A method for recognizing floating targets using centroid temporal information

Abstract

This invention relates to the field of radar signal processing technology, and mainly to a method for identifying floating targets and ships on the sea surface. This invention utilizes the temporal evolution of the centroid of the floating target's Doppler spectrum, improving the utilization rate of target echo information and solving the problem that previous methods relying solely on one-dimensional range profile features of the target could not achieve refined identification. This invention proposes a dual-threshold method to remove outliers from the centroid sequence; this method is applicable to general outlier removal scenarios. With a second-level observation time, the method proposed in this invention can achieve an accuracy of over 90% in identifying targets with significant mass differences, such as ships and common floating targets.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, and mainly to a method for identifying floating targets on the sea surface. Background Technology

[0002] Under the premise that maritime targets can be effectively detected by maritime surveillance radar, accurate identification and classification of them to determine their specific categories, uses, and threat levels is of great significance in both civilian and military fields. Common floating targets on the sea surface mainly include small fishing boats, buoys, and ice floes, which generally have different structures and physical dimensions from large cruise ships, cargo ships, and ro-ro cargo ships. High-resolution radar can obtain one-dimensional range profile features of targets, which reflect the distribution of strong scattering points along the distance. This is an important target characteristic and the main basis for distinguishing floating targets from large ships. However, one-dimensional range profile features are closely related to the observation angle, especially when the radar beam illuminates along the side of the ship. Ship targets and floating targets have similar one-dimensional range profile features, which seriously affects the identification performance. In addition, for low-resolution radar systems, it is particularly difficult to identify floating targets without relying on auxiliary information because one-dimensional range profile features cannot be obtained.

[0003] Starting from the physical mechanisms of ship targets and floating targets, and delving into the differences between them, is the fundamental way to solve the problems faced by the one-dimensional range image method in identifying floating targets. Obviously, ship targets generally have a larger density and mass, while floating targets generally have a smaller density and mass. Therefore, under normal sea conditions, ship targets are usually in a relatively stable state and rarely fluctuate violently with the waves, while floating targets often fluctuate up and down with the waves.

[0004] Therefore, ship targets and floating targets have different wave characteristics, so this invention aims to use the Doppler information of the target radar echo to distinguish between the two. Summary of the Invention

[0005] The purpose of this invention is to provide a floating target identification method that utilizes Doppler spectrum centroid timing information, particularly for floating targets on the sea surface and ship targets that cannot be distinguished by one-dimensional range profiles due to the inability of low-resolution radar to obtain one-dimensional range profiles, or targets with similar physical sizes.

[0006] A floating target recognition method utilizing centroid temporal information, characterized by the following steps:

[0007] Step 1: Extraction of Doppler spectral centroid sequence

[0008] A marine radar detects targets on the sea surface. A coherent pulse train x(n), n = 1, 2, ..., N, is received at a target range cell. x(n), n = 1, 2, ..., N is truncated into short, non-overlapping vectors u of length I. iAs shown below:

[0009] [x(1),x(1),…,x(N)] T =[u1,u2,…,u N / I ] T (1)

[0010] Calculate each short vector u separately i The Doppler spectrum is obtained, and the centroid features of the Doppler spectrum are extracted. This invention uses the power spectrum as the Doppler spectrum and employs nonparametric methods from modern signal spectrum analysis to estimate the average and short-time Doppler spectra. Windowing is used during estimation to effectively reduce sidelobes.

[0011] The Doppler spectrum estimation method is as follows: Assuming the length of a single data segment is L, for the φ-th (φ=1,2,…,Φ) data segment, the short-time Doppler spectrum estimate is expressed as:

[0012]

[0013]

[0014]

[0015] in, and Let ω(n) represent the short-time amplitude spectrum and short-time power spectrum, respectively; let T represent the pulse repetition interval (PRI); and let c represent the pulse repetition interval (PRI). (φ) (n), n=1,2,…,L represents the radar echo sequence of the φ-th data segment, and Δ is the window function power, expressed as:

[0016]

[0017] By averaging the short-time Doppler spectrum along the frequency dimension, the average Doppler spectrum estimate can be obtained, expressed as:

[0018]

[0019] Doppler center of mass f C The Doppler shift of the target is described, and the estimation method is as follows:

[0020]

[0021] Where Q(φ) represents the power level of the short-time Doppler spectrum of the φ-th data segment, that is:

[0022]

[0023] The centroid sequence of the Doppler spectrum of a certain sea target finally obtained is denoted as X1, X2, …, X N / I .

[0024] Step 2, Outlier correction of the centroid sequence

[0025] Due to random interference such as co-frequency interference, noise, and electromagnetic waves of unknown origin, the centroid sequence of the Doppler spectrum extracted in Step 1 often has outliers. The present invention proposes a double-threshold method to correct the centroid sequence of the Doppler spectrum extracted in Step 1 and reduce the influence of outliers. The double-threshold method is used to process the centroid sequence X1, X2, …, X of the Doppler spectrum of a certain sea target as follows: N / I The processing flow is as follows:

[0026] 1) Initialization: According to the general change trend of the centroid sequence, set two decision thresholds standard1 and standard2. When the dimension of the centroid is the normalized frequency, the value range of standard1 is between 0.03 and 0.15, and the range of standard2 is between 0.05 and 0.3. For example, let standard1 = i+1 0.03 and standard2 = i-1 0.05.

[0027] 2) Processing of X1: Judge whether |X1 - mean([X2, X3])| is less than standard1, where mean(·) represents the mean operation and || represents taking the absolute value. If |X1 - mean([X2, X3])| < standard1, then X1 is a normal value; otherwise, X1 is an outlier, and let X1 = mean([X2, X3]).​​​​​​​​​​​​​​​​​​​​​​​​​​​​i = mean([X i-1 , X i+1 ), otherwise X i = mean([X i-2 , X i-1 ).

[0030] 5) Processing of X N / I : Determine whether |X N / I - X N / I-1 | is less than standard1. If |X N / I - X N / I-1 | < standard1, then X N / I is a normal value, otherwise X N / I is an abnormal value. Let X N / I = mean([X N / I-1 , X N / I-2 ).

[0031] Step 3, AR modeling of the centroid sequence and secondary feature extraction

[0032] Use a centered p-order autoregressive (AR) model to fit the centroid sequence after removing outliers for all targets. The fitting results of the centered AR(p) model for the centroid sequence X1, X2,..., X N / I after removing outliers are as follows:

[0033] X t = c1X t-1 +... + c p X t-p + ε t (9)

[0034] In the formula, the model order p is a known quantity, and the model parameters c = (c1, c2,..., c p ) T are parameters to be estimated, and their moment estimation values can be calculated by Equation (10).

[0035]

[0036] In the formula, is the estimated value of the autocorrelation coefficient. Equation (10) is called the Yule-Walker equation estimation method for parameters. The moment estimation value of the white noise variance is as follows:

[0037]

[0038] In the formula, is the variance of the centroid feature sequence, Let be the autocovariance function of interval j.

[0039] Equation (9) is the AR model fitting result for the centroid sequence of a certain sea surface target. The AR model order p is generally set to 3, and the estimated AR model coefficients are then... These are denoted as AR(1), AR(2), and AR(3), which are the secondary features corresponding to the centroid sequence of a certain sea surface target.

[0040] Step 4, Classifier Construction and Recognition

[0041] During the classifier training phase, a large number of secondary features AR(1), AR(2), and AR(3) of the Doppler spectral centroid are extracted from the labeled ship target echoes to form the required training set feature samples. Standard deviation normalization is then performed on these features. A ship target error probability is preset, and the fast convex hull learning algorithm is used for training to obtain the decision region, thus completing the classifier training. During the classifier recognition phase, the target echo samples [AR(1), AR(2), AR(3)] acquired by the radar are input into the already trained convex hull single classifier to identify the target category.

[0042] Compared with existing technologies, the method for identifying floating targets on the sea surface using centroid time-series information described in this technical solution has the following advantages:

[0043] (1) The method proposed in this invention utilizes the temporal evolution law of the centroid of the Doppler spectrum of the floating target, improves the utilization rate of the target echo information, and solves the problem that the previous method could not accurately identify the target by relying solely on the one-dimensional range image features of the target.

[0044] (2) The present invention proposes a dual threshold method for removing outliers in centroid sequences. This method is applicable to general outlier removal scenarios.

[0045] (3) In terms of observation time in seconds, the method proposed in this patent can achieve an accuracy of over 90% in identifying targets with large mass differences, such as ship targets and common floating targets. Attached Figure Description

[0046] Appendix Figure 1 This is a flowchart illustrating the implementation of the present invention;

[0047] Appendix Figure 2 This is a typical Doppler spectral centroid sequence of ship targets and floating targets proposed in this invention;

[0048] Appendix Figure 3 These are typical Doppler spectral centroid sequences of ship targets and floating targets before and after outlier removal proposed in this invention.

[0049] Appendix Figure 4 This refers to the distribution of ship targets and floating targets in the AR model coefficient domain of the centroid sequence, as proposed in this invention.

[0050] Appendix Figure 5 This is a schematic diagram of the training and decision-making process for a convex hull classifier. Detailed Implementation

[0051] To better understand and implement this invention, specific embodiments are provided below with reference to the accompanying drawings to illustrate a floating target identification method utilizing Doppler spectral centroid timing information.

[0052] The specific implementation process of this target recognition method is attached. Figure 1 As shown in the accompanying drawings, the processing flow of the present invention will be described in detail below:

[0053] 1) Extraction of Doppler spectral centroid sequence

[0054] Suppose a maritime radar is detecting a target on the sea surface. A coherent pulse train x(n), n = 1, 2, ..., N, is received at a target range cell. x(n), n = 1, 2, ..., N is then truncated into a short, non-overlapping vector u of length I according to equation (1). i .

[0055] Calculate each short vector u according to equations (2) to (8). i The Doppler spectrum of a certain sea surface target was obtained, and the centroid features of the Doppler spectrum were extracted. The resulting Doppler spectrum centroid sequence of a certain sea surface target was denoted as X1, X2, ..., X NI The Doppler centroid sequences of typical ship targets and floating targets (buoys) obtained from measured data are attached. Figure 2 As shown, the Doppler centroid sequences of ship targets and floating targets exhibit significantly different variation patterns. The centroid sequence of ship targets is generally concentrated around 0Hz and is relatively stable with small fluctuations; the centroid sequence of floating targets shows obvious fluctuations and is approximately periodic. Therefore, the temporal evolution pattern of the centroid sequence can be used to distinguish between ship targets and floating targets.

[0056] 2) Outlier correction for centroid sequences

[0057] During radar operation, it is often affected by random interference such as co-channel interference, noise, and electromagnetic waves from unknown sources, which often leads to outliers in the Doppler spectrum centroid sequence extracted in step 1). This patent proposes a dual-threshold method to correct the Doppler spectrum centroid sequence extracted in step 1), making the Doppler spectrum centroid sequence smoother and eliminating the influence of outliers. The dual-threshold method is used to analyze the Doppler spectrum centroid sequence X1, X2, ..., X of a sea surface target. N / I The processing procedure is as follows:

[0058] 1) Initialization: Set two decision thresholds, standard1 and standard2, according to the general trend of the centroid sequence. When the dimension of the centroid is the normalized frequency, the value range of standard1 is between 0.03 and 0.15, and the range of standard2 is between 0.05 and 0.3. For example, let standard1 = 0.03 and standard2 = 0.05.

[0059] 2) Processing of X1: Determine whether |X1 - mean([X2, X3])| is less than standard1, where mean(·) represents the mean operation and || represents taking the absolute value. If |X1 - mean([X2, X3])| < standard1, then X1 is a normal value; otherwise, X1 is an abnormal value, and let X1 = mean([X2, X3]).

[0060] 3) Processing of X2: Determine whether |X2 - X1| is less than standard1. If |X2 - X1| < standard1, then X2 is a normal value; otherwise, X2 is an abnormal value, and let X2 = mean([X1, X3]).

[0061] 4) Processing of X3, X4, …, X N / I-1 : Determine whether |X i - X i-1 | is less than standard1, where i ∈ {3, 4, …, N / I - 1}. If |X i - X i-1 | < standard1, then X<00所0054> is a normal value; otherwise, X i is an abnormal value, and it is necessary to further determine whether |X i-1 - X i+1 | is less than standard2. If |X i-1 - X i+1 | < standard2, then it is a non - continuous abnormal value, and let X i = mean([X i-1 , X i+1 ); otherwise, X i = mean([X i-2 , X i-1 ).

[0062] 5) Processing of X N / I : Determine whether |X N / I - X N / I-1 | is less than standard1. If |X N / I - X N / I-1 | < standard1, then X N / I is a normal value; otherwise, X N / ILet X be an outlier. N / I =mean([X N / I-1 ,X N / I-2 ]).

[0063] The above procedure can also be used to correct outliers in other common feature sequences. The Doppler centroid sequences of typical ship targets and floating targets (buoys) before and after outlier correction obtained from measured data are attached. Figure 3 As shown. Clearly, the dual-threshold method proposed in this patent can achieve excellent outlier smoothing and solve the problem of correcting continuous outliers.

[0064] 3) AR modeling and secondary feature extraction of centroid sequences

[0065] The centroid sequence X1, X2, ..., X is corrected for outliers using a centered AR(p) model. N / I The fitting results are shown in Equation (9), and the model parameters are calculated using the Yul-Walker equation estimation method introduced in Equation (10).

[0066] The order p of the AR model is generally set to 3, then the estimated AR model coefficients are... Generally denoted as AR(1), AR(2), AR(3), these are the secondary features corresponding to the centroid sequence of a certain sea surface target.

[0067] The typical distributions of ship targets and floating targets (buoys) obtained from measured data in the AR model coefficient domain of the centroid sequence are shown in the appendix. Figure 4 As shown. Obviously, ship targets and floating targets have good separability in the three-dimensional feature space composed of AR(1), AR(2), and AR(3).

[0068] 4) Classifier construction and recognition

[0069] After extracting the secondary features AR(1), AR(2), and AR(3) corresponding to the Doppler spectral centroid sequences of different target echoes, the identification problem becomes a classification problem in a 3D feature space. This invention places this problem within an anomaly detection framework and designs a single-class classifier using the convex hull learning algorithm to achieve the identification of ship targets and floating targets.

[0070] A single-class classifier can be trained using only samples from one class of data, that is, based solely on the set of labeled ship target echo feature vectors S, it can be trained in a 3D feature space according to a preset ship target error probability P. fBy progressively training, the convex hull of the ship target echo is obtained, effectively solving the problem of imbalance between the two types of samples. According to the Minimal-Volume Criterion, the standard deviation of the set S needs to be normalized beforehand to avoid excessive variation in the range of one feature value affecting the recognition of other features. Assume the training set is represented as:

[0071] S0=[AR(1)0,AR(2)0,AR(3)0] (12)

[0072] In the formula, AR(1)0, AR(2)0, and AR(3)0 are column vectors composed of the second-order features AR(1), AR(2), and AR(3) values ​​of the centroid sequence of a large number of ship target units, respectively, and the length of the vector is denoted as Q. The standard deviation of each feature component can be estimated by the following formula:

[0073]

[0074] In the formula, AR(1) q AR(2) q AR(3) q Let represent the q-th eigenvalues ​​of the eigenvectors AR(1)0, AR(2)0, and AR(3)0, respectively, and mean(·) represent the mean operator. Then the normalized sample set can be represented as:

[0075]

[0076] In the formula, AR(1), AR(2), and AR(3) represent column vectors composed of the corresponding normalized features.

[0077] Clearly, ship target samples and floating target samples should cluster in different regions within the 3D feature space. Therefore, during training, feature points within the convex hull formed by ship target samples that are closer to the clustering region of floating target samples are more likely to become outliers and be removed. Thus, the process of forming the classifier's decision region is as follows:

[0078] 1) Initialization: Let W be the number of ship target echo feature vectors, and calculate the number of ship target anomalies F. num =W·P f , where P f Let l = 0, representing the preset probability of incorrect ship target identification.

[0079] 2) Find the maximum value of AR(1) feature and the minimum values ​​of AR(2) and AR(3) feature in set S to form a new spatial vertex v0 = [max(AR(1)), min(AR(2)), min(AR(3))].

[0080] 3) Generate the convex hull CH(S) of the current data point, with vertices {v1, v2, ..., v...} r}. Count the number of feature points falling into the convex hull CH(S), let it be n. all .

[0081] 4) Calculate the Euclidean distance from all feature points to vertex v0, and find the feature point v that is farthest from v0. i And remove;

[0082] 5) Generate a new convex hull CH(S-{v i Then calculate the number of feature points in the new convex hull, let it be n. q .

[0083] 6) Let S-{v i}=S,l+n all -n q =l.

[0084] 7) If l <F num If the process ends, return to step 2) and continue removing the next vertex. Otherwise, terminate the removal process and output the final decision region Ω = CH(S).

[0085] After the convex hull single classifier is trained, the target echo samples [AR(1), AR(2), AR(3)] are input to complete the recognition. The recognition rules are as follows:

[0086]

[0087] The diagrams illustrating the training and decision processes of the convex hull classifier obtained from the measured data are attached. Figure 5 As shown, the centroid secondary features of the ship target echo are used as samples to train a convex hull single classifier, with a preset ship target misclassification probability P. f The accuracy rate is 0%, and the correct identification probability of floating targets (buoys) is 98.09%.

Claims

1. A method for identifying floating targets using centroid temporal information, characterized in that... Includes the following steps: Step 1: Extraction of Doppler spectral centroid sequence A marine radar detects targets on the sea surface. A target range cell receives a phase signal of length N. Given a pulse train x(n), n = 1, 2, ..., N, cut x(n), n = 1, 2, ..., N into short, non-overlapping vectors u of length I. i As shown below: [x(1),x(1),…,x(N)] T =[u1,u2,…,u NI ] T (1) Calculate each short vector u separately i The Doppler spectrum was obtained, and the centroid features of the Doppler spectrum were extracted. The power spectrum was used as the Doppler spectrum. The average and short-time Doppler spectra were estimated using nonparametric methods in modern signal spectrum analysis. Windowing was used during the estimation to effectively reduce sidelobes. The Doppler spectrum estimation method is as follows: Assuming the length of a single data segment is L, for the φ-th (φ=1,2,…,Φ) data segment, the short-time Doppler spectrum estimate is expressed as: in, and Let ω(n) represent the short-time amplitude spectrum and the short-time power spectrum, respectively. Time-domain window function, where T represents the pulse repetition period, c (φ) (n), n=1,2,…,L represents the radar of the φ-th data segment. The echo sequence, where Δ is the window function power, is expressed as: By averaging the short-time Doppler spectrum along the frequency dimension, the average Doppler spectrum estimate can be obtained, expressed as: Doppler center of mass f C The Doppler shift of the target is described, and the estimation method is as follows: Where Q(φ) represents the power level of the short-time Doppler spectrum of the φ-th data segment, that is: The final Doppler spectrum centroid sequence of a certain sea surface target is denoted as X1, X2, ..., X N / I ; Step 2, outlier correction of the centroid sequence Due to random interference such as co-frequency interference, noise, and electromagnetic waves from unknown sources, outliers often appear in the Doppler spectrum centroid sequence extracted in step 1. A dual threshold method is proposed to correct the Doppler spectrum centroid sequence extracted in step 1 and reduce the impact of outliers. Step 3: AR modeling and secondary feature extraction of centroid sequences The centroid sequence after outlier removal for all targets is fitted using a centered p-order auto-regressive (AR) model. The centroid sequence after outlier removal is X1, X2, ..., X... N / I The fitting results of the centered AR(p) model are as follows: X t =c1X t-1 +…+c p X t-p +ε t (9) In the formula, the model order p is a known quantity, and the model parameters c = (c1, c2, ..., c...). p ) T For the parameter to be estimated, its moment estimate is... It can be calculated using equation (10); In the formula, ρ is the estimated value of the autocorrelation coefficient, and equation (10) is called the Yul-Walker equation estimation method for the parameters, and the white noise variance The moment estimate as follows: In the formula, γ0 is the variance of the centroid feature sequence, γ j Let j be the autocovariance function of the interval j; Equation (9) is the AR model fitting result for the centroid sequence of a certain sea surface target. The AR model order p is generally set to 3, and the estimated AR model coefficients are then obtained as follows: These are denoted as AR(1), AR(2), and AR(3), which are the secondary features corresponding to the centroid sequence of a certain sea surface target. Step 4, Classifier Construction and Recognition During the classifier training phase, a large number of secondary features AR(1), AR(2), and AR(3) of the Doppler spectral centroid are extracted from the labeled ship target echoes to form the required training set feature samples, and their standard deviation is normalized. A ship target error decision probability is preset, and the fast convex hull learning algorithm is used for training to obtain the decision region and complete the classifier training. During the classifier recognition phase, the target echo samples [AR(1), AR(2), AR(3)] collected by the radar are input into the trained convex hull single classifier to identify the target category.

2. The floating target identification method using centroid temporal information according to claim 1, characterized in that... The dual-threshold method described in step 2 is used to analyze the centroid sequence X1, X2, ..., X of the Doppler spectrum of a sea surface target. N / I The processing procedure is as follows: 1) Initialization: Based on the general trend of the centroid sequence, set two decision thresholds, standard1 and standard2. When the dimension of the centroid is the normalized frequency, the value of standard1 is between 0.03 and 0.15, and the value of standard2 is between 0.05 and 0.

3. For example, set standard1 = 0.03 and standard2 = 0.

05. 2) Processing X1: Determine whether |X1-mean([X2,X3])| is less than standard1, where mean() represents the mean operation and || represents taking the absolute value. If |X1-mean([X2,X3])| < standard1, then X1 is a normal value; otherwise, X1 is an outlier. Let X1 = mean([X2,X3]). 3) Processing X2: Determine if |X2-X1| is less than standard1. If |X2-X1| < standard1, then X2 is a normal value; otherwise, X2 is an outlier. Let X2 = mean([X1,X3]). 4) X3, X4, ..., X N / I-1 Processing: Determine |X i -X i-1 Is |X less than standard1, where i∈{3,4,…,N / I-1}, if |X i -X i-1 |<standard1, then X i If it is a normal value, otherwise X i This is an outlier and needs to be further evaluated for |X. i-1 -X i+1 Is |X| less than standard2? i-1 -X i+1 If | < standard2, then it is a discontinuous outlier. Let X i =mean([X i-1 ,X i+1 ]), otherwise X i =mean([X i-2 ,X i-1 ]); 5)X N / I Processing: Determine |X N / I -X N / I-1 Is |X| less than standard1? N / I -X N / I-1 |<standard1, then X N / I If it is a normal value, otherwise X N / I Let X be an outlier. N / I =mean([X N / I-1 ,X N / I-2 ]).

Images