CN116340533B - Satellite-borne electromagnetic spectrum big data intelligent processing system based on knowledge graph - Google Patents
Satellite-borne electromagnetic spectrum big data intelligent processing system based on knowledge graph Download PDFInfo
- Publication number
- CN116340533B CN116340533B CN202310189348.9A CN202310189348A CN116340533B CN 116340533 B CN116340533 B CN 116340533B CN 202310189348 A CN202310189348 A CN 202310189348A CN 116340533 B CN116340533 B CN 116340533B
- Authority
- CN
- China
- Prior art keywords
- module
- spectrum
- knowledge graph
- knowledge
- signal
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Abstract
The invention belongs to the technical field of electromagnetic data information acquisition and processing, and particularly relates to an electromagnetic spectrum data intelligent processing system. A knowledge-graph-based satellite-borne electromagnetic spectrum big data intelligent processing system comprises: the system comprises a frequency spectrum knowledge spectrum module as a basic module, a blind source separation module, a feature extraction module, a rule mining module, a feature identification module and a passive positioning module which are involved in a data processing flow; the spectrum knowledge graph module is used for serving each module of the upper layer data processing flow, and after each module of the upper layer data processing flow processes data, the content of the spectrum knowledge graph module is reversely supplemented to form a complementary mode. The system can improve the accuracy of classification and identification and the accuracy of passive positioning.
Description
Technical Field
The invention belongs to the technical field of electromagnetic data information acquisition and processing, and particularly relates to an electromagnetic spectrum data intelligent processing system.
Background
With the development of science and technology and the importance of various countries to electronic reconnaissance satellites, the variety and quantity of relevant devices are rapidly increased, electromagnetic data is explosively increased, the capability of satellite ground application systems is continuously enhanced, the generated information is continuously enriched, the reconnaissance data and the information data quantity in the life cycle of each satellite system are also increased by several times, the quantity of satellites is increased, and the total data quantity reaches the PB (gigabytes) scale. Meanwhile, as the electronic reconnaissance capability of the satellite is improved, the quality of reconnaissance information is greatly improved, and the reconnaissance data information fusion among different satellites is assisted, so that more information is obtained.
While electronic reconnaissance satellites are rapidly evolving, the various satellite application systems described above have significant drawbacks in terms of: firstly, the processing depth of electronic reconnaissance data is insufficient, and particularly the depth mining capability based on multi-star combination is lacking; secondly, the processing level is insufficient, the data discrimination, single-target simple discrimination, platform simple discrimination and the like are basically performed at present, the analysis capability of the behavior layer information of the electronic target is weaker, and the electronic combat intention information, the prediction information and the group target task information are basically in blank states; thirdly, the self-adaption and self-learning capabilities of the application system are weak, and interaction, feedback capability and self-adaption capability with a reconnaissance load or an external environment are lacked; fourth, the tactical support capability of the application system is weaker, which is reflected in insufficient close relation with the joint combat action task and insufficient real-time support capability. Big data technology is developed to the present day, a plurality of technologies are relatively mature, and the technical characteristics of abundant models, self-adaption and the like can be well applied to solving approaches of the problems. Therefore, research on electronic reconnaissance data mining and intelligent processing technologies based on big data technology is particularly urgent.
Disclosure of Invention
The purpose of the invention is that: aiming at the defects of the prior various satellite application systems, based on the data of the satellite-borne electromagnetic spectrum and the knowledge spectrum of the electromagnetic data, and combining the characteristics of the electromagnetic spectrum, a set of intelligent processing system suitable for the electromagnetic spectrum data is provided.
The technical scheme of the invention is as follows: a knowledge-graph-based satellite-borne electromagnetic spectrum big data intelligent processing system comprises: the system comprises a frequency spectrum knowledge spectrum module as a basic module, a blind source separation module, a feature extraction module, a rule mining module, a feature identification module and a passive positioning module which are involved in a data processing flow; the spectrum knowledge graph module is used for serving each module of the upper layer data processing flow, and after each module of the upper layer data processing flow processes data, the content of the spectrum knowledge graph module is reversely supplemented to form a complementary mode.
Based on the scheme, the method specifically comprises the following steps:
the blind source separation module is used for processing the aliased satellite-borne electromagnetic signals to realize separation of source signals; in the processing process of the blind source separation module, the spectrum knowledge graph module provides signal characteristic comparison support; after the processing is finished, the blind source separation module sends the separated signals to the spectrum knowledge graph module to complement the signals perfectly.
The feature extraction module extracts target signal feature parameters of the separation signals output by the blind source separation module by using priori knowledge in the frequency spectrum knowledge graph module; the extracted features are also sent to a spectrum knowledge graph module to perfect and supplement the features, and enough priori information is provided for a rule mining module.
The rule mining module is used for mining mass data by utilizing a data model trained by the frequency spectrum knowledge graph module, and the mining result is sent to the frequency spectrum knowledge graph module for perfect supplementation, so that support is provided for the feature recognition module and the passive positioning module.
The feature recognition module is used for recognizing the modulation mode of the signal based on priori knowledge in the frequency spectrum knowledge graph module and the feature information extracted by the feature extraction module; after the identification is completed, the obtained signal modulation mode is sent to a spectrum knowledge graph module to be perfectly supplemented.
The passive positioning module realizes passive positioning of signals based on priori knowledge in the frequency spectrum knowledge graph module and the characteristic information extracted by the characteristic extraction module; after the positioning is finished, if the rule mining module has completed the position information of the target signal in the spectrum knowledge graph module, comparing the calculated target position information with the data in the spectrum knowledge graph module, so as to improve the positioning precision and perfect the signal positioning information; if the rule mining module does not complement the position information of the target signal in the spectrum knowledge graph module, the positioning information of the single signal is directly supplemented to the spectrum knowledge graph module.
The beneficial effects are that: (1) The invention combines the characteristics of satellite-borne electromagnetic data to carry out layered design on the processing flow of electromagnetic spectrum, and is provided with a spectrum knowledge spectrum module serving as a basic module, a blind source separation module, a feature extraction module, a rule mining module, a feature identification module and a passive positioning module which are involved in the data processing flow, thereby realizing decoupling among functional modules through a whole set of layered architecture design. Finally, the modulation type analysis of the target signal and the position analysis of the target signal are realized.
(2) The invention applies the spectrum knowledge graph module to the whole system to serve as a basic support of the system. Firstly, a knowledge spectrum layer is constructed in a supplementing mode based on a data acquisition layer, and then the knowledge spectrum layer provides support for an upper blind source separation, feature extraction, rule mining, feature identification and passive positioning module. Meanwhile, each module reversely supplements the knowledge graph module, perfects the triplet information in the knowledge graph module, extracts valuable information from the triplet information, and improves the accuracy of classification and identification and the accuracy of passive positioning.
Drawings
Fig. 1 is a block diagram of the components of the present invention.
Detailed Description
Example 1: referring to fig. 1, an intelligent processing system for satellite-borne electromagnetic spectrum big data based on a knowledge graph comprises: the system comprises a frequency spectrum knowledge spectrum module as a basic module, a blind source separation module, a feature extraction module, a rule mining module, a feature identification module and a passive positioning module which are involved in a data processing flow; the spectrum knowledge graph module is used for serving each module of the upper layer data processing flow, and after each module of the upper layer data processing flow processes data, the content of the spectrum knowledge graph module is reversely supplemented to form a complementary mode. Wherein:
the blind source separation module is used for processing the aliased satellite-borne electromagnetic signals to realize separation of source signals; in the processing process of the blind source separation module, the spectrum knowledge graph module provides signal characteristic comparison support; after the processing is finished, the blind source separation module sends the separated signals to the spectrum knowledge graph module to complement the signals perfectly.
The feature extraction module extracts target signal feature parameters of the separation signals output by the blind source separation module by using priori knowledge in the frequency spectrum knowledge graph module; the extracted features are sent to a spectrum knowledge graph module to be perfectly supplemented, and enough priori information is provided for a rule mining module.
The rule mining module is used for mining mass data by utilizing a data model trained by the frequency spectrum knowledge graph module, and the mining result is sent to the frequency spectrum knowledge graph module for perfect supplementation, so that support is provided for the feature recognition module and the passive positioning module.
The feature recognition module is used for recognizing the modulation mode of the signal based on priori knowledge in the frequency spectrum knowledge graph module and the feature information extracted by the feature extraction module; after the identification is completed, the modulation mode of the obtained signal is sent to a spectrum knowledge graph module to be perfectly supplemented.
The passive positioning module realizes passive positioning of signals based on priori knowledge in the frequency spectrum knowledge graph module and the characteristic information extracted by the characteristic extraction module; after the positioning is finished, if the rule mining module has completed the position information of the target signal in the spectrum knowledge graph module, comparing the calculated target position information with the data in the spectrum knowledge graph module, so as to improve the positioning precision and perfect the signal positioning information; if the rule mining module does not complement the position information of the target signal in the spectrum knowledge graph module, the positioning information of the single signal is directly supplemented to the spectrum knowledge graph module.
Example 2: on the basis of embodiment 1, a spectrum knowledge graph module, a blind source separation module, a feature extraction module, a rule mining module, a feature recognition module and a passive positioning module are further described respectively:
(1) Spectrum knowledge graph module
The spectrum knowledge graph module comprises: a mode layer and a data layer; the mode layer comprises: knowledge entry class, background knowledge class, real-time information class, and content of a single signal class; the data layer is a specific implementation of the mode layer and is a scheme for data presentation.
Further, the knowledge entry class contains 3 knowledge units: time, space and frequency band; wherein, time refers to the starting time to be searched, space refers to different geographic positions to be monitored, and frequency band refers to the frequency band range to be searched. The attributes in the knowledge entry class are labels for each knowledge entity, and the monitoring node IDs are added to these attributes to generate unique identification IDs.
Background knowledge classes contain 3 subclasses: registering information class, hardware information class and signal template class; wherein the registration information class includes: modulation type, frequency band and working time; the hardware information class includes: manufacturer, antenna, transmit power; the signal template class includes: a frequency band template, a signal template and an abnormal template; background knowledge classes can provide rich a priori information for knowledge mining and reasoning.
The real-time information class includes: background noise power, electromagnetic wave power, spectrogram, signal number, idle frequency band and frequency band practical ratio 7 attributes; the real-time information class is used to collect knowledge units contained in the real-time frequency band.
A single signal class contains 3 subclasses: static information class, real-time change class and knowledge reasoning class; the static information class includes: frequency band, modulation mode, bandwidth; the real-time change class includes: decoding content, IQ waveforms, symbol rates; the knowledge reasoning class includes: doppler rate of change, signal arrival angle, signal reception strength, signal localization, and anomaly type. A single signal class is a categorization of the detailed attributes of a single signal.
The method comprises the steps of storing main standard data of a data layer in a frequency spectrum knowledge graph module, defining a knowledge graph of a radio according to a mode layer, establishing 8 tables in total of the data layer of the knowledge graph, including a knowledge entry table, a table corresponding to a real-time information class, and respectively establishing three tables by a background knowledge class and a single signal class. The knowledge entry class stores the data of time, space and frequency band of the field by taking the primary key id as a unique identifier. The real-time information class contains 6 fields of a table, namely background noise power, electromagnetic wave power, a spectrogram, the number of signals, idle frequency bands and frequency band utilization rate. The background knowledge class contains three tables, namely registration information of legal stations, hardware information of different nodes and history information of different templates. A single signal class contains three tables. A basic information table of signals, a detail information table of signals and an inference information table of signals. The signal basic information table stores signal basic information such as frequency bands, bandwidths and the like, the signal detail information table stores signal content information obtained by decoding signals, and the signal reasoning information table stores information obtained by in-depth analysis of the signals. Because the monitoring of a single signal cannot separate the frequency band information, the three tables all contain field external keys corresponding to the main keys of the real-time information data table.
(2) Blind source separation module
The data source mainly processed by the system consists of electromagnetic data on the satellite and other electromagnetic data. The electromagnetic data on board the satellite are mainly composed of reconnaissance electromagnetic data (ships, radars, airplanes). Other data mainly comprise data such as internet, weather hydrology, information and the like.
Electromagnetic data signals received by the blind source separation module are usually displayed as aliasing signals, and the blind source separation module is used for processing the electromagnetic data signals respectively by using prior information of each type of data, so that the multi-target positioning problem can be converted into a single target positioning problem. Generally, the blind source separation model can be divided into an overdetermined model, a positive model and an underdetermined model according to the number of signal sources which are smaller, equal to or larger than the number of sensors, but for a low-rail electromagnetic monitoring system, satellites are always fewer than radiation sources, and are often suitable for the underdetermined blind source separation model. The blind source separation module adopts a two-step method to realize the separation of source signals. The idea of the two-step method is that a mixing matrix is obtained by estimating according to observation signals, and then the source signals are separated by an optimization algorithm by combining the estimated mixing matrix.
The spectrum knowledge graph module provides support for blind source separation of signals. During interaction, the blind source separation module separates the aliased signals through a blind source separation algorithm to obtain the information such as the frequency band, the bandwidth, the IQ waveform and the like of each independent component signal. And then, inquiring and obtaining the information such as frequency bands, bandwidths, IQ waveforms and the like of the single signals in the single signal class in the frequency spectrum knowledge graph module through the space-time frequency information of each component signal. And then comparing the frequency bands, bandwidths and IQ waveforms of the signals obtained by the blind source separation module with the frequency bands, bandwidths and IQ waveforms of the single signals obtained by the knowledge base, so that the accuracy of calculation of the blind source separation algorithm is improved. And finally, the signals separated by the blind source separation module are used for perfecting specific parameter information of single signals in the spectrum knowledge graph module.
(3) Feature extraction module
The feature extraction module is based on the spectrum knowledge graph module, utilizes the triplet information existing in the signals in the spectrum knowledge graph module, relies on the separation signals obtained by the blind source separation module, and aims at information requirements, and the extraction of the feature parameters of the target signals is realized by utilizing key technologies such as machine learning, rule mining and the like.
In the feature extraction module, the electromagnetic signal contains a plurality of feature parameters reflecting the attribute of the electromagnetic signal, including working frequency, pulse repetition frequency, pulse width, signal modulation mode, frequency change mode and the like. After the detected electromagnetic signals are subjected to processing such as correlation, filtering and feature extraction, a joint feature vector is obtained, wherein each parameter component of the vector represents each feature of the electromagnetic signals, and each given set of feature parameters represents an electromagnetic signal observation sample to be identified. Different signal characteristics can be obtained by the characteristic extraction method based on different principles, and the main characteristic extraction method is as follows:
A. feature extraction based on instantaneous value statistics
The characteristics of the instantaneous amplitude, frequency, phase and the like of the communication signal comprise useful modulation information, and the reasonable selection of the characteristics is a good means for acquiring the signal identification characteristics. Meanwhile, the extracted parameters such as time difference and frequency difference can also provide support for the subsequent passive positioning module.
B. Constellation-based feature extraction
The shape and state relationship of the digital modulation signal are hidden in the constellation diagram. The constellation diagram contains important characteristics capable of identifying the signal types, and different types of digital modulation signals have unique constellation diagrams, so that the relation is a one-to-one mapping relation, and modulation mode identification can be well carried out by using the mapping relation.
C. Feature extraction based on time-frequency analysis
The time-frequency distribution can represent the intensity and density conditions of the signals at different time and frequency, and the time-frequency two-dimensional distribution characteristics are used for representing the signals, so that different modulation types can be effectively identified. The wavelet transformation not only can well extract the time domain and frequency domain characteristics of signals, but also can distinguish waveform mutation, and is a mainstream technology of non-stationary signal time-frequency analysis.
D. Feature extraction based on higher order cumulants
The high-order cumulant characteristic can reflect the category characteristics of different modulation signals, has strong anti-noise interference capability, and plays a good role in modulation recognition. The time delay difference, namely TDOA, can be caused by different propagation distances among interference signals of the same source received by different satellites, and the TDOA parameters can be well estimated based on the high-order accumulation of the signals. The key precondition for passive positioning is to accurately and rapidly estimate and measure the TDOA value between the interference source signals transparently forwarded by each satellite and select a proper positioning algorithm.
The feature extraction module takes the features of a plurality of signals obtained by interaction of the blind source separation module and the frequency spectrum knowledge graph module as the input of the feature extraction module, can extract the features of the signals such as instantaneous time domain information, constellation diagram information, wavelet transformation information, signal cycle statistics features, high-order accumulation amount and the like, provides input parameters for a subsequent classifier, and finally realizes pattern classification based on feature extraction in modulation classification. Meanwhile, the signal features extracted by the feature extraction module can be used as the perfect information of the single signal in the spectrum knowledge graph module, and enough priori information is provided for a rule mining module in the next step.
(4) Rule mining module
The feature extraction module extracts a large amount of relevant information of each independent signal, such as time-frequency information, constellation diagram information, high-order accumulation amount and the like of the signals, and the information supplements and perfects knowledge graph information of the single signal. And then training a data model on the basis of the data obtained by the frequency spectrum knowledge graph module, wherein the trained data model is used by a rule mining module, and the rule mining result provides support for passive positioning and feature recognition in an upper application layer.
The core objective of the rule mining system is to mine information of interest of a user in mass data by analyzing information such as user behaviors, requirements and the like. The mining method is generally based on a large amount of interaction data between users and objects, and the knowledge graph technology is applied to the mining system, so that the problem of sparse data of object behavior relation of users which is difficult to be qualified by the traditional mining method can be solved, and the interpretation of mining is improved. The regular mining is based on the knowledge graph, and the mining result can complement the data of the triplet missing in the knowledge graph, thereby providing a large amount of prior information for the passive positioning and feature recognition.
The rule mining adopts a knowledge graph embedding method, and projects entities and relations in the knowledge graph to a low-dimensional and dense vector space, so that a computer can conveniently and quickly calculate. Because of the relation of a large number of more than one spectrum knowledge graphs, a TransR model is selected for learning. The entity, relationship, vector representation of the mapping matrix, and a scoring function are learned by the model. Taking link prediction as an example (knowledge graph is often subject to the problem of link missing, i.e. knowledge graph incompleteness, and identifying these missing links is called link prediction), given that the head entity, the relationship is given, the goal of link prediction is to predict the missing tail entity. In actual operation, all candidate tail entities under the head entity, the relation and the corresponding category are combined to form a triplet set to be tested, then score functions are calculated respectively to obtain the score ranking of the candidate tail entities, and the entity with the highest score ranking and larger than a set threshold value is used as a rule mining result. And finally storing the perfect triples into the knowledge graph to finish the tasks of completion of the knowledge graph and the like.
(5) Passive positioning module
After feature selection and regular mining by the upper layer data processing module, electromagnetic spectrum data are subjected to preliminary screening and feature parameter extraction, and the data provide a basis for passive positioning and feature recognition of an application service layer.
Passive positioning is a technique in which the receiver does not emit electromagnetic signals and the position of the source is determined using only the signals intercepted by a single or multiple receiving stations. The passive positioning has the advantages of good electromagnetic concealment, long positioning distance and the like, and can be effectively applied to a low-rail electromagnetic monitoring system. According to the positioning steps, the passive positioning technology can be divided into a two-step method and a direct method.
The basic idea of the two-step positioning method is as follows: and step 1, estimating positioning parameters which contain the position information of the radiation source for the intercepted radiation source signals, and step 2, establishing an equation and solving by utilizing the relation between the positioning parameters and the position of the radiation source to realize positioning.
Parameters to be estimated in the two-step method can be divided into two types, the first type is time delay and frequency difference obtained by calculating signals collected by a single antenna, and the second type is AOA obtained by measuring a special antenna array. The first kind of delay estimation method comprises the following steps: a matched filtering method, a time delay estimation algorithm (such as a maximum likelihood method) using a cost function, an adaptive time delay estimation algorithm (such as ETDE), a subspace time delay estimation algorithm (such as a MUSIC algorithm) and a mutual blurring function method. The frequency offset estimation method comprises the following steps: differential measurement, a mutual blurring function method, equivalent pulse compression and the like. The AOA in the second class needs to use an array antenna, and when far-field signals reach different array elements of the same array antenna, time delay is generated, so that phase differences between signals received by the different array elements are caused. Early AOA estimation is mostly realized based on beam forming, but the measurement accuracy is limited, and high accuracy is realized by using a spectrum estimation method based on a characteristic structure nowadays.
The direct localization method is to directly process an original sampling signal, construct a cost function only related to the position of a radiation source by utilizing the position information of the radiation source contained in the signal, and realize localization by using an optimization algorithm such as an exhaustive search, and is generally called as a direct localization method because the localization method realizes direct estimation from the signal to the position of the radiation source.
In the two-step positioning method, time difference, frequency difference, doppler change rate, signal arrival angle and received signal strength of the signals are required. The knowledge entry class in the knowledge base of the knowledge graph contains three fields of time, space and frequency band of the signal, and the stored parameters such as Doppler change rate, signal arrival angle, received signal strength and the like can be found in a single signal class through the knowledge entry class in the knowledge base. And then, using the parameters to perform positioning calculation by using a two-step algorithm to obtain a corresponding target position. In the upper layer rule mining module, if the rule mining module complements the position information of the target signal in the spectrum knowledge graph module, the calculated target position information can be compared with the data in the spectrum knowledge graph module, so that the positioning accuracy is improved, and meanwhile, the signal positioning information in the single signal class in the knowledge base is perfected. If the rule mining module does not complement the position information of the target signal in the spectrum knowledge graph module, the positioning information of the single signal can be directly supplemented to the spectrum knowledge graph module.
(6) Feature recognition module
The automatic identification of modulated signals is in essence to solve the pattern recognition problem for unknown variables in a given modulation set. Analysis summarizes that the existing modulation methods can be divided into two large categories, one is a method of recognition completed by using a decision theory, and the other is a method of pattern recognition completed by using a feature extraction method.
The modulation mode identification problem is regarded as a composite hypothesis test problem by the decision theory method, and the modulation mode identification is realized by selecting a proper threshold value as a decision basis through analysis and statistics. It has roughly three branches: likelihood ratio method, high order moment high order cumulant method, constellation diagram method. The triplet knowledge stored in the knowledge graph can be used as priori knowledge of a decision theory method.
The pattern recognition method based on the feature extraction is used for completing recognition of a modulation mode by extracting the features of a signal modulation mode and then comparing the features with the results of theoretical features; the method has the advantages of low calculation complexity, high working efficiency and adaptation to various models. The identification method of the target feature generally comprises two processes: classifier design and classifier identification. In the process of designing the classifier, model training data mined based on the rule of the knowledge graph in an upper module can be used as training samples to train the classifier. Training the classifier is in effect making the classifier adaptive to learn the classifier parameters. After the classifier training is completed, inputting the sample to be classified into the classifier, comparing and analyzing the data in the sample to be classified and the knowledge graph, and then completing the classification and identification. The radiation source classification recognition process is also actually a pattern recognition process, i.e. after completing the step of rule mining, the subsequent feature classification is performed by means of a classifier.
The feature recognition often requires parameters such as a higher-order accumulation of signals, time-frequency distribution of signals, wavelet amplitude difference of signals, and the like. The high-order accumulation parameters are extracted in the upper-layer feature extraction module, the law mining module perfects the high-order accumulation data of the missing signals, and the knowledge graph contains the high-order accumulation parameters of the signals, so that a large amount of priori data is formed, and support is provided for feature recognition of the signals. And a frequency spectrum knowledge graph module for perfecting the corresponding signal according to the modulation mode of the obtained signal.
Embodiment 3 provides an intelligent processing method for satellite-borne electromagnetic spectrum data based on embodiment 1 or 2.
Firstly, collected data is separated from aliased signals by a blind source separation mode, so that independent component signals are obtained, and the multi-target positioning problem is converted into a single-target positioning problem. And then extracting characteristic components of the signals through a characteristic extraction module, and perfecting the corresponding satellite-borne electromagnetic knowledge graph. On the basis of the feature extraction module, regular mining is performed, and a large amount of extracted signal features and priori data existing in the knowledge graph are combined to perform regular mining, so that missing information in the knowledge graph is filled. After passing through the upper layer feature extraction module and the rule mining module, the electromagnetic spectrum data are subjected to preliminary screening and feature parameter extraction, and the data provide a basis for a feature recognition module. The feature recognition module uses the data as a training sample to train a classifier, and then uses the machine learning-based classifier recognition to judge the type of the target so as to realize the recognition of the type of the target modulation signal. In the passive positioning module, the passive positioning of the signals is realized by a two-step method or a direct method based on the information such as space-time frequency information, doppler change rate, signal arrival angle, signal receiving intensity and the like of the signals in the knowledge graph and the feature extraction module.
While the invention has been described in detail in the foregoing general description and specific examples, it will be apparent to those skilled in the art that modifications and improvements can be made thereto. Accordingly, such modifications or improvements may be made without departing from the spirit of the invention and are intended to be within the scope of the invention as claimed.
Claims (3)
1. A satellite-borne electromagnetic spectrum big data intelligent processing system based on a knowledge graph is characterized by comprising: the system comprises a frequency spectrum knowledge spectrum module as a basic module, a blind source separation module, a feature extraction module, a rule mining module, a feature identification module and a passive positioning module which are involved in a data processing flow; the spectrum knowledge graph module is used for serving each module of the upper layer data processing flow, and after each module of the upper layer data processing flow processes data, the content of the spectrum knowledge graph module is reversely supplemented to form a complementary mode;
wherein:
the blind source separation module is used for processing the aliased satellite-borne electromagnetic signals to realize separation of source signals; in the processing process of the blind source separation module, the spectrum knowledge graph module provides signal characteristic comparison support; after the processing is finished, the blind source separation module sends the separated signals to the spectrum knowledge graph module for perfect supplementation;
the feature extraction module extracts target signal feature parameters of the separation signals output by the blind source separation module by using priori knowledge in the spectrum knowledge graph module; the extracted features are sent to the spectrum knowledge graph module for perfect supplementation, and sufficient priori information is provided for the rule mining module;
the rule mining module is used for mining mass data by utilizing the data model training provided by the frequency spectrum knowledge graph module, and the mining result is sent to the frequency spectrum knowledge graph module for perfect supplementation to provide support for the feature recognition module and the passive positioning module;
the characteristic recognition module is used for recognizing the modulation mode of the signal based on priori knowledge in the spectrum knowledge graph module and the characteristic information extracted by the characteristic extraction module; after the identification is completed, the modulation mode of the obtained signal is sent to the spectrum knowledge graph module for perfect supplementation;
the passive positioning module realizes passive positioning of signals based on priori knowledge in the spectrum knowledge graph module and the characteristic information extracted by the characteristic extraction module; after positioning is finished, if the rule mining module has completed the position information of the target signal in the spectrum knowledge graph module, comparing the calculated target position information with the data in the spectrum knowledge graph module, so as to improve the positioning precision and perfect the signal positioning information; and if the rule mining module does not complement the position information of the target signal in the spectrum knowledge graph module, directly supplementing the positioning information of the single signal to the spectrum knowledge graph module.
2. The knowledge-based on-board electromagnetic spectrum big data intelligent processing system according to claim 1, wherein the spectrum knowledge-graph module comprises: a mode layer and a data layer;
the mode layer comprises: knowledge entry class, background knowledge class, real-time information class, and content of a single signal class;
the data layer is a specific implementation of the mode layer.
3. The knowledge-based on-board electromagnetic spectrum big data intelligent processing system according to claim 2, wherein the knowledge entry class comprises 3 knowledge units: time, space and frequency band; wherein the time refers to the starting time to be searched, the space refers to different geographic positions to be monitored, and the frequency band refers to the frequency band range to be searched;
the background knowledge class contains 3 subclasses: registering information class, hardware information class and signal template class; wherein the registration information class includes: modulation type, frequency band and working time; the hardware information class includes: manufacturer, antenna, transmit power; the signal template class includes: a frequency band template, a signal template and an abnormal template;
the real-time information class includes: background noise power, electromagnetic wave power, spectrogram, signal number, idle frequency band and frequency band utilization rate;
a single signal class contains 3 subclasses: static information class, real-time change class and knowledge reasoning class; the static information class includes: frequency band, modulation mode, bandwidth; the real-time change class includes: decoding content, IQ waveforms, symbol rates; the knowledge reasoning class includes: doppler rate of change, signal arrival angle, signal reception strength, signal localization, and anomaly type.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310189348.9A CN116340533B (en) | 2023-02-23 | 2023-02-23 | Satellite-borne electromagnetic spectrum big data intelligent processing system based on knowledge graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310189348.9A CN116340533B (en) | 2023-02-23 | 2023-02-23 | Satellite-borne electromagnetic spectrum big data intelligent processing system based on knowledge graph |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116340533A CN116340533A (en) | 2023-06-27 |
| CN116340533B true CN116340533B (en) | 2023-09-01 |
Family
ID=86878304
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310189348.9A Active CN116340533B (en) | 2023-02-23 | 2023-02-23 | Satellite-borne electromagnetic spectrum big data intelligent processing system based on knowledge graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116340533B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111707260A (en) * | 2020-06-24 | 2020-09-25 | 电子科技大学 | Positioning method based on frequency domain analysis and convolutional neural network |
| WO2020228141A1 (en) * | 2019-05-13 | 2020-11-19 | 清华大学 | Electromagnetic signal identification method and device for constructing graph convolutional network on basis of implicit knowledge |
| CN113038614A (en) * | 2021-03-09 | 2021-06-25 | 中国人民解放军陆军工程大学 | Intelligent spectrum management and control framework based on spectrum knowledge graph |
| CN114417939A (en) * | 2022-01-27 | 2022-04-29 | 中国人民解放军32802部队 | Interference strategy generation method based on knowledge graph |
-
2023
- 2023-02-23 CN CN202310189348.9A patent/CN116340533B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020228141A1 (en) * | 2019-05-13 | 2020-11-19 | 清华大学 | Electromagnetic signal identification method and device for constructing graph convolutional network on basis of implicit knowledge |
| CN111707260A (en) * | 2020-06-24 | 2020-09-25 | 电子科技大学 | Positioning method based on frequency domain analysis and convolutional neural network |
| CN113038614A (en) * | 2021-03-09 | 2021-06-25 | 中国人民解放军陆军工程大学 | Intelligent spectrum management and control framework based on spectrum knowledge graph |
| CN114417939A (en) * | 2022-01-27 | 2022-04-29 | 中国人民解放军32802部队 | Interference strategy generation method based on knowledge graph |
Non-Patent Citations (1)
| Title |
|---|
| 基于知识图谱的电磁干扰源定位方法;张清等;太赫兹科学与电子信息学报;第20卷(第4期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116340533A (en) | 2023-06-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113156391B (en) | Radar signal multi-dimensional feature intelligent sorting method | |
| CN108924736B (en) | PCA-Kalman-based passive indoor personnel state detection method | |
| Liu | Multi-feature fusion for specific emitter identification via deep ensemble learning | |
| Cain et al. | Convolutional neural networks for radar emitter classification | |
| CN108051781A (en) | A kind of radar signal operating mode recognition methods based on DBN model | |
| CN106483514B (en) | Airplane motion mode identification method based on EEMD and support vector machine | |
| CN111722188A (en) | PRI (pulse repetition index) conversion radar signal sorting method based on STFT (space time Fourier transform) pre-sorting | |
| CN107729916A (en) | A kind of interference source classification and identification algorithm and device based on ISODATA | |
| CN111050315B (en) | Wireless transmitter identification method based on multi-core two-way network | |
| CN113759323B (en) | Signal sorting method and device based on improved K-Means joint convolution self-encoder | |
| CN112036239B (en) | Radar signal working mode identification method and system based on deep learning network | |
| Lin et al. | Unknown radar waveform recognition based on transferred deep learning | |
| CN114114166A (en) | Radar pulse de-interlacing method based on DTM algorithm | |
| CN115438708A (en) | Classification and identification method based on convolutional neural network and multi-mode fusion | |
| Nuhoglu et al. | Image segmentation for radar signal deinterleaving using deep learning | |
| CN108153519B (en) | Universal design framework for target intelligent tracking method | |
| CN116340533B (en) | Satellite-borne electromagnetic spectrum big data intelligent processing system based on knowledge graph | |
| CN103076595B (en) | Abnormal type identifying method for multivariate discrete radar emitter signal | |
| Ma et al. | Radar working mode recognition based on hierarchical feature representation and clustering | |
| Dudczyk et al. | Applying the distance and similarity functions to radar signals identification | |
| CN112666528A (en) | Multi-station radar system interference identification method based on convolutional neural network | |
| Horky et al. | Unsupervised time series pattern recognition for purpose of electronic surveillance | |
| Venkatasubramanian et al. | Data-driven target localization using adaptive radar processing and convolutional neural networks | |
| CN115902804B (en) | Unmanned aerial vehicle cluster type identification method and system | |
| CN112543428B (en) | Indoor positioning method based on ELM-AdaBoost classification learning |
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 |