Radar sea surface small target detection method based on heterogeneous graph strong product topological features, medium and equipment

Abstract

The invention provides a radar sea surface small target detection method based on heterogeneous graph strong product topological characteristics, a medium and equipment. The method comprises the following steps: calculating a frequency domain module value sequence and an autocorrelation function module value sequence of an observation signal; performing quantization graph transformation on the frequency domain modulus value sequence, and performing ordinal number graph transformation on the autocorrelation function modulus value sequence; performing strong graph product operation on the adjacent matrixes obtained by transformation of the two graphs to obtain a composite graph and an adjacent matrix thereof; and the maximum characteristic value of the A1 / 2 spectrum of the composite image is used as a test statistic and is compared with the calculated range interval, so that whether the signal is a target signal or a clutter signal is judged. According to the method, topological features of different types of graphs are fused in a graph strong product mode, the transfer relation between single samples contained in a quantized graph can be utilized, correlation information of a sample state space embodied by ordinal graph transformation can be fused, the difference between a target sequence and a clutter sequence can be better embodied, and the method has the advantages of being high in robustness and high in robustness. And the detection performance is obviously improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, and in particular relates to a radar sea surface small target detection method, medium and equipment based on heterogeneous graph intensity product topological features. Background Art

[0002] Small target detection in the sea clutter background has become a hot and difficult topic in the field of radar detection. In recent years, with the rise of graph signal processing technology, detection algorithms based on graph domain signal processing have been widely used in radar small target detection and have become increasingly mature. The basic idea is to transform the signal from the traditional time domain, frequency domain or other transform domains to the graph domain through specific rules. The graph domain features can often provide potential structural features of the signal that traditional statistical signal processing cannot provide, providing a new basis for improving the performance of the detection algorithm. However, the existing radar small target detection algorithms based on graph domain processing only use a single type of graph features, such as quantized graphs, visible graphs, ordinal graphs, etc. For example: construct a graph through quantized graph transformation, complete detection through the connectivity of the graph and the maximum eigenvalue of the Laplace matrix that reflects the connectivity of the graph; consider the connectivity density of the graph, and further improve the detection performance by introducing the edge weight of the graph; after the signal is transformed into a visible graph, the graph neural network is used to train its visible graph adjacency matrix, extract features, and complete detection. In essence, each type of graph transformation only extracts the topological features of a certain feature from one aspect. The detection performance of this detection algorithm based on a single graph transformation is relatively one-sided, and the detection effect needs to be improved. If different graph transformation methods can be integrated and the advantages of different graph transformations can be combined, it will be of great value to improve the effectiveness and reliability of graph domain detection statistics. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a radar sea surface small target detection method, medium and device based on the topological characteristics of the strong product of heterogeneous graphs. Graph product is a way to combine two different graphs. The commonly used graph products are Cartesian product and Kronecker product, while the strong product combines the characteristics of Cartesian product and Kronecker product. The present invention combines the quantitative graph with the ordinal graph by means of the strong product of the graph, effectively improving the performance of the existing detection algorithm based on single graph transformation.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] In a first aspect, the present invention provides a radar sea surface small target detection method based on heterogeneous graph intensity product topological features, characterized in that it comprises the following steps:

[0006] Step 1: Obtain the echo sequence received by the radar as an observation signal, perform frequency domain transformation on the observation signal to obtain a spectrum sequence, set the quantization level γ, and perform quantization map transformation on the spectrum sequence;

[0007] Step 2: Calculate the autocorrelation function modulus sequence of the observed signal, set the embedding dimension m and the embedding delay τ, and perform ordinal graph transformation on the autocorrelation function modulus sequence;

[0008] Step 3: Perform graph product synthesis on the quantization graph and the sequence graph in a strong product manner to extract A of the synthesized graph. 1 / 2 spectrum, and for the A 1 / 2 The spectrum is subjected to eigendecomposition to obtain the maximum eigenvalue λ1;

[0009] Step 4: Collecting X-band radar detection data of the sea, wherein the detection data includes a plurality of clutter units and a target unit, and calculating the range interval of the clutter unit;

[0010] Step 5: According to the relationship between the maximum eigenvalue λ1 and the range of the clutter unit, determine whether the observed signal is a target signal or a clutter signal.

[0011] Optionally, in step 1, the calculation process of the quantization map transformation is as follows:

[0012] Step 1.1: Perform fast Fourier transform on the observed signal x(n) and take its modulus value to obtain the spectrum sequence F(k);

[0013] Step 1.2: Normalize the maximum and minimum values ​​of the spectrum sequence F(k) to obtain the normalized spectrum sequence Y(k);

[0014] Step 1.3: Set the quantization level γ to quantize Y(k) at equal intervals to obtain the quantized spectrum sequence Q(k);

[0015] Step 1.4: Map Q(k) to a directed weighted graph and obtain the corresponding adjacency matrix A1.

[0016] Optionally, in step 2, the calculation process of the ordinal diagram transformation is as follows:

[0017] Step 2.1: De-mean the observed signal x(n) to obtain a zero-mean observed sequence y(n), and calculate the autocorrelation function modulus sequence r(k) of y(n);

[0018] Step 2.2: Reconstruct r(k) in phase space with embedding dimension m and embedding delay τ to obtain the state vector:

[0019] R(i)={r(i), r(i+τ),..., r(i+(m-1)τ)};

[0020] Where R(i) is the subsequence of the reconstructed state vector, 1≤i≤L-(m-1)τ, and L is the length of the echo sequence

[0021] Step 2.3: Re-sort the data in each reconstructed subsequence R(i) in ascending order, and extract the original index value of the sorted subsequence. In the m-dimensional phase space, there are m! sorting modes for the subsequence R(i), forming a sorting set {π(i));

[0022] Step 2.4: Convert each reordered subsequence R(i) into a graph G = (V, E), where E and V represent the edge set and vertex set of the graph respectively, and the vertices of the graph are given by the embedding dimension N = {1, 2, ..., i, ..., m!};

[0023] Step 2.5: Get the edge set based on the adjacent relationship of the vertices in the vertex set V. For the repeated edge e αβ The number of occurrences of is accumulated, and the accumulated value is used as the edge e αβ The weight of each edge is w αβ Constitute the elements in the adjacency matrix A2.

[0024] Optionally, in step 3, the maximum eigenvalue is calculated as follows:

[0025] Step 3.1: Calculate the strong product of the adjacency matrix A1 of the quantized graph and the adjacency matrix A2 of the ordinal graph to obtain the adjacency matrix A:

[0026]

[0027] In the formula, represents the Kronecker product, and denote the identity matrices of N1 and N2 dimensions respectively;

[0028] Step 3.2: Sum each row of the adjacency matrix A to obtain the degree d of each vertex α for:

[0029]

[0030] In the formula, γ represents the quantization level of the quantization map;

[0031] The degree matrix D of the graph is defined as:

[0032] D = diag(d1, d2, ..., d α , ..., d γ*m！ );

[0033] Where diag(·) is a α A diagonal matrix of vectors;

[0034] Get the A of the figure 1 / 2 The spectrum is:

[0035] A 1 / 2 =γD+(1-γ)A;

[0036] Step 3.3: For A 1 / 2 Perform eigendecomposition of the spectrum and take its maximum eigenvalue λ1:

[0037] λ1=max(λ1,λ2,...,λ γ*m！ );

[0038] Where λ1, λ2, ..., λ γ*m！ are the features obtained after decomposition.

[0039] Optionally, in step 4, the calculation process of the range interval of the clutter unit is as follows:

[0040] Step 4.1: Collect X-band radar detection data of the sea;

[0041] Step 4.2: Calculate the maximum eigenvalue of all clutter units in the detection data and obtain the vector

[0042]

[0043] In the formula, λ 1，1， λ 2，1 , ..., λ i，1 , ..., λ k，1 They represent the maximum eigenvalues ​​corresponding to each clutter unit, and there are k clutter units in total;

[0044] Step 4.3: Calculate the range interval ξ of the clutter unit as:

[0045]

[0046] Where E(·) represents the mean and std(·) represents the variance.

[0047] Optionally, in step 5, the method of determining whether the observed signal is a target signal or a clutter signal is as follows:

[0048] The judgment is made based on the relationship between the maximum eigenvalue λ1 and the range interval ξ of the clutter unit: if λ1∈ξ, the observed signal is a clutter signal; otherwise, the observed signal is a target signal.

[0049] In a second aspect, the present invention provides a computer-readable storage medium storing a computer program, characterized in that the computer program enables a computer to execute the radar sea surface small target detection method based on heterogeneous graph intensity product topological features as described in the first aspect.

[0050] In a third aspect, the present invention provides an electronic device, characterized in that it includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the radar sea surface small target detection method based on heterogeneous graph intensity product topological features as described in the first aspect.

[0051] The beneficial effects of the present invention are:

[0052] The present invention fully considers that sea clutter is generally a non-Gaussian, nonlinear, and non-stationary signal. It is difficult to distinguish the target from the sea clutter by traditional time domain or frequency domain. In addition, the method based on graph transformation and quantization graph transformation cannot be well applied to the data set with the signal-to-noise ratio lower than 10dB. The ordinal graph has poor data detection performance for HH polarization and VV polarization. Therefore, the signal is converted into an ordinal graph and a quantization graph in a targeted manner, and the graph product is combined, and the A of the graph product is used. 1 / 2 The maximum eigenvalue of the spectrum matrix is ​​used to detect targets and clutter. This method has the following advantages:

[0053] (1) The quantitative graph and the ordinal graph are combined by strong product through the graph product method. The strong product combines the characteristics of Cartesian product and Kronecker product, making the combined graph have stronger connectivity and at the same time have the mutual correlation between the nodes of the quantitative graph and the correlation between the state spaces of the ordinal graph.

[0054] (2)A 1 / 2 The spectrum is a linear combination of the degree matrix and the adjacency matrix. It is usually used to study the eigenvalues ​​of a graph and can better reveal the connectivity, clustering and other information of the graph. And through this linear combination, a smooth transition of the graph signal can be achieved.

[0055] The present invention realizes the fusion of different types of graph topological features by means of graph strength product. It can not only utilize the transfer relationship between single samples contained in the quantization graph, but also fuse the correlation information of the sample state space reflected by the ordinal graph transformation, so as to better reflect the difference between the target sequence and the noise sequence. Its detection performance is significantly better than the existing method. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flow chart of the radar sea surface small target detection method based on heterogeneous graph intensity product topological features in this embodiment.

[0057] Figure 2a and Figure 2b They are schematic diagrams of the target unit and the clutter unit generating spectrum sequences in this embodiment.

[0058] Figure 2c and Figure 2d They are schematic diagrams of the target unit and the clutter unit generating the autocorrelation function sequence in this embodiment.

[0059] Figure 3a and Figure 3b They are schematic diagrams of the target unit and the clutter unit after quantization map conversion in this embodiment.

[0060] Figure 3c and Figure 3d They are schematic diagrams of the target unit and the clutter unit after ordinal diagram conversion in this embodiment.

[0061] Figure 4a and Figure 4b It is a graph drawn by the adjacency matrix before and after strong graph product synthesis in this embodiment.

[0062] Figure 5 It is a comparative schematic diagram of the receiver operating characteristic curve (ROC) using the method proposed in this embodiment and the ordinal diagram method. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0064] In one embodiment, the present invention proposes a radar sea surface small target detection method based on heterogeneous graph intensity product topological features, such as Figure 1 As shown, the following steps are included.

[0065] Step 1: Obtain the observed signal, perform Fourier transform on the observed signal to obtain the spectrum, and perform quantization map transformation with an appropriate quantization level.

[0066] The observation signal is the echo sequence received by the radar. The observation signal can be described as a binary hypothesis test:

[0067]

[0068] Where x(n) represents the signal of the pulse echo data received by the radar, w(n) is the pure sea clutter signal, s(n) is the target signal, and L is the total length of the echo sequence.

[0069] In step 1, the calculation process of the quantization map transformation is as follows:

[0070] Step 1.1: Perform frequency domain transformation on the echo sequence x(n) received by the radar, i.e. Fast Fourier Transform (FFT), and take its modulus value to obtain the spectrum sequence F(k):

[0071]

[0072] Step 1.2: Normalize the maximum and minimum values ​​of the spectrum modulus, that is:

[0073]

[0074] In the formula, θ max and θ min are the maximum and minimum values ​​of F(k) respectively.

[0075] Step 1.3: Set the quantization level γ to quantize Y(k) at equal intervals, that is:

[0076]

[0077] Step 1.4: Map Q(k) to a directed weighted graph G = {E, V}, where E and V represent the edge set and vertex set of the graph respectively; and obtain the adjacency matrix A1.

[0078] Figure 3a and Figure 3b They are schematic diagrams of the target unit and the clutter unit after quantization map conversion.

[0079] Step 2: Calculate the autocorrelation function modulus sequence of the observed signal and perform ordinal graph transformation with appropriate embedding dimension and embedding delay.

[0080] In step 2, the calculation process of the ordinal map transformation is as follows:

[0081] Step 2.1: De-mean the echo sequence x(n) received by the radar to obtain the zero-mean observation sequence y(n):

[0082]

[0083] In the formula, is the mean value of the echo sequence x(n), L is the length of the echo sequence;

[0084] The modulus sequence r(k) of the autocorrelation function is:

[0085]

[0086] Step 2.2: Reconstruct the phase space of the autocorrelation function modulus sequence r(k) with the embedding dimension size m and the embedding delay τ to obtain the state vector:

[0087] R(i)={r(i), r(i+τ),..., r(i+(m-1)τ)};

[0088] Where R(i) is the subsequence of the reconstructed state vector, 1≤i≤L-(m-1)τ, and L is the length of the echo sequence.

[0089] Step 2.3: Re-sort the data in each reconstructed subsequence R(i) in ascending order, and extract the original index value of the sorted subsequence. In the m-dimensional phase space, there are m! sorting modes for the subsequence R(i), forming a sorting set {π(i)).

[0090] Step 2.4: Convert each reordered subsequence into a corresponding graph G = (V, E), where the vertices are given by the embedding dimension N = {1, 2, ..., i, ..., m!}.

[0091] Figure 3c and Figure 3d They are schematic diagrams of the target unit and the clutter unit after ordinal graph conversion.

[0092] Step 2.5: Get the edge set based on the adjacent relationship of the vertices in the vertex set V. For the repeated edge e αβ The number of occurrences of is accumulated, and the accumulated value is used as the edge e αβ The weight of each edge is w αβ Constitute the elements in the adjacency matrix A2.

[0093] In this embodiment, Figures 2a to 2d The following are schematic diagrams of the spectrum sequence and autocorrelation function sequence generated by the H1 target unit and the H0 clutter unit. By observing the spectrum sequence and time domain autocorrelation function sequence of the target unit and the clutter unit, it can be found that the amplitude fluctuation difference between the two is more prominent than the fluctuation difference of the time domain amplitude sequence.

[0094] Step 3: Perform graph product synthesis on the obtained ordinal graph and quantization graph in a strong product manner to obtain a new composite graph, and extract the A of the graph 1 / 2 Spectrum, for A 1 / 2 The spectrum is decomposed to obtain the maximum eigenvalue λ1, which is used as the detection statistic.

[0095] In step 3, the calculation process of the strong product of the two adjacency matrices and the test statistic is as follows:

[0096] Step 3.1: Calculate the strong product of the two adjacency matrices, as follows:

[0097]

[0098] in, represents the Kronecker product, and denote the identity matrices of N1 and N2 dimensions respectively.

[0099] Step 3.2: Sum each row of the adjacency matrix A to obtain the degree d of each vertex α for:

[0100]

[0101] The degree matrix D of the graph is defined as:

[0102] D = diag(d1, d2, ..., d α , ..., d γ*m！ );

[0103] Where diag(·) is a α A diagonal matrix of vectors;

[0104] Get the A of the figure 1 / 2 The spectrum is:

[0105]

[0106] A 1 / 2 The spectrum is a fusion of the graph degree matrix and the graph adjacency matrix, which more effectively and fully reflects the topological information of the graph.

[0107] Step 3.3: For A 1 / 2 Perform eigendecomposition of the spectrum and take its maximum eigenvalue λ1:

[0108] λ1=max(λ1,λ2,...,λ γ*m！ ).

[0109] Figure 4a and Figure 4b It is a graph drawn by the adjacency matrix before and after strong graph product synthesis in this embodiment.

[0110] Step 4: Collect X-band radar detection data of the sea, which includes multiple clutter units and one target unit, and calculate the range of the clutter units.

[0111] In step 4, the calculation process of the range interval of the clutter unit is as follows:

[0112] Step 4.1: The source of the detection data is the "Radar Sea Detection Data Sharing Program". This group of data was organized by Professor Liu Ningbo of the Naval Aviation University of the Chinese People's Liberation Army. The X-band radar was used to conduct sea detection tests to obtain sea clutter and target data under different radar resolutions, ground-grazing angles and different sea conditions, and record the corresponding meteorological and hydrological data and the target's true trajectory. The data contains multiple clutter units and one target unit, and the data signal is named "20221112180016_stare_HH.mat". For the LFM pulse signal in T2 mode, the target is located at the 502nd unit.

[0113] Step 4.2: For all clutter units, the A can be calculated using the above method. 1 / 2The detection statistic of the spectrum (i.e., the maximum eigenvalue) can also be calculated using other existing methods to obtain the vector

[0114]

[0115] Among them, λ i，1 A represents the i-th clutter unit 1 / 2 The maximum eigenvalue of the spectrum, there are k clutter units in total.

[0116] Step 4.3: Define the range interval ξ of the clutter unit as:

[0117]

[0118] Among them, E(.) represents the mean of the vector, and std(.) represents the variance of the vector.

[0119] Step 5: According to the relationship between the maximum eigenvalue λ1 and the range interval ξ of the clutter unit, determine whether the observed signal is a target signal or a clutter signal. If λ1∈ξ, it is considered to be H0; otherwise, it is H1.

[0120] Figure 5 It is a schematic diagram of the ROC curve comparison between the method of this embodiment and the ordinal graph method. (1) Changes in the ordinal graph: Since the length of the signal will become 2N-1 of the original signal after the autocorrelation sequence is converted, in order to make full use of the data, the simulation conditions are: 1000 simulations are performed, 512 points are taken for each segment, and 256 points are slid each time (i.e., overlap 256), and the embedding dimension size m=4 and the embedding delay τ=1. (2) Changes in the quantization graph: 512 points are taken for each segment of the time domain sequence, 128 points are slid each time, and the frequency domain conversion is performed on the segment of the time domain sequence respectively, and at the same time, the quantization level γ=7. It can be seen from the figure that the area under the ROC curve (AUC) of this method is larger than that of other methods, which means that the detection performance of the method of this embodiment is more superior. Therefore, the method of this embodiment can be well applied to the radar sea surface small target detection under the background of sea clutter, and the detection probability is greatly improved compared with the previous methods.

[0121] In another embodiment, the present invention proposes a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the radar sea surface small target detection method based on heterogeneous graph intensity product topological features as described in the first embodiment.

[0122] In another embodiment, the present invention proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the radar sea surface small target detection method based on heterogeneous graph intensity product topological features as described in Example 1 is implemented.

[0123] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0124] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0125] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A radar sea surface small target detection method based on heterogeneous graph intensity product topological features, characterized in that: The steps include: Step 1: Obtain the echo sequence received by the radar as an observation signal, perform frequency domain transformation on the observation signal to obtain a spectrum sequence, set the quantization level γ, and perform quantization map transformation on the spectrum sequence; Step 2: Calculate the autocorrelation function modulus sequence of the observed signal, set the embedding dimension m and the embedding delay τ, and perform ordinal graph transformation on the autocorrelation function modulus sequence; Step 3: Perform graph product synthesis on the quantization graph and the sequence graph in a strong product manner to extract A of the synthesized graph. 1 / 2 spectrum, and for the A 1 / 2 The spectrum is subjected to eigendecomposition to obtain the maximum eigenvalue λ1; Step 4: Collecting X-band radar detection data of the sea, wherein the detection data includes a plurality of clutter units and a target unit, and calculating the range interval of the clutter unit; Step 5: According to the relationship between the maximum eigenvalue λ1 and the range of the clutter unit, determine whether the observed signal is a target signal or a clutter signal.

2. The radar sea surface small target detection method based on heterogeneous graph intensity product topological features as claimed in claim 1 is characterized by: In step 1, the calculation process of the quantization map transformation is as follows: Step 1.1: Perform fast Fourier transform on the observed signal x(n) and take its modulus value to obtain the spectrum sequence F(k); Step 1.2: Normalize the maximum and minimum values ​​of the spectrum sequence F(k) to obtain the normalized spectrum sequence Y(k); Step 1.3: Set the quantization level γ to quantize Y(k) at equal intervals to obtain the quantized spectrum sequence Q(k); Step 1.4: Map Q(k) to a directed weighted graph and obtain the corresponding adjacency matrix A1.

3. The radar sea surface small target detection method based on heterogeneous graph intensity product topological features as claimed in claim 1 is characterized by: In step 2, the calculation process of the ordinal diagram transformation is as follows: Step 2.1: De-mean the observed signal x(n) to obtain a zero-mean observed sequence y(n), and calculate the autocorrelation function modulus sequence r(k) of y(n); Step 2.2: Reconstruct r(k) in phase space with embedding dimension m and embedding delay τ to obtain the state vector: R(i)={r(i), r(i+τ),..., r(i+(m-1)τ)}; Where R(i) is the subsequence of the reconstructed state vector, 1≤i≤L-(m-1)τ, and L is the length of the echo sequence Step 2.3: Re-sort the data in each reconstructed subsequence R(i) in ascending order, and extract the original index value of the sorted subsequence. In the m-dimensional phase space, there are m! sorting modes for the subsequence R(i), forming the sorting set {π(i)}; Step 2.4: Convert each reordered subsequence R(i) into a graph G = (V, E), where E and V represent the edge set and vertex set of the graph respectively, and the vertices of the graph are given by the embedding dimension N = {1, 2, ..., i, ..., m!}; Step 2.5: Get the edge set based on the adjacent relationship of the vertices in the vertex set V. For the repeated edge e αβ The number of occurrences of is accumulated, and the accumulated value is used as the edge e αβ The weight of each edge is w αβ Constitute the elements in the adjacency matrix A2.

4. The radar sea surface small target detection method based on heterogeneous graph intensity product topological features as claimed in claim 1 is characterized by: In step 3, the calculation process of the maximum eigenvalue is as follows: Step 3.1: Calculate the strong product of the adjacency matrix A1 of the quantized graph and the adjacency matrix A2 of the ordinal graph to obtain the adjacency matrix A: In the formula, represents the Kronecker product, and denote the identity matrices of N1 and N2 dimensions respectively; Step 3.2: Sum each row of the adjacency matrix A to obtain the degree d of each vertex α for: In the formula, γ represents the quantization level of the quantization map; The degree matrix D of the graph is defined as: D=diag(d1,d2,...,d α ,...,d γ*m！ ); In the formula, diag(·) is a α A diagonal matrix of vectors; Get the A of the figure 1 / 2 The spectrum is: A 1 / 2 =γD+(1-γ)A; Step 3.3: For A 1 / 2 Perform eigendecomposition of the spectrum and take its maximum eigenvalue λ1: λ1=max(λ1,λ2,...,λ γ*m！ ); Where λ1, λ2, ..., λ γ*m！ are the features obtained after decomposition.

5. The radar sea surface small target detection method based on heterogeneous graph intensity product topological features as claimed in claim 1 is characterized by: In step 4, the calculation process of the range interval of the clutter unit is as follows: Step 4.1: Collect X-band radar detection data of the sea; Step 4.2: Calculate the maximum eigenvalue of all clutter units in the detection data and obtain the vector In the formula, λ 1，1 ,λ 2，1 , ..., λ i，1 , ..., λ k，1 They represent the maximum eigenvalues ​​corresponding to each clutter unit, and there are k clutter units in total; Step 4.3: Calculate the range interval ξ of the clutter unit as: Where E(·) represents the mean and std(·) represents the variance.

6. The radar sea surface small target detection method based on heterogeneous graph intensity product topological features as claimed in claim 5 is characterized by: In step 5, the method of determining whether the observed signal is a target signal or a clutter signal is as follows: The judgment is made based on the relationship between the maximum eigenvalue λ1 and the range interval ξ of the clutter unit: if λ1∈ξ, the observed signal is a clutter signal; otherwise, the observed signal is a target signal.

7. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables the computer to execute the radar sea surface small target detection method based on heterogeneous graph intensity product topological features as described in any one of claims 1-6.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the radar sea surface small target detection method based on heterogeneous graph intensity product topological features as described in any one of claims 1 to 6 is implemented.

Images