CN111353379A - Signal measurement feature matching and labeling method based on weight clustering - Google Patents
Signal measurement feature matching and labeling method based on weight clustering Download PDFInfo
- Publication number
- CN111353379A CN111353379A CN202010010182.6A CN202010010182A CN111353379A CN 111353379 A CN111353379 A CN 111353379A CN 202010010182 A CN202010010182 A CN 202010010182A CN 111353379 A CN111353379 A CN 111353379A
- Authority
- CN
- China
- Prior art keywords
- characteristic
- signal measurement
- data
- signal
- label
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R29/00—Arrangements for measuring or indicating electric quantities not covered by groups G01R19/00 - G01R27/00
- G01R29/02—Measuring characteristics of individual pulses, e.g. deviation from pulse flatness, rise time or duration
- G01R29/023—Measuring pulse width
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Probability & Statistics with Applications (AREA)
- Signal Processing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a signal measurement characteristic matching and labeling method based on weight clustering, which can realize automatic labeling of measurement characteristics and remarkably improve labeling efficiency and accuracy. The invention is realized by the following technical scheme: inputting detected radiation source signal measurement characteristic data, respectively carrying out preprocessing in the aspects of normalization, abnormal value elimination and data compression, and storing and managing according to a uniform internal format; then, a multi-dimensional characteristic weight clustering analysis algorithm is adopted to respectively make coarse judgment on the characteristic of each dimensional signal and comprehensive similarity judgment on the multi-dimensional characteristic weight to form a cluster center set after clustering analysis; calculating comprehensive fuzzy membership degrees of the association of the clustering center set and the label knowledge base by using a fuzzy matching algorithm, and selecting the maximum comprehensive fuzzy membership degree as an optimal associated label result; and updating and maintaining the sample data of the tag library according to the correlation result, and completing automatic labeling of the attribute tag for the input unknown signal measurement characteristic data.
Description
Technical Field
The invention relates to a radiation source signal characteristic detection and analysis technology, in particular to a cluster analysis and automatic labeling method for unknown signal measurement characteristics under the background of detected big data.
Background
The signal measurement characteristic marking is to adopt methods such as manual or machine processing and the like to automatically fill attribute labels in the detected unknown signal measurement characteristics and key words capable of describing the signal measurement characteristics. The automatic marking of the signal measurement characteristics has extremely wide application in the fields of radiation source/target mining analysis such as characteristic retrieval, characteristic knowledge extraction and the like, and meanwhile, with the rapid development of intelligent processing technology, the marked signal measurement characteristics can be used as a sample set for deep learning model training. The signal characteristics refer to parameters which can characterize the electromagnetic characteristics of the radiation source and are obtained after signal processing. The signal measurement characteristics refer to that the detection equipment detects the signal emitted by the radiation source in real time, and analyzes and processes the measurement signal after receiving the signal, such as parameter measurement, characteristic extraction and the like, so as to form characteristic descriptions of the signal of the radiation source, such as characteristic parameters of radio frequency, bandwidth, pulse width, modulation mode and the like. In practical applications, to effectively transmit the original baseband signal generated by the source, various processes are usually performed on the original baseband signal, including: the signal modulation makes the signal form a plurality of signal patterns such as amplitude modulation and frequency modulation, the signal coding makes the signal have anti-interference capability, and the like, and the intentional processing forms the most basic characteristics of the signal and can be used as one of the main bases for judging the attribute of a radiation source/target.
The existing signal measurement feature marking method mainly adopts the following steps: (1) manually labeling a sample method; (2) a classification labeling method of a support vector machine. The traditional manual labeling method (1) mainly comprises the step of labeling measurement sample attributes after identification through expert experience knowledge, but with the coming of a big data era, the manual labeling method has the main defects that: firstly, the multi-platform multi-frame detection data is increased explosively, and the measurement of characteristic data by marking signals manually is difficult to complete; secondly, the understanding and comprehension of the signal measurement characteristics are inconsistent due to different expert experiences, and the sample labeling results are different due to the lack of uniform judgment standards. The method (2) is a classification and labeling method for a support vector machine, which is based on a supervised learning mode, firstly uses a labeled sample as a training sample, trains and learns the constructed classifier parameters, and then labels the measurement sample categories, and has the main defects that: firstly, a large amount of sample data with labels is needed, and when a limited number of training samples exist, the effect of the classifier is difficult to obtain the optimal state; secondly, the classifier has complex structure and various parameters, and the parameters are not easy to adjust to adapt to the labeling of a new class of objects, so that the accuracy of the labeling of the sample is influenced.
The signal measurement feature labeling process can also be converted to a data clustering or classification problem. The cluster analysis originates from taxonomy, but clustering is not equal to classification, which differs from classification in that: the class divided by the clustering is unknown, and the class formed by the classification is known. Clustering is the process of dividing a data set into different classes according to a certain set specific criterion, the generated class is a set of data objects, the objects are similar to the objects in the same class, and the differences of the data objects in different classes are as large as possible. At present, a large number of clustering algorithms exist, and for specific application, the selection of the clustering algorithm depends on the type of data, the purpose of clustering and the like, and common clustering algorithms are divided into: density-based methods, mesh-based methods, model-based methods, and the like. However, the research on the clustering problem is not limited to the above hard clustering, and the fuzzy clustering is also a branch of the cluster analysis. Fuzzy clustering determines how well each measurement data belongs to a class by means of membership functions, rather than rigidly classifying a data object into a class. More fuzzy clustering, such as algorithms of K-means clustering, fuzzy C-means clustering, hierarchical clustering and the like are applied, but each algorithm has advantages and disadvantages, such as that the K-means clustering and the fuzzy C-means clustering are sensitive to an initial clustering center, the clustering number needs to be determined artificially, and the local optimal solution is easy to fall into; hierarchical clustering may not be used to determine the number of classes, but once some splitting or merging is performed, it cannot be corrected, and clustering quality is limited.
Disclosure of Invention
The invention aims to provide an unknown signal measurement characteristic automatic matching and labeling method based on weight clustering aiming at radiation source signal characteristic big data of multi-platform multi-detection, so as to improve signal measurement characteristic labeling efficiency and accuracy.
In order to achieve the above object, the present invention provides a signal measurement feature matching and labeling method based on weight clustering, which is characterized by comprising the following steps: constructing a weight clustering-based automatic matching and labeling software architecture for signal measurement characteristics by taking a signal measurement characteristic data preprocessing module, a multi-dimensional characteristic weight clustering analysis module, a multi-dimensional characteristic fuzzy matching association module, a tag library sample data updating module and a signal characteristic data attribute labeling module as units, wherein the signal measurement characteristic data preprocessing module receives detected radiation source signal measurement characteristic data, respectively performs preprocessing in the aspects of normalization, abnormal value elimination and data compression, and stores the data into a cache queue to be processed according to a uniform internal format; a multi-dimensional characteristic weight cluster analysis module sequentially traverses all elements of the signal measurement characteristic data set and the class center set, respectively uses a multi-dimensional characteristic weight cluster analysis algorithm to make rough judgment on each dimensional signal characteristic and comprehensive similarity judgment on multi-dimensional characteristic weight, and then performs cluster analysis on the same radiation source signal measurement characteristic data of multi-platform multi-detection to form a class center set after cluster analysis; the multi-dimensional characteristic fuzzy matching association module associates the clustering center set with the label knowledge base by adopting a membership association algorithm based on fuzzy matching, calculates comprehensive fuzzy membership, judges to form a label result set to be confirmed, and selects the maximum comprehensive fuzzy membership as an optimal associated label result; the tag library sample data updating module updates the tag library sample data according to the association result of the multi-dimensional characteristic fuzzy matching association module to realize maintenance and management of the tag library sample data; and the signal characteristic data attribute labeling module performs corresponding processing on the signal measurement characteristic data, if the label knowledge with optimal association exists, the radiation source attribute information corresponding to the label knowledge data is used for complementing the attribute of the signal measurement characteristic, and if the label knowledge with optimal association does not exist, the generated serial number of the class label to be confirmed is assigned to the class label of the signal measurement characteristic, so that the automatic labeling of the attribute label on the input unknown signal measurement characteristic data is completed.
Compared with the prior art, the invention has the following beneficial effects:
the invention takes a signal measurement characteristic data preprocessing module, a multi-dimensional characteristic weight clustering analysis module, a multi-dimensional characteristic fuzzy matching association module, a label library sample updating maintenance module and a signal characteristic data attribute labeling module as units, constructs a signal measurement characteristic automatic labeling software framework based on weight clustering and a complete processing flow. The whole process is automatically realized in a data processing system based on a distributed parallel computing software mode, so that the operation efficiency is improved, and the problem of batch processing of a large amount of data is solved. Experimental data verification shows that the automatic signal measurement characteristic labeling method can effectively realize automatic labeling of large signal measurement characteristic data, improves signal measurement characteristic labeling efficiency and accuracy, and has high engineering practical value.
The invention adopts the signal measurement characteristic data preprocessing module to preprocess data aiming at the signal measurement characteristic big data, effectively compresses the scale of the input data set, and properly reduces the subsequent processing pressure on the premise of keeping the distribution of the original sample. Through a multi-dimensional characteristic weight clustering analysis algorithm, rough judgment of each dimensional signal characteristic and comprehensive similarity judgment of multi-dimensional characteristic weights are carried out, the clustering number is not required to be input, classification can be automatically realized, and compared with the traditional K-means clustering and hierarchical clustering algorithms, the defects that the clustering number needs to be manually determined and local optimal solutions are easy to fall into are overcome; in addition, multi-feature clustering of cross-domain dimensionality can be supported by adopting the normalized similarity, the calculated amount can be reduced by adopting twice judgment, and the clustering efficiency is improved.
According to the invention, a multi-dimensional characteristic fuzzy matching module adopts a membership degree association algorithm based on fuzzy matching to associate a clustering center set with a tag knowledge base, calculate comprehensive fuzzy membership degree, judge to form a tag result set to be confirmed, select the maximum comprehensive fuzzy membership degree as an optimal associated tag result from the tag result set, obtain knowledge information associated with the tag base, realize conversion of expert knowledge into a quantifiable evaluation standard, support configurable modification of characteristic weight, and facilitate subsequent expansion.
The invention adopts a label library sample updating and maintaining module, provides a sample data updating and managing mechanism, makes the sample information of the label library richer and more perfect, and provides a solution and a solid foundation for realizing the cluster analysis and automatic labeling method of the unknown signal measurement characteristics under the background of big data.
Drawings
For a more clear understanding of the invention, it will now be further elucidated by way of the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of a signal measurement feature matching and labeling method based on weight clustering according to the present invention.
Fig. 2 is a flow chart of the signal measurement characteristic data preprocessing module of fig. 1.
FIG. 3 is a flow diagram of the multidimensional feature weight cluster analysis module of FIG. 1.
FIG. 4 is a flow diagram of the multi-dimensional feature fuzzy matching correlation module of FIG. 1.
The invention will be further explained with reference to the drawings.
Detailed Description
See fig. 1. The invention takes a signal measurement characteristic data preprocessing module, a multidimensional characteristic weight clustering analysis module, a multidimensional characteristic fuzzy matching correlation module, a label library sample data updating module and a signal characteristic data attribute labeling module as units to construct a signal measurement characteristic automatic matching labeling software architecture based on weight clustering, wherein (S1) the signal measurement characteristic data preprocessing module receives detected radiation source signal measurement characteristic data, respectively carries out preprocessing such as normalization, abnormal value elimination and data compression, and stores the data into a cache queue to be processed according to a uniform internal format; (S2) the multi-dimensional characteristic weight cluster analysis module sequentially traverses all elements of the signal measurement characteristic data set and the class center set, respectively makes rough judgment on each dimensional signal characteristic and comprehensive similarity judgment on the multi-dimensional characteristic weight, and then performs cluster analysis on the same radiation source signal measurement characteristic data of multi-platform multi-detection to form a class center set after cluster analysis; (S3) the multi-dimensional characteristic fuzzy matching correlation module adopts a membership correlation algorithm based on fuzzy matching to correlate the clustering center set with the label knowledge base, calculates comprehensive fuzzy membership, judges to form a label result set to be confirmed, and selects the maximum comprehensive fuzzy membership as an optimal correlation label result; (S4) the label library sample data updating module updates the label library sample data according to the correlation result of the multi-dimensional characteristic fuzzy matching correlation module to realize the maintenance and management of the label library sample data; (S5) the signal characteristic data attribute labeling module carries out corresponding processing on the signal measurement characteristic data, if the label knowledge with optimal association exists, the radiation source attribute information corresponding to the label knowledge data is used for complementing the attribute of the signal measurement characteristic, if the label knowledge with optimal association does not exist, the generated serial number of the class label to be confirmed is assigned to the class label of the signal measurement characteristic, and the automatic labeling of the attribute label on the input unknown signal measurement characteristic data is completed.
Each module comprises the following specific steps:
see fig. 2. In the data preprocessing, a signal measurement characteristic data preprocessing module receives detected radiation source signal measurement characteristic data, respectively carries out preprocessing such as normalization, abnormal value elimination and data compression, and stores the data into a cache queue to be processed according to a uniform internal format, so that the subsequent module can be conveniently operated and processed, and the module specifically comprises the following steps:
step S11: the signal measurement characteristic data preprocessing module carries out standardized processing on input signal measurement characteristic data, mainly converts input time, input positions and external interfaces of signal measurement parameters into an internally processed data format, and unifies all characteristic parameter units into standard units;
step S12: the signal measurement characteristic data preprocessing module eliminates abnormal values, namely eliminates parameters exceeding the appointed range of certain one-dimensional characteristic parameters and mutation values in input data, and does not participate in subsequent processing;
step S13: the signal measurement characteristic data preprocessing module adopts a j-dimension signal measurement characteristic value obtained after data mean compression aiming at the radiation source signal measurement characteristic data with the same number
Reducing the input signal measurement characteristic data quantity to improve the real-time processing speed of the system, wherein L represents the total number of the radiation source signal measurement characteristics with the same number, Zj(l) Representing the characteristic value of the ith j-th dimension signal measurement;
step S14: the signal measurement characteristic data preprocessing module stores normal data into a buffer queue to be processed and pushes the normal data to a subsequent module for corresponding processing.
See fig. 3. The multidimensional characteristic weight cluster analysis module takes out data from a buffer queue to be processed stored by the measurement characteristic data preprocessing module, and carries out cluster analysis on the same radiation source signal measurement characteristic data of multi-platform multi-detection, and the module comprises the following specific steps:
step S21: setting an initial class center by a multi-dimensional characteristic weight clustering analysis module, and selecting the 1 st data of the signal measurement characteristics after preprocessing as an initial class center C (1);
step S22: sequentially traversing all elements of the signal measurement characteristic data set and the class center set, making rough judgment, and judging whether the rough judgment condition is met: using the nth signal measurement feature data element X (n) and the kth class center element C (k), and making a rough decision | X for each dimension of signal featurej(n)-Cj(k)|≤δjWherein, deltajIs a class decision threshold of j-th dimension signal characteristic parameter, n is a signal measurement characteristic data set serial number, k is a class center set serial number, Xj(n) a j-th dimension of X (n), Cj(k) The j dimension characteristic parameter of C (k);
step S23: in the calculation of the normalized similarity, if any dimension signal feature does not satisfy the decision condition, the signal measurement feature data x (n) and the class center c (k) are not in the same class, that is, the step S22 is returned to continue traversing all the elements of the measurement feature data set and the class center set to make rough decision, and if each dimension signal feature satisfies the decision condition, the similarity after the jth dimension normalization is calculated
Step S24, in calculating the comprehensive similarity of multi-dimensional feature weight, using each dimension similarity βjAnd the characteristic weight value mujCalculating the comprehensive similarity of the multi-dimensional feature weights
D is the dimension of cluster analysis, and the characteristic weight value satisfies
Step S25, according to the calculation result of the comprehensive similarity of multi-dimensional feature weight, judging whether the judgment of comprehensive similarity satisfies the condition, if yes, then the comprehensive similarity β of multi-dimensional feature weight is judgedΣAnd integrated decision threshold value βthMaking a determination when βΣ≥βthThen, generating a result set of the class to be confirmed, and putting the center C (k) into the result set of the class to be confirmed of the signal measurement characteristic data element X (n);
step S26: if the class result set to be confirmed of x (n) is empty, the multidimensional feature weight cluster analysis module newly establishes a class center C (k +1) ═ x (n); otherwise, from the class result set to be confirmed, selecting the class center with the maximum comprehensive similarity as the class to which the signal measurement characteristic belongs, and storing the serial number of the class center in a class sub-source linked list;
step S27: then, the class center value is updated by using the new signal measurement characteristic X (n), and the updated class center j dimension characteristic parameter is calculated
For updating the characteristic value, X, of the signal in dimension j of the center of the preceding classj(n) is the j-th dimension signal characteristic value of the measurement characteristic X (n) of the new signal, h is the updated times of the class center C (k); otherwise, building a class center; judging whether the traversal of the signal measurement characteristic data set is finished or not, if so, finishing the processing to form a cluster center set after cluster analysis; if not traverseAnd returning to the step S22 to continue traversing all the elements of the signal measurement feature data set and the class center set to make rough judgment until the traversal is finished to form the class center set after the cluster analysis.
See fig. 4. The multidimensional characteristic fuzzy matching correlation module uses a class center set output by the multidimensional characteristic weight clustering analysis module, adopts a membership correlation algorithm based on fuzzy matching to be correlated with a label knowledge base, and calculates comprehensive fuzzy membership to obtain optimal correlated label knowledge, and the module comprises the following specific steps:
step S31: the multidimensional characteristic fuzzy matching correlation module measures a characteristic class central value C by using a kth class jth dimension signalj(k) As input, a tag database query value (C) is setj(k)-εj,Cj(k)+εj)，εjInquiring a threshold value for the j-dimension characteristic parameter;
step S32: the multidimensional characteristic fuzzy matching correlation module queries the label database by adopting an OR method aiming at M-dimensional signal characteristics needing to be matched, and queries the label database meeting Cj(k)-εj＜φj＜Cj(k)+εjThe label knowledge data of the condition is processed by the following fuzzy membership degree calculation, wherein phi isjThe nominal value of the j dimension characteristic knowledge parameter in the label library is obtained;
step S33: multidimensional characteristic fuzzy matching correlation module utilizes class center value Cj(k) And nominal value phi of inquired label knowledge datajCalculating the fuzzy membership B of the j-th dimension signal featurej＝exp(-α|φj-Cj(k) I), wherein the membership modulation coefficient α meets α > 0, if the label database has no query return value, the association with the label database fails;
step S34: the multi-dimensional characteristic fuzzy matching correlation module utilizes the fuzzy membership B of each dimensional signal characteristiciAnd its M-dimensional signal feature weight wjAnd satisfy
Calculating signal characteristic comprehensive fuzzy membership degree
Step S35: comprehensive confidence B calculated by multi-dimensional characteristic fuzzy matching association modulesumAnd a decision threshold value BthMaking a determination when Bsum≥BthThen, outputting the label knowledge data to a label result set to be confirmed;
step S36: the multi-dimensional characteristic fuzzy matching correlation module selects the label knowledge data with the maximum fuzzy membership in the label result set to be confirmed as the successful correlation label; if the tag result set is empty, the association fails.
The tag library sample updating and maintaining module updates tag library sample data according to the association result of the multi-dimensional characteristic fuzzy matching association module to realize the maintenance and management of the tag library sample data, and the module specifically operates as follows: if the correlation with the tag knowledge base is successful, the tag base sample updating and maintaining module takes the central signal characteristic value as a knowledge sample and stores the knowledge sample into the tag base, and meanwhile, correlation times of the correlated tag knowledge data are accumulated; if the association with the label knowledge base fails, the label base sample updating maintenance module generates a new label serial number to be confirmed, the new label serial number is also stored in the label base, and the attribute of the radiation source is to be further confirmed subsequently; then, for the label knowledge data which is not correlated for a long time, the signal characteristic of the radiation source can be considered to be changed, and the label knowledge data needs to be confirmed through other means.
The signal characteristic data attribute labeling module carries out corresponding processing on the signal measurement characteristic data by utilizing the output results of the multi-dimensional characteristic fuzzy matching correlation module and the label library sample updating maintenance module as follows: if the label knowledge with optimal association exists, complementing the attribute of the signal measurement characteristic with the radiation source attribute information corresponding to the label knowledge data; if the label knowledge of optimal association does not exist, assigning the generated serial number of the class label to be confirmed to the class label of the signal measurement characteristic; and the automatic labeling of the attribute label on the input unknown signal measurement characteristic data can be completed.
The foregoing is directed to the preferred embodiment of the present invention and it is noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit and substance of the invention, and these modifications and improvements are also considered to be within the scope of the invention.
Claims (10)
1. A signal measurement feature matching and labeling method based on weight clustering is characterized by comprising the following steps: constructing a weight clustering-based automatic matching and labeling software architecture for signal measurement characteristics by taking a signal measurement characteristic data preprocessing module, a multi-dimensional characteristic weight clustering analysis module, a multi-dimensional characteristic fuzzy matching association module, a tag library sample data updating module and a signal characteristic data attribute labeling module as units, wherein the signal measurement characteristic data preprocessing module receives detected radiation source signal measurement characteristic data, respectively performs preprocessing in the aspects of normalization, abnormal value elimination and data compression, and stores the data into a cache queue to be processed according to a uniform internal format; a multi-dimensional characteristic weight cluster analysis module sequentially traverses all elements of the signal measurement characteristic data set and the class center set, respectively uses a multi-dimensional characteristic weight cluster analysis algorithm to make rough judgment on each dimensional signal characteristic and comprehensive similarity judgment on multi-dimensional characteristic weight, and then performs cluster analysis on the same radiation source signal measurement characteristic data of multi-platform multi-detection to form a class center set after cluster analysis; the multi-dimensional characteristic fuzzy matching association module associates the clustering center set with the label knowledge base by adopting a membership association algorithm based on fuzzy matching, calculates comprehensive fuzzy membership, judges to form a label result set to be confirmed, and selects the maximum comprehensive fuzzy membership as an optimal associated label result; the tag library sample data updating module updates the tag library sample data according to the association result of the multi-dimensional characteristic fuzzy matching association module to realize maintenance and management of the tag library sample data; and the signal characteristic data attribute labeling module performs corresponding processing on the signal measurement characteristic data, if the label knowledge with optimal association exists, the radiation source attribute information corresponding to the label knowledge data is used for complementing the attribute of the signal measurement characteristic, and if the label knowledge with optimal association does not exist, the generated serial number of the class label to be confirmed is assigned to the class label of the signal measurement characteristic, so that the automatic labeling of the attribute label on the input unknown signal measurement characteristic data is completed.
2. The signal measurement feature matching labeling method based on weight clustering as claimed in claim 1, characterized in that: the signal measurement characteristic data preprocessing module converts the input time, position and external interfaces of signal measurement parameters into an internal processing data format, and simultaneously unifies all characteristic parameter units into standard units; and eliminating abnormal values, and eliminating parameters exceeding the appointed range of certain one-dimensional characteristic parameters and mutation values in the input data without participating in subsequent processing.
3. The signal measurement feature matching labeling method based on weight clustering as claimed in claim 1, characterized in that: the signal measurement characteristic data preprocessing module adopts a j-dimension signal measurement characteristic value obtained after data mean compression aiming at the radiation source signal measurement characteristic data with the same number
Reducing the input signal measurement characteristic data quantity to improve the real-time processing speed of the system, wherein L represents the total number of the radiation source signal measurement characteristics with the same number, Zj(l) And the characteristic value of the ith j-th dimension signal measurement is represented.
4. The signal measurement feature matching labeling method based on weight clustering as claimed in claim 1, characterized in that: the multidimensional characteristic weight clustering analysis module takes out data from a cache queue to be processed stored by the measurement characteristic data preprocessing module, performs clustering analysis on the same radiation source signal measurement characteristic data of multi-platform multi-detection, sets an initial class center, and selects the 1 st data of the signal measurement characteristic after preprocessing as the initial class center C (1); sequentially traversing signal measurement feature data set and class center setHaving elements, and making rough decision, i.e. using nth signal measurement characteristic data element X (n) and kth class center element C (k), making decision | X for each dimension of signal characteristicj(n)-Cj(k)|≤δjWherein, deltajIs a class decision threshold of j-th dimension signal characteristic parameter, n is a signal measurement characteristic data set serial number, k is a class center set serial number, Xj(n) a j-th dimension of X (n), Cj(k) The j dimension characteristic parameter of C (k).
5. The signal measurement feature matching labeling method based on weight clustering as claimed in claim 4, characterized in that: the multi-dimensional characteristic weight clustering analysis module calculates each-dimensional normalized similarity and multi-dimensional characteristic weight comprehensive similarity as follows: if the signal characteristics of any dimension do not meet the judgment condition, the signal measurement characteristic data X (n) and the class center C (k) are not in the same class, namely, returning to continuously traverse all elements of the measurement characteristic data set and the class center set to make rough judgment; if each dimension of signal features meets the judgment condition, calculating the similarity after the jth dimension normalization
Wherein deltajA class decision threshold for j-th dimension signal characteristic parameters and using β degree of similarity in each dimensionjAnd its characteristic weight value mujCalculating the comprehensive similarity of the multi-dimensional feature weights
And the characteristic weight value satisfies
D is the clustering dimension.
6. The signal measurement feature matching labeling method based on weight clustering as claimed in claim 5, wherein the multidimensional feature weight clustering analysis module calculates the multidimensional feature weight comprehensive similarity β according to the multidimensional feature weight comprehensive similarityΣAnd integrated decision gateLimit value βthMaking a determination when βΣ≥βthWhen the result set to be confirmed is empty, the multidimensional characteristic weight cluster analysis module creates a class center C (k +1) ═ x (n); otherwise, the multidimensional characteristic weight clustering analysis module selects the class center with the maximum comprehensive similarity from the class result set to be confirmed as the class to which the signal measurement characteristic belongs, and simultaneously stores the serial number of the class center in the class sub-source linked list.
7. The signal measurement feature matching labeling method based on weight clustering as claimed in claim 6, characterized in that: the multidimensional characteristic weight clustering analysis module updates the class center value by using the new signal measurement characteristic X (n), and updates the class center C according to the new signal measurement characteristic X (n) and the updated previous class center0(k) Calculating the updated class center j dimension characteristic parameter
Otherwise, establishing a class center, judging whether the traversal of the signal measurement characteristics is finished, if so, finishing the processing, and forming a class center set after the clustering analysis; if the traversal is not finished, returning to continuously execute the coarse judgment of all the elements of the traversal measurement feature data set and the class center set until the traversal is finished to form the class center set after the cluster analysis, wherein,
for updating the characteristic value, X, of the signal in dimension j of the center of the preceding classj(n) is the j-th dimension signal feature value of the measurement feature X (n) of the new signal, and h is the updated number of times of the class center C (k).
8. The signal measurement feature matching labeling method based on weight clustering as claimed in claim 1, characterized in that: the multidimensional characteristic fuzzy matching correlation module measures a characteristic class central value C by using a kth class jth dimension signalj(k) As input, a tag database query value (C) is setj(k)-εj,Cj(k)+εj) And querying the label database by adopting an OR method aiming at the M-dimensional signal characteristics needing to be matched to obtain the label database meeting the requirement Cj(k)-εj＜φj＜Cj(k)+εjThe label knowledge data of the condition is subjected to subsequent fuzzy membership calculation processing, wherein epsilonjQuerying a threshold value, phi, for the j-th dimension characteristic parameterjAnd the nominal value of the j-th dimension characteristic knowledge parameter in the label library.
9. The signal measurement feature matching labeling method based on weight clustering as claimed in claim 8, characterized in that: multidimensional characteristic fuzzy matching correlation module utilizes class center value Cj(k) And nominal value phi of inquired label knowledge datajCalculating the fuzzy membership B of the j-th dimension signal featurej＝exp(-α|φj-Cj(k) And |) and membership modulation coefficient α satisfies α > 0, and association with the tag database fails if the tag database has no query return value.
10. The signal measurement feature matching labeling method based on weight clustering as claimed in claim 1, characterized in that: the multi-dimensional characteristic fuzzy matching module utilizes the fuzzy membership B of each dimensional signal characteristiciAnd its M-dimensional signal feature weight wjAnd satisfy
Calculating signal characteristic comprehensive fuzzy membership degree
And for the calculated BsumAnd a decision threshold value BthMaking a determination when Bsum≥BthThen, outputting the label knowledge data to a label result set to be confirmed, and selecting the label knowledge data with the maximum comprehensive fuzzy membership in the label result set to be confirmed as the related successful label; if the tag result set is empty, association fails; the tag library sample updating maintenance module updates the tag library sample data of the correlation result to realize the object matchingMaintaining and managing the sample data of the sign library; if the correlation with the tag knowledge base is successful, the central signal characteristic value is used as a knowledge sample and stored in the tag knowledge base, and correlation times of the correlated tag knowledge data are accumulated; and if the association with the label knowledge base fails, generating a new serial number of the label to be confirmed, storing the serial number into the label base, and subsequently further confirming the attribute of the radiation source.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010010182.6A CN111353379B (en) | 2020-01-06 | 2020-01-06 | Signal measurement feature matching and labeling method based on weight clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010010182.6A CN111353379B (en) | 2020-01-06 | 2020-01-06 | Signal measurement feature matching and labeling method based on weight clustering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111353379A true CN111353379A (en) | 2020-06-30 |
| CN111353379B CN111353379B (en) | 2023-04-07 |
Family
ID=71193975
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010010182.6A Active CN111353379B (en) | 2020-01-06 | 2020-01-06 | Signal measurement feature matching and labeling method based on weight clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111353379B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112131447A (en) * | 2020-09-29 | 2020-12-25 | 中国船舶重工集团公司第七二四研究所 | Radar signal detection data sample labeling method |
| CN112633051A (en) * | 2020-09-11 | 2021-04-09 | 博云视觉（北京）科技有限公司 | Online face clustering method based on image search |
| CN114124536A (en) * | 2021-11-24 | 2022-03-01 | 四川九洲电器集团有限责任公司 | Multi-station detection signal tracing method |
| CN115392376A (en) * | 2022-08-25 | 2022-11-25 | 广东工业大学 | Method, system and device for labeling heterogeneous fuzzy membership degree matrix |
| CN116137551A (en) * | 2023-04-14 | 2023-05-19 | 西安晟昕科技股份有限公司 | Communication reconnaissance performance test control method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107301430A (en) * | 2017-06-29 | 2017-10-27 | 常州工学院 | Broad sense Multivariable Fuzzy c means clustering algorithms |
| CN107330458A (en) * | 2017-06-27 | 2017-11-07 | 常州信息职业技术学院 | A kind of fuzzy C-means clustering method of minimum variance clustering of optimizing initial centers |
-
2020
- 2020-01-06 CN CN202010010182.6A patent/CN111353379B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107330458A (en) * | 2017-06-27 | 2017-11-07 | 常州信息职业技术学院 | A kind of fuzzy C-means clustering method of minimum variance clustering of optimizing initial centers |
| CN107301430A (en) * | 2017-06-29 | 2017-10-27 | 常州工学院 | Broad sense Multivariable Fuzzy c means clustering algorithms |
Non-Patent Citations (1)
| Title |
|---|
| 李长磊等: "一种改进模糊C均值聚类的图像标注方法", 《小型微型计算机系统》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112633051A (en) * | 2020-09-11 | 2021-04-09 | 博云视觉（北京）科技有限公司 | Online face clustering method based on image search |
| CN112131447A (en) * | 2020-09-29 | 2020-12-25 | 中国船舶重工集团公司第七二四研究所 | Radar signal detection data sample labeling method |
| CN114124536A (en) * | 2021-11-24 | 2022-03-01 | 四川九洲电器集团有限责任公司 | Multi-station detection signal tracing method |
| CN115392376A (en) * | 2022-08-25 | 2022-11-25 | 广东工业大学 | Method, system and device for labeling heterogeneous fuzzy membership degree matrix |
| CN115392376B (en) * | 2022-08-25 | 2024-02-02 | 广东工业大学 | Heterogeneous fuzzy membership matrix labeling method, system and device |
| CN116137551A (en) * | 2023-04-14 | 2023-05-19 | 西安晟昕科技股份有限公司 | Communication reconnaissance performance test control method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111353379B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111353379B (en) | Signal measurement feature matching and labeling method based on weight clustering | |
| CN107515895B (en) | Visual target retrieval method and system based on target detection | |
| CN108875816A (en) | Merge the Active Learning samples selection strategy of Reliability Code and diversity criterion | |
| CN107798033B (en) | Case text classification method in public security field | |
| CN111221920B (en) | Case base construction method and device for power transformation equipment operation and maintenance device and computer storage medium | |
| CN109581339B (en) | Sonar identification method based on automatic adjustment self-coding network of brainstorming storm | |
| CN110134719B (en) | Identification and classification method for sensitive attribute of structured data | |
| CN113920472B (en) | Attention mechanism-based unsupervised target re-identification method and system | |
| CN114861705A (en) | Electromagnetic target intelligent perception identification method based on multi-feature heterogeneous fusion | |
| CN111367801A (en) | Data transformation method for cross-company software defect prediction | |
| CN112329884A (en) | Zero sample identification method and system based on discriminant visual attributes | |
| CN116662832A (en) | Training sample selection method based on clustering and active learning | |
| CN114093445B (en) | Patient screening marking method based on partial multi-marking learning | |
| CN115700558A (en) | Abnormal flow detection method based on space-time feature analysis | |
| CN111967501B (en) | Method and system for judging load state driven by telemetering original data | |
| CN100446001C (en) | Remote sensing image classify method combined case-based reasoning with Fuzzy ARTMAP network | |
| CN111950652A (en) | Semi-supervised learning data classification algorithm based on similarity | |
| CN116342857A (en) | Weak supervision target positioning method based on category correction | |
| CN114372181B (en) | Equipment production intelligent planning method based on multi-mode data | |
| CN115017671B (en) | Industrial process soft measurement modeling method and system based on online cluster analysis of data flow | |
| CN114037866B (en) | Generalized zero sample image classification method based on distinguishable pseudo-feature synthesis | |
| CN115082726A (en) | Ceramic biscuit product classification method for toilet based on PointNet optimization | |
| CN109886340A (en) | A kind of Remote Image Classification | |
| CN118154985A (en) | Unsupervised industrial data classification method | |
| CN118536062A (en) | Load classification method based on deep 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 |