CN114219963A - Multi-scale capsule network remote sensing ground feature classification method and system guided by geoscience knowledge - Google Patents

Multi-scale capsule network remote sensing ground feature classification method and system guided by geoscience knowledge Download PDF

Info

Publication number
CN114219963A
CN114219963A CN202111663292.3A CN202111663292A CN114219963A CN 114219963 A CN114219963 A CN 114219963A CN 202111663292 A CN202111663292 A CN 202111663292A CN 114219963 A CN114219963 A CN 114219963A
Authority
CN
China
Prior art keywords
remote sensing
knowledge
scale
model
ontology
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.)
Pending
Application number
CN202111663292.3A
Other languages
Chinese (zh)
Inventor
邵振峰
吴文福
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.)
Wuhan University WHU
Original Assignee
Wuhan University WHU
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 Wuhan University WHU filed Critical Wuhan University WHU
Priority to CN202111663292.3A priority Critical patent/CN114219963A/en
Publication of CN114219963A publication Critical patent/CN114219963A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a method and a system for classifying remote sensing ground features of a multiscale capsule network guided by geological knowledge, which comprises the steps of preprocessing a remote sensing image and labeling ground feature labels to construct a remote sensing ground feature classification sample library; establishing a remote sensing interpretation ontology of a ground surface covering entity according to the main land covering type of a city, and then completing the construction of a geoscience knowledge graph through an instantiation ontology and attributes thereof on the basis of the remote sensing interpretation ontology; carrying out explicit symbolic representation on the geological knowledge contained in the geological knowledge map, and mining implicit geological knowledge contained in the geological knowledge map; constructing a multi-scale convolution capsule network as a deep learning network, inputting the processed image as sample data into the deep learning network, and constructing model loss constraint on the basis of geoscience knowledge; and training the constructed multi-scale convolution capsule model to obtain a training model, and testing the training model to realize the classification of the remote sensing ground objects. The invention can improve the precision and the automation degree of remote sensing ground feature classification.

Description

Multi-scale capsule network remote sensing ground feature classification method and system guided by geoscience knowledge
Technical Field
The invention belongs to the field of remote sensing, and particularly relates to a geo-knowledge-guided multi-scale capsule network remote sensing ground feature classification scheme.
Background
The current global urban sustainable development faces serious challenges, and urban diseases such as urban waterlogging, urban heat island, urban ecological function degradation and the like become global problems. The technology support can be provided for the sustainable development of the city by extracting the high-resolution information of the typical elements of the city by using the remote sensing technology. However, due to the influence of cultural differences, human activities, geographical positions and the like, serious phenomena of 'same-object different spectrum' and 'same-spectrum foreign matter' are encountered when urban typical element remote sensing information extraction is carried out on a global scale. In addition, the urban high-resolution images also have information loss caused by building shadows, tree occlusion, moving target interference and the like. The accuracy and the automation degree of the urban remote sensing information extraction are low due to the problems, and the increasingly fine management requirements of the cities cannot be met.
The method mainly studies how to obtain high-precision ground object type information from the high-resolution remote sensing image under the dual drive of data and knowledge. With the development of computer technology, the technology of computer-aided remote sensing image ground feature classification has gradually replaced manual interpretation methods, such as scale-invariant feature transformation, directional gradient histogram, and the like. In addition, traditional machine learning methods such as decision trees, random forests, support vector machines, hidden Markov random fields and the like are also applied to remote sensing image ground feature classification, and the ground feature classification effect is improved. However, these conventional methods require manual design and feature extraction, and only several layers of linear or nonlinear processing can be performed on the features, so that the image features cannot be sufficiently learned, the terrain expression capability of the high-resolution remote sensing image in a complex urban scene is limited, under-fitting or over-fitting phenomena easily occur in other related tasks, and real automation cannot be achieved.
In recent years, the deep learning technology is vigorously developed and rapidly brought to the world, and is also introduced into the remote sensing field to classify the ground object targets of the remote sensing images, and gradually becomes the mainstream method for interpreting the remote sensing images. In remote sensing image surface feature classification, the existing deep neural network mainly takes a convolutional neural network as a basic network, and various improvements are carried out according to the characteristics of remote sensing images, such as constructing a DCNNs semantic segmentation method based on the idea of a full convolutional neural network to carry out remote sensing image classification, modifying an activation function of the DCNNs semantic segmentation method based on the full convolutional neural network, and the like. The remote sensing ground feature classification method based on deep learning has strong feature learning and expression capability, does not need manual feature design and selection, and can realize automatic end-to-end output results. However, deep learning is a data-driven method, highly dependent on a large number of data samples, reversely optimizes network parameters by reducing loss between output results and labels, does not fully utilize the geoscience knowledge in the remote sensing field, has poor reliability and interpretability of output results of the remote sensing image ground feature classification method based on deep learning, and belongs to a black box operation. In addition, the remote sensing image data contains abundant information such as space, spectrum, time and the like, and the traditional convolutional neural network abandons the relation between the relative position and the spatial hierarchy among the ground objects in the process of processing the remote sensing image, so that the classification precision of the remote sensing ground objects is hindered from being further improved.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a geography knowledge guided multi-scale capsule network remote sensing geography classification scheme, which effectively utilizes the geography knowledge to establish a knowledge constraint mechanism to enable a deep learning model to have certain interpretability and good generalization capability.
In order to achieve the purpose, the technical scheme adopted by the invention is a geo-knowledge-guided multi-scale capsule network remote sensing land feature classification method, which comprises the following steps,
step a, constructing a remote sensing ground object classification sample library by preprocessing a remote sensing image and labeling a ground object label;
b, establishing a remote sensing interpretation ontology of the land cover entities of buildings, roads, bare soil, green lands and water bodies 5 according to the main land cover types of the cities, and then completing the construction of the geography knowledge map by instantiating the ontology and the attributes thereof on the basis of the remote sensing interpretation ontology;
step c, explicit symbolization representation is carried out on the geological knowledge contained in the geological knowledge map, and implicit geological knowledge contained in the geological knowledge map is mined;
d, constructing a multi-scale convolution capsule network as a deep learning network, inputting the processed image obtained in the step a into the deep learning network as sample data, constructing model loss constraint on the basis of the geological knowledge obtained in the step c, and guiding the training of the multi-scale capsule network;
e, extracting and classifying urban multi-source remote sensing information, wherein the method comprises the steps of training the constructed multi-scale convolution capsule model under the constraint of geological knowledge based on the loss function established in the step d to obtain a training model; and testing the test data according to the model obtained by training, and after obtaining a qualified model, realizing the classification of the remote sensing ground features.
In step b, a network ontology language OWL is used for describing a remote sensing interpretation ontology and the attribute relation thereof, the surface feature target in the sample set is used as an example object of the interpretation ontology, the attribute of an information instantiation object corresponding to the surface feature target, such as the category, the shape feature and the space semantic, is extracted, and the surface knowledge graph is constructed.
Furthermore, in step e, the model test results are evaluated using four indexes based on the overall accuracy of the confusion matrix, the Kappa coefficient, the user accuracy, and the producer accuracy.
The invention provides a geography knowledge guided multi-scale capsule network remote sensing geography classification system which is used for realizing the geography knowledge guided multi-scale capsule network remote sensing geography classification method.
And, including the following modules,
the remote sensing ground object classification system comprises a first module, a second module and a third module, wherein the first module is used for preprocessing a remote sensing image and labeling a ground object label to construct a remote sensing ground object classification sample library;
the second module is used for establishing a remote sensing interpretation ontology of a building, a road, bare soil, a green land and a water body 5 surface covering entity according to the main land covering type of the city, and then completing the construction of a geoscience knowledge graph through an instantiation ontology and attributes thereof on the basis of the remote sensing interpretation ontology;
the third module is used for carrying out explicit symbolization representation on the geological knowledge contained in the geological knowledge map and mining implicit geological knowledge contained in the geological knowledge map;
a fourth module, configured to construct a multi-scale convolution capsule network as a deep learning network, input the processed image obtained in step a into the deep learning network as sample data, construct a model loss constraint on the basis of the geological knowledge obtained in step c, and guide training of the multi-scale capsule network;
a fifth module for extracting and classifying the urban multi-source remote sensing information, which comprises training the constructed multi-scale convolution capsule model under the constraint of the geological knowledge based on the loss function established in the step d to obtain a training model; and testing the test data according to the model obtained by training, and after obtaining a qualified model, realizing the classification of the remote sensing ground features.
Or, the system comprises a processor and a memory, wherein the memory is used for storing program instructions, and the processor is used for calling the stored instructions in the memory to execute the geography knowledge guided multi-scale capsule network remote sensing geography classification method as described above.
Or, comprises a readable storage medium, on which a computer program is stored, which when executed, implements a geo-knowledge-guided multi-scale capsule network remote sensing ground object classification method as described above.
The invention has the beneficial effects that: the invention provides a ground-knowledge-guided multi-scale capsule network remote sensing ground feature classification scheme, which fully considers the structure and space distribution information among ground feature targets, effectively utilizes the ground knowledge in the field of geography, establishes a knowledge constraint mechanism, guides the autonomous training of a model, enables a deep learning model to have certain interpretability and good generalization capability, and further improves the precision and the automation degree of remote sensing ground feature classification.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention.
Fig. 2 is a network structure diagram according to an embodiment of the present invention.
Detailed Description
The technical solution of the present invention is further explained with reference to the drawings and the embodiments.
The invention discloses a geo-knowledge-guided multi-scale capsule network remote sensing terrain classification scheme, which comprises the steps of constructing an urban geo-knowledge map; through the expression learning of the urban geography knowledge map, the domain knowledge of the geographic concept-geographic entity-geographic relation contained in the geography knowledge map is expressed and learned into a low-dimensional vector containing semantic information; according to the characteristics of the remote sensing image and the existing scale effect, a multi-scale three-dimensional convolution capsule network remote sensing information extraction model is constructed, a knowledge vector learned from a geographical knowledge map is embedded into the model, and an effective domain knowledge constraint mechanism is established to guide the learning of the model, so that the model has certain interpretability and good global generalization capability.
Through the mode, the method can uniformly learn the space-time spectrum characteristics of the urban multi-source heterogeneous data, effectively utilize the geoscience knowledge, establish an effective knowledge constraint mechanism, enable the deep learning model to have certain interpretability and good generalization capability, and improve the extraction precision of urban remote sensing information.
Referring to fig. 1, the method for classifying the terrain remotely sensed by the multi-scale capsule network guided by the geological knowledge provided by the embodiment of the invention comprises the following steps:
a. constructing a remote sensing sample library: and constructing a remote sensing ground object classification sample library by preprocessing the remote sensing image and labeling ground object labels.
During specific implementation, a series of preprocessing such as geometric correction and radiation correction can be firstly carried out on the obtained remote sensing data, then the corresponding existing historical interpretation data such as the general survey of geographic national conditions, the multi-source survey and monitoring results and the third national state survey are collected, and the existing historical interpretation data is matched to select and label the ground feature samples so as to construct the remote sensing ground feature classification sample library. The step can be performed in advance, and the existing remote sensing sample library is directly input during the process.
b. Construction of a geo-knowledge map: building a remote sensing interpretation ontology of 5 surface covering entities of buildings, roads, bare soil, greenbelts and water bodies according to main land covering types of cities, and then completing the construction of a geo-knowledge map through instantiation ontology and attributes thereof on the basis of the remote sensing interpretation ontology.
In specific implementation, the urban geography knowledge map can be constructed from knowledge carriers such as urban geographic positions, basic geographic information data, crowd-sourced geographic information, open network texts and the like through knowledge extraction, knowledge fusion and knowledge reasoning. Preferably, a network ontology language OWL is used for describing a remote sensing interpretation ontology and an attribute relation thereof, a surface feature target in a sample set is used as an example object of the interpretation ontology, attributes of information instantiation objects such as the category, shape characteristics and space semantics of the corresponding surface feature target are extracted, and a geoscience knowledge graph is constructed.
In the embodiment, according to the main surface feature types of a common city, a remote sensing interpretation ontology of a building, a road, bare soil, a green land and a water body 5 surface covering entity is constructed, and an ontology network language OWL is used for describing the ontology and the attribute relation thereof. In the field of remote sensing, the attribute relationship of the remote sensing interpretation ontology mainly comprises upper and lower layer attributes such as subordinate attributes, spatial relationship attributes such as adjacent, surrounding and azimuth attributes and statistical attributes such as majority types. The images in the remote sensing ground feature classification sample library are labeled in advance and contain information such as ground feature target category, shape characteristics, space semantics and the like. Therefore, when the geography knowledge graph is constructed, the geography target is used as an instance object of the ontology, and the attributes of the information instantiation object such as the corresponding geography target category, the shape feature, the space semantic and the like are extracted to construct the geography knowledge graph.
c. Extraction of geographical knowledge: in specific implementation, the Semantic Web Rule Language (SWRL) can be used for explicitly symbolizing the geoscience knowledge contained in the geoscience knowledge graph, and mining the implicit geoscience knowledge contained in the geoscience knowledge graph in a logical reasoning mode.
The embodiment expresses spatial symbiosis knowledge implied in the geoscience knowledge graph through symbiosis condition probability on the basis of the geoscience knowledge graph constructed in the step b, and the concrete method is as follows:
assume that there are two ground object classes, respectively CiAnd CjRepresenting, statistically, the terrain category C in the mapiProbability of occurrence P (C)i) Then counting the object adjacent object class appearance CjProbability of class P (C)i,Cj) And finally, calculating at C according to formula (1)iC appears in the neighborhood of the class target under the condition that the class target appearsiProbability P (C) of class objectj|Ci)：
Figure BDA0003450752050000051
d. And (c) constructing a multi-scale convolution capsule network as a deep learning network, constructing the multi-scale convolution capsule network as the deep learning network, inputting the processed image obtained in the step a into the deep learning network as sample data, constructing model loss constraint on the basis of the geological knowledge obtained in the step c, guiding the training of the multi-scale capsule network, and obtaining a training model.
1) It is considered that the conventional convolutional neural network CNN ignores the relationship of the position and spatial hierarchy between the image pixels, and such information is advantageous for the classification of the ground object target. In addition, on the remote sensing image, the size of the ground object target is different, an obvious multi-scale effect exists, and the characteristics learned by the model are too single due to the use of the convolution layer with a fixed size, so that the training of the model and the classification of the ground object are not facilitated. Therefore, the multi-scale convolution capsule network is adopted as a deep learning network, and referring to fig. 2, the deep learning network mainly comprises a multi-scale convolution module, a capsule module and a full connection layer.
In specific implementation, the space-time spectrum characteristics of the remote sensing image can be extracted by utilizing a multi-scale three-dimensional convolution kernel according to the characteristics and the existing scale effect of the remote sensing image; and then, extracting deeper features by utilizing a capsule network on the basis of the shallow features extracted by the convolutional neural network structure, and considering the structural and spatial distribution information among the urban elements.
A multi-scale convolution module: in the multi-scale convolution module, an embodiment adopts three convolution kernels conv1-3 with the sizes of 9 × 9, 7 × 7 and 5 × 5 respectively, which are respectively marked as k9, k7 and k5, the step size is set to 1, the number of channels is set to 16, the convolution kernels conv1-3 are used for extracting multi-scale features in a remote sensing image, normalized batch processing (BatchNorm) and a ReLu activation function are respectively carried out after three convolution operations, the nonlinear expression capability of the model is enhanced, and finally the multi-scale features are connected and input into the capsule module through a Concatenate operation, and the operation of the convolution module can be represented by the following formula:
Figure BDA0003450752050000052
in the above formula, X is the input remote sensing image, Convi×i(. cndot.) is a convolution operation with convolution kernel size of i x i, BN (. cndot.) is normalized batch, ReLu (. cndot.) is an activation function, Cont (. cndot.) is a Concatenate operation, layer1Representing the output of the first layer of the network model, Featurei×iI is 5,7, 9.
A capsule module: the capsule module is composed of an initial capsule layer and a digital capsule layer. In the initial capsule layer, there are n capsules u1, u2, … un, where n is 32 in the embodiment, each capsule is a 1 × 8 vector, which is equivalent to performing convolution operation on the feature map input by the multi-scale volume block by using 8 sets of convolution kernels with the size of 7 × 7 × 32 and the step size of 1, and the obtained feature map is used as the input of the digital capsule layer. The output of the digital capsule layer can be obtained according to the dynamic routing algorithm, the number of the capsules is equal to the final classification number, the number of the capsules is set to be 5, each capsule v1, v2 and … v5 is a 16-dimensional vector, and the final output of the digital capsule layer is 5 multiplied by 16. The capsule network layers are connected through a dynamic routing algorithm, which specifically comprises the following steps:
suppose the output of the ith capsule of the (l-1) th layer is
Figure BDA0003450752050000061
Then, the strength of the connection of the ith capsule of the (l-1) th layer to the jth capsule of the l layer
Figure BDA0003450752050000062
Comprises the following steps:
Figure BDA0003450752050000063
wherein the content of the first and second substances,
Figure BDA0003450752050000064
is a weight matrix.
Input vector for jth capsule at l layer
Figure BDA0003450752050000065
The strength of k capsules in the (l-1) th layer
Figure BDA0003450752050000066
And coefficient of coupling
Figure BDA0003450752050000067
The weighted sum of (a) can be represented by the following formula:
Figure BDA0003450752050000068
if the l-th layer has p capsules, the coupling coefficient is
Figure BDA0003450752050000069
The calculation formula of (2) is as follows:
Figure BDA00034507520500000610
wherein the content of the first and second substances,
Figure BDA00034507520500000611
for similarity score, exp (-) is an exponential function based on a natural constant eAnd (4) counting.
Finally, the output vector of the jth capsule of the ith layer
Figure BDA00034507520500000612
Comprises the following steps:
Figure BDA00034507520500000613
where | l | · | |, represents a norm.
During the iterative process of training the network, the change is carried out
Figure BDA00034507520500000614
To continuously update the coupling coefficient, the updating process can be represented by the following equation:
Figure BDA00034507520500000615
full connection layer: the last layer of the model, which is the fully-connected layer, acts as a classifier and can be represented by the following formula:
y＝Wx (8)
where W is the weight matrix and x is the input, where y is the characteristic of the digital capsule layer output.
2) Establishing a loss function constrained by geoscience knowledge:
constructing a loss function constrained by the geoscience knowledge on the basis of the geoscience knowledge graph constructed in the steps b and c, and guiding the training of the multi-scale convolution capsule network model constructed in the step 1). Thus, the loss function of the present patent design includes edge loss LmarginGeoscience knowledge constraint loss Lknowledge：
Figure BDA0003450752050000071
Wherein, TkFor the indicator, T is when the input vector belongs to class kk1, otherwise Tk＝0。m-And m+Lower and upper penalties, respectively, typically taken as m+＝0.9，m-＝0.1。||vkI is the norm of the output vector of the fully-connected layer, when category k exists, | vk||≥m+Else, | | vk||≤m-. λ is an adjustment coefficient used to adjust the ratio of the effect of the correct result to the incorrect result on the final function value. max () is the maximum function and n is the number of classification classes.
In the geoscience knowledge constraint, the class confidence vector Y from the output of the modeliTaking the maximum value in the confidence vector and the corresponding maximum value sequence number (i.e. class number) as the classification confidence y of the targetjAnd a classification category k. Then, taking 7 × 7 neighborhoods with the target as the center, obtaining the class confidence of the target in the current neighborhood space by the weighted summation of the classification confidence of the target in the neighborhoods and the space symbiosis conditional probability (formula 1), and using HiRepresenting and then calculating a confidence H of the category based on the spatial distributioniAnd the true category confidence vector YiThe loss therebetween can be represented by the following formula:
Figure BDA0003450752050000072
wherein N is the total number of samples, i is the sample target (i is more than or equal to 1 and less than or equal to N), miIs the number of targets in the neighborhood, j denotes the targets in the neighborhood, C is the number of classes, q denotes the class (1 ≦ q ≦ C), P (Cq|Ck) Is the spatial symbiotic conditional probability.
Model final loss function LtotalComprises the following steps:
Ltotal＝Lmargin+βLknowledge (11)
wherein beta is a hyperparameter used to adjust LmarginAnd LknowledgeThe weight of (c).
e. Urban multi-source remote sensing information extraction and classification: on the basis of the sample set constructed in the step a, training the constructed multi-scale convolution capsule model under the constraint of the geological knowledge based on the loss function established in the step d to obtain a training model; and testing the test data according to the model obtained by training, and evaluating the test result of the model by using four indexes of overall precision, Kappa coefficient, user precision and producer precision based on the confusion matrix. And after the qualified model is obtained, the method can be used for realizing the classification of the remote sensing ground object.
In specific implementation, a person skilled in the art can implement the automatic operation process by using a computer software technology, and a system device for implementing the method, such as a computer-readable storage medium storing a corresponding computer program according to the technical solution of the present invention and a computer device including a corresponding computer program for operating the computer program, should also be within the scope of the present invention.
In some possible embodiments, a geo-knowledge-guided multi-scale capsule network remote sensing ground object classification system is provided, which comprises the following modules,
the remote sensing ground object classification system comprises a first module, a second module and a third module, wherein the first module is used for preprocessing a remote sensing image and labeling a ground object label to construct a remote sensing ground object classification sample library;
the second module is used for establishing a remote sensing interpretation ontology of a building, a road, bare soil, a green land and a water body 5 surface covering entity according to the main land covering type of the city, and then completing the construction of a geoscience knowledge graph through an instantiation ontology and attributes thereof on the basis of the remote sensing interpretation ontology;
the third module is used for carrying out explicit symbolization representation on the geological knowledge contained in the geological knowledge map and mining implicit geological knowledge contained in the geological knowledge map;
the fourth module is used for constructing a multi-scale convolution capsule network as a deep learning network, inputting the processed image obtained by the first module into the deep learning network as sample data, constructing model loss constraint on the basis of the geoscience knowledge obtained by the third module, and guiding the training of the multi-scale capsule network;
the fifth module is used for extracting and classifying the urban multi-source remote sensing information, and comprises a loss function established based on the fourth module, and a multi-scale convolution capsule model established under the constraint of geological knowledge is trained to obtain a training model; and testing the test data according to the model obtained by training, and after obtaining a qualified model, realizing the classification of the remote sensing ground features.
In some possible embodiments, a geo-knowledge-guided multi-scale capsule network remote sensing terrain classification system is provided, which comprises a processor and a memory, wherein the memory is used for storing program instructions, and the processor is used for calling the stored instructions in the memory to execute the geo-knowledge-guided multi-scale capsule network remote sensing terrain classification method.
In some possible embodiments, a geo-knowledge-guided multi-scale capsule network remote sensing ground object classification system is provided, which includes a readable storage medium, on which a computer program is stored, and when the computer program is executed, the method for geo-knowledge-guided multi-scale capsule network remote sensing ground object classification is implemented.
The specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.

Claims (7)

1. A geo-knowledge-guided multi-scale capsule network remote sensing land feature classification method is characterized by comprising the following steps: the method comprises the following steps of a, preprocessing a remote sensing image and labeling a ground feature label to construct a remote sensing ground feature classification sample library;
b, establishing a remote sensing interpretation ontology of the land cover entities of buildings, roads, bare soil, green lands and water bodies 5 according to the main land cover types of the cities, and then completing the construction of the geography knowledge map by instantiating the ontology and the attributes thereof on the basis of the remote sensing interpretation ontology;
step c, explicit symbolization representation is carried out on the geological knowledge contained in the geological knowledge map, and implicit geological knowledge contained in the geological knowledge map is mined;
d, constructing a multi-scale convolution capsule network as a deep learning network, inputting the processed image obtained in the step a into the deep learning network as sample data, constructing model loss constraint on the basis of the geological knowledge obtained in the step c, and guiding the training of the multi-scale capsule network;
e, extracting and classifying urban multi-source remote sensing information, wherein the method comprises the steps of training the constructed multi-scale convolution capsule model under the constraint of geological knowledge based on the loss function established in the step d to obtain a training model; and testing the test data according to the model obtained by training, and after obtaining a qualified model, realizing the classification of the remote sensing ground features.
2. The geo-knowledge-guided multi-scale capsule network remote sensing terrain classification method according to claim 1, characterized in that: in the step b, a network ontology language OWL is used for describing a remote sensing interpretation ontology and the attribute relation thereof, the surface feature target in the sample set is used as an example object of the interpretation ontology, the attributes of information instantiation objects such as the category, shape characteristics, space semantics and the like of the corresponding surface feature target are extracted, and the surface science knowledge map is constructed.
3. The geo-knowledge-guided multi-scale capsule network remote sensing terrain classification method according to claim 1 or 2, characterized in that: in step e, the model test results are evaluated using four indexes based on the overall accuracy of the confusion matrix, the Kappa coefficient, the user accuracy and the producer accuracy.
4. A geography knowledge guided multi-scale capsule network remote sensing geography classification system is characterized in that: the remote sensing land feature classification method of the multi-scale capsule network for realizing the guidance of the geological knowledge according to any one of claims 1 to 3.
5. The geo-knowledge-guided multi-scale capsule network remote sensing terrain classification system of claim 4, wherein: comprises the following modules which are used for realizing the functions of the system,
the remote sensing ground object classification system comprises a first module, a second module and a third module, wherein the first module is used for preprocessing a remote sensing image and labeling a ground object label to construct a remote sensing ground object classification sample library;
the second module is used for establishing a remote sensing interpretation ontology of a building, a road, bare soil, a green land and a water body 5 surface covering entity according to the main land covering type of the city, and then completing the construction of a geoscience knowledge graph through an instantiation ontology and attributes thereof on the basis of the remote sensing interpretation ontology;
the third module is used for carrying out explicit symbolization representation on the geological knowledge contained in the geological knowledge map and mining implicit geological knowledge contained in the geological knowledge map;
the fourth module is used for constructing a multi-scale convolution capsule network as a deep learning network, inputting the processed image obtained by the first module into the deep learning network as sample data, constructing model loss constraint on the basis of the geoscience knowledge obtained by the third module, and guiding the training of the multi-scale capsule network;
the fifth module is used for extracting and classifying the urban multi-source remote sensing information, and comprises a loss function established based on the fourth module, and a multi-scale convolution capsule model established under the constraint of geological knowledge is trained to obtain a training model; and testing the test data according to the model obtained by training, and after obtaining a qualified model, realizing the classification of the remote sensing ground features.
6. The geo-knowledge-guided multi-scale capsule network remote sensing terrain classification system of claim 4, wherein: comprising a processor and a memory, wherein the memory is used for storing program instructions, and the processor is used for calling the stored instructions in the memory to execute the geography classification method of the multi-scale capsule network remote sensing guided by the geography knowledge according to any one of claims 1-3.
7. The geo-knowledge-guided multi-scale capsule network remote sensing terrain classification system of claim 4, wherein: comprising a readable storage medium having stored thereon a computer program which, when executed, implements a geo-knowledge-guided multi-scale capsule network remote sensing geo-feature classification method as claimed in any one of claims 1 to 3.
CN202111663292.3A 2021-12-31 2021-12-31 Multi-scale capsule network remote sensing ground feature classification method and system guided by geoscience knowledge Pending CN114219963A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111663292.3A CN114219963A (en) 2021-12-31 2021-12-31 Multi-scale capsule network remote sensing ground feature classification method and system guided by geoscience knowledge

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111663292.3A CN114219963A (en) 2021-12-31 2021-12-31 Multi-scale capsule network remote sensing ground feature classification method and system guided by geoscience knowledge

Publications (1)

Publication Number Publication Date
CN114219963A true CN114219963A (en) 2022-03-22

Family

ID=80707334

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111663292.3A Pending CN114219963A (en) 2021-12-31 2021-12-31 Multi-scale capsule network remote sensing ground feature classification method and system guided by geoscience knowledge

Country Status (1)

Country Link
CN (1) CN114219963A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114723999A (en) * 2022-05-24 2022-07-08 航天宏图信息技术股份有限公司 Garbage identification method and device based on unmanned aerial vehicle orthographic image
CN116385841A (en) * 2023-02-28 2023-07-04 南京航空航天大学 Multi-mode ground object target identification method based on knowledge graph
CN117151222A (en) * 2023-09-15 2023-12-01 大连理工大学 Domain knowledge guided emergency case entity attribute and relation extraction method thereof, electronic equipment and storage medium
CN117151222B (en) * 2023-09-15 2024-05-24 大连理工大学 Domain knowledge guided emergency case entity attribute and relation extraction method thereof, electronic equipment and storage medium

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114723999A (en) * 2022-05-24 2022-07-08 航天宏图信息技术股份有限公司 Garbage identification method and device based on unmanned aerial vehicle orthographic image
CN114723999B (en) * 2022-05-24 2022-08-23 航天宏图信息技术股份有限公司 Garbage identification method and device based on unmanned aerial vehicle orthographic image
CN116385841A (en) * 2023-02-28 2023-07-04 南京航空航天大学 Multi-mode ground object target identification method based on knowledge graph
CN116385841B (en) * 2023-02-28 2023-11-21 南京航空航天大学 Multi-mode ground object target identification method based on knowledge graph
CN117151222A (en) * 2023-09-15 2023-12-01 大连理工大学 Domain knowledge guided emergency case entity attribute and relation extraction method thereof, electronic equipment and storage medium
CN117151222B (en) * 2023-09-15 2024-05-24 大连理工大学 Domain knowledge guided emergency case entity attribute and relation extraction method thereof, electronic equipment and storage medium

Similar Documents

Publication Publication Date Title
CN111191736B (en) Hyperspectral image classification method based on depth feature cross fusion
Zhang et al. A feature difference convolutional neural network-based change detection method
Zhou et al. BOMSC-Net: Boundary optimization and multi-scale context awareness based building extraction from high-resolution remote sensing imagery
US20230169623A1 (en) Synthetic aperture radar (sar) image target detection method
Jin et al. Object-oriented method combined with deep convolutional neural networks for land-use-type classification of remote sensing images
CN106909924A (en) A kind of remote sensing image method for quickly retrieving based on depth conspicuousness
Zhang et al. Unsupervised difference representation learning for detecting multiple types of changes in multitemporal remote sensing images
Ge et al. Geoscience-aware deep learning: A new paradigm for remote sensing
Dai et al. RADANet: Road augmented deformable attention network for road extraction from complex high-resolution remote-sensing images
CN114219963A (en) Multi-scale capsule network remote sensing ground feature classification method and system guided by geoscience knowledge
CN109859209B (en) Remote sensing image segmentation method and device, storage medium and server
CN108537121A (en) The adaptive remote sensing scene classification method of environment parament and image information fusion
CN110991284B (en) Optical remote sensing image statement description generation method based on scene pre-classification
CN110334656A (en) Multi-source Remote Sensing Images Clean water withdraw method and device based on information source probability weight
Liu et al. Survey of road extraction methods in remote sensing images based on deep learning
Singh et al. Detection and mapping of agriculture seasonal variations with deep learning–based change detection using Sentinel-2 data
CN105809200A (en) Biologically-inspired image meaning information autonomous extraction method and device
CN111598140A (en) Remote sensing image classification method based on capsule network
Deepan et al. Comparative analysis of scene classification methods for remotely sensed images using various convolutional neural network
Lu et al. Crop classification in high-resolution remote sensing images based on multi-scale feature fusion semantic segmentation model
Shi et al. Intelligent classification of land cover types in open-pit mine area using object-oriented method and multitask learning
Ahmed et al. DeepRoadNet: A deep residual based segmentation network for road map detection from remote aerial image
Shi et al. Deep change feature analysis network for observing changes of land use or natural environment
Alshahrani et al. Optimal Deep Convolutional Neural Network for Vehicle Detection in Remote Sensing Images.
Su et al. Mask-vit: an object mask embedding in vision transformer for fine-grained visual classification

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