CN111538776B - Multilayer cognitive constraint high-dimensional geographic spatial data focusing visualization method - Google Patents
Multilayer cognitive constraint high-dimensional geographic spatial data focusing visualization method Download PDFInfo
- Publication number
- CN111538776B CN111538776B CN202010197344.1A CN202010197344A CN111538776B CN 111538776 B CN111538776 B CN 111538776B CN 202010197344 A CN202010197344 A CN 202010197344A CN 111538776 B CN111538776 B CN 111538776B
- Authority
- CN
- China
- Prior art keywords
- data
- feature
- layer
- cognitive
- information
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/26—Visual data mining; Browsing structured data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion 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
- G06N3/04—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
- G06N3/088—Non-supervised learning, e.g. competitive learning
Abstract
The invention relates to a multilayer cognitive constraint high-dimensional geographic space data focusing visualization method, which comprises the following steps: a) the method comprises the steps that high-dimensional spatial data with different expression characteristics are oriented, and based on semantic information and an autonomous structure of the high-dimensional spatial data, focusing attributes, spatial characteristics and time characteristics are used as cognitive characteristics and are used for constructing a constraint item set of a semi-supervised mechanism dimension reduction model; b) constructing a deep self-encoder of an adaptive semi-supervised mechanism depending on cognitive features, and reducing the dimension of high-dimensional data by using a model to obtain an adaptive feature matrix set; c) according to the cognitive process of human beings from superficial visual information to deep emotional information on a visual interface, focusing the perception and attention of the human beings to establish a feature matrix normalization mapping rule, and summarizing a feature mapping rule set; d) and outputting a knowledge focused high-dimensional visual expression. The invention can overcome the problems of low accuracy, poor stability, high redundancy and the like easily caused by the conventional dimension reduction processing method.
Description
Technical Field
The invention belongs to the technical field of geospatial data processing, and particularly relates to a multi-layer cognitive constraint high-dimensional geospatial data focusing visualization method.
Background
The rapid development of the fields of earth observation systems, global change simulation and the like accumulates massive space-time data, presents the characteristics of multidimensional and multiattribute, irregular structure and the like, and effectively extracts and analyzes the characteristics of irregular space-time data, thereby being a hotspot in the technical field of geospatial data processing. Geospatial data is a digital representation of a geographic object, is generally suitable for describing continuously changing objects or phenomena, is mostly used for expressing environmental data distribution, statistical distribution of characteristic indexes, probability distribution of geoscience phenomena and the like, and is obtained by adding time attributes to the characteristics so as to increase the complexity of data organization and expression.
Because the direct processing of high-dimensional data faces the problems of "dimension disaster", "algorithm failure" and the like, research has been carried out to provide a series of effective feature learning methods for these problems, such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and the like, but these methods still have various problems in the scenes of complexity, high nonlinearity, multiple features and the like, and how to fully utilize the original feature information to realize the reduction and fusion of high-dimensional features is still a very challenging problem.
Research shows that the existing feature learning model based on data dimension reduction (such as the commonly used t-SNE for data dimension reduction and visualization) is difficult to completely realize the expression of complex geographic objects and continuous geographic phenomena and the analysis, modeling and simulation of the time-space change process of the complex geographic objects and the continuous geographic phenomena. Moreover, most of the models are linear methods for reducing dimensions based on a global structure, have higher complexity in the aspects of feature extraction of different dimensions, data visualization and the like, are difficult to completely express complex geographic objects and continuous geographic phenomena, and are not beneficial to high-dimensional expansion.
In order to adapt to the requirements and development of the times of space big data and cloud computing, the computer field and the GIS field are combined to become the current research direction. A deep self-encoder is built by combining the idea of constructing an artificial neural network, and the input data can be learned to be efficiently represented through unsupervised learning. This efficient representation of the input data is called coding (codings), which is typically much smaller in dimension than the input data, making the self-coder useful for dimension reduction. The self-encoder can be used as a powerful Feature Detector (FD) and applied to pre-training of a deep neural network, deep mining is carried out on complex high-dimensional geospatial data, deep features are extracted, applications such as visualization are further carried out, efficient semantic cognition and information comparison are achieved, and feedback of prediction, judgment and the like of a machine based on human cognition is facilitated.
A large amount of valuable information is hidden in high-dimensional data with high complexity, the high-dimensional data is mined and visually presented, people can be helped to acquire more information and deeper meanings of the information, but due to the complexity and multi-dimensional attributes of the large data, the amount of information which is represented and presented is too large, the hierarchical structure of data information is complex, and high fatigue and low efficiency of people in the cognitive process are caused.
Disclosure of Invention
The invention aims to provide a multi-layer cognition constraint high-dimensional geographic space data focusing visualization method aiming at space-time field object targets with various data sources and complex structures. The method comprises the following steps:
Further, the data feature extraction processing according to the hierarchical division rule includes:
taking the attribute features, the spatial features and the time features as hierarchical division standards, and sequentially extracting the data features in an attribute feature layer, a spatial feature layer and a time feature layer;
on the attribute feature layer, taking components of qualitative or quantitative indexes for describing natural or human attributes of the geographic space target as consideration, and combining a big data feature analysis method, fusing multi-source information including instrument measurement and model calculation so as to establish a relation between geographic target information and knowledge in different fields;
in a spatial feature layer, the extraction of spatial features is pertinently supplemented by combining data and knowledge including mathematical statistics, mathematical morphology and computer vision by taking the geometric features and the positioning features of geographic spatial targets as basic items;
in the time characteristic layer, a multi-dimensional time series characteristic layer data set is constructed on the basis of data processed by the attribute characteristic layer and the space characteristic layer, so that the development change of the feature is dynamically represented through multi-period data of the same feature.
Further, step 2 comprises:
step 2.1, unifying different scale characteristics: adopting a data normalization universal method, namely combining a support vector machine which is divided according to a basic structure and is suitable for dynamic change feature fusion with a Gaussian kernel function method, and uniformly mapping feature items to be input into a network layer;
step 2.2, building a deep network structure: the self-encoder consists of three layers of neural networks, including an input layer, a hidden layer and an output layer, wherein the encoding stage is from the input layer to the hidden layer, and the decoding stage is from the hidden layer to the output layer; in the encoding stage, using function fθFor input vector
Mapping to obtain hidden layer representation
Using function g in the decoding stageθFor implicit layer representation
Mapping to obtain reconstructed data
Coding function in the sequence adaptive model
Will be provided with
Mapping onto an l-dimensional hidden layer to obtain an intermediate representation
Decoding function
Reconstruction
To obtain
Step 2.3, embedding a cognitive constraint mechanism: constructing a constrained semi-supervised self-encoder by utilizing the uniform feature cognition element set generated in the step (1) and taking the error between actual output cognition and expected output cognition as a target reconstruction cost function to show a feature cognition constraint network process;
step 2.4, training network adjustment parameters: and constructing an objective optimization function of the semi-supervised sequence adaptive model according to the steps, and minimizing the objective function by using a gradient descent method to optimize network parameters.
Further, the step 3 comprises:
step 3.1, inducing cognitive dimensionality and interface information dimensionality expression information based on the interface information cognitive process of gradually enhancing human perception attention;
step 3.2, establishing a mapping relation from the user cognitive dimension to the information presentation dimension in the big data visualization interface facing different adaptive requirements, and mapping out the corresponding relation: let U be a universe of discourseThe Objective cognition (SC) set and the Objective Information (OI) set are respectively two different sets, namely a cognition dimension set and an Information dimension set, if an element aiC is formed by SC and has a unique element biE.o, such that bi＝f(ai) Then, f is the mapping from SC to OI, and is denoted as f: SC → OI.
According to the method, the focusing process of intelligent dimensionality reduction of high-dimensional data from a machine to emotional visual expression is realized by utilizing multilayer cognitive constraints, semantic features of the data are expressed by combining a self-encoder through layer-by-layer unsupervised learning, the problems of low accuracy, poor stability, high redundancy and the like easily caused when the conventional dimensionality reduction processing method is used in the visualization field are solved by depending on the internal structure and the autonomous features of the high-dimensional data, and the research method for realizing automatic dimensionality reduction of the conventional high-dimensional geospatial data is supplemented to a certain extent; in the process, the data visualization interface is hierarchically divided based on emotion cognition, so that the limitation of the traditional big data visualization method on data surface layer cognition is broken through, the progressive understanding from objective information dimensionality to subjective information dimensionality is normalized, the human-computer interaction facing to a targeted target is enhanced, and the deep meaning of data can be rapidly understood; the method can support the visual expression of high-dimensional geographic data with various irregular structures, realize efficient semantic cognition and information comparison, and is also beneficial to realizing the feedback of prediction, judgment and the like of a machine based on human cognition.
Drawings
FIG. 1 is a flow chart of the method steps of the present invention.
Fig. 2(a) is a schematic diagram of a feature recognition hierarchy for partitioning a data-based autonomous structure.
Fig. 2(b) is a schematic diagram of a landslide disaster feature classification cognitive hierarchy.
Fig. 2(c) is a schematic diagram of feature mapping rule construction based on emotion visualization.
Fig. 3 is a schematic view of visualization of initial input data according to an embodiment of the present invention.
Fig. 3(a) -3(e) are schematic views of the visualization of the data of each layer in fig. 3.
FIG. 4 is a visualization of the data of FIG. 3 using t-SNE processing.
Fig. 5 is a visualization of the data of fig. 3 processed using the method of the present invention.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the attached drawings:
the invention provides a multi-layer cognition constraint high-dimensional geographic space data focusing visualization method. The high-dimensional geographic space data is oriented to continuously-changed geographic objects or phenomena and is used for expressing an attribute-space-time integrated digital expression form of environmental data distribution, statistical distribution of characteristic indexes and geoscience phenomenon probability distribution; the visualization method for the multilayer progressive constraint high-dimensional data focusing emotion cognition realizes integrated display: the semantic features of the data are expressed by combining an auto-encoder through layer-by-layer unsupervised learning, and the internal structure and the autonomous features of the high-dimensional data are relied on; the data visualization interface is hierarchically divided based on emotion cognition, the limitation of the traditional big data visualization method on data surface layer cognition is broken through, the progressive understanding from objective information dimensionality to subjective information dimensionality is normalized, and finally high-dimensional visualization expression of knowledge focusing is output.
As shown in fig. 1-2, the method comprises the steps of:
The method aims to aim at high-dimensional spatial data with different expression characteristics, focus attributes, spatial and temporal characteristics are used as cognitive characteristics based on semantic information and an autonomous structure of the high-dimensional spatial data, and the cognitive characteristics are used for constructing a constraint item set of a semi-supervised mechanism dimension reduction model.
Further, the data feature extraction processing is carried out according to the hierarchy division rule, and the data feature extraction processing method comprises the following steps:
taking the attribute features, the spatial features and the time features as hierarchical division standards, and sequentially extracting the data features in an attribute feature layer, a spatial feature layer and a time feature layer; according to the target data expression and the hierarchical processing requirement, the target data expression and the hierarchical processing requirement can be refined into corresponding target information, sample distance and time sequence.
In the attribute feature layer, the components of qualitative or quantitative indexes describing the natural or human attributes of the geographic space target are taken as consideration, and multi-source information including instrument measurement and model calculation is fused by combining a big data feature analysis method, so that the relationship between the geographic target information and knowledge in different fields is established. The processing can break through the traditional geographic space equal domain data attribute feature description limitation.
As shown in fig. 2(b), landslide disasters are taken as an example, and the attribute features are refined into landslide information by combining with knowledge in the field of remote sensing data processing, wherein the landslide information comprises spectral features, textural features, topographic features and vegetation indexes; and extracting essential information which is a partition rule of the attribute feature layer by taking the enhanced target self-adaptive semantic features as the purpose.
In the spatial feature layer, the extraction of spatial features is pertinently supplemented by combining data and knowledge including mathematical statistics, mathematical morphology and computer vision on the basis of geometric features and positioning features of geospatial targets.
In a spatial feature layer, multi-professions cooperate to realize multi-azimuth and multi-angle deep analysis facing to ground object spatial information and exploration of the relevance of the ground objects and the environment elements, so that the limitation of expressing traditional data types such as spatial coordinates and relative positions is broken through, and understanding and expression of sample spatial features are expanded to a certain extent.
As shown in fig. 2(b), the landslide hazard is taken as an example, the spatial features are refined to be the ground object distance based on a spatial measuring instrument and a remote sensing imaging instrument, the depth excavation disaster causing environmental factor is calculated through a knowledge driving model such as mathematical statistics, mathematical morphology and the like, and the landslide and river distance, the road distance and the fault distance are quantitatively analyzed by using qualitative indexes; and extracting spatial information as a partition criterion of a spatial feature layer by taking the self-adaptive semantic features of the salient objects as the purposes.
In the time characteristic layer, a multi-dimensional time series characteristic layer data set is constructed on the basis of data processed by the attribute characteristic layer and the space characteristic layer, so that the development change of the feature is dynamically represented through multi-period data of the same feature.
The space-time transformation of the ground object target can be comprehensively described by innovatively utilizing the priori knowledge in the time characteristic layer, and the space-time transformation is sent to the model to complete specific data analysis, so that the time complexity and the data analysis difficulty are effectively reduced, and the requirements of various fields such as real-time anomaly detection, risk potential discovery and the like are met.
As shown in fig. 2(b), a landslide disaster is taken as an example, the reason and the expression of landslide morphological change are analyzed based on a time sequence and geomechanical knowledge, and time sequence analysis is performed on a lag term, a period term and a trend term through refining; and extracting time sequence information as a division rule of a time characteristic layer by taking the self-adaptive semantic characteristics of the mined target as the aim.
The step is to construct a deep self-encoder of an adaptive semi-supervised mechanism by depending on cognitive features, and to reduce the dimension of high-dimensional data by using a model to obtain an adaptive feature matrix set.
The limited Boltzmann machine is composed of two layers of neural networks, can be regarded as a process from encoding to decoding, constructs a deep self-encoder by stacking the machine, and realizes the maximum reconstruction from an input signal to an output signal.
The construction of the focusing sequence model in step 2 and the obtaining of the adaptive feature matrix set are further described with reference to fig. 2(a), which includes the following steps:
and 2.1, unifying the characteristics of different scales. By adopting a data normalization universal method, namely combining a support vector machine which is divided according to a basic structure and is suitable for dynamic change feature fusion with a Gaussian kernel function method (Gaussian SVM), inputting feature cognition to a network layer and uniformly mapping:
gaussian kernel method: expanding the data dimension to an infinite dimension by using a Gaussian kernel function mode to obtain a boundary, wherein the result is only related to the calculation of the distance between the x and the central point xn and is not related to other data; the following formula is adopted:
and 2.2, building a deep network structure. The basic self-encoder consists of a three-layer neural network, which comprises an input layer, a hidden layer and an output layer, wherein the encoding stage is from the input layer to the hidden layer, and the decoding stage is from the hidden layer to the output layer. In the encoding stage, using function fθFor input vector
Mapping to obtain hidden layer representation
Using function g in the decoding stageθFor implicit layer representation
Mapping to obtain reconstructed data
Coding function in the sequence adaptive model described in the patent
Will be provided with
Mapping onto an l-dimensional hidden layer to obtain an intermediate representation
Decoding function
Reconstruction
To obtain
And 2.3, embedding a cognitive constraint mechanism. A step of utilizing the unified feature cognition element set generated in the step 1 to reduce errors between actual output cognition and expected output cognition to serve as a target reconstruction cost function to show a feature cognition constraint network; the constrained semi-supervised self-encoder combines the advantages of unsupervised learning and supervised learning, overcomes the defect of low classification accuracy of unsupervised learning, and better improves the dimension reduction accuracy and the network generalization capability.
And 2.4, training network adjustment parameters. Constructing an objective optimization function of the semi-supervised sequence adaptive model according to the steps, wherein J is the target optimization functionAE，JWD，JFEAThe three parts are as follows: j. the design is a squareAEInput and output data calculated by unsupervised learning; j. the design is a squareWDIs a weighted two-norm regularization term that prevents overfitting; and JFEAThe network parameters are adjusted by minimizing the cognitive error term described in step 2.3.
The cognitive interaction between a person and a data visualization interface is firstly embodied in surface layer information of the interface, and is mainly expressed in an interface information cognitive process gradually enhanced by perceiving attention; the cognitive process can be refined from a visual information layer to an emotional information layer: firstly, human perception of a big data visualization interface mainly comes from vision, and human eyes firstly receive stimulation of visual elements such as interface colors, shapes, symbols and the like, namely objective information dimensionality of patents; secondly, the human brain selects a large amount of sensed information, and performs overall and local shallow processing on interface information in the brain to generate deep attention information, namely the subjective information dimension;
through the processing of the step, the characteristic matrix normalization mapping rule is established according to the cognitive process from the superficial visual information to the deep emotional information of the human body on the visual interface and focusing the perception and attention of the human body, and the characteristic mapping rule set is summarized. The mapping rule set for the geospatial data features in this embodiment may be as shown in fig. 2 (c).
In step 3, the implementation of the focusing of the perception and attention information further comprises the following steps:
step 3.1, inducing cognitive dimensionality and interface information dimensionality expression information based on the interface information cognitive process of gradually enhancing human perception attention; the main information source of the visual information dimension is visual information in a visual interface, and various information attribute codes are correspondingly presented, wherein some common attribute codes have colors, sizes, shapes, textures and the like; the visual information is combined in series to obtain more complex emotional information, such as the cold-warm contrast of blue and red, and the like, which have certain emotional implications and directivity.
Step 3.2, establishing a cognitive dimension-interface information mapping rule, establishing a mapping relation from the user cognitive dimension to the information presentation dimension in the big data visualization interface facing different adaptive requirements according to the step 3.2, and mapping out the corresponding relation: setting U as a domain, and respectively setting a Subjective Cognitive (SC) set and an Objective Information (OI) set as two different sets, namely a cognitive dimension set and an Information dimension set, if an element aiC is formed by SC and has a unique element biE.o, such that bi＝f(ai) Then, f is the mapping from SC to OI, and is denoted as f: SC → OI.
After the three layers of focusing, the visual expression of the geospatial data can be realized.
The visualization method of the present invention is further described below in conjunction with fig. 3-5.
FIG. 3 is a visual representation of a set of high-dimensional geospatial data to be processed. Fig. 3(a) - (e) are present representations of the classes of cognitive feature vector sample sets in fig. 3. Based on multispectral data, DEM data and other products in a research area, various continuous ground feature information is acquired by a space analysis method and comprises five cognitive feature vector sample sets of slope direction (Aspect), slope (slope), Elevation (Elevation), vegetation index (NDVI) and Distance (Distance); and visualizing the complex high-dimensional multi-element data by utilizing an Andrews curve, wherein N indexes of each type of feature vector sample form a point in an N-dimensional space and are represented by a curve in a two-dimensional space.
the t-SNE dimension reduction model is a common method for dimension reduction and visualization of high-dimensional geospatial data in the prior art. As shown in fig. 4, as a result of performing dimension reduction analysis on the initial cognitive feature sample shown in fig. 3 by using the t-SNE dimension reduction model, various feature vectors after dimension reduction are represented by points of different colors in a two-dimensional space, and as a result, fig. 4 can find that the conventional t-SNE dimension reduction model has disadvantages: the dimensionality reduction result of the large-scale multivariate data has randomness, and a part of sample information is lost; the distance between the samples has no specific significance, and the method is not suitable for the dimension reduction processing of the geographic data with strong spatial correlation; the requirement for super-parameter setting in the model is high, irregular shapes are easy to appear as a result, sample distribution is not obvious, and the research on the characteristic distribution of the nature of ground objects is not facilitated.
FIG. 5 is a model constructed by applying the method of the present invention, and a deep semantic feature vector capable of well representing complex ground objects is mined by performing a comprehensive deep analysis processing result of multi-source data-adaptive model-knowledge driving aiming at the geospatial target of FIG. 3; and performing clustering analysis while reducing the dimensions of the data, and realizing target identification and classification for the ground objects in the research area by combining feature cognition of landslide disasters with emotion visualization cognition. In fig. 5: the red highlights the cognition of the landslide risk, the cognition of the expression of the Y-shaped point objects with different sizes on the lithology of the landslide and the like. Compared with a t-SNE dimension reduction method, the method disclosed by the patent not only focuses on the processing of initial data on a dimension space, but also focuses on the correlation among multi-source features through cognitive constraint, and provides information such as target classification identification and the difference degree between homogeneous foreign matters by using variables such as colors, shapes and sizes in the upper graph.
Claims (2)
1. A multi-layer cognition constraint high-dimensional geographic spatial data focusing visualization method is characterized by comprising the following steps:
step 1, focusing semantic data feature items: dividing feature items of high-dimensional geographic spatial data based on the semantic relevance of the essential features of the spatial data, extracting data features according to a hierarchical division rule, and extracting the features of the data to be processed into uniform feature cognitive elements to obtain a uniform feature cognitive element set;
step 2, focusing of sequence model adaptability: sequentially constructing a restricted Boltzmann machine, constructing an autonomous structure presented by analyzing high-dimensional data by a deep self-encoder, and training a model to realize optimal dimensionality reduction expression of target data; the feature cognition is integrated into an optimization target or a constraint function through a supervised learning mode, so that the aim of feature training is guaranteed while the intrinsic features of the sample are effectively extracted; inputting the unified feature cognitive elements in the step 1 to perform machine dimension reduction to obtain a self-adaptive feature matrix set; the step 2 comprises the following steps:
step 2.1, unifying different scale characteristics: adopting a data normalization universal method, namely combining a support vector machine which is divided according to a basic structure and is suitable for dynamic change feature fusion with a Gaussian kernel function method, and uniformly mapping feature items to be input into a network layer;
step 2.2, building a deep network structure: the self-encoder consists of three layers of neural networks, including an input layer, a hidden layer and an output layer, wherein the encoding stage is from the input layer to the hidden layer, and the decoding stage is from the hidden layer to the output layer; in the encoding stage, using function fθFor input vector
Mapping to obtain hidden layer representation
Using function g in the decoding stageθFor implicit layer representation
Mapping to obtain reconstructed data
Coding function in the sequence model
Will be provided with
Mapping onto an l-dimensional hidden layer to obtain an intermediate representation
Decoding function
Reconstruction
To obtain
Step 2.3, embedding a cognitive constraint mechanism: constructing a constrained semi-supervised self-encoder by utilizing the uniform feature cognition element set generated in the step (1) and taking the error between actual output cognition and expected output cognition as a target reconstruction cost function to show a feature cognition constraint network process;
step 2.4, training network adjustment parameters: constructing a target optimization function of the semi-supervised sequence adaptive model according to the steps, and minimizing the target function to optimize network parameters by using a gradient descent method;
step 3, focusing of perception and attention information: analyzing the cognitive process of the data visualization interface based on human, and establishing a dynamic normalized mapping rule of a feature matrix improved by taking a data visualization platform as a consideration; inputting the self-adaptive feature matrix set in the step 2 to obtain a feature mapping rule set; the step 3 comprises the following steps:
step 3.1, inducing cognitive dimensionality and interface information dimensionality expression information based on the interface information cognitive process of gradually enhancing human perception attention;
step 3.2, establishing a mapping relation from the user cognitive dimension to the information presentation dimension in the big data visualization interface facing different adaptive requirements, and mapping out the corresponding relation: setting U as a discourse domain, and respectively setting the subjective cognition set SC and the objective information set OI as two different sets, namely a cognition dimension set and an information dimension set, if the element aiBelongs to a subjective cognition set SC and has a unique element biBelongs to the objective information set OI, such that bi＝f(ai) If so, f is the mapping from the subjective cognition set SC to the objective information set OI, and is written as f: SC → OI;
step 4, visual expression of emotion cognition: and (4) performing visual expression on the input high-dimensional geospatial data based on the focusing of the step 1-3, and outputting the result to a spatial target visual interface.
2. The method according to claim 1, wherein the performing data feature extraction processing according to the hierarchical division rule comprises:
taking the attribute features, the spatial features and the time features as hierarchical division standards, and sequentially extracting the data features in an attribute feature layer, a spatial feature layer and a time feature layer;
on the attribute feature layer, taking components of qualitative or quantitative indexes for describing natural or human attributes of the geographic space target as consideration, and combining a big data feature analysis method, fusing multi-source information including instrument measurement and model calculation so as to establish a relation between geographic target information and knowledge in different fields;
in a spatial feature layer, the extraction of spatial features is pertinently supplemented by combining data and knowledge including mathematical statistics, mathematical morphology and computer vision by taking the geometric features and the positioning features of geographic spatial targets as basic items;
in the time characteristic layer, a multi-dimensional time series characteristic layer data set is constructed on the basis of data processed by the attribute characteristic layer and the space characteristic layer, so that the development change of the feature is dynamically represented through multi-period data of the same feature.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010197344.1A CN111538776B (en) | 2020-03-19 | 2020-03-19 | Multilayer cognitive constraint high-dimensional geographic spatial data focusing visualization method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010197344.1A CN111538776B (en) | 2020-03-19 | 2020-03-19 | Multilayer cognitive constraint high-dimensional geographic spatial data focusing visualization method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111538776A CN111538776A (en) | 2020-08-14 |
| CN111538776B true CN111538776B (en) | 2021-10-01 |
Family
ID=71974974
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010197344.1A Active CN111538776B (en) | 2020-03-19 | 2020-03-19 | Multilayer cognitive constraint high-dimensional geographic spatial data focusing visualization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111538776B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112070366B (en) * | 2020-08-19 | 2022-03-29 | 核工业湖州勘测规划设计研究院股份有限公司 | Regional landslide risk quantitative measuring and calculating method based on multi-source monitoring data correlation analysis |
| CN112069200A (en) * | 2020-08-31 | 2020-12-11 | 武汉大学 | Multi-detail-level visualization method for high-dimensional space-time big data |
| CN112215083B (en) * | 2020-09-17 | 2021-11-09 | 中国科学院沈阳应用生态研究所 | Multi-geographic-video self-adaptive event detection method based on abnormal change modeling |
| CN115658045B (en) * | 2022-12-06 | 2023-03-21 | 西安恒歌数码科技有限责任公司 | System and method for rapidly building geographic information data application |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855662A (en) * | 2012-07-25 | 2013-01-02 | 中国科学院对地观测与数字地球科学中心 | Ocean environment visualization method |
| CN105426461A (en) * | 2015-11-12 | 2016-03-23 | 中国科学院遥感与数字地球研究所 | Map visualization system and method for performing knowledge mining based on big spatial data |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7450736B2 (en) * | 2005-10-28 | 2008-11-11 | Honda Motor Co., Ltd. | Monocular tracking of 3D human motion with a coordinated mixture of factor analyzers |
| CN105630897B (en) * | 2015-12-18 | 2019-12-24 | 武汉大学 | Content-aware geographic video multilevel correlation method |
| CN107330357A (en) * | 2017-05-18 | 2017-11-07 | 东北大学 | Vision SLAM closed loop detection methods based on deep neural network |
| JP6856557B2 (en) * | 2018-01-22 | 2021-04-07 | 株式会社日立製作所 | Optimization device and hyperparameter optimization method |
| CN109858509A (en) * | 2018-11-05 | 2019-06-07 | 杭州电子科技大学 | Based on multilayer stochastic neural net single classifier method for detecting abnormality |
-
2020
- 2020-03-19 CN CN202010197344.1A patent/CN111538776B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855662A (en) * | 2012-07-25 | 2013-01-02 | 中国科学院对地观测与数字地球科学中心 | Ocean environment visualization method |
| CN105426461A (en) * | 2015-11-12 | 2016-03-23 | 中国科学院遥感与数字地球研究所 | Map visualization system and method for performing knowledge mining based on big spatial data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111538776A (en) | 2020-08-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111538776B (en) | Multilayer cognitive constraint high-dimensional geographic spatial data focusing visualization method | |
| Huang et al. | A review on dimensionality reduction techniques | |
| WO2022041678A1 (en) | Remote sensing image feature extraction method employing tensor collaborative graph-based discriminant analysis | |
| Ou et al. | A CNN framework with slow-fast band selection and feature fusion grouping for hyperspectral image change detection | |
| CN107636693A (en) | Assigned for the relevance scores of artificial neural network | |
| Zhai et al. | BeautyNet: Joint multiscale CNN and transfer learning method for unconstrained facial beauty prediction | |
| WO2024087639A1 (en) | Remote sensing image recommendation method based on content understanding | |
| Forestier et al. | Coastal image interpretation using background knowledge and semantics | |
| Maciejewski | Data representations, transformations, and statistics for visual reasoning | |
| Xu et al. | Feature-based constraint deep CNN method for mapping rainfall-induced landslides in remote regions with mountainous terrain: An application to Brazil | |
| Juliani et al. | Deep learning of terrain morphology and pattern discovery via network-based representational similarity analysis for deep-sea mineral exploration | |
| Zhang et al. | Feature extraction of ancient Chinese characters based on deep convolution neural network and big data analysis | |
| Zhang et al. | Complex image recognition algorithm based on immune random forest model | |
| Xing et al. | Deep network with irregular convolutional kernels and self-expressive property for classification of hyperspectral images | |
| Gao | Application of convolutional neural network in emotion recognition of ideological and political teachers in colleges and universities | |
| CN112686830B (en) | Super-resolution method of single depth map based on image decomposition | |
| Qiao et al. | Group behavior recognition based on deep hierarchical network | |
| Li | Multimodal visual image processing of mobile robot in unstructured environment based on semi-supervised multimodal deep network | |
| He et al. | Range likelihood tree: A compact and effective representation for visual exploration of uncertain data sets | |
| Cai et al. | Real-time leaf recognition method based on image segmentation and feature extraction | |
| Liao et al. | Video Face Detection Technology and Its Application in Health Information Management System | |
| Ma et al. | Image segmentation based on differential immune clone clustering algorithm | |
| Yu et al. | Clustering-enhanced pointcnn for point cloud classification learning | |
| Shim et al. | Content-based image retrieval for industrial material images with deep learning and encoded physical properties | |
| Jian et al. | Uncertainty Aware Graph Self-Supervised Learning for Hyperspectral Image Change Detection |
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 | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A Focused Visualization Method for High Dimensional Geospatial Data with Multilayer Cognitive Constraints Effective date of registration: 20230625 Granted publication date: 20211001 Pledgee: Zhejiang Deqing Rural Commercial Bank Co.,Ltd. science and technology sub branch Pledgor: ZHEJIANG ZHONGHAIDA SPACE INFORMATION TECHNOLOGY Co.,Ltd. Registration number: Y2023980045468 |