CN111931833B - Multi-source data driving-based space-based multi-dimensional information fusion method and system - Google Patents
Multi-source data driving-based space-based multi-dimensional information fusion method and system Download PDFInfo
- Publication number
- CN111931833B CN111931833B CN202010754344.7A CN202010754344A CN111931833B CN 111931833 B CN111931833 B CN 111931833B CN 202010754344 A CN202010754344 A CN 202010754344A CN 111931833 B CN111931833 B CN 111931833B
- Authority
- CN
- China
- Prior art keywords
- target
- information
- fusion
- decision
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides a multi-source data drive-based space-based multi-dimensional information fusion system and method, which comprises the following steps: module M1: carrying out target information association on target information in the same-phase multi-information-source information; module M2: performing multi-source information fusion based on the associated target information; module M3: and performing decision-making re-fusion on the output result of the multi-source information fusion to obtain a fusion product, and packaging the fusion product to output important target products and regional comprehensive information. The invention has strong self-adaptive capacity and good system robustness, can flexibly intervene in the information source meeting the input format requirement, and improves the product dimension. Therefore, the invention is particularly suitable for the flexible space-based multi-dimensional information fusion based on multi-source data driving.
Description
Technical Field
The invention relates to the technical field of space-based intellectualization, in particular to an elastic space-based multi-dimensional information fusion system based on multi-source data driving.
Background
The modern remote sensing technology provides various types of observation data for earth observation, however, even though the space-based remote sensing data are processed and detected in real time in an on-orbit manner to form signal description, image slices and target characteristic data, secondary interpretation and comprehensive association are still needed, and the scene application of high time resolution requirements such as marine vessel navigation guarantee, earthquake and flood disaster area search and rescue, airport port and expressway real-time congestion early warning and the like cannot be met. In order to solve the problems, multi-source information fusion is carried out on target information in the same-time multi-source information on track, target characteristics in a designated area can be revealed based on a decision-level fusion means and a feature-level fusion means, and target information is extracted, so that high-confidence recognition and multi-dimensional state information synthesis of a target are realized. Product dimensionality can be enriched, algorithm complexity is low, secondary interpretation is not needed, text and image slice products of important targets are provided for various scenes with high time resolution requirements, and regional comprehensive information description is formed.
Through the literature technology of the prior art and through the literature search of the prior art, the articles Multtisenor fusion using Hopfield neural network in INS/SMGS integrated system, Signal Proeessing, Vol.2, 2002: 1199A 1202 presents a method for multi-sensor information fusion using a HoPfiltered neural network. However, this method requires a large number of training samples to be collected when the amount of sensor data is large, and requires a large amount of training time to adjust parameters of the neural network for practical use. The method is limited by the existing satellite-borne computing capability and cannot be applied to a space-based system.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a multi-source data-driven space-based multi-dimensional information fusion system and method.
The invention provides a multi-source data drive-based space-based multi-dimensional information fusion system, which comprises:
module M1: carrying out target information association on target information in the same-phase multi-information-source information;
module M2: performing multi-source information fusion based on the associated target information;
module M3: and performing decision-making re-fusion on the output result of the multi-source information fusion to obtain a fusion product, and packaging the fusion product to output important target products and regional comprehensive information.
Preferably, said module M1 comprises: according to different information source types, the target information association comprises the following steps: target information association based on platform positioning information and target information association based on registration information;
the target information association based on the platform positioning information is used for performing target information association on an area with low target intensity and low requirement on space information precision;
the target information association based on the registration information is used for performing target information association on an area with higher target intensity and higher requirement on spatial information precision;
the target information correlation based on the registration information comprises the positioning precision provided by the information source, wherein the target positioning error range after the registration with the geocoded base map is adopted.
Preferably, the target information provided by the target information association based on platform positioning information comprises target geographical position, positioning accuracy, direction, aspect ratio and area;
the target information provided by the target information correlation based on the registration information comprises target position, positioning precision, direction, aspect ratio and area obtained by registration with the base map.
Preferably, said module M1 comprises:
module M1.1: performing target information association on every two information sources in all the information sources according to the target positioning precision range provided by the information sources;
module M1.2: selecting a precision error with large precision error as a precision error range within the positioning precision error range of every two information sources, and directly establishing an association relationship between targets when only one target has a corresponding relationship with the other target within the selected precision error range; when a plurality of targets are in a corresponding relation with one target within a selected precision error range, establishing a correlation relation based on the target position, the geometric attribute and the topological relation of target distribution by adopting an Eigenvector graph matching algorithm to realize accurate correlation between different targets provided by different information sources, and forming a corresponding data table of target serial numbers pointing to the same target in different target information provided by different information sources;
module M1.3: and when the matching chains conflict, establishing an association relation by using the matching chain with the highest sum of the target detection confidence degrees in the matching chains.
Preferably, the Eigenvector graph matching algorithm in the module M1.2 includes:
module M1.2.1: establishing distance measurement between targets, including target position distance and attribute distance;
the target position distance is an Euclidean distance between target absolute coordinates under the constraint of positioning errors;
the attribute distance measurement is the compatibility of decision attributes of targets detected by all information sources, and the formula is as follows:
D(x i ，x’ j )＝w d d(x i ，x’ j )+(1-w d )a(x i ，x’ j )
wherein, D (x) i ，x′ j ) Representing a measure of the distance of the attribute, w d Is the location distance weight, d (x) i ，x’ j ) Is the position Euclidean distance, a (x) i ，x’ j ) Is the attribute distance, x i ∈X＝{x 1 ，x 2 ，…x m }，x′ j ∈X′＝{x′ 1 ，x′ 2 ，…x′ n }, X, X' are sets of objects detected by two sources;
module M1.2.2: establishing a target similarity incidence matrix H-H according to the attribute distance metric value ij ]The formula is as follows:
h ij ＝exp(-D(x i ，x’ j ) 2 /2σ 2 )
wherein h is ij Representing the combined distance between objects, the element in row i and column j being an objectThe distance between the target i and the target j; the sigma represents a distance weight parameter and represents distance sensitivity under different target density scenes, the sigma is a constant, an initial value is a preset value, and the sigma is adjusted and optimized in real time according to space-based application performance;
module M1.3.3: solving the incidence matrix by adopting an Eigenvector graph matching algorithm;
performing singular value decomposition on the incidence matrix H to obtain G-TDU, wherein T and U are m-dimension and n-dimension orthogonal matrixes respectively, D is a non-negative diagonal matrix, and G represents a matrix obtained after singular value decomposition; when m is less than n, U only has significance in the first m rows; replacing D with an identity matrix I to obtain another orthogonal matrix P ═ TIU, when P is ij And when the element is the ith row and j columns maximum element in the matrix P, the point pair is the potential correct match, and in the determined potential matching relationship, the unsatisfied matched pair is removed by utilizing the position accuracy constraint.
Preferably, the module M2 includes: performing fusion decision including a target existence layer and a target identity recognition layer based on the associated target information;
and performing target existence layer fusion decision based on the associated target information to obtain a target existence layer fusion decision result, and performing target identity recognition layer fusion decision on the target existence layer fusion decision result.
Preferably, the target presence layer fusion decision comprises: performing decision-level fusion;
reading n information sources which all participate in the fusion, wherein the credibility parameter of the ith information source is d i The detection confidence c of the current detection target t of the ith information source it ；
Let (1,0, -1) denote the decision space, where 1 denotes target, 0 denotes non-target, -1 denotes uncertain;
obtaining a basic probability assignment function of a source i to a target t: m is it (1)＝c it d i 、m it (0)＝c it (1-d i )、m it (-1)＝1-c it ；
And setting K as a conflict coefficient, and obtaining a mixed basic probability assignment function of all the information sources to the target t as follows:
selection of m t (1)、m t (0) And m t And (1) taking the decision branch with the maximum median as the decision result of the target.
Preferably, the target identification layer fusion decision includes: performing decision-level fusion and feature-level fusion;
the decision-level fusion establishes a decision space for the decision event of the target identity by using each information source, and matches the target attribute information provided by each information source with the reference attribute data of the target type and/or model corresponding to each decision event in the decision space in a satellite-borne target attribute knowledge base;
performing comprehensive target identification analysis by using target characteristic information provided by each information source in characteristic level fusion; the input of the feature level fusion is the target feature information provided by the associated member star, and the fusion recognition result is output after the classification by the multi-level cascade classifier.
Preferably, the decision-level re-fusion makes a re-fusion decision using DSmT theory.
The invention provides a multi-source data drive-based space-based multi-dimensional information fusion method, which comprises the following steps:
step M1: carrying out target information association on target information in the same-phase multi-information-source information;
step M2: performing multi-source information fusion based on the associated target information;
step M3: and performing decision-making re-fusion on the output result of the multi-source information fusion to obtain a fusion product, and packaging the fusion product to output important target products and regional comprehensive information.
Compared with the prior art, the invention has the following beneficial effects:
1. the system has high miniaturization degree and is suitable for small satellite clusters;
2. the method has high information interaction efficiency and low system complexity, does not need secondary interpretation, provides text and image slice products of important targets for various scenes with high time resolution requirements, and forms regional comprehensive information description;
3. the invention has strong self-adaptive capacity and good system robustness, can flexibly intervene in the information source meeting the input format requirement, and improves the product dimension. Therefore, the invention is particularly suitable for the flexible space-based multi-dimensional information fusion based on multi-source data driving.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:
FIG. 1 is a flow of flexible space-based multi-dimensional information fusion based on multi-source data driving;
FIG. 2 is a typical hardware architecture of a space-based multi-dimensional information fusion system;
FIG. 3 is a process flow of a target information association sub-module of platform positioning information;
FIG. 4 is a process flow of a target information association sub-module based on base map registration information;
FIG. 5 is a process flow of an information fusion identification submodule of the daytime open sea search and rescue ship;
FIG. 6 is a processing flow of a target information fusion identification submodule of a search and rescue ship in open sea at night;
FIG. 7 is a flow of a target presence layer decision fusion algorithm;
FIG. 8 basic flow of training and classification for cascaded classifiers
FIG. 9 shows a flow of a re-decision algorithm based on the DSMt theory
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that it would be obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit of the invention. All falling within the scope of the present invention.
Aiming at the defects in the prior art, the invention aims to provide a multi-source data-driven space-based multi-dimensional information fusion method and system. And performing multi-source information fusion on target information in the same-phase multi-source information, and mutually supplementing and mutually verifying the target information through a decision-level fusion means and a feature-level fusion means.
The invention discloses a multi-source data-driven space-based multi-dimensional information fusion method and system, which are used for carrying out multi-source information fusion on target information in the same-phase multi-information-source information, and mutually supplementing and mutually verifying the target information through a decision-level fusion means and a feature-level fusion means. The target characteristics in the designated area can be revealed, and the target information is extracted, so that high-confidence-degree identification and multi-dimensional state information synthesis of the target are realized. The system adopts Eigenvector graph matching algorithm to form target association, carries out target existing layer decision-making level fusion and target identity recognition layer decision-making level feature level fusion, and carries out mixed decision-making level fusion recognition through DSmT reasoning.
Example 1
The invention provides a multi-source data drive-based space-based multi-dimensional information fusion system, which comprises:
module M1: carrying out target information association on target information in the same-phase multi-information-source information;
module M2: performing multi-source information fusion based on the associated target information;
module M3: and performing decision-making re-fusion on the output result of the multi-source information fusion to obtain a fusion product, packaging the fusion product to output an important target product and regional comprehensive information, and transmitting the important target product and the regional comprehensive information to a satellite platform memory through a network.
The target information association comprises information such as target position and geometric attributes provided by the information source, and the association relation between the same target information from different information sources is established;
specifically, according to the number and the types of the information sources participating in the fusion determined by the fusion mode management scheduling module, a single task target information data packet is acquired from a satellite memory, the information source data is analyzed, target related attributes or characteristic information are extracted, and the target related attributes or characteristic information are converted into a data format defined by the local computer. And if the data packet analysis is not passed, the target data to be fused of the information source is set to be null.
And meanwhile, the multi-information-source information comprises electromagnetic information source information, and redundancies of the received electromagnetic information such as electromagnetic radiation sources, AIS (automatic identification system), ADS-B (automatic dependent surveillance-broadcast) and the like are removed through the target marks, positions and the like, clustering is carried out, and repeated information is removed.
Specifically, the module M1 includes: according to different source types, the target information association comprises the following steps: target information association based on platform positioning information and target information association based on registration information;
the target information association based on the platform positioning information is used for performing target information association on an area with low target intensity and low requirement on space information precision; mainly for vehicle, aircraft and ship validation in the sea or in areas with few targets.
The target information association based on the registration information is used for performing target information association on an area with higher target intensity and higher requirement on spatial information precision; the method is mainly used for confirming the states of key roads and disaster areas.
The target information correlation based on the registration information comprises the positioning precision provided by the information source, wherein the target positioning error range after the registration with the geocoded base map is adopted.
Specifically, the target information provided by the target information association based on the platform positioning information comprises a target geographic position, positioning accuracy, a direction, an aspect ratio and an area;
the target information provided by the target information correlation based on the registration information comprises target position, positioning precision, direction, aspect ratio and area obtained by registration with the base map.
Specifically, the module M1 includes:
module M1.1: performing target information association on every two information sources in all the information sources according to the target positioning precision range provided by the information sources;
module M1.2: selecting a precision error with large precision error as a precision error range within the positioning precision error range of every two information sources, and directly establishing an association relationship between targets when only one target has a corresponding relationship with the other target within the selected precision error range; when a plurality of targets are in a corresponding relation with one target within a selected precision error range, establishing a correlation relation based on the target position, the geometric attribute and the topological relation of target distribution by adopting an Eigenvector graph matching algorithm to realize accurate correlation between different targets provided by different information sources, and forming a corresponding data table of target serial numbers pointing to the same target in different target information provided by different information sources;
module M1.3: and when the matching chains conflict, establishing an association relation by using the matching chain with the highest sum of the target detection confidence degrees in the matching chains.
Specifically, the Eigenvector graph matching algorithm in the module M1.2 includes:
module M1.2.1: establishing distance measurement between targets, including target position distance and attribute distance;
the target position distance is an Euclidean distance between target absolute coordinates under the constraint of positioning errors;
the attribute distance measurement is the compatibility of decision attributes of targets detected by all information sources, and the formula is as follows:
D(x i ，x’ j )＝w d d(x i ，x’ j )+(1-w d )a(x i ，x’ j )
wherein, D (x) i ，x′ j ) Representing a measure of the distance of the attribute, w d Is the location distance weight, d (x) i ，x’ j ) Is the position Euclidean distance, a (x) i ，x’ j ) Is the attribute distance, x i ∈X＝{x 1 ，x 2 ，…x m }，x′ j ∈X′＝{x′ 1 ，x′ 2 ，…x′ n }, X, X' are sets of objects detected by two sources;
module M1.2.2: establishing a target similarity incidence matrix H-H according to the attribute distance metric value ij ]The formula is as follows:
h ij ＝exp(-D(x i ，x’ j ) 2 /2σ 2 )
wherein h is ij Representing the integrated distance between the targets, the element of the ith row and the jth column is the distance between the target i and the target j; the sigma represents a distance weight parameter and represents distance sensitivity under different target density scenes, the sigma is a constant, an initial value is a preset value, and the sigma is adjusted and optimized in real time according to space-based application performance;
module M1.3.3: solving the incidence matrix by adopting an Eigenvector graph matching algorithm;
performing singular value decomposition on the incidence matrix H to obtain G-TDU, wherein T and U are m-dimension and n-dimension orthogonal matrixes respectively, D is a non-negative diagonal matrix, and G represents a matrix obtained after singular value decomposition; when m is less than n, U only has significance in the first m rows; replacing D with an identity matrix I to obtain another orthogonal matrix P ═ TIU, when P is ij And when the maximum element is the ith row and the j column in the matrix P, the point pair is a potential correct match, and in the determined potential matching relationship, the unsatisfied matched pair is removed by using the position accuracy constraint.
Specifically, the module M2 includes: performing fusion decision including a target existence layer and a target identity recognition layer based on the associated target information;
and performing target existence layer fusion decision based on the associated target information to obtain a target existence layer fusion decision result, and performing target identity recognition layer fusion decision on the target existence layer fusion decision result.
Specifically, the target presence layer fusion decision includes: performing decision-level fusion;
reading n information sources which all participate in the fusion, wherein the credibility parameter of the ith information source is d i The detection confidence c of the current detection target t of the ith information source ir ；
Let (1,0, -1) denote the decision space, where 1 denotes target, 0 denotes non-target, -1 denotes uncertainty;
obtaining a basic probability assignment function of a source i to a target t: m is it (1)＝c it d i 、m it (0)＝c it (1-d i )、m it (-1)＝1-c it ；
And setting K as a conflict coefficient, and obtaining a mixed basic probability assignment function of all the information sources to the target t as follows:
selection of m t (1)、m t (0) And m t And (1) taking the decision branch with the maximum median as the decision result of the target.
Specifically, the target identity recognition layer fusion decision includes: performing decision-level fusion and feature-level fusion;
the decision-level fusion establishes a decision space for the decision event of the target identity by using each information source, and matches the target attribute information provided by each information source with the reference attribute data of the target type and/or model corresponding to each decision event in the decision space in a satellite-borne target attribute knowledge base; the decision-making steps are the same as the decision-level fusion of the target presence layer fusion decisions.
Performing comprehensive target identification analysis by using target characteristic information provided by each information source in characteristic level fusion; the input of the feature level fusion is the target feature information provided by the associated member star, and the fusion recognition result is output after the classification by the multi-level cascade classifier.
In particular, the decision-level re-fusion makes use of DSmT theory for re-fusion decisions.
The invention provides a multi-source data drive-based space-based multi-dimensional information fusion method, which comprises the following steps:
step M1: carrying out target information association on target information in the same-phase multi-information-source information;
step M2: performing multi-source information fusion based on the associated target information;
step M3: and performing decision-making re-fusion on the output result of the multi-source information fusion to obtain a fusion product, packaging the fusion product to output an important target product and regional comprehensive information, and transmitting the important target product and the regional comprehensive information to a satellite platform memory through a network.
The target information association comprises information such as target position and geometric attributes provided by the information source, and the association relation between the same target information from different information sources is established;
specifically, according to the number and the types of the information sources participating in the fusion determined by the fusion mode management scheduling module, a single task target information data packet is acquired from a satellite memory, the information source data is analyzed, target related attributes or characteristic information are extracted, and the target related attributes or characteristic information are converted into a data format defined by the local computer. And if the data packet analysis is not passed, the target data to be fused of the information source is set to be null.
And meanwhile, the multi-information-source information comprises electromagnetic information source information, and redundancies of the received electromagnetic information such as electromagnetic radiation sources, AIS (automatic identification system), ADS-B (automatic dependent surveillance-broadcast) and the like are removed through the target marks, positions and the like, clustering is carried out, and repeated information is removed.
Specifically, the step M1 includes: according to different information source types, the target information association comprises the following steps: target information association based on platform positioning information and target information association based on registration information;
the target information association based on the platform positioning information is used for performing target information association on an area with low target intensity and low requirement on space information precision; mainly for vehicle, aircraft and ship validation in the sea or in areas with few targets.
The target information association based on the registration information is used for performing target information association on an area with higher target intensity and higher requirement on spatial information precision; the method is mainly used for confirming the states of key roads and disaster areas.
The target information correlation based on the registration information comprises the positioning precision provided by the information source, wherein the target positioning error range after the registration with the geocoded base map is adopted.
Specifically, the target information provided by the target information association based on the platform positioning information comprises a target geographic position, positioning accuracy, a direction, an aspect ratio and an area;
the target information provided by the target information correlation based on the registration information comprises target position, positioning precision, direction, aspect ratio and area obtained by registration with the base map.
Specifically, the step M1 includes:
step M1.1: according to the target positioning precision range provided by the information sources, carrying out target information association on every two information sources in all the information sources;
step M1.2: selecting a precision error with large precision error as a precision error range within the positioning precision error range of every two information sources, and directly establishing an association relationship between targets when only one target has a corresponding relationship with the other target within the selected precision error range; when a plurality of targets are in a corresponding relation with one target within a selected precision error range, establishing a correlation relation based on the target position, the geometric attribute and the topological relation of target distribution by adopting an Eigenvector graph matching algorithm to realize accurate correlation between different targets provided by different information sources, and forming a corresponding data table of target serial numbers pointing to the same target in different target information provided by different information sources;
step M1.3: and when the matching chains conflict, establishing an association relation by using the matching chain with the highest sum of the target detection confidence degrees in the matching chains.
Specifically, the Eigenvector graph matching algorithm in the step M1.2 includes:
step M1.2.1: establishing distance measurement between targets, including target position distance and attribute distance;
the target position distance is an Euclidean distance between target absolute coordinates under the constraint of positioning errors;
the attribute distance measurement is the compatibility of decision attributes of targets detected by all information sources, and the formula is as follows:
D(x i ，x’ j )＝w d d(x i ，x’ j )+(1-w d )a(x i ，x’ j )
wherein, D (x) i ，x′ j ) Representing a measure of the distance of the attribute, w d Is the location distance weight, d (x) i ，x’ j ) Is the position Euclidean distance, a (x) i ，x’ j ) Is the attribute distance, x i ∈X＝{x 1 ，x 2 ，…x m }，x′ j ∈X′＝{x′ 1 ，x′ 2 ，…x′ n }, X, X' are sets of objects detected by two sources;
step M1.2.2: establishing a target similarity incidence matrix H-H according to the attribute distance metric value ij ]The formula is as follows:
h ij ＝exp(-D(x i ，x’ j ) 2 /2σ 2 )
wherein h is ij Representing the integrated distance between the targets, the element of the ith row and the jth column is the distance between the target i and the target j; the sigma represents a distance weight parameter and represents distance sensitivity under different target density scenes, the sigma is a constant, an initial value is a preset value, and the sigma is adjusted and optimized in real time according to space-based application performance;
step M1.3.3: solving the incidence matrix by adopting an Eigenvector graph matching algorithm;
for the moment of correlationPerforming singular value decomposition on the matrix H to obtain G-TDU, wherein T and U are m-dimension and n-dimension orthogonal matrixes respectively, D is a non-negative diagonal matrix, and G represents a matrix obtained after singular value decomposition; when m is less than n, U only has significance in the first m rows; replacing D with an identity matrix I to obtain another orthogonal matrix P ═ TIU, when P is ij And when the element is the maximum element of the ith row and the ith column in the matrix P, the point pair is a potential correct match, and in the determined potential matching relationship, the unsatisfied matched pair is removed by using the position accuracy constraint.
Specifically, step M2 includes: performing fusion decision of a target existence layer and fusion decision of a target identity recognition layer based on the associated target information;
and performing target existing layer fusion decision based on the associated target information to obtain a target existing layer fusion decision result, and performing target identity recognition layer fusion decision on the target existing layer fusion decision result.
Specifically, the target presence layer fusion decision includes: performing decision-level fusion;
reading n information sources which all participate in the fusion, wherein the credibility parameter of the ith information source is d i The detection confidence c of the current detection target t of the ith information source it ；
Let (1,0, -1) denote the decision space, where 1 denotes target, 0 denotes non-target, -1 denotes uncertain;
obtaining a basic probability assignment function of a source i to a target t: m is it (1)＝c it d i 、m it (0)＝c it (1-d i )、m it (-1)＝1-c it ；
And setting K as a conflict coefficient, and obtaining a mixed basic probability assignment function of all the information sources to the target t as follows:
selection of m t (1)、m t (0) And m t And (1) taking the decision branch with the maximum median as the decision result of the target.
Specifically, the target identity recognition layer fusion decision includes: performing decision-level fusion and feature-level fusion;
the decision-level fusion establishes a decision space for the decision event of the target identity by using each information source, and matches the target attribute information provided by each information source with the reference attribute data of the target type and/or model corresponding to each decision event in the decision space in a satellite-borne target attribute knowledge base; the decision-making steps are the same as the decision-level fusion of the target presence layer fusion decisions.
Performing comprehensive target identification analysis by using target characteristic information provided by each information source in characteristic level fusion; the input of the feature level fusion is the target feature information provided by the associated member star, and the fusion recognition result is output after the classification by the multi-level cascade classifier.
In particular, the decision-level re-fusion makes use of DSmT theory for re-fusion decisions.
Example 2
Example 2 is a modification of example 1
As shown in fig. 1, an embodiment of the present invention provides an elastic space-based multidimensional information fusion process based on multi-source data driving, and for target characteristics revealed by SAR information source, optical information source, and electromagnetic information source information at the same time phase, a decision-level fusion means and a feature-level fusion means are adopted to perform multi-source information fusion, and a decision-level fusion is performed on an output result, and target information of each information source is associated, supplemented, and verified with each other, so that high confidence identification and multidimensional state information synthesis of a target are achieved, and an important target product and a regional comprehensive information product are formed. The typical hardware architecture of the space-based multi-dimensional information fusion system shown in fig. 2 is adopted. The hardware platform comprises processing board and power strip, and the form that the two adopted range upon range of is in order to adopt the connector connection between the board, and processing board contains: a core processing unit and a management unit.
The information correlation realizes the correlation calculation of (suspected) target information from a plurality of information sources, establishes the corresponding relation among a plurality of target information, and provides positioning information with two types according to the different types of the information sources: the space geographic information of the target is obtained through calculation based on the satellite platform information, and the space information of the (suspected) target is obtained through registration based on the public base map, wherein the space geographic information is suitable for the sea surface or the area with few targets, and the space geographic information is suitable for the area with high target intensity and higher requirement on space information precision.
For the confirmation of vehicles, airplanes and ships on the sea or in a region with few targets, the processing flow is as shown in fig. 3, and due to the requirement of a large-range view field, the association of target information is realized by using large-range optical, SAR information source information and electromagnetic information source information, and the positioning information of the target given by a satellite platform, the attribute and the spatial distribution characteristic of the target are required to be used. The general satellite platform has the positioning accuracy of 500m-600m, the probability of gathering a plurality of targets in the area satisfying the condition is not high, and even if a plurality of targets exist, the targets can be associated through the attribute association of the targets, the spatial distribution topology of the targets and the like.
Aiming at the region with higher target density and higher requirement on spatial information precision, the processing flow is shown in fig. 4, the environment prior information provided by the ground is effectively utilized as a support, and the searching range of the key target is effectively restricted to the prior ROI through the technical links of base map (high-resolution remote sensing image after geographic fine correction, such as an ortho-image) registration and ROI mapping, so that the space of information searching is reduced, the target can be detected and identified only in the ROI, and the calculating speed is improved; more importantly, the spatial registration of the data of multiple information sources at the same time phase is realized through the registration of the information of each information source and the base map, so that the spatial alignment of the characteristic information and the target information is indirectly realized, and a foundation is laid for establishing the association relationship.
And according to the precision range of each target, performing information association of every two information sources and targets. If a plurality of targets appear in the position error range, in order to realize accurate association between different targets provided by different information sources, an Eigenvector graph matching algorithm is adopted, and an association relationship is established by considering the topological relationship of the target position, the geometric attribute and the target distribution. On the basis, a complete matching chain is formed by pairwise correlation matching, and target correlation among multiple information sources is realized. And if the matching chains have conflicts, establishing a matching relation by using the highest matching chain of the sum of the target detection confidences in the matching chains, so as to solve the conflicts among the association relations.
On the basis of information association, information fusion and decision-level re-fusion are completed by means of decision-level fusion and feature-level fusion, and fusion classification recognition is performed on the targets, so that the recognition accuracy and the confidence coefficient are improved. In consideration of different application scenes, a plurality of scene fusion modes can be specified in advance, and in different fusion modes, the fusion process is the same, but the information sources participating in fusion are different. Taking the processing flow of the information fusion and identification submodule for searching and rescuing in open sea at daytime in fig. 5 and the processing flow of the target information fusion and identification submodule for searching and rescuing in open sea at night in fig. 6 as an example, it can be seen that the preferred information source will generate corresponding changes in different scenes.
Specifically, the existence judgment of the target is realized by adopting a target existence layer fusion decision, the confidence coefficient of the target existence judgment (namely the confidence coefficient of the final fusion detection) is given, the false alarm is further removed through information fusion, and the detection confidence coefficient is improved. Firstly, establishing a basic probability function for the information source to judge aiming at the target existing layer by utilizing the statistical indexes such as target detection rate, false alarm rate and the like provided by each related information source and the detection confidence coefficient provided for each related target in real time; and then, establishing a mixed probability function about each associated target by using an inference criterion of the D-S evidence theory so as to obtain a judgment result of the existence of all related source information on a certain target.
The information of the presence layer fusion decision input includes: credibility parameters (described by indexes such as target detection rate, false alarm rate and the like in the statistical sense of each information source) of the related information sources participating in fusion and detection confidence degree provided by each information source for detecting the target in real time; the output is fused vessel target presence confidence (i.e., confidence of fused detection).
Specifically, a target identity recognition layer fusion decision is adopted, and the process is as shown in fig. 7, and all related information sources participating in fusion are used for fusing identity information such as target types, models and the like with high confidence in the description of the associated ship target attribute information. The input information is attribute information (including type, confidence and other related attributes) provided by information sources related to the targets judged to exist through the existence decision, and reference attribute data of the types/models of the targets in the satellite-borne target attribute knowledge base. The output is a target identity judgment result comprising a target type (model), a judgment confidence degree and the like.
Specifically, the fusion decision process of the target identity recognition layer comprises the following steps:
(1) the target identity recognition layer performs decision level fusion, and a decision space is established by utilizing the judgment events (specific target types, models and the like) of all information sources for the target identity; then, matching the target attribute information provided by each information source with reference attribute data of the target type/model corresponding to each decision event in the decision space in a satellite-borne target attribute knowledge base, and establishing a basic probability function of each information source for judging the target identity by using matching measurement; and finally, establishing a mixed probability function according to the reasoning criterion of the D-S evidence theory, thereby obtaining a decision-level fusion recognition result of the target identity fusing information of each information source. The specific algorithmic process is analogous to target presence decision fusion.
(2) And (4) performing feature level fusion on the target identity recognition layer, and performing comprehensive target recognition analysis by using feature information extracted from the original image information by each information source. The input is the target characteristic information provided by the associated information source, and the output is the fusion recognition result. The feature level fusion process taken is shown in fig. 8.
The process comprises training of a cascade classifier and real-time feature level fusion target recognition. The construction of the classifier mainly comprises the steps of establishing a training sample set and extracting features as input of classifier training. The training process is completed by using a Gentle Adaboost algorithm and an SVM algorithm, and the final training target is to form a cascade classifier consisting of an Adaboost classifier and an SVM classifier. And in the on-orbit feature level fusion stage, importing the well-correlated information source target feature data into a cascade classifier, and outputting an identification result after classification and identification.
(3) Based on the re-decision of the DSmt theory, as shown in fig. 9, the decision-level fusion identity recognition and the feature-level fusion identity recognition both obtain identity recognition results, and for two groups of results obtained from different discrimination methods, the DSmt theory is further utilized to perform re-fusion decision, and finally a fusion result output product is obtained.
Those skilled in the art will appreciate that, in addition to implementing the system and its various devices, modules, units provided by the present invention as pure computer readable program code, the system and its various devices, modules, units provided by the present invention can be fully implemented by logically programming method steps in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Therefore, the system and various devices, modules and units thereof provided by the invention can be regarded as a hardware component, and the devices, modules and units included in the system for realizing various functions can also be regarded as structures in the hardware component; means, modules, units for realizing various functions can also be regarded as structures in both software modules and hardware components for realizing the methods.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The embodiments and features of the embodiments of the present application may be combined with each other arbitrarily without conflict.
Those skilled in the art will appreciate that, in addition to implementing the systems, apparatus, and various modules thereof provided by the present invention in purely computer readable program code, the same procedures can be implemented entirely by logically programming method steps such that the systems, apparatus, and various modules thereof are provided in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Therefore, the system, the device and the modules thereof provided by the present invention can be considered as a hardware component, and the modules included in the system, the device and the modules thereof for implementing various programs can also be considered as structures in the hardware component; modules for performing various functions may also be considered to be both software programs for performing the methods and structures within hardware components.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The embodiments and features of the embodiments of the present application may be combined with each other arbitrarily without conflict.
Claims (7)
1. A multi-source data-driven space-based multi-dimensional information fusion system is characterized by comprising:
module M1: carrying out target information association on target information in the same-phase multi-information-source information;
module M2: performing multi-source information fusion based on the associated target information;
module M3: performing decision-level re-fusion on the output result of the multi-source information fusion to obtain a fusion product, and packaging the fusion product to output important target products and regional comprehensive information;
the module M2 includes: performing fusion decision including a target existence layer and a target identity recognition layer based on the associated target information;
performing target existence layer fusion decision based on the associated target information to obtain a target existence layer fusion decision result, and performing target identity recognition layer fusion decision on the target existence layer fusion decision result;
the target presence layer fusion decision comprises: performing decision-level fusion;
reading n information sources which all participate in the fusion, wherein the credibility parameter of the ith information source is d i The detection confidence c of the current detection target t of the ith information source it ；
Let (1,0, -1) denote the decision space, where 1 denotes target, 0 denotes non-target, -1 denotes uncertain;
obtaining a basic probability assignment function of a source i to a target t: m is it (1)＝c it d i 、m it (0)＝c it (1-d i )、m it (-1)＝1-c it ；
And setting K as a collision coefficient, and obtaining a mixed basic probability assignment function of all the information sources to the target t as follows:
selection of m t (1)、m t (0) And m t The decision branch with the maximum median value (-1) is taken as the decision result of the target;
the target identity recognition layer fusion decision comprises the following steps: performing decision-level fusion and feature-level fusion;
the decision-level fusion establishes a decision space for the decision event of the target identity by using each information source, and matches the target attribute information provided by each information source with the reference attribute data of the target type and/or model corresponding to each decision event in the decision space in a satellite-borne target attribute knowledge base;
performing comprehensive target identification analysis by using target characteristic information provided by each information source in characteristic level fusion; the input of the feature level fusion is the target feature information provided by the associated member satellite, and the fusion identification result is output after the classification by the multi-level cascade classifier.
2. The multi-source-data-driven-based space-based multi-dimensional information fusion system according to claim 1, wherein the module M1 comprises: according to different source types, the target information association comprises the following steps: target information association based on platform positioning information and target information association based on registration information;
the target information correlation based on the registration information comprises the positioning accuracy provided by the information source, wherein the target positioning error range after the target positioning accuracy is registered with the geocoded base map is adopted.
3. The multi-source data-driven-based space-based multi-dimensional information fusion system of claim 2, wherein the target information provided by the target information association based on platform positioning information comprises target geographic position, positioning accuracy, direction, aspect ratio and area;
the target information provided by the target information correlation based on the registration information comprises target position, positioning accuracy, direction, aspect ratio and area obtained by registering with the base map.
4. The multi-source-data-driven-based space-based multi-dimensional information fusion system according to claim 2, wherein the module M1 comprises:
module M1.1: performing target information association on every two information sources in all the information sources according to the target positioning precision range provided by the information sources;
module M1.2: selecting a precision error with large precision error as a precision error range within the positioning precision error range of every two information sources, and directly establishing an association relationship between targets when only one target has a corresponding relationship with the other target within the selected precision error range; when a plurality of targets are in a corresponding relation with one target within a selected precision error range, establishing a correlation relation based on the target position, the geometric attribute and the topological relation of target distribution by adopting an Eigenvector graph matching algorithm to realize accurate correlation between different targets provided by different information sources, and forming a corresponding data table of target serial numbers pointing to the same target in different target information provided by different information sources;
module M1.3: and when the matching chains conflict, establishing an association relation by using the matching chain with the highest sum of the target detection confidence degrees in the matching chains.
5. The multi-source data-driven space-based multi-dimensional information fusion system according to claim 4, wherein the Eigenvector graph matching algorithm in the module M1.2 comprises:
module M1.2.1: establishing distance measurement between targets, including target position distance and attribute distance;
the target position distance is an Euclidean distance between target absolute coordinates under the constraint of positioning errors;
the attribute distance measurement is the compatibility of decision attributes of targets detected by all information sources, and the formula is as follows:
D(x i ,x’ j )＝w d d(x i ,x’ j )+(1-w d )a(x i ,x’ j )
wherein, D (x) i ,x′ j ) Representing a measure of the distance of the attribute, w d Is the location distance weight, d (x) i ,x’ j ) Is the position Euclidean distance, a (x) i ,x’ j ) Is the attribute distance, x i ∈X＝{x 1 ,x 2 ,…x m }，x' j ∈X'＝{x' 1 ,x' 2 ,…x' n }, X, X' are sets of objects detected by two sources;
module M1.2.2: establishing a target similarity incidence matrix H-H according to the attribute distance metric value ij ]The formula is as follows:
h ij ＝exp(-D(x i ,x’ j ) 2 /2σ 2 )
wherein h is ij Representing the integrated distance between the targets, the element of the ith row and the jth column is the distance between the target i and the target j; the sigma represents a distance weight parameter and represents distance sensitivity under different target density scenes, the sigma is a constant, an initial value is a preset value, and the sigma is adjusted and optimized in real time according to space-based application performance;
module M1.2.3: solving the incidence matrix by adopting an Eigenvector graph matching algorithm;
performing singular value decomposition on the incidence matrix H to obtain G-TDU, wherein T and U are m-dimension and n-dimension orthogonal matrixes respectively, D is a non-negative diagonal matrix, and G represents a matrix obtained after singular value decomposition; when m is<When n, U only has significance in the first m rows; replacing D with an identity matrix I to obtain another orthogonal matrix P ═ TIU, when P is ij And when the maximum element is the ith row and the j column in the matrix P, the point pair is a potential correct match, and in the determined potential matching relationship, the unsatisfied matched pair is removed by using the position accuracy constraint.
6. The multi-source data-driven-based space-based multi-dimensional information fusion system of claim 1, wherein the decision-level re-fusion utilizes DSmT theory for re-fusion decisions.
7. A multi-source data-driven space-based multi-dimensional information fusion method is characterized by comprising the following steps:
step M1: carrying out target information association on target information in the same-phase multi-information-source information;
step M2: performing multi-source information fusion based on the associated target information;
step M3: performing decision-level re-fusion on the output result of the multi-source information fusion to obtain a fusion product, and packaging the fusion product to output important target products and regional comprehensive information;
the step M2 includes: performing fusion decision including a target existence layer and a target identity recognition layer based on the associated target information;
performing target existence layer fusion decision based on the associated target information to obtain a target existence layer fusion decision result, and performing target identity recognition layer fusion decision on the target existence layer fusion decision result;
the target presence layer fusion decision comprises: performing decision-level fusion;
reading n information sources which all participate in the fusion, wherein the credibility parameter of the ith information source is d i This detection of the ith sourceConfidence of detection c of target t it ；
Let (1,0, -1) denote the decision space, where 1 denotes target, 0 denotes non-target, -1 denotes uncertain;
obtaining a basic probability assignment function of a source i to a target t: m is it (1)＝c it d i 、m it (0)＝c it (1-d i )、m it (-1)＝1-c it ；
And setting K as a conflict coefficient, and obtaining a mixed basic probability assignment function of all the information sources to the target t as follows:
selection of m t (1)、m t (0) And m t The decision branch with the maximum median value (-1) is taken as the decision result of the target;
the target identity recognition layer fusion decision comprises the following steps: performing decision-level fusion and feature-level fusion;
the decision-level fusion establishes a decision space for the decision event of the target identity by using each information source, and matches the target attribute information provided by each information source with the reference attribute data of the target type and/or model corresponding to each decision event in the decision space in a satellite-borne target attribute knowledge base;
performing comprehensive target identification analysis by using target characteristic information provided by each information source in characteristic level fusion; the input of the feature level fusion is the target feature information provided by the associated member star, and the fusion recognition result is output after the classification by the multi-level cascade classifier.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010754344.7A CN111931833B (en) | 2020-07-30 | 2020-07-30 | Multi-source data driving-based space-based multi-dimensional information fusion method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010754344.7A CN111931833B (en) | 2020-07-30 | 2020-07-30 | Multi-source data driving-based space-based multi-dimensional information fusion method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111931833A CN111931833A (en) | 2020-11-13 |
| CN111931833B true CN111931833B (en) | 2022-08-12 |
Family
ID=73315721
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010754344.7A Active CN111931833B (en) | 2020-07-30 | 2020-07-30 | Multi-source data driving-based space-based multi-dimensional information fusion method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111931833B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113298006B (en) * | 2021-06-04 | 2024-01-19 | 西北工业大学 | Novel abnormal target detection method based on brain-computer fusion cognition and decision |
| CN115600158B (en) * | 2022-12-08 | 2023-04-18 | 奥特贝睿(天津)科技有限公司 | Unmanned vehicle multi-sensor fusion method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108802741A (en) * | 2018-06-22 | 2018-11-13 | 西安电子科技大学 | Mobile robot sonar data fusion method based on DSmT theories |
| CN110186820A (en) * | 2018-12-19 | 2019-08-30 | 河北中科遥感信息技术有限公司 | Multisource data fusion and environomental pollution source and pollutant distribution analysis method |
| CN110207671A (en) * | 2018-12-29 | 2019-09-06 | 中国科学院软件研究所 | A kind of space-based intelligence imaging system |
| CN110443201A (en) * | 2019-08-06 | 2019-11-12 | 哈尔滨工业大学 | The target identification method merged based on the shape analysis of multi-source image joint with more attributes |
| CN111353450A (en) * | 2020-03-06 | 2020-06-30 | 北京波尔通信技术股份有限公司 | Target identification system and method based on heterogeneous electromagnetic perception information fusion |
-
2020
- 2020-07-30 CN CN202010754344.7A patent/CN111931833B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108802741A (en) * | 2018-06-22 | 2018-11-13 | 西安电子科技大学 | Mobile robot sonar data fusion method based on DSmT theories |
| CN110186820A (en) * | 2018-12-19 | 2019-08-30 | 河北中科遥感信息技术有限公司 | Multisource data fusion and environomental pollution source and pollutant distribution analysis method |
| CN110207671A (en) * | 2018-12-29 | 2019-09-06 | 中国科学院软件研究所 | A kind of space-based intelligence imaging system |
| CN110443201A (en) * | 2019-08-06 | 2019-11-12 | 哈尔滨工业大学 | The target identification method merged based on the shape analysis of multi-source image joint with more attributes |
| CN111353450A (en) * | 2020-03-06 | 2020-06-30 | 北京波尔通信技术股份有限公司 | Target identification system and method based on heterogeneous electromagnetic perception information fusion |
Non-Patent Citations (5)
| Title |
|---|
| Data fusion of multi-source satellite data sets for cost-effective disaster management studies;B. Gokaraju 等;《SoutheastCon 2017》;20170511;全文 * |
| 基于DSmT的城市道路多源交通状态信息融合的应用研究;宋晓鹏 等;《中国水运》;20191031;全文 * |
| 基于D-S证据理论的无线传感器网络决策融合研究;张雅媛;《中国优秀硕士学位论文全文数据库 信息科技辑》;20191015(第10期);第2章 * |
| 天基信息港及其多源信息融合应用;李斌 等;《中国电子科学研究院学报》;20170630;第12卷(第3期);全文 * |
| 电子载荷与成像载荷协同高效应用思考;陈锋 等;《航天电子对抗》;20160830;第32卷(第4期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111931833A (en) | 2020-11-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Multiscale visual attention networks for object detection in VHR remote sensing images | |
| Al-qaness et al. | An improved YOLO-based road traffic monitoring system | |
| US5963653A (en) | Hierarchical information fusion object recognition system and method | |
| Zhang et al. | Data-driven intelligent transportation systems: A survey | |
| CN114269620A (en) | Performance testing of robotic systems | |
| US11651302B2 (en) | Method and device for generating synthetic training data for an artificial-intelligence machine for assisting with landing an aircraft | |
| Sun et al. | Vehicle reidentification using multidetector fusion | |
| CN112233097A (en) | Road scene other vehicle detection system and method based on space-time domain multi-dimensional fusion | |
| CN111931833B (en) | Multi-source data driving-based space-based multi-dimensional information fusion method and system | |
| Bendich et al. | Topological and statistical behavior classifiers for tracking applications | |
| Abdelraouf et al. | Using vision transformers for spatial-context-aware rain and road surface condition detection on freeways | |
| Ji et al. | Incremental online object learning in a vehicular radar-vision fusion framework | |
| CN112613668A (en) | Scenic spot dangerous area management and control method based on artificial intelligence | |
| Seeger et al. | Towards road type classification with occupancy grids | |
| CN117725448A (en) | Cluster analysis method for meteorological navigation signal characteristics | |
| Kochenderfer et al. | Uncorrelated encounter model of the National Airspace System version 1.0 | |
| Luo et al. | Traffic signal transition time prediction based on aerial captures during peak hours | |
| Li et al. | Real-time monocular joint perception network for autonomous driving | |
| Uzkent et al. | Feature matching with an adaptive optical sensor in a ground target tracking system | |
| Zhao | Exploring the fundamentals of using infrastructure-based LiDAR sensors to develop connected intersections | |
| Mendez et al. | Automatic label creation framework for fmcw radar images using camera data | |
| Helgesen | Sensor fusion for detection and tracking of maritime vessels | |
| Makarau et al. | Alphabet-based multisensory data fusion and classification using factor graphs | |
| Greenberg et al. | Neural-network classifiers for automatic real-world aerial image recognition | |
| Bouhsine et al. | Atmospheric visibility image-based system for instrument meteorological conditions estimation: A deep learning approach |
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 |