A method for detecting small targets on the sea surface based on relative sample entropy

By using the relative sample entropy method in the detection of small sea surface targets, the contradiction between performance improvement and calculation amount in the existing technology is solved, and the detection performance of small sea surface targets is improved, especially under low-information and miscellaneous ratio conditions, the more High detection accuracy.

CN114755649BActive Publication Date: 2024-11-22NANJING UNIV OF INFORMATION SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210300265.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-25
Publication Date
2024-11-22
Estimated Expiration
2042-03-25

AI Technical Summary

Technical Problem

The prior art has a contradiction between performance improvement and calculation amount reduction in sea surface small target detection, especially under low information and miscellaneous ratio conditions, it is difficult to effectively increase the detection probability of sea surface small target detection.

Method used

The sea surface small target feature detection method based on relative sample entropy is used. By whitening the radar signal in the Doppler domain, the sample entropy is calculated, and the relative sample entropy is used as the test statistic, combined with the Monte Carlo experimental method, the judgment threshold is obtained. Determine whether there is a goal.

Benefits of technology

It improves the detection performance of small targets on the sea surface, reduces the calculation amount, increases the spectrum difference between sea clutter and target-containing echoes, and improves the detection probability under low signal-to-mism ratio conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114755649B_ABST
    Figure CN114755649B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for detecting features of small targets on the sea surface based on relative sample entropy, comprising the following steps: obtaining a CUT observation vector; calculating a Doppler spectrum; obtaining a whitened spectrum; extracting sample entropy; obtaining relative sample entropy; and detecting and judging whether a target exists in the CUT. The present invention defines a whitened spectrum in the frequency domain to increase the difference in spectrum between sea clutter and echoes containing a target; improves the sample entropy algorithm to extract relative sample entropy as an efficient single feature, integrating multivariate information of the spectrum in terms of energy and geometry; the present invention can solve the contradiction between performance improvement and calculation reduction, and improve the detection probability of small targets on the sea surface under low signal-to-clutter ratio conditions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of radar signal processing, and in particular to a method for detecting features of small sea targets based on relative sample entropy. Background Art

[0002] At present, small targets on the sea surface, such as boats, speedboats, frogmen, and aircraft wreckage, have become the focus and difficulty of marine radar in performing early warning, detection, and tracking. Since the signal-to-clutter ratio (SCR) of small targets on the sea surface is at the critical observable level, long-term observation accumulation technology is often required during radar signal processing to improve target detection performance.

[0003] Under long-term observations of seconds, scholars have developed a variety of theories such as chaos theory, neural networks, fractal theory, and feature detection for small target detection on the sea surface. In the framework of chaos theory, whether sea clutter can be modeled as a chaotic dynamic system is still controversial. With the rapid development of artificial intelligence, methods based on neural network learning have been newly developed, and information is obtained by learning a large number of samples. However, this type of method is limited to the need for a large number of training samples, especially the samples of target echoes are very scarce. Considering the requirements for rapid implementation of actual radars, robust and efficient methods for small target detection on the sea surface can be roughly divided into two categories. The first category is detection methods based on fractal theory. Fractal theory can better characterize the nonlinearity of dynamic sea surfaces. A large number of scholars have studied the Hurst exponential characteristics, multi-fractal characteristics, and fractal characteristics in the transform domain, and used these characteristics in detection. This type of detector is simple, but its observation time usually takes more than several seconds. The second category is feature-based detection methods, the key is to find the difference between sea clutter and target echoes. Based on the statistical characteristics of radar echoes, in order to eliminate the influence of non-stationary texture components in the composite Gaussian model under long-term observation, some scholars have proposed a feature detector based on speckle consistency factor. Based on the physical scattering mechanism, some scholars have proposed a detector based on three polarization features. The combination of multiple features can indeed improve performance, but at the same time it increases the complexity of the algorithm and reduces the real-time and speed of the algorithm. Summary of the invention

[0004] Purpose of the invention: In order to overcome the defects existing in the background technology, the present invention discloses a method for detecting features of small targets on the sea surface based on relative sample entropy. The method introduces and improves the sample entropy algorithm, designs efficient single features, and improves the detection performance of small targets on the sea surface. It can solve the contradiction between performance improvement and reduction of computational complexity, and aims to improve the radar's detection capability of small targets on the sea surface.

[0005] Technical solution: The method for detecting small sea surface target features based on relative sample entropy of the present invention comprises the following steps:

[0006] S1. Assume that the radar receives N consecutive pulses in a range unit. These N pulses constitute an observation vector z = [z(1), z(2), ..., z(N)] T , called the unit to be detected CUT; at the same time, the observation vector z of P reference units around CUT is obtained p ,p=1,2,...,P, construct the target detection problem;

[0007] S2, the CUT observation vector z in S1 and the reference unit observation vector z p Convert to the Doppler domain and calculate the Doppler amplitude spectrum, which is denoted as DAS(f d |z) and DAS(f d |z p );

[0008] S3, for DAS(f d |z) is whitened to obtain the whitened spectrum, denoted as WS(f d |z); at the same time, the whitened spectrum WS (f d |z p );

[0009] S4, for the CUT whitening spectrum sequence WS (f d |z) and reference cell WS(f d |z p ), and the sample entropy is calculated and recorded as SE(z) and SE(z) respectively. p );

[0010] S5, SE(z) in S4 and the surrounding P unit sample entropy SE(z p ) means, defined as the relative sample entropy, as the final test statistic ξ;

[0011] S6, at a given false alarm rate P fa Next, the decision threshold γ is obtained by the Monte Carlo test method, and the size of the test statistic ξ in S5 and the decision threshold γ are compared to determine whether there is a target in the CUT:

[0012] If ξ≤γ, it indicates that there is a target in the CUT; if ξ>γ, it indicates that there is no target in the CUT.

[0013] Among them, S1 classifies the target detection problem into the following binary hypothesis testing problem:

[0014]

[0015] Where c represents the sea clutter vector, s represents the target echo vector, the H0 hypothesis indicates that the observation vector contains only sea clutter, and the H1 hypothesis indicates that the observation vector contains the target.

[0016] The essence of detection is to determine which category the CUT observation vector belongs to. The detection problem is regarded as a two-classification problem, where H0 is assumed to be the first category and H1 is assumed to be the second category.

[0017] Furthermore, the formula for calculating the Doppler amplitude spectrum in S2 is as follows:

[0018]

[0019] Among them, f r Indicates the pulse repetition frequency of the radar.

[0020] Furthermore, in S3, the DAS (f d |z p ), the average Doppler spectrum is obtained, and the whitened spectrum WS (f d |z) is:

[0021]

[0022] Further, S4 specifically includes the following steps:

[0023] S4.1 Assume that the whitened spectrum sequence WS(f d |z) length is N, and is represented by {u(i), i=1,2,…,N}. Given the phase space reconstruction dimension m, we get a set of m-dimensional time series

[0024]

[0025] S4.2 When the i-th time series and the jth time series When the distance is greater than the similarity tolerance r, the similarity of the two sequences is 0, otherwise it is 1; thus, the similarity function is obtained:

[0026]

[0027] S4.3 Statistics of the i-th time series Similar probability values ​​to other time series

[0028]

[0029] Traverse all time series The cumulative function is denoted by φ m (r)

[0030]

[0031] S4.4 Considering the internal self-similarity of the sequence, φ is recalculated when m is increased to m+1. m+1 (r), the sample entropy SE is defined as:

[0032]

[0033] Furthermore, S5 is obtained from the CUT sample entropy SE in S4 m,r (WS(f d |z)) and the surrounding reference unit sample entropy SE m ,r (WS(f d |z p )), calculate the relative sample entropy RSE and use it as the test statistic ξ:

[0034]

[0035] Furthermore, S6 specifically includes the following steps:

[0036] S6.1 Under the H0 hypothesis, assume that M sea clutter samples are obtained, calculate the relative sample entropy respectively, and construct the sample set Ω = {ξ1,ξ2,…,ξ M}, sort the M samples from small to large, satisfying ξ1≤ξ2≤…≤ξ M ;

[0037] S6.2 Given a false alarm rate P fa Under this condition, the decision threshold γ is:

[0038]

[0039] Among them, [] means taking an integer;

[0040] S6.3 compares the test statistic ξ and the decision threshold γ to determine whether there is a target in the CUT:

[0041] If ξ≤γ, it means there is a target in the CUT;

[0042] If ξ>γ, it indicates that there is no target in the CUT.

[0043] Beneficial effects: Compared with the prior art, the advantages of the present invention are: defining a whitened spectrum in the frequency domain, increasing the difference in spectrum between sea clutter and echoes containing targets; improving the sample entropy algorithm, extracting relative sample entropy as an efficient single feature, integrating multivariate information of the spectrum in terms of energy and geometry; through this invention, the contradiction between performance improvement and calculation reduction can be solved, and the detection probability of small targets on the sea surface under low signal-to-clutter ratio conditions is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of the method of the present invention;

[0045] Figure 2 The present invention simulates 1 clutter unit and target unit spectrum and whitening spectrum;

[0046] Figure 3 The distribution diagram of the relative sample entropy values ​​of the simulated 2 sea clutter and target echo whitening spectrum sequence of the present invention;

[0047] Figure 4 This is a detection result diagram of the present invention simulation 3 based on the Hurst exponent detector;

[0048] Figure 5 This is a test result diagram of simulation 3 based on the consistency factor detector of the present invention;

[0049] Figure 6 This is a detection result diagram of the sample entropy detector based on simulation 3 of the present invention;

[0050] Figure 7 This is a diagram of the detection results of the detector proposed in simulation 3 of the present invention. DETAILED DESCRIPTION

[0051] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0052] like Figure 1 The method for detecting small sea surface target features based on relative sample entropy includes the following steps:

[0053] Step 1: Get the CUT observation vector

[0054] Assume that the radar receives N consecutive pulses in a range unit. These N pulses constitute an observation vector z = [z(1), z(2), ..., z(N)] T , called the cell under test (CUT). At the same time, the observation vectors z of P reference cells around the CUT are obtained p ,p=1,2,...,P, construct the target detection problem. Then, the target detection problem can be classified into the following binary hypothesis testing problem:

[0055]

[0056] Where c represents the sea clutter vector and s represents the target echo vector. The H0 hypothesis indicates that the observation vector contains only sea clutter; the H1 hypothesis indicates that the observation vector contains the target. The essence of detection is to determine which category the CUT observation vector belongs to. Therefore, the detection problem can be regarded as a two-classification problem, with the H0 hypothesis being the first category and the H1 hypothesis being the second category.

[0057] Step 2: Calculate the Doppler spectrum

[0058] The CUT observation vector z in step 1 and the reference unit observation vector z p , converted to the Doppler domain. Calculate the Doppler amplitude spectrum, the formula is as follows:

[0059]

[0060] Among them, f r Indicates the pulse repetition frequency (PRF) of the radar.

[0061] Step 3: Whiten the spectrum

[0062] In actual detection environments, sea clutter has a local uniformity in space. d |z p ), the average Doppler spectrum is obtained, and the DAS (f d |z) is whitened to obtain the whitened spectrum WS(f d |z)

[0063]

[0064] Step 4: Extract sample entropy

[0065] 4.1) Assume that the whitened spectrum sequence WS(f d |z) length is N, and is represented by {u(i), i=1,2,…,N}. Given the phase space reconstruction dimension m, we get a set of m-dimensional time series

[0066]

[0067] 4.2) When the i-th time series and the jth time series When the distance is greater than the similarity tolerance r, the similarity of the two sequences is 0, otherwise it is 1. Thus, the similarity function is obtained

[0068]

[0069] 4.3) Count the i-th time series Similar probability values ​​to other time series

[0070]

[0071] Then, traverse all time series The cumulative function is denoted by φ m (r)

[0072]

[0073] 4.4) Considering the internal self-similarity of the sequence, we increase m to m+1 and recalculate φ m+1 (r), the sample entropy (SE) is defined as

[0074]

[0075] Step 5: Obtain relative sample entropy

[0076] The CUT sample entropy SE obtained in step 4 m,r (WS(f d |z)) and the surrounding reference unit sample entropy SE m,r (WS(f d |z p )) and calculate the relative sample entropy (RSE) as the test statistic ξ

[0077]

[0078] Step 6: Detection and judgment

[0079] At a given false alarm rate P fa The false alarm controllable decision threshold γ is obtained offline by the Monte Carlo test method.

[0080] 6.1) Under the H0 hypothesis, assume that M sea clutter samples are obtained, calculate the relative sample entropy respectively, and construct the sample set Ω = {ξ1,ξ2,…,ξ M}, sort the M samples from small to large, satisfying ξ1≤ξ2≤…≤ξ M .

[0081] 6.2) Given a false alarm rate P fa Under this condition, the decision threshold γ is

[0082]

[0083] Among them, [] indicates taking an integer.

[0084] 6.3) Compare the test statistic ξ and the decision threshold γ to determine whether there is a target in the CUT:

[0085] If ξ≤γ, it means there is a target in the CUT;

[0086] If ξ>γ, it indicates that there is no target in the CUT.

[0087] The effects of the present invention are further described below in conjunction with the experimental results of measured data.

[0088] This example uses a database collected by the IPIX radar in 1993. In the experiment, the data name is 19931111_163625_starea54. The pulse repetition frequency of the radar is 1000Hz, and the range resolution is 30m. The test target is a small ball with a diameter of 1m.

[0089] Simulation 1, we get Figure 2 The clutter unit and target unit spectrum and whitening spectrum are shown. In which, the pulse accumulation number N is set to 512, and the pulse repetition frequency f r =1000Hz. Figure 2 In (a), the spectrum offset of the clutter unit (H0 assumption) is 100 Hz, the main clutter band occupies approximately [20,180] Hz, and the rest is noise area. Figure 2 In (b), the spectrum of the target unit (H1 hypothesis) consists of a linear superposition of clutter and target, the Doppler shift of the target is 0 Hz, and the energy is concentrated in a narrow band centered at 0 Hz. For the clutter unit, the spectrum sequence presents a "hillside" pattern. For the target unit, the spectrum sequence presents a "hillside + a tree" pattern. Therefore, it is difficult to distinguish the two from the perspective of spectrum confusion. Figure 2 In (c), the main clutter band is completely suppressed, the whitened spectrum of the clutter is randomly distributed in the entire frequency domain, and the entire sequence is chaotic. Figure 2 In (d), the whitened spectrum of the target unit only contains the target spectrum. The whitening process has an impact on the target, but the suppression degree is small, while the sea clutter is completely suppressed. The entire sequence presents a "one tree" pattern with obvious regularity. Therefore, the whitened spectrum sequences under the two types of assumptions have great differences, which is more conducive to the subsequent feature extraction.

[0090] Simulation 2, such as Figure 3 As shown, the distribution of relative sample entropy values ​​of sea clutter and whitened spectrum sequence containing target echo is given. Set the sequence length N = 512, the phase space reconstruction dimension m = 2, and the adjustment factor β = 0.2. It can be seen that sea clutter has a larger relative sample entropy, reflecting the irregularity and randomness of its whitened spectrum sequence, which is determined by the noise-like characteristics of the whitened spectrum of sea clutter. On the contrary, the relative sample entropy of the target echo is smaller, which is determined by the fixed peak pattern of its whitened spectrum. The higher the peak, the higher the SCR, the more obvious the fixed pattern, and the smaller the relative sample entropy. From Figure 3 It can be seen that the relative sample entropy based on the whitened spectrum can better distinguish sea clutter and target echoes.

[0091] Simulation 3 verifies the detection algorithm proposed by the present invention and compares its detection performance with that of other existing detectors. The pulse accumulation number is set to N = 512, that is, the observation time is 0.512s, and the reference unit P = 9. The results are as follows: Figure 4, Figure 5 , Figure 6 , Figure 7 As shown, where: Figure 4 This is the detection result based on the Hurst exponent detector, with a detection probability of 0.72. The results show that: small targets can be basically detected, but the target trajectory appears intermittently; Figure 5 This is the detection result based on the consistency factor detector, with a detection probability of 0.73, and the performance is consistent with the Hurst exponent-based detector. It should be noted that these two detectors often require an accumulation time of more than a few seconds to achieve good performance and are suitable for high SCR situations; Figure 6 The detection result based on the sample entropy detector is 0.47. The present invention directly calculates the sample entropy from the Doppler spectrum for detection, and the performance loss is obvious. Even under high SCR conditions, it is still difficult for the sample entropy to find the difference between sea clutter and target echo from the spectrum, which is caused by the wide main clutter band of sea clutter. Figure 7 The detection result of the proposed detector is 0.91, and it can be clearly observed that the small target is located in unit 8. Compared with the detector based on sample entropy, the performance of the proposed detector is significantly improved. Its performance advantage mainly comes from the whitening spectrum and relative sample entropy. The former realizes the suppression of the main clutter band in the spectrum, and the latter ensures the robustness to the marine environment.

[0092] In summary, it can be seen that under the condition of low signal-to-clutter ratio, the detection performance of the present invention for sea surface targets is better than that of the existing detection methods, and can solve the contradiction between performance improvement and reduction of calculation amount.

Claims

1. A method for detecting small targets on the sea surface based on relative sample entropy. It is characterized in that The following steps are involved: S1. Assume that the radar receives N consecutive pulses in a range unit. These N pulses constitute an observation vector z = [z(1), z(2), ..., z(N)] T , called the unit to be detected CUT; at the same time, the observation vector z of P reference units around CUT is obtained p ,p=1,2,...,P, construct the target detection problem; S2, the CUT observation vector z in S1 and the reference unit observation vector z p Convert to the Doppler domain and calculate the Doppler amplitude spectrum, which is denoted as DAS(f d |z) and DAS(f d |z p ); S3, for DAS(f d |z) is whitened to obtain the whitened spectrum, denoted as WS(f d |z); at the same time, the whitened spectrum WS (f d |z p ); S4, for the CUT whitening spectrum sequence WS (f d |z) and reference cell WS(f d |z p ), and the sample entropy is calculated and recorded as SE(z) and SE(z) respectively. p ); S5, SE(z) in S4 and the surrounding P unit sample entropy SE(z p ) means, defined as the relative sample entropy, as the final test statistic ξ; S6, at a given false alarm rate P fa Next, the decision threshold γ is obtained by the Monte Carlo test method, and the size of the test statistic ξ in S5 and the decision threshold γ are compared to determine whether there is a target in the CUT: If ξ≤γ, it indicates that there is a target in the CUT; if ξ>γ, it indicates that there is no target in the CUT; Among them, S1 classifies the target detection problem into the following binary hypothesis testing problem: Where c represents the sea clutter vector, s represents the target echo vector, and H 0 Assume that the observation vector contains only sea clutter; H 1 Assume that the observation vector contains the target; The essence of detection is to determine which category the CUT observation vector belongs to. The detection problem is regarded as a two-classification problem. 0 Assuming the first type, H 1 Assume the second category; The formula for calculating the Doppler amplitude spectrum in S2 is as follows: Among them, f r Indicates the pulse repetition frequency of the radar; In S3, the DAS (f d |z p ), the average Doppler spectrum is obtained, and the whitened spectrum WS (f d |z) is:

2. The method for detecting small sea surface targets based on relative sample entropy according to claim 1, It is characterized in that S4 specifically includes the following steps: S4.1 Assume that the whitened spectrum sequence WS(f d |z) length is N, and is represented by u(i), i = 1, 2, ..., N. Given the phase space reconstruction dimension m, a set of m-dimensional time series is obtained S4.2 When the i-th time series and the jth time series When the distance is greater than the similarity tolerance r, the similarity of the two sequences is 0, otherwise it is 1; thus, the similarity function is obtained: S4.3 Statistics of the i-th time series Similar probability values ​​to other time series Traverse all time series The cumulative function is denoted by φ m (r) S4.4 Considering the internal self-similarity of the sequence, φ is recalculated when m is increased to m+1. m+1 (r), the sample entropy SE is defined as:

3. The method for detecting small sea surface targets based on relative sample entropy according to claim 2, Features: S5 is the CUT sample entropy SE obtained from S4 m,r (WS(f d |z)) and the surrounding reference unit sample entropy SE m,r (WS(f d |z p )), calculate the relative sample entropy RSE and use it as the test statistic ξ:

4. The method for detecting small sea surface targets based on relative sample entropy according to claim 3, Features: S6 specifically includes the following steps: S6.1 in H 0 Assume that M sea clutter samples are obtained, calculate the relative sample entropy respectively, and construct the sample set Ω={ξ 1 ,ξ 2 ,…,ξ M }, sort the M samples from small to large to satisfy ξ 1 ≤ξ 2 ≤…≤ξ M ; S6.2 Given a false alarm rate P fa Under this condition, the decision threshold γ is: Among them, [] means taking an integer; S6.3 compares the test statistic ξ and the decision threshold γ to determine whether there is a target in the CUT: If ξ≤γ, it means there is a target in the CUT; If ξ>γ, it indicates that there is no target in the CUT.