A method and device for detecting weak targets in sea clutter based on graph entropy

Abstract

The present application belongs to the technical field of radar target detection, and discloses a weak target detection method under sea clutter background based on graph entropy, comprising the following steps: extracting information entropy features from the adjacent matrix of graph signals in the frequency domain by using echo data received by a radar; and obtaining target detection results for comparative analysis based on the realization of a weak target detector of graph entropy. The present application has the beneficial effect of further exploring the characteristics between clutter and targets, and proposes information entropy features based on graphs. In the experimental analysis of actual sea clutter data combined with the performance curve analysis of other entropy feature detection, it is shown that the weak target detection device based on graph entropy proposed by the present application has better detection performance.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of radar target detection, and relates to a weak target detection method and device based on graph entropy under sea clutter background. BACKGROUND

[0002] Under the sea clutter background, since the sea clutter has the characteristics of non-Gaussian, non-linear and non-stationary, the radar will inevitably be affected by the sea state, wind speed and other environments when detecting small targets such as small boats, icebergs and aircraft wreckage floating on the sea surface. Therefore, small target detection has always been a difficult point in sea surface target detection. The traditional detection algorithm based on statistical model will cause certain errors in the statistical model due to the complexity of the sea state. With the wide application of graph structure in biology, transportation, speech signal processing and neural network, the graph structure as a general data representation form is also applied to the weak target detection of sea clutter. The edges and edge weights between the graph vertices contain the correlation information between the signals on the graph vertices, that is, the correlation between the data is obtained through the graph.

[0003] In IEEE Communication Letter, Vol. 57, published in 2019, YAN Kun et al. proposed a graph-based sea clutter target detection method, in which the radar echo data is converted into a graph to realize the echo data detection function, and the detection performance of the method is better under low signal-to-noise ratio (SNR). The deficiency of this paper is that the connectivity density between the data and the sparsity of the data in the graph are not considered when constructing the graph, and there is room for improvement. In addition, the invention patent application discloses a sea surface target detection method and device based on graph connectivity density, application number: 202110196211.7, publication number: CN113009443A. The invention patent application constructs the frequency domain signal of the radar echo into a graph, uses the Laplacian matrix composed of the graph connectivity density to extract the maximum value of the Laplacian matrix eigenvalue, not only considers the correlation between the data, but also considers the connectivity between the data, but does not further explore the characteristics between the sea clutter and the target, and there is a better weak target detection method. SUMMARY

[0004] In order to solve the problem of weak target detection difficulty under sea clutter background in the prior art, a better detection method needs to be sought, and the application provides a weak target detection method and device based on graph entropy under sea clutter background. The information entropy feature of the radar received echo data is extracted from the graph signal in the frequency domain, the isolated points in the graph signal are removed, and the information entropy in the graph signal after removing the isolated points is calculated. Good detection performance is obtained through experiments on actual sea clutter data.

[0005] To achieve the above purpose, the application provides the following technical scheme:

[0006] In a first aspect, a weak target detection method based on graph entropy (GE) in sea clutter background includes the following steps:

[0007] Extracting information entropy features from the adjacent matrix of the graph signal in the frequency domain using the echo data received by the radar;

[0008] Implementing a weak target detector based on graph entropy to obtain a target detection result for comparative analysis.

[0009] In combination with the first aspect, further, the method for extracting information entropy features from the adjacent matrix of the graph signal in the frequency domain is:

[0010] The graph signal in the frequency domain of the radar echo is represented as:

[0011] G = (V, E)

[0012] Wherein, V is the vertex set of the graph, E is the edge set, e δβ represents the edge from vertex v β to vertex v δ , v β ∈V, δ, β ∈ [0, γ]; the edge set E = {e

[0013]

[0014] The number of occurrences of each edge in the edge set E calculates the connectivity density between each vertex in the vertex set:

[0015] ωδβ=∑eδβ

[0016] Wherein, ω δβ represents the connectivity density between vertex v β δ and vertex v

[0017] Obtaining the adjacent matrix A of the sea surface target graph according to the connectivity density between the vertices:

[0018]

[0019] Converting the adjacent matrix A into a vector a

[0020] a = vec(A)

[0021] Removing the isolated points with a value of 0 in a to form a new vector

[0022]

[0023] denotes a point in the deletion vector whose element value is 0, with a length of L, The probability p(l) of each element in the deletion vector is calculated as:

[0024]

[0025] wherein, The information entropy GE of the connected graph is expressed as:

[0026]

[0027] In combination with the first aspect, further, the weak target detector based on the graph entropy is implemented to obtain a weak target detection result, the graph entropy is taken as a feature of target detection to show the difference between the clutter and the target, and then a sea surface target detection result is obtained. An expression of the weak target detection result is:

[0028]

[0029] wherein, H0 and H1 are sea surface target detection results, H0 denotes that there is no target in a sea surface echo amplitude sequence obtained by the radar, H1 denotes that there is a target in the sea surface echo amplitude sequence obtained by the radar, and ζ is a detection statistic determined based on a clutter feature.

[0030] In combination with the first aspect, further, a specific method for obtaining the echo data received by the radar is: obtaining a sea surface echo amplitude sequence by using the radar, and pre-processing the sea surface echo amplitude sequence to obtain pre-processed echo amplitude data.

[0031] In combination with the first aspect, further, the pre-processing includes frequency domain conversion, normalization processing, and mean value quantization processing.

[0032] In combination with the first aspect, further, the step of pre-processing the sea surface echo amplitude sequence includes:

[0033] Based on the sea surface echo amplitude sequence, fast Fourier transform is performed to obtain frequency domain converted echo amplitude data, and the formula is as follows:

[0034]

[0035] wherein, F(k) denotes the kth echo amplitude data after frequency domain conversion, x(n) denotes an amplitude of the nth pulse echo data in the sea surface echo amplitude sequence, k=1, 2, …, N, n=1, 2, …, N, and N is a quantity of pulse echo data in the sea surface echo amplitude sequence;

[0036] Based on the maximum minimum standard, the frequency domain converted echo amplitude data is normalized to obtain normalized echo amplitude data, and the formula is as follows:

[0037]

[0038] wherein U(k) represents the kth echo amplitude data after normalization, θ max represents the maximum value in the echo amplitude data after frequency domain transformation, θ min represents the minimum value in the echo amplitude data after frequency domain transformation,

[0039] According to the quantization interval 1 / γ, the normalized echo amplitude data is uniformly quantized to obtain quantized echo amplitude data, and the formula is as follows:

[0040]

[0041] wherein Q(k) represents the kth echo amplitude data after quantization, i.e., the kth preprocessed echo amplitude data, i is the quantization value of the echo amplitude data, and γ is the quantization level number.

[0042] In a second aspect, a weak target detection device based on graph entropy in sea clutter background comprises:

[0043] A feature extraction module is configured to extract information entropy features from the graph signal in the frequency domain of the echo data received by the radar.

[0044] A target detection module is configured to obtain a sea surface target detection result based on the weak target detector based on graph entropy.

[0045] Compared with the prior art, the present application provides a weak target detection method and device based on graph entropy in sea clutter background, which has the following beneficial effects:

[0046] (1) The present application further explores the characteristics between clutter and target on the basis of constructing the graph connected density matrix, and proposes the information entropy features based on the graph. In the actual sea clutter data experiment, the performance curve analysis combined with other entropy feature detection shows that the weak target detection device based on graph entropy proposed by the present application has better detection performance.

[0047] (2) The weak target detection method proposed by the present application extracts information entropy features from the graph signal in the frequency domain of the echo data received by the radar, and has lower calculation complexity and faster calculation speed compared with the prior art.

[0048] (3) The weak target detection method of the present application can remove isolated points in the graph signal, and then calculate the information entropy of the graph signal after removing the isolated points. Due to the difference between the target and the clutter, the number of isolated points in the target graph and the clutter graph is also very different. The number of isolated points in the clutter is more than that in the target, and the connected graph formed by the clutter is more aggregated than the connected graph formed by the target. Therefore, the information entropy obtained after removing the isolated points can better reflect the sparsity of the data, and further improves the target detection performance, and can better eliminate the influence of the clutter data.

[0049] (4) The weak target detection device provided by the present application has constant false alarm characteristics and more robust detection performance. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 The step flow chart of the weak target detection method under the sea clutter background based on graph entropy of the present application is shown in the figure.

[0051] Figure 2 The target and clutter connected graph under the sea clutter background of the present application is shown in the figure.

[0052] Figure 3 The performance curve comparison figure of the detection combined with other types of entropy characteristics under the VH polarization in the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0054] The present application provides a weak target detection method under the sea clutter background based on graph entropy, as shown in the figure, specifically including the following steps: Figure 1 As shown in the figure, specifically including the following steps:

[0055] Step one, extracting information entropy features from the graph signal in the frequency domain according to the echo data received by the radar:

[0056] In step one, the sea echo amplitude sequence is a time sequence composed of N pulse echo data received by the radar, and each pulse echo data includes the amplitude of the echo. The specific operation of preprocessing the sea echo amplitude sequence is as follows:

[0057] Based on the sea echo amplitude sequence, fast Fourier transform is performed to obtain the echo amplitude data after frequency domain transformation, and the specific formula is as follows:

[0058]

[0059] Wherein, F(k) represents the kth echo amplitude data after frequency domain transformation, x(n) represents the amplitude of the nth pulse echo data in the sea surface echo amplitude sequence, k=1, 2,..., N, n=1, 2,..., N, and N is the number of pulse echo data in the sea surface echo amplitude sequence.

[0060] In order to simplify the calculation method and reduce the difference between data, the echo amplitude data after frequency domain transformation is normalized based on the maximum minimum standard, the data is mapped between [0, 1], and the normalized echo amplitude data is obtained, and the specific formula is as follows:

[0061]

[0062] Wherein, U(k) represents the kth echo amplitude data after normalization, and θ max represents the maximum value in the echo amplitude data after frequency domain transformation, θ min represents the minimum value in the echo amplitude data after frequency domain transformation,

[0063] In order to reduce noise and improve the signal-to-noise ratio (SRN) of data, the normalized echo amplitude data is uniformly quantized according to the quantization interval 1 / γ, and the quantized echo amplitude data is obtained, and the specific formula is as follows:

[0064]

[0065] Wherein, Q(k) represents the kth echo amplitude data after quantization, that is, the kth preprocessed echo amplitude data, i is the quantization value of the echo amplitude data, and γ is the quantization level number.

[0066] When F(k) is equal to θ min , U(k)=0, at this time, i=0, that is, Q(k)=0; when U(k) is not equal to 0, the value of i is determined according to the value of U(k), for example, if γ is 6, U(k)=0.1773, then 1 / 6

[0067] In step one, the graph signal on the radar echo frequency domain is represented as:

[0068] G=(V, E) (4)

[0069] Wherein, V is the vertex set of the graph, E is the edge set, e δβ represents the edge from vertex v δ to vertex v β , v δ , v β∈ V, δ, β ∈ [0, γ]; according to e δβ The edge set E of the generated graph is {e δβ | <δ, β> ∈ (γ + 1) × (γ + 1)}, and γ is the number of quantization levels;

[0070]

[0071] The number of times each edge in the edge set E appears calculates the connectivity density between each vertex in the vertex set V:

[0072] ω δβ = ∑e δβ (6)

[0073] where ω δβ represents the connectivity density between vertex v δ and vertex v β .

[0074] According to the connectivity density between vertices, the adjacency matrix A of the sea surface target graph is obtained:

[0075]

[0076] Through analysis of Figure 2 , it is found that the isolated points in the clutter are more than those in the target, and the connected graph formed by the clutter data is more aggregated than the connected graph formed by the target data. Therefore, the adjacency matrix A is converted into a vector a

[0077] a = vec(A) (8)

[0078] Remove the isolated points with a value of 0 in a to form a new vector

[0079]

[0080] represents the length of the vector , and the length of is L, The probability p(l) of each element in

[0081] is calculated as:

[0082] where The information entropy (GE) of the connected graph is represented as:

[0083]

[0084] Step two, the implementation of the weak target detector based on graph entropy, obtains the weak target detection result:

[0085] The graph entropy (GE) is taken as a feature of target detection, and the difference between clutter and target is shown, and then the sea surface target detection result is obtained.

[0086]

[0087] H0 and H1 are sea surface target detection results, H0 represents that there is no target in the sea surface echo amplitude sequence obtained by the radar, H1 represents that there is a target in the sea surface echo amplitude sequence obtained by the radar, and ζ is a detection statistic based on the clutter feature.

[0088] Step three, combining the performance curves of other entropy feature detection and comparative analysis:

[0089] Figure 3 In order to compare the performance curves of the weak target detector based on the graph entropy and other entropy feature detectors, the effectiveness of the weak target detection method based on the graph entropy in the sea clutter background is verified.

[0090] A specific experiment is given below to verify the effect of the method:

[0091] In the embodiment of the application, the data measured by Professor S. Haykin of McMaster University in Canada using IPIX radar in a real marine environment is taken as the sea surface echo amplitude sequence obtained by the radar, and the data used in the experiment is named 19931107_135603_starea17. The sea surface echo amplitude sequence includes four polarizations obtained according to different transmission and reception signal modes, namely HH, VV, HV and VH. Each polarization contains 14 distance units, and the sampling data of one distance unit is 131072 (i.e. 131.072s). The 9th distance unit of the target in the embodiment of the application is the unit to be detected, 8, 10 and 11 are the affected units, and the rest are clutter units. In the experiment, the method of the application, the frequency domain Shannon entropy detector (SE), the relative vector entropy (RVE) and the non-extended entropy (Tsallis Entropy, TE) are used for sea surface target detection respectively, and the detection results of different algorithms under VH polarization are as shown in the table. Figure 3 As can be seen from the figure, when the false alarm probability is 10 -3 , the detection probability of the feature detection method proposed in the application can reach 92.2%, while the detection probabilities of the frequency domain Shannon entropy detector (SE), the relative vector entropy (RVE) and the non-extended entropy (Tsallis Entropy, TE) are 25.2%, 30.5% and 32.6% respectively; when the false alarm probability is 10 -2 , the detection probability of the other three algorithms does not improve very obviously, so the feature proposed in the application still has good target detection performance under low false alarm probability.

[0092] The application further provides a weak target detection device based on graph entropy under sea clutter background, comprising a feature extraction module and a target detection module, the feature extraction module is used for extracting information entropy features from the graph signal of the echo data received by the radar in the frequency domain; the target detection module is used for realizing the weak target detector based on the graph entropy to obtain the sea surface target detection result.

[0093] It should be noted that, in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0094] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, alternatives, and variations can be made in the embodiments without departing from the spirit and scope of the present application as defined by the appended claims and their equivalents.

Claims

1. A method for weak target detection in sea clutter background based on graph entropy, characterized in that, Includes the following steps: Information entropy features are extracted from the adjacency matrix of graph signals in the frequency domain using radar echo data. A weak target detector based on graph entropy is used to obtain comparative analysis of target detection results; The method for extracting information entropy features from the adjacency matrix of graph signals in the frequency domain is as follows: The radar echo signal in the frequency domain is represented as: ； in, Let E be the set of vertices of the graph, and E be the set of edges. Represents vertices To the top The edge, , ;according to Edge set of the generated graph , To quantify the number of levels; ； Calculate the connectivity density between vertices in the vertex set by counting the occurrences of each edge in edge set E. ； in, Represents vertices With vertex Connectivity density between them; Obtain the adjacency matrix of the sea surface target map based on the connectivity density between vertices. : ； Adjacency matrix Transform into a vector ; ； Remove Isolated points with a value of 0 in the vector are used to form a new vector. , ； This indicates deleting points from the vector whose element value is 0. The length is , The probability of each element The calculation is as follows: ； in, The information entropy GE of a connected graph is represented as: ； A weak target detector based on graph entropy is implemented to obtain weak target detection results, the expression of which is: ； in, and For the detection results of sea surface targets, This indicates that there is no target in the radar-acquired sea surface echo amplitude sequence. This indicates the presence of a target in the radar-acquired sea surface echo amplitude sequence. This refers to the detection statistics determined based on clutter characteristics; The specific method for obtaining radar echo data is as follows: use radar to obtain the sea surface echo amplitude sequence, and preprocess the sea surface echo amplitude sequence to obtain the preprocessed echo amplitude data.

2. The method for weak target detection in sea clutter background based on graph entropy according to claim 1, characterized in that, The preprocessing includes frequency domain transformation, normalization, and mean quantization.

3. The method for weak target detection in sea clutter background based on graph entropy according to claim 2, characterized in that, The steps for preprocessing the sea surface echo amplitude sequence include: The echo amplitude data after frequency domain transformation is obtained by performing a Fast Fourier Transform on the sea surface echo amplitude sequence, as shown in the following formula: ； in, This represents the k-th echo amplitude data after frequency domain transformation. This represents the amplitude of the nth pulse echo data in the sea surface echo amplitude sequence. , N represents the number of pulse echo data in the sea surface echo amplitude sequence; The echo amplitude data after frequency domain transformation is normalized based on the maximum and minimum standards to obtain the normalized echo amplitude data, as shown in the following formula: ； in, This represents the normalized k-th echo amplitude data. This represents the maximum value in the echo amplitude data after frequency domain transformation. , This represents the minimum value in the echo amplitude data after frequency domain transformation. ; According to quantization interval The normalized echo amplitude data is then subjected to uniform quantization to obtain the quantized echo amplitude data, as shown in the following formula: ； in, This represents the k-th echo amplitude data after quantization, i.e., the k-th preprocessed echo amplitude data. The quantized value of the echo amplitude data. This is for quantifying the number of levels.

4. A weak target detection device based on graph entropy in a sea clutter background, characterized in that, include: The feature extraction module is used to extract information entropy features from the graph signal in the frequency domain of the echo data received by the radar. The target detection module is used for weak target detectors based on graph entropy to obtain sea surface target detection results; The method for extracting information entropy features from the adjacency matrix of graph signals in the frequency domain is as follows: The radar echo signal in the frequency domain is represented as: ； in, Let E be the set of vertices of the graph, and E be the set of edges. Represents vertices To the top The edge, , ;according to Edge set of the generated graph , To quantify the number of levels; ； Calculate the connectivity density between vertices in the vertex set by counting the occurrences of each edge in edge set E. ； in, Represents vertices With vertex Connectivity density between them; Obtain the adjacency matrix of the sea surface target map based on the connectivity density between vertices. : ； Adjacency matrix Transform into a vector ; ； Remove Isolated points with a value of 0 in the vector are used to form a new vector. , ； This indicates deleting points from the vector whose element value is 0. The length is , The probability of each element The calculation is as follows: ； in, The information entropy GE of a connected graph is represented as: ； A weak target detector based on graph entropy is implemented to obtain weak target detection results, the expression of which is: ； in, and For the detection results of sea surface targets, This indicates that there is no target in the radar-acquired sea surface echo amplitude sequence. This indicates the presence of a target in the radar-acquired sea surface echo amplitude sequence. This refers to the detection statistics determined based on clutter characteristics; The specific method for obtaining radar echo data is as follows: use radar to obtain the sea surface echo amplitude sequence, and preprocess the sea surface echo amplitude sequence to obtain the preprocessed echo amplitude data.

Images