention can be achieved by adopting the following technical scheme:
an unsupervised satellite image classification device fusing multi-source data, comprising:
the acquisition unit is used for acquiring satellite image data sets of a plurality of data sources, preprocessing the data and separating sample information and category information of the data;
the coding unit is used for constructing a codec model and designing a codec suitable for the data type and the data dimension of the data from each data source;
the input unit is used for inputting satellite image data of a plurality of data sources into the coding and decoding model, the coding and decoding model extracts key features of the images, introduces K-means clustering loss into each coder and decoder, fuses the extracted key features of the images with a clustering target, and adds constraints among different data sources;
the determining unit is used for inputting the sample information of the data into a deep clustering model, training the deep clustering model through a back propagation algorithm and guiding the generation of hidden layer features;
and the output unit is used for clustering the hidden layer characteristics of each data source generated in the training process in the determination unit by using a K-means algorithm to obtain a plurality of clustering results, and finally obtaining the final category distribution result in a voting mode.
Compared with the prior art, the invention has the following advantages and effects:
1) According to the invention, an encoding and decoding framework is adopted, and original satellite image data is projected into a hidden layer space, so that the main characteristics and properties of an image can be extracted, and the influence caused by the noise of the image is reduced;
2) According to the method, the K-means loss is added into the model in a trainable mode to train the hidden layer feature representation and guide the generation of the hidden layer feature, so that the formation of a feature cluster structure is facilitated, clustering operation can be better performed, and the clustering precision is improved;
3) The invention integrates the satellite image characteristics of a plurality of data sources and applies consistency constraint, thereby extracting more abundant and comprehensive characteristics and achieving the effective unsupervised classification target.
Drawings
FIG. 1 is a specific flowchart of the unsupervised satellite image classification method with multi-source data fusion disclosed in the present invention
FIG. 2 is a model structure diagram of the unsupervised satellite image classification method fusing multisource data according to the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, 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 some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
Example one
The embodiment discloses a satellite image unsupervised classification method fusing multi-source data, and as shown in fig. 1, the unsupervised classification method comprises the following steps:
s1, acquiring satellite image data sets of a plurality of data sources, preprocessing the data and separating sample information and category information of the data;
in specific application, the satellite image data has different data sources, such as visible light images, infrared images, water vapor images, short infrared images, microwave images and the like. As shown in FIG. 2, the model has multiple data source inputs, but its categories are consistent.
S2, constructing a codec model, and designing a codec suitable for the data type and the data dimension of each data source for the data of each data source;
in specific application, the encoding and decoding processes of the encoding and decoding model are as follows:
Figure BDA0002841030730000061
Figure BDA0002841030730000062
one of the data sources, i.e., the visible light image data source, is described here as an example.
Figure BDA0002841030730000063
The ith sample representing the visible light image,
Figure BDA0002841030730000064
representing the encoder and decoder corresponding to the visible light image data source,
Figure BDA0002841030730000065
parameters of an encoder and a decoder corresponding to the visible light image data source are represented,
Figure BDA0002841030730000066
a corresponding feature representation of the ith sample representing the visible light image,
Figure BDA0002841030730000067
representing the ith sample of the reconstructed visible image data source. The encoder shown in fig. 2 uses a full connection layer, and then adds an activation function, and the decoder and the encoder are in a mirror structure.
Reconstructing visible light image sample data through the encoding and decoding operation in the step S2, wherein the reconstruction loss is as follows:
Figure BDA0002841030730000068
wherein the content of the first and second substances,
Figure BDA0002841030730000069
representing the reconstruction loss of visible light image sample data,
Figure BDA00028410307300000610
the ith sample representing the visible light image,
Figure BDA00028410307300000611
the ith sample representing the source of reconstructed visible image data,
Figure BDA00028410307300000612
representing the two-norm of the vector.
S3, inputting satellite image data of a plurality of data sources into an encoding and decoding model, extracting key features of the images by the encoding and decoding model, introducing K-means clustering loss into each encoder and decoder, fusing the key features of the extracted images with a clustering target, and adding constraints among different data sources to enable information among the features to be complementary and consistent;
in specific application, the K-means clustering loss is as follows:
Figure BDA0002841030730000071
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0002841030730000072
representing the K-means clustering loss of the visible light image data source, tr representing the trace of the matrix, H v Hidden layer feature representing a visible light image data source, F v Is a pseudo label matrix under the data source ofAn orthonormal matrix, I denotes an identity matrix.
As shown in fig. 2, with the help of K-means clustering loss, the feature representations of the samples are more discriminative, i.e., the sample features of the same class are closer in distance, and the sample features of different classes are farther in distance.
The constraints between different data sources are:
Figure BDA0002841030730000073
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0002841030730000074
denotes the loss of consistency of the v-th data source, F v A pseudo tag matrix representing the v-th data source, Y representing a consistency class matrix, I representing an identity matrix,
Figure BDA0002841030730000075
representing the Frobsenius norm of the matrix.
As shown in fig. 2, sample feature learning of each data source can be assisted by other data source information through alignment of pseudo tag matrixes of multiple data sources. With the help of other data source information, the sample characteristics of each data source are better learned, namely the sample characteristics of the same type are more compact, the sample characteristics of different types are more far away, and the subsequent type distribution is more facilitated.
S4, inputting the sample information of the data into a deep clustering model, training through a back propagation algorithm, and guiding the generation of hidden layer characteristics, so that the original sample can be reconstructed, and a cluster-shaped structure can be formed, thereby better clustering operation can be performed, and the clustering precision is improved;
in specific application, satellite images of a plurality of data sources are input into the deep clustering model, training is carried out through a back propagation algorithm, and the generation process of guiding the hidden layer features is as follows:
constructing an overall loss function according to the reconstruction loss, the K-means clustering loss and the consistency loss, and determining an overall training target of the deep clustering model through the overall loss function, wherein the overall loss function is as follows:
Figure BDA0002841030730000081
in the formula (I), the compound is shown in the specification,
Figure BDA0002841030730000082
for the reconstruction loss of the vth codec model,
Figure BDA0002841030730000083
indicating a loss of consistency for the vth data source,
Figure BDA0002841030730000084
and representing the loss of the K-means clustering of the v-th data source, m representing the number of the data sources, and lambda representing the weight of the loss of the K-means clustering.
And minimizing the total loss function by using a BP algorithm so as to achieve the aim of optimizing network parameters.
And S5, clustering the hidden layer characteristics of each data source generated in the training process in the step S4 by using a K-means algorithm to obtain a plurality of clustering results, and finally obtaining a final category distribution result in a voting mode.
In specific application, the loss function of the original K-means algorithm is expressed as follows:
Figure BDA0002841030730000085
wherein, J v Representing the raw K-means loss for the v-th data source,
Figure BDA0002841030730000086
implicit layer feature representation of the ith sample representing the vth data source, C i Indicates the cluster of classes to which the ith sample belongs,
Figure BDA0002841030730000087
respectively represent the total number of samples and
Figure BDA0002841030730000088
the center of the closest cluster is the center of the cluster,
Figure BDA0002841030730000089
representing the two-norm of the vector. And obtaining the cluster category by taking the loss function of the minimized K-means algorithm as a target. And finally, obtaining a final classification result in a voting mode of different data sources. Such as a total of three data sources: visible light images, infrared images, water vapor images. Among the three data sources, if a pair of samples belongs to the same class according to the clustering results of two or three data sources (namely more than half), the pair of samples is finally judged to belong to the same class, otherwise, the pair of samples is judged to be in different classes.
Example two
The embodiment discloses a satellite image unsupervised classification device fusing multi-source data, which comprises:
the acquisition unit is used for acquiring satellite image data sets of a plurality of data sources, preprocessing the data and separating sample information and category information of the data;
the coding unit is used for constructing a codec model and designing a codec suitable for the data type and the data dimension of the data from each data source;
the input unit is used for inputting satellite image data of a plurality of data sources into the coding and decoding model, the coding and decoding model extracts key features of the images, introduces K-means clustering loss into each coder and decoder, fuses the extracted key features of the images with a clustering target, and adds constraints among different data sources;
the determining unit is used for inputting the sample information of the data into a deep clustering model, training the deep clustering model through a back propagation algorithm and guiding the generation of hidden layer features;
and the output unit is used for clustering the hidden layer characteristics of each data source generated in the training process in the determination unit by using a K-means algorithm to obtain a plurality of clustering results, and finally obtaining the final category distribution result in a voting mode.
It should be noted that, in the foregoing embodiment, each included unit is only divided according to functional logic, but is not limited to the above division as long as the corresponding function can be implemented; in addition, specific names of the functional units are only for convenience of distinguishing from each other, and are not used for limiting the protection scope of the present invention.
In addition, those skilled in the art can understand that all or part of the steps in the foregoing embodiments may be implemented by a program to instruct related hardware, and the corresponding program may be stored in a computer readable storage medium, such as a ROM/RAM, a magnetic disk, or an optical disk.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims (4)

1. A satellite image unsupervised classification method fusing multi-source data is characterized by comprising the following steps:
s1, satellite image data sets of a plurality of data sources are obtained, and sample information and category information of the data are separated by preprocessing the data;
s2, constructing a codec model, and designing a codec suitable for the data type and the data dimension of the data source for the data of each data source;
the encoding and decoding process of the encoding and decoding model is as follows:
Figure FDA0003933299400000011
Figure FDA0003933299400000012
wherein, the first and the second end of the pipe are connected with each other,
Figure FDA0003933299400000013
the ith sample representing the vth data source,
Figure FDA0003933299400000014
which represents the v-th encoder, decoder,
Figure FDA0003933299400000015
represents the parameters of the v-th encoder and decoder,
Figure FDA0003933299400000016
a hidden layer is shown,
Figure FDA0003933299400000017
an ith sample representing the reconstructed vth data source;
reconstructing sample data through encoding and decoding operations of the encoding and decoding model, wherein in the process, the reconstruction loss of the v-th encoding and decoding model is as follows:
Figure FDA0003933299400000018
wherein the content of the first and second substances,
Figure FDA0003933299400000019
representing the reconstruction loss of the v-th codec model,
Figure FDA00039332994000000110
the ith sample representing the v-th data source,
Figure FDA00039332994000000111
represents the reconstructed secondThe ith sample of the v data sources, n represents the number of samples of the data set,
Figure FDA00039332994000000112
a two-norm representation of a vector;
s3, inputting satellite image data of a plurality of data sources into a coding and decoding model, extracting key features of the images by the coding and decoding model, introducing K-means clustering loss into each coder and decoder, fusing the key features of the extracted images with a clustering target, and adding constraints among different data sources; the process of introducing the K-means clustering loss into each codec is as follows:
define K-means clustering loss as:
Figure FDA0003933299400000021
wherein the content of the first and second substances,
Figure FDA0003933299400000022
k-means clustering loss representing the v-th data source, tr representing the trace of the matrix, H v Hidden layer representation representing the v-th data source, F v Is a pseudo tag matrix for the v-th data source, and F v Is an orthonormal matrix, I represents an identity matrix;
the constraints between different data sources are established as follows:
Figure FDA0003933299400000023
wherein the content of the first and second substances,
Figure FDA0003933299400000024
denotes the loss of consistency of the v-th data source, F v A pseudo tag matrix representing the v-th data source, Y representing a consistency class matrix, I representing an identity matrix,
Figure FDA0003933299400000025
a Frobsenius norm representing a matrix;
s4, inputting the sample information of the data into a deep clustering model, training through a back propagation algorithm, and guiding the generation of hidden layer features;
and S5, clustering the hidden layer characteristics of each data source generated in the training process in the step S4 by using a K-means algorithm to obtain a plurality of clustering results, and finally obtaining a final category distribution result in a voting mode.
2. The unsupervised satellite image classification method fusing multisource data according to claim 1, wherein the procedure of step S4 is as follows
Constructing an overall loss function according to the reconstruction loss, the K-means clustering loss and the consistency loss, and determining an overall training target of the deep clustering model through the overall loss function, wherein the overall loss function is as follows:
Figure FDA0003933299400000031
in the formula (I), the compound is shown in the specification,
Figure FDA0003933299400000032
for the reconstruction loss of the v-th codec model,
Figure FDA0003933299400000033
indicating a loss of consistency for the vth data source,
Figure FDA0003933299400000034
representing the K mean value clustering loss of the v-th data source, wherein m represents the number of the data sources, and lambda represents the weight of the K mean value clustering loss;
and minimizing the total loss function by using a BP algorithm so as to achieve the aim of optimizing network parameters.
3. The unsupervised classification method for satellite images fused with multi-source data according to claim 1, wherein the step S5 comprises the following steps:
defining a loss function according to the original K-means algorithm, expressed as follows:
Figure FDA0003933299400000035
wherein, J v Representing the raw K-means loss, C, of the v-th data source i Indicates the cluster of the class to which the ith sample belongs, n,
Figure FDA0003933299400000036
respectively represent the total number of samples and
Figure FDA0003933299400000037
the center of the closest cluster is the center of the cluster,
Figure FDA0003933299400000038
a two-norm representation of a vector;
and obtaining a clustering class by taking a loss function of the minimized K-means algorithm as a target, and finally obtaining a final class result in a voting mode through different data sources.
4. A satellite image unsupervised classification device based on the satellite image unsupervised classification method fusing multisource data according to any one of claims 1 to 3, wherein the unsupervised classification device comprises:
the acquisition unit is used for acquiring satellite image data sets of a plurality of data sources, preprocessing the data and separating sample information and category information of the data;
the coding unit is used for constructing a codec model and designing a codec suitable for the data type and the data dimension of each data source for the data of each data source;
the input unit is used for inputting satellite image data of a plurality of data sources into the coding and decoding model, the coding and decoding model extracts key features of the images, introduces K-means clustering loss into each coder and decoder, fuses the extracted key features of the images with a clustering target, and adds constraints among different data sources;
the determining unit is used for inputting the sample information of the data into a deep clustering model, training the deep clustering model through a back propagation algorithm and guiding the generation of hidden layer features;
and the output unit is used for clustering the hidden layer characteristics of each data source generated in the training process in the determination unit by using a K-means algorithm to obtain a plurality of clustering results, and finally obtaining the final category distribution result in a voting mode.
CN202011492176.5A 2020-12-17 2020-12-17 Satellite image unsupervised classification method and device fusing multi-source data Active CN112507937B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011492176.5A CN112507937B (en) 2020-12-17 2020-12-17 Satellite image unsupervised classification method and device fusing multi-source data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011492176.5A CN112507937B (en) 2020-12-17 2020-12-17 Satellite image unsupervised classification method and device fusing multi-source data

Publications (2)

Publication Number Publication Date
CN112507937A CN112507937A (en) 2021-03-16
CN112507937B true CN112507937B (en) 2023-02-10

Family

ID=74921690

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011492176.5A Active CN112507937B (en) 2020-12-17 2020-12-17 Satellite image unsupervised classification method and device fusing multi-source data

Country Status (1)

Country Link
CN (1) CN112507937B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112766425A (en) * 2021-03-25 2021-05-07 浙江师范大学 Deep missing clustering machine learning method and system based on optimal transmission
CN113240024A (en) * 2021-05-20 2021-08-10 贾晓丰 Data classification method and device based on deep learning clustering algorithm
CN116578874B (en) * 2023-07-10 2023-11-21 武汉能钠智能装备技术股份有限公司四川省成都市分公司 Satellite signal attribute appraising method and device based on network protocol

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020232905A1 (en) * 2019-05-20 2020-11-26 平安科技（深圳）有限公司 Superobject information-based remote sensing image target extraction method, device, electronic apparatus, and medium

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10373056B1 (en) * 2018-01-25 2019-08-06 SparkCognition, Inc. Unsupervised model building for clustering and anomaly detection
CN109919189A (en) * 2019-01-29 2019-06-21 华南理工大学 A kind of depth K mean cluster method towards time series data
CN109919204B (en) * 2019-02-23 2023-06-23 华南理工大学 Noise image-oriented deep learning clustering method
CN111259979B (en) * 2020-02-10 2022-11-18 大连理工大学 Deep semi-supervised image clustering method based on label self-adaptive strategy
CN111832627B (en) * 2020-06-19 2022-08-05 华中科技大学 Image classification model training method, classification method and system for suppressing label noise

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020232905A1 (en) * 2019-05-20 2020-11-26 平安科技（深圳）有限公司 Superobject information-based remote sensing image target extraction method, device, electronic apparatus, and medium

Also Published As

Publication number Publication date
CN112507937A (en) 2021-03-16

Similar Documents

Publication Publication Date Title
CN112507937B (en) Satellite image unsupervised classification method and device fusing multi-source data
CN111581405B (en) Cross-modal generalization zero sample retrieval method for generating confrontation network based on dual learning
CN115033670A (en) Cross-modal image-text retrieval method with multi-granularity feature fusion
CN113609285B (en) Multimode text abstract system based on dependency gating fusion mechanism
CN111460883B (en) Video behavior automatic description method based on deep reinforcement learning
CN113889179A (en) Compound-protein interaction prediction method based on multi-view deep learning
CN116030018A (en) Incoming material qualification inspection system and method for door processing
CN116304984A (en) Multi-modal intention recognition method and system based on contrast learning
CN115658934A (en) Image-text cross-modal retrieval method based on multi-class attention mechanism
CN117217368A (en) Training method, device, equipment, medium and program product of prediction model
CN112434514A (en) Multi-granularity multi-channel neural network based semantic matching method and device and computer equipment
CN113537322B (en) Zero sample visual classification method for cross-modal semantic enhancement generation countermeasure network
CN114595700A (en) Zero-pronoun and chapter information fused Hanyue neural machine translation method
CN113920379A (en) Zero sample image classification method based on knowledge assistance
CN117036833A (en) Video classification method, apparatus, device and computer readable storage medium
CN113076421A (en) Social noise text entity relation extraction optimization method and system
CN117521641A (en) Automatic text proofreading system and method based on natural language processing
CN116775929A (en) Cross-modal retrieval method based on multi-level fine granularity semantic alignment
CN116186350A (en) Power transmission line engineering searching method and device based on knowledge graph and topic text
CN116775855A (en) Automatic TextRank Chinese abstract generation method based on Bi-LSTM
Zhang et al. Wikitag: Wikipedia-based knowledge embeddings towards improved acoustic event classification
CN115809438A (en) Multi-modal emotion analysis method, system, device and storage medium
CN114692604A (en) Deep learning-based aspect-level emotion classification method
CN115272857B (en) Multi-source remote sensing image target identification method based on attention mechanism
Osaid et al. A Survey On Image Captioning

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