CN115774248A - Sea surface small target detection method and system based on multiple characteristics of graph - Google Patents
Sea surface small target detection method and system based on multiple characteristics of graph Download PDFInfo
- Publication number
- CN115774248A CN115774248A CN202211513958.1A CN202211513958A CN115774248A CN 115774248 A CN115774248 A CN 115774248A CN 202211513958 A CN202211513958 A CN 202211513958A CN 115774248 A CN115774248 A CN 115774248A
- Authority
- CN
- China
- Prior art keywords
- graph
- unit
- training
- detected
- matrix
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 48
- 238000012549 training Methods 0.000 claims abstract description 83
- 239000013598 vector Substances 0.000 claims abstract description 51
- 238000000034 method Methods 0.000 claims abstract description 30
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 12
- 238000007781 pre-processing Methods 0.000 claims abstract description 11
- 239000011159 matrix material Substances 0.000 claims description 51
- 238000001228 spectrum Methods 0.000 claims description 51
- 238000013139 quantization Methods 0.000 claims description 15
- 238000010606 normalization Methods 0.000 claims description 6
- 238000002156 mixing Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 241001235534 Graphis <ascomycete fungus> Species 0.000 claims description 3
- 238000004590 computer program Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 2
- 239000000284 extract Substances 0.000 abstract description 3
- 230000010287 polarization Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 238000005388 cross polarization Methods 0.000 description 1
- 238000002592 echocardiography Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a sea surface small target detection method and a system based on multiple image characteristics, wherein the method and the system are divided into a training part and a detection part, the training part comprises the steps of preprocessing radar echo data of a training unit, extracting image characteristics from the preprocessed data, constructing image characteristic vectors, and calculating a judgment area by utilizing a convex hull learning algorithm; the detection part comprises preprocessing radar echo data of a unit to be detected, extracting graph features from the preprocessed data, constructing graph feature vectors of the unit to be detected, and judging whether a radar target exists or not by constructing detection statistics through a judgment area and the graph feature vectors of the unit to be detected; the invention extracts three different image characteristics from the sea clutter sequence, and uses the three image characteristics to distinguish the pure clutter data from the echo data containing the target to complete the sea surface target detection.
Description
Technical Field
The invention relates to the technical field of radar target detection, in particular to a sea surface small target detection method and system based on multiple characteristics of a graph.
Background
In recent years, small sea-surface targets such as boats, yachts, and airplane debris have become important targets for radar warning along with miniaturization and stealth of sea-surface targets. Generally, the radar scattering cross-sectional area (RCS) of a small floating target on the sea surface is weak, the moving speed is slow, and a coherent or incoherent constant false alarm rate detection algorithm for detecting by using echo energy has an obvious performance bottleneck when the small floating target is detected. In order to avoid energy detection, some researchers provide feature-based target detection methods, the methods extract features by means of differences between clutter and target echoes, the extracted features comprise single-fractal and fractional Fourier transform domain fractal and other fractal features, relative average amplitude and time domain information entropy and other time domain features, frequency domain features, time-frequency domain features, polarization features and the like, detectors based on the features all obtain good detection effects to a certain extent, but the features do not consider the relevance among data in the extraction process, so that some key information existing among the data is ignored, and the detection performance is limited.
The graph theory and the graph neural network are used as an effective method for describing the relevance between data, and have wide application in the aspects of computer science, the economic field, discrete signal processing and the like. In order to utilize the amplitude correlation among echo data, a method based on graph theory is greatly introduced into target detection in the background of sea clutter. On volume 57 of IEEE Transactions on Geoscience and Remote Sensing published in 2019, YAN Kun et al propose a method for detecting a small sea surface floating target based on a graph, which directly processes radar echo signals in a time domain, constructs a directed graph by using the amplitude of the echo, calculates the maximum eigenvalue of a Laplace matrix of the directed graph in consideration of the connectivity of edges, obtains a detection threshold through a door-to-door Carlo experiment and completes target detection, and compares the detection with a detector based on a single fractal characteristic, wherein the detection method has excellent performance, especially when the signal-to-noise ratio is small. However, since the graph features extracted by this method are single, and feature decomposition is required when obtaining the eigenvalue of the laplacian matrix, the time required for feature extraction is long, and thus the detection effect is yet to be improved.
Disclosure of Invention
In view of this, the present invention provides a method and a system for detecting a small sea surface target based on multiple features of a map, where the method obtains better target detection performance by using correlation between echo data of a sea clutter radar.
In order to achieve the purpose, the invention provides the following technical scheme:
the invention provides a sea surface small target detection method based on multiple characteristics of a graph, which comprises the following steps:
selecting P distance units from pure clutter echo data as training units, the training unit time sequence z p Comprises the following steps: z is a radical of formula p =[z p (1),z p (2),…,z p (N)]Wherein, P =1,2, \8230, P is the number of training units, and N is the time sequence length; will train the unit time sequence z p The average is divided into short vectors of length L, i.e.: z is a radical of formula p =[z p,1 ,z p,2 ,…,z p,l ,…,z p,N/L ]Wherein P =1,2, \ 8230;, P, z p,l The L short vector representing the training unit time sequence, L =1,2, \8230;, N/L;
selecting a unit r to be detected from echo data containing a target;
step 3, training unit time sequence z p Normalized Doppler power spectrum ofModeling graphs and graph signals in which a time series of elements z is trained p Each short sequence z of p,l Normalized Doppler power spectrum ofTwo undirected graphs and signals on the graphs can be obtained through modeling, and three graph characteristics are obtained through calculation: graph Laplace regularization η 1 (z p,l ) Trace η of graph laplace matrix 2 (z p,l ) Variance eta of graph vertex in degree 3 (z p,l ) And constructing a map three feature vector by the three features:
η(z p,l )=[η 1 (z p,l ),η 2 (z p,l ),η 3 (z p,l )] T ,
wherein [ ·] T Representing transposing the matrix; combining P training units and N/L graph three-feature vectors of each training unit into a training sample set
Step 4, according to the training sample setClosing boxObtaining a three-dimensional convex hull in a three-dimensional feature space and determining a false alarm probability P f Next, calculating a decision region omega of the detector by using a greedy convex hull learning algorithm;
step 5, calculating the unit r to be detected to obtain graph Laplace regularization eta 1 (r), trace η of the graph Laplace matrix 2 (r) variance η of the in-degree of the graph vertex 3 (r) three graph characteristics, and constructing a graph three characteristic vector eta (r) of the unit to be detected, namely eta (r) = [ eta = 1 (r),η 2 (r),η 3 (r)] T ;
Step 6, calculating the detection statistic of the unit to be detected according to the decision region omega of the detector and the map three characteristic vectors eta (r) of the unit to be detectedBased on detection statisticsWhether the target exists is judged by the size of (1): if the statistic is detectedIf the value is larger than zero, the result shows that the map three feature vectors eta (r) of the unit to be detected are outside the decision area of the detector, and a target exists in the unit to be detected, if the detection statistic quantity isAnd if the result is less than zero, the map three feature vector eta (r) of the unit to be detected is in the decision area of the detector, and no target exists in the unit to be detected.
Further, step 2 is to train unit time sequence z p =[z p,1 ,z p,2 ,…,z p,l ,…,z p,N/L ]Carrying out pretreatment, and specifically comprising the following substeps:
Where Δ f is the Doppler frequency sampling interval, T r For the pulse repetition interval, L is the number of FFT sample points, exp () represents an exponential function with e as the base, and k represents the doppler unit;
2.2 calculating the mean spectrum from the P training elements and the N/L Doppler power spectra of each training elementSum standard deviation spectrum
Wherein k = - [ L/2], [ L/2] -1;
2.3 spectra from the meanStandard deviation spectrumAnd a short vector z for each training unit p,l Doppler power spectrum ofCalculating a short vector z for each training unit p,l Normalized Doppler power spectrum of
Further, in step 3, the training unit time sequence z p Normalized Doppler power spectrum ofModeling of maps and map signals, wherein a time sequence z of training elements is p Short sequence of (a) z p,l Normalized Doppler power spectrum ofModeling into two undirected graphs and signals on the graphs, and calculating three graph characteristics according to the following specific sub-steps:
wherein k = - [ L/2],...,[L/2]-1,The energy data representing the largest and smallest normalized kth doppler cell,
3.2 mixingModeling as an undirected graphDefining a Doppler Unit { - [ L/2 { [ L/2]],...,[L/2]-1} is an undirected graphGet an undirected graphSet of vertices ofAnd considering each vertex to be connected with d adjacent vertexes, wherein the weight value between the connected vertexes is 1, and obtaining the undirected graphOf a neighboring matrixAnd laplacian matrixDefining the energy data of the k-th Doppler unit after maximum and minimum normalization as a vertex v k Graph signal x of k And will be located in an undirected graphIs represented by
wherein k = - [ L/2],...,[L/2]-1,Representing the quantized energy data of the kth Doppler unit, i representing the quantized value of the energy data of the Doppler unit, and gamma representing the number of quantization levels;
3.4 mixingModeling as an undirected graphDefining quantized values {0, \8230;, i, \8230;. Gamma } as undirected graphsGet an undirected graphSet of vertices ofDefining the vertex corresponding to the quantization value in each Doppler unit to be connected with the vertex corresponding to the quantization value in the adjacent Doppler unit, defining the connection times between two vertexes as the weight between the two vertexes, and obtaining the undirected graphOf a neighboring matrix
To undirected graphOf (2) an adjacent matrixSumming the elements of each row to obtain an undirected graphIn-degree matrix of
Wherein,representing a vertex v i The degree of entry is calculated to obtain an undirected graphLaplacian matrix of
3.5 according to undirected graphLaplacian matrix ofUndirected graphSignal onUndirected graphLaplacian matrix ofAnd undirected graphIn-degree matrix ofThree graph features were calculated: graph Laplace regularization η 1 (z p,l ) Trace η of graph laplace matrix 2 (z p,l ) Variance eta of graph vertex in degree 3 (z p,l ):
Wherein Tr (·) represents the trace of the matrix;
3.6 Laplace regularization η from the graph 1 (z p,l ) Trace η of graph laplace matrix 2 (z p,l ) Variance eta of graph vertex in degree 3 (z p,l ) And constructing a feature vector of the graph: eta (z) p,l )=[η 1 (z p,l ),η 2 (z p,l ),η 3 (z p,l )] T 。
4. The method for detecting small sea surface targets based on multiple features of map as claimed in claim 1, wherein in step 4, the method is based on a training sample setAt a given false alarm probability P f Next, the specific sub-step of calculating the decision region Ω of the detector by using the greedy convex hull learning algorithm is as follows:
4.1 define a convex hull CH (S) containing all samples of the training sample set S:
wherein SP {. Cndot } represents a closed space surrounded by triangles,three vertices representing the qth triangular face constituting the surface of the convex hull CH (S), Q representing the total number of triangular faces on the convex hull CH (S);
4.2 according to a given false alarm probability P f And greedy convex hull learning algorithm, sequentially deleting to minimize convex hull volumeA false alarm training sample to obtain a residual training sample set S', wherein [ ·]Represents rounding down;
4.3 forming a convex hull CH (S ') from the remaining training sample set S ' as a decision region Ω = CH (S ') of the detector.
Further, in step 6, according to the decision region Ω = CH (S') of the detector and the map three feature vectors η (r) of the unit to be detected, the detection statistic of the unit to be detected is calculated
Where det (-) denotes the determinant of the matrix,three vertices of the F-th triangular face constituting the surface of the convex hull CH (S ') are indicated, and F represents the total number of triangular faces on the convex hull CH (S').
The invention provides a sea surface small target detection system based on multiple graph features, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the method when executing the program.
The invention has the beneficial effects that:
the invention provides a method and a system for detecting a small sea surface floating target based on multiple characteristics of a graph, which are characterized in that a radar transmitter is used for transmitting a pulse signal, a radar receiver is used for receiving a multi-pulse signal reflected by the sea surface to obtain radar echo data containing a target, and a unit r to be detected is selected; for training unit time sequence z p Preprocessing is carried out to obtain a training unit time sequence z p Normalized Doppler power spectrum ofTraining unit time sequence z p Normalized Doppler power spectrum ofModeling a graph and a graph signal, calculating to obtain three graph features and constructing a graph three feature vector: obtaining a three-dimensional convex hull in a three-dimensional feature space according to a training sample set and giving a false alarm probability P f Next, calculating a decision region omega of the detector by using a greedy convex hull learning algorithm; calculating three graph characteristics of the unit to be detected, constructing a graph three characteristic vector eta (r) of the unit to be detected, and calculating the detection statistic of the unit to be detected according to the decision region omega of the detector and the graph three characteristic vector eta (r) of the unit to be detectedBased on detection statisticsDetermines the target.
Three different image characteristics are extracted from the sea clutter sequence, and the three image characteristics are combined and utilized to distinguish pure clutter data from echo data containing targets so as to complete sea surface target detection.
The method utilizes the relevance between the sea clutter radar echo data, considers some key information existing between the data, and can obtain better target detection performance compared with the existing multi-feature-based detector. The method considers the correlation among the radar echo data from different angles, carries out composition, extracts a plurality of graph features, avoids the feature decomposition calculation of a matrix, and has higher feature extraction speed and better radar target detection effect.
The method models the normalized Doppler power spectrum into two undirected graphs and signals on the graphs, and jointly utilizes three graph characteristics: the graph Laplace regularization, the variance of the trace of the graph Laplace matrix and the image vertex incoherence consider the relevance among the sea clutter data, and distinguish the pure clutter data from the echo data containing the target.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims thereof.
Drawings
In order to make the purpose, technical scheme and beneficial effect of the invention more clear, the invention provides the following drawings for explanation:
fig. 1 is a flow chart of an implementation of the present invention according to an embodiment.
Figure 2 is a flow chart of modeling the normalized doppler power spectrum of an embodiment as two undirected graphs and an on-graph signal and deriving three graph features.
FIG. 3 is a graph comparing the detection performance of the example with the present invention at four polarizations of the three prior art detectors at an observation time that is multiplied from 0.064 seconds to 4.096 seconds.
Detailed Description
The present invention is further described below in conjunction with the drawings and the embodiments so that those skilled in the art can better understand the present invention and can implement the present invention, but the embodiments are not to be construed as limiting the present invention.
Referring to fig. 1, the method for detecting the small sea surface floating target based on the multiple graph features comprises a training part and a detection part, wherein the training part comprises preprocessing radar echo data of a training unit, extracting graph features from the preprocessed data, constructing graph feature vectors, and calculating a decision area by using a convex hull learning algorithm; the detection part comprises preprocessing radar echo data of a unit to be detected, extracting graph features from the preprocessed data, constructing graph feature vectors of the unit to be detected, and judging whether a radar target exists or not by constructing detection statistics through a judgment area and the graph feature vectors of the unit to be detected; the method comprises the following specific steps:
selecting P distance units from pure clutter echo data as training units, the training unit time sequence z p Comprises the following steps: z is a radical of p =[z p (1),z p (2),…,z p (N)]Wherein, P =1,2, \8230, P is the number of training units, and N is the time sequence length; will train the unit time sequence z p The average is divided into short vectors of length L, i.e.: z is a radical of formula p =[z p,1 ,z p,2 ,…,z p,l ,…,z p,N/L ]Wherein P =1,2, \ 8230;, P, z p,l The L short vector representing the training unit time sequence, L =1,2, \8230;, N/L;
selecting a unit r to be detected from echo data containing a target;
Where Δ f is the Doppler frequency sampling interval, T r For the pulse repetition interval, L is the number of FFT sample points, exp () represents an exponential function with e as the base, and k represents the doppler unit;
2.2 calculating the mean spectrum from the P training units and the N/L Doppler power spectrums of each training unitSum standard deviation spectrum
Wherein k = - [ L/2], [ L/2] -1;
2.3 spectra from the meanStandard deviation spectrumAnd a short vector z for each training unit p,l Doppler power spectrum ofComputing a short vector z for each training unit p,l Normalized Doppler power spectrum of
According to the short vector z of each training unit p,l Normalized Doppler power spectrum ofObtaining a training unit time sequence z p Normalized Doppler power spectrum ofWherein P =1,2, \ 8230, and P is the number of training units;
step 3, training unit time sequence z p Normalized Doppler power spectrum ofModelling of graphs and graph signals, in which a time sequence of elements z is trained p Each short sequence z of p,l Normalized Doppler power spectrum ofTwo undirected graphs and signals on the graphs can be obtained through modeling, and three graph features and a graph three feature vector are obtained through calculation by the two undirected graphs and the signals on the graphs, and the method comprises the following steps with reference to fig. 2:
wherein k = - [ L/2],...,[L/2]-1,The energy data representing the largest and smallest normalized kth doppler cell,
3.2 mixingModeling as an undirected graphDefining a Doppler Unit { - [ L/2 { [ L/2]],...,[L/2]-1} is an undirected graphTo get an undirected graphSet of vertices ofAnd considering each vertex to be connected with d adjacent vertexes, wherein the weight value between the connected vertexes is 1, and obtaining the undirected graphOf a neighboring matrixAnd laplacian matrixDefining the energy data of the k-th Doppler unit after maximum and minimum normalization as a vertex v k Graph signal x of k And will be located in an undirected graphIs represented by
wherein k = - [ L/2],...,[L/2]-1,Represents the quantized energy data of the kth Doppler cell, i represents the quantized value of the energy data of the Doppler cell, and gamma represents the number of quantization levels;
3.4 willModeling as an undirected graphDefining quantized values {0, \8230;, i, \8230;. Gamma } as undirected graphsTo get an undirected graphSet of vertices ofDefining the vertex corresponding to the quantization value in each Doppler unit to be connected with the vertex corresponding to the quantization value in the adjacent Doppler unit, defining the connection times between two vertexes as the weight between the two vertexes, and obtaining the undirected graphOf a neighboring matrix
To undirected graphOf (2) an adjacent matrixSumming the elements of each row to obtain an undirected graphIn-degree matrix of
3.5 according to undirected graphLaplacian matrix ofUndirected graphSignal onUndirected graphLaplacian matrix ofAnd undirected graphIn-degree matrix ofThree graph features were calculated: graph Laplace regularization η 1 (z p,l ) Trace η of graph laplace matrix 2 (z p,l ) Variance eta of graph vertex in degree 3 (z p,l ):
Wherein Tr (·) represents the trace of the matrix;
3.6 regularization η according to Laplace of the graph 1 (z p,l ) Trace η of graph laplace matrix 2 (z p,l ) Variance eta of graph vertex in degree 3 (z p,l ) And constructing a graph three feature vectors: eta (z) p,l )=[η 1 (z p,l ),η 2 (z p,l ),η 3 (z p,l )] T Therein, [ ·] T Indicating that the matrix is transposed.
Forming a training sample set by P training units and N/L graph three-feature vectors of each training unitIn the embodiment, 10000 training samples are taken;
step 4, according to the training sample setObtaining a three-dimensional convex hull in a three-dimensional feature space and determining a false alarm probability P f Next, a greedy convex hull learning algorithm is used to calculate a decision region Ω of the detector, which includes the following steps:
4.1 define a convex hull CH (S) containing all samples of the training sample set S:
wherein SP {. Is } represents a closed space surrounded by triangles,three vertices representing the qth triangular face constituting the surface of the convex hull CH (S), Q representing the total number of triangular faces on the convex hull CH (S);
4.2 according to a given false alarm probability P f And a greedy convex hull learning algorithm, which deletes [ P N/L P ] in turn to make the convex hull volume reduce most f ]A false alarm training sample to obtain a residual training sample set S', wherein [ ·]Represents rounding down;
4.3 compose a convex hull CH (S ') from the remaining training sample set S ' as the decision region Ω = CH (S ') of the detector.
Step 5, utilizing the time sequence z of the unit r to be detected r And the mean spectrum obtained in step 2.2Sum standard deviation spectrumPreprocessing to obtain a time sequence z of a unit to be detected r Normalized Doppler power spectrum ofAnd calculating to obtain graph Laplacian regularization eta of the detection unit 1 (r), trace η of graph Laplace matrix 2 (r) variance η of graph vertex incomes 3 (r) the three map features, and then obtaining a map three feature vector eta (r) of the unit to be detected, namely eta (r) = [ eta (r) = 1 (r),η 2 (r),η 3 (r)] T ;
Step 6, calculating the detection statistic of the unit to be detected according to the decision region omega of the detector and the map three characteristic vectors eta (r) of the unit to be detected
Wherein det (-) represents the determinant of the matrix,three vertices representing the F-th triangular face constituting the surface of the convex hull CH (S '), F representing the total number of triangular faces on the convex hull CH (S');
and based on the detected statisticsWhether the target exists is judged: if the statistic is detectedIf the value is larger than zero, the result shows that the map three feature vectors eta (r) of the unit to be detected are outside the decision area of the detector, and a target exists in the unit to be detected, if the detection statistic quantity isAnd if the value is less than zero, the map three feature vectors eta (r) of the unit to be detected are in a decision area of the detector, and no target exists in the unit to be detected.
Based on the steps 1 to 6, the detection of the small floating targets on the sea surface based on the multiple characteristics of the graph is realized.
The system for detecting the small sea surface target based on the multiple features of the graph comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein the processor executes the program to realize the method.
The 10 IPIX radar actual measurement sea clutter data used in the embodiment are collected in 1993 years, the data are current internationally recognized small target actual measurement data sets, the height of the radar erection is 30 meters, the pulse repetition frequency is 1000Hz, and the distance resolution is 30 meters; each set of data comprises four kinds of polarization data, wherein two kinds of polarization data are homopolar data HH and VV, and two kinds of polarization data are cross polarization data HV and VH; each polarization data comprises 14 distance units and the data length is 2 17 The experimental target was a metal sphere of 1 meter diameter. In an experiment, the method, the detector based on graph link density, the detector based on three characteristics and the detector based on fractal characteristics are respectively utilized to detect the sea surface target, and referring to fig. 3, when the observation time is multiplied from 0.064 seconds to 4.096 seconds, the average detection probability of 10 groups of data in each polarization direction is far greater than that of three existing detectors. In conclusion, the method of the present invention has better detection performance than the conventional detectors.
The above-mentioned embodiments are merely preferred embodiments for fully illustrating the present invention, and the scope of the present invention is not limited thereto. The equivalent substitution or change made by the technical personnel in the technical field on the basis of the invention is all within the protection scope of the invention. The protection scope of the invention is subject to the claims.
Claims (6)
1. The sea surface small target detection method based on the multiple characteristics of the graph is characterized in that: the method comprises the following steps:
step 1, transmitting a pulse signal through a radar transmitter, receiving a multi-pulse signal reflected by the sea surface through a radar receiver, and obtaining radar echo data after the multi-pulse signal is subjected to matching filtering, wherein the echo data is divided into pure clutter echo data and echo data containing a target;
selecting P distance units from pure clutter echo data as training units, the training unit time sequence z p Comprises the following steps: z is a radical of formula p =[z p (1),z p (2),…,z p (N)]Wherein, P =1,2, \8230, P is the number of training units, and N is the time sequence length; training unit time sequence z p The average is divided into short vectors of length L, i.e.: z is a radical of p =[z p,1 ,z p,2 ,…,z p,l ,…,z p,N/L ]Wherein P =1,2, \ 8230;, P, z p,l The L short vector representing the training unit time sequence, L =1,2, \8230;, N/L;
selecting a unit r to be detected from echo data containing a target;
step 2, training unit time sequence z p Preprocessing is carried out to obtain a training unit time sequence z p Normalized Doppler power spectrum ofNamely, it isWherein P =1,2, \ 8230;, P,representing a time sequence z of training elements p The normalized Doppler power spectrum obtained by the first short vector preprocessing;
step 3, training unit time sequence z p Normalized Doppler power spectrum ofModelling of graphs and graph signals, in which a time sequence of elements z is trained p Each short sequence z of p,l Normalized Doppler power spectrum ofTwo undirected graphs and signals on the graphs can be obtained through modeling, and three graph characteristics are obtained through calculation: graph Laplace regularization η 1 (z p,l ) Trace η of graph laplace matrix 2 (z p,l ) Variance eta of graph vertex in degree 3 (z p,l ) And constructing a map three feature vector by the three features:
η(z p,l )=[η 1 (z p,l ),η 2 (z p,l ),η 3 (z p,l )] T ,
wherein [ ·] T Representing transposing the matrix; combining P training units and N/L graph three-feature vectors of each training unit into a training sample set
Step 4, according to the training sample setObtaining a three-dimensional convex hull in a three-dimensional feature space and determining a false alarm probability P f Next, calculating a decision region omega of the detector by using a greedy convex hull learning algorithm;
step 5, calculating the unit r to be detected to obtain graph Laplace regularization eta 1 (r), trace η of graph Laplace matrix 2 (r) variance η of graph vertex incomes 3 (r) three graph characteristics, and constructing a graph three characteristic vector eta (r) of the unit to be detected, namely eta (r) = [ eta = 1 (r),η 2 (r),η 3 (r)] T ;
Step 6, calculating the detection statistic of the unit to be detected according to the decision region omega of the detector and the map three characteristic vectors eta (r) of the unit to be detectedBased on detection statisticsWhether the target exists is judged: if the statistic is detectedIf the value is larger than zero, the result shows that the map three feature vectors eta (r) of the unit to be detected are outside the decision area of the detector, and a target exists in the unit to be detected, if the detection statistic quantity isAnd if the value is less than zero, the map three feature vectors eta (r) of the unit to be detected are in a decision area of the detector, and no target exists in the unit to be detected.
2. The method as claimed in claim 1, wherein the training unit time sequence z is processed in step 2 p =[z p,1 ,z p,2 ,…,z p,l ,…,z p,N/L ]Performing pretreatment, wherein the specific substeps are as follows:
Where Δ f is the Doppler frequency sampling interval, T r For the pulse repetition interval, L is the number of FFT sample points, exp () represents an exponential function with e as the base, and k represents the doppler unit;
2.2 training units based on PAnd calculating the mean spectrum of N/L Doppler power spectrums of each training unitSum standard deviation spectrum
Wherein k = - [ L/2], [ L/2] -1;
2.3 spectra from the meanStandard deviation spectrumAnd a short vector z for each training unit p,l Doppler power spectrum ofCalculating a short vector z for each training unit p,l Normalized Doppler power spectrum of
3. The method for detecting small sea surface targets based on multiple features of map as claimed in claim 1, characterized by the steps ofTraining unit time sequence z in step 3 p Normalized Doppler power spectrum ofModeling graphs and graph signals, wherein a training unit time sequence z p Short sequence of (a) z p,l Normalized Doppler power spectrum ofModeling into two undirected graphs and signals on the graphs, and calculating three graph characteristics according to the following specific sub-steps:
wherein k = - [ L/2],...,[L/2]-1,The energy data representing the largest and smallest normalized kth doppler cell,
3.2 mixingModeling as an undirected graphDefining a Doppler Unit { - [ L/2 { [ L/2]],...,[L/2]-1 is an undirected graphGet undirectedDrawing (A)Set of vertices ofAnd considering each vertex to be connected with d adjacent vertexes, wherein the weight value between the connected vertexes is 1, and obtaining the undirected graphOf a neighboring matrixAnd laplacian matrixDefining the energy data of the k-th Doppler unit after maximum and minimum normalization as a vertex v k Graph signal x of k And will be located in an undirected graphIs represented by
wherein k = - [ L/2],...,[L/2]-1,Represents the quantized energy data of the kth Doppler cell, i representsA quantized value of the energy data of the doppler cell, γ representing a quantization level number;
3.4 mixingModeling as an undirected graphDefining quantized values {0, \8230;, i, \8230;. Gamma } as undirected graphTo get an undirected graphSet of vertices ofDefining the vertex corresponding to the quantization value in each Doppler unit to be connected with the vertex corresponding to the quantization value in the adjacent Doppler unit, defining the connection times between two vertexes as the weight between the two vertexes, and obtaining the undirected graphOf a neighboring matrix
Wherein,representing the vertex v i The degree of entry is calculated to obtain an undirected graphLaplacian matrix of
3.5 according to undirected graphLaplacian matrix ofUndirected graphSignal onUndirected graphLaplacian matrix ofAnd undirected graphIn-degree matrix ofThree graph features were calculated: graph Laplace regularization η 1 (z p,l ) Trace η of graph laplace matrix 2 (z p,l ) And the variance eta of the figure vertex in-degree 3 (z p,l ):
Wherein Tr (·) represents the trace of the matrix;
3.6 regularization η according to Laplace of the graph 1 (z p,l ) Trace η of graph laplace matrix 2 (z p,l ) And the variance eta of the figure vertex in-degree 3 (z p,l ) And constructing a feature vector of the graph: eta (z) p,l )=[η 1 (z p,l ),η 2 (z p,l ),η 3 (z p,l )] T 。
4. The method for detecting sea surface small targets based on multiple graph features according to claim 1, wherein in step 4, the method is based on a training sample setAt a given false alarm probability P f Next, the specific sub-steps of calculating the decision region Ω of the detector by using the greedy convex hull learning algorithm are as follows:
4.1 define a convex hull CH (S) containing all samples of the training sample set S:
wherein SP {. Cndot } represents a closed space surrounded by triangles,three vertices representing the qth triangular face constituting the surface of the convex hull CH (S), Q representing the total number of triangular faces on the convex hull CH (S);
4.2 according to a given false alarm probability P f And a greedy convex hull learning algorithm, which deletes [ P N/L P ] in turn to make the convex hull volume reduce most f ]A false alarm training sample to obtain a residual training sample set S', wherein [ ·]Represents rounding down;
4.3 compose a convex hull CH (S ') from the remaining training sample set S ' as the decision region Ω = CH (S ') of the detector.
5. The method for detecting small sea surface targets based on multiple map features as claimed in claim 1, wherein in step 6, the detection statistic of the unit to be detected is calculated according to the decision region Ω = CH (S') of the detector and the map three feature vector η (r) of the unit to be detected
6. Sea surface small target detection system based on multiple features of the figure, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of any of the preceding claims 1 to 5 when executing the program.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211513958.1A CN115774248A (en) | 2022-11-29 | 2022-11-29 | Sea surface small target detection method and system based on multiple characteristics of graph |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211513958.1A CN115774248A (en) | 2022-11-29 | 2022-11-29 | Sea surface small target detection method and system based on multiple characteristics of graph |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115774248A true CN115774248A (en) | 2023-03-10 |
Family
ID=85391449
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211513958.1A Pending CN115774248A (en) | 2022-11-29 | 2022-11-29 | Sea surface small target detection method and system based on multiple characteristics of graph |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115774248A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117388816A (en) * | 2023-09-15 | 2024-01-12 | 中国人民解放军海军航空大学 | CA/GO/OS three-dimensional combination-based adaptive CFAR detection method |
-
2022
- 2022-11-29 CN CN202211513958.1A patent/CN115774248A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117388816A (en) * | 2023-09-15 | 2024-01-12 | 中国人民解放军海军航空大学 | CA/GO/OS three-dimensional combination-based adaptive CFAR detection method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105137498B (en) | A kind of the underground objects detection identifying system and method for feature based fusion | |
Chen et al. | Spatial–temporal convolutional gated recurrent unit network for significant wave height estimation from shipborne marine radar data | |
CN106569193B (en) | The small targets detection in sea clutter method filtered based on anterior-posterior to income reference particle | |
Park et al. | Unsupervised classification of scattering mechanisms in polarimetric SAR data using fuzzy logic in entropy and alpha plane | |
CN111707999B (en) | Sea surface floating small target detection method based on combination of multiple features and ensemble learning | |
Golbon-Haghighi et al. | Ground clutter detection for weather radar using phase fluctuation index | |
Huang et al. | Wave height estimation from X-band nautical radar images using temporal convolutional network | |
Yang et al. | Evaluation and mitigation of rain effect on wave direction and period estimation from X-band marine radar images | |
CN115774248A (en) | Sea surface small target detection method and system based on multiple characteristics of graph | |
Chen et al. | LFM signal detection and estimation based on deep convolutional neural network | |
Long et al. | A comprehensive deep learning-based outlier removal method for multibeam bathymetric point cloud | |
CN117436329A (en) | Machine learning-based synthetic aperture radar image wind direction inversion method | |
Yuan | A time-frequency feature fusion algorithm based on neural network for HRRP | |
Bai et al. | Enhanced local sparsity coefficient-based sea-surface floating target detection | |
CN114594463A (en) | Sea surface small target feature detection method based on combined convex hull | |
Liang et al. | A multi-featured detection method for small target on sea surface based on GSP | |
CN113673355A (en) | Method for classifying echo waveforms of wave spectrometer | |
Chen et al. | Integrated processing of radar detection and classification for moving target via time-frequency graph and CNN learning | |
Chen et al. | Automatic identification of rain-contaminated regions in X-band marine radar images | |
Liu et al. | Arctic Sea-ice type recognition based on the surface wave investigation and monitoring instrument of the China-French Ocean satellite | |
Chen et al. | Gaussian process regression for estimating wind speed from X-band marine radar images | |
Narayanan et al. | Classification of SHOALS 3000 bathymetric LiDAR signals using decision tree and ensemble techniques | |
Wang et al. | Evaluating the Adaptability of Deep Learning-Based Multi-feature Sonar Image Detection Algorithms. | |
Adolfsson et al. | Machine learning for categorization of small boats and sea clutter | |
Zhang et al. | Small target detection in sea-clutter based on time-doppler spectrum |
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 |