A stable tracking method of radar targets in complex sea environment based on multi-feature correlation

Abstract

The present invention relates to a radar target stable tracking method in a complex sea environment based on multi-feature association, including: using the tracking result at the previous moment to estimate the position of the target at the current moment, and retaining the four types of features of the target at the previous moment, namely, time domain-fractal, frequency domain-Doppler entropy, energy domain-relative average amplitude and modulation domain-time-frequency ridge as feature templates, extracting the above features for all scattering points in the suspicious target area at the current moment, adjusting the feature matrix considering the difference in the ability of each feature to distinguish between the target and clutter, and reducing the dimension of the feature matrix based on the amplitude characteristics of the scattering points, and selecting the target being tracked at the current moment by comparing the correlation distance between the feature matrix after dimensionality reduction and the feature template. In order to ensure the effectiveness of the target feature template, the present invention introduces a target determination area, and realizes the effective update of the target feature template by judging the relationship between the correlation distance and the target determination area, thereby further improving the stability of target tracking.

Description

Technical Field

[0001] The invention belongs to the field of radar target tracking, and in particular relates to a radar target stable tracking method in a complex sea environment based on multi-feature association. Background Art

[0002] When facing dynamic and complex scenes at sea, radar often has insufficient tracking capabilities due to natural environmental interference such as strong clutter and the small size of the target itself. When the target switches from detection state to tracking state, the radar will focus the high-power beam on the target, which will greatly improve the resolution of the target echo. Improving the target resolution can obtain richer detailed information, which is conducive to improving the tracking performance of targets in complex sea environments.

[0003] The current tracking algorithm is mainly based on the energy information of the target, and selects the point with the strongest energy between a certain distance unit and Doppler channel as the target measurement. However, when there is strong sea clutter around the target, especially when there are swells, the target energy may be weaker than the clutter. Selecting the strongest point will reduce the tracking accuracy due to the wrong measurement. Therefore, the idea of ​​tracking while detecting is introduced, and the target tracking algorithm based on multi-feature association is used. With the help of feature information from other domains, it is analyzed from multiple feature dimensions whether the target is the target tracked at the last moment, so as to obtain correct measurement and ensure stable tracking. At present, through feature screening, for floating targets on the sea surface, we have selected 4 typical features from the time domain, frequency domain, energy domain and modulation domain, namely time domain-fractal, frequency domain-Doppler entropy, energy domain-relative average amplitude and modulation domain-time-frequency ridge. The above features have obvious differences in sea clutter and sea surface targets, and can effectively distinguish clutter from targets.

[0004] Although the multi-feature values ​​extracted at different times are different, the separability and relative strength of the features are universal. Therefore, the tracking algorithm is used to estimate the target's position at the next moment, and multiple features are used to make judgments around the predicted position to improve tracking accuracy.

[0005] There are also some shortcomings in the use of multi-feature tracking algorithms: (1) In the target recognition stage, SVM classifiers are often used to determine the category to which suspicious areas belong. However, different bandwidths are used in the recognition and tracking stages. The features of the same target are different under different bandwidths. The larger the bandwidth, the finer the echo and the greater the difference in feature values ​​at different times. The SVM classifier used in the recognition stage cannot be directly used in the tracking stage, and the SVM classifier method is difficult to cover various feature changes under the tracking state; (2) Under large bandwidth, the target shows scalability and attitude sensitivity. A single target contains multiple scattering points. As the target moves and the waves rise and fall, the position and number of scattering points will change, and stable tracking cannot be achieved by relying on a certain point; (3) Different features have different abilities to distinguish targets from sea clutter, and the identification ability of each feature needs to be comprehensively considered. Therefore, there is still a lot of room for improvement in how to use multiple features to effectively enhance target tracking performance.

[0006] In the above background, due to the complex sea surface environment, there are two problems in multi-feature target tracking technology. The first is that the target and clutter scattering points are prone to flicker, the number of scattering points of the same target at different times is inconsistent, and the extracted features do not come from the same scattering points, which makes it difficult to effectively match the features extracted at previous and subsequent times; the second is that different features have different degrees of distinction between clutter and targets. Using the same measurement standard may cause features with low differences to weaken features with high differences, and the ability to distinguish between targets and clutter is not obvious, and ultimately the wrong measurement is selected. Summary of the invention

[0007] In order to solve the above technical problems, the present invention proposes a radar target stable tracking method in a complex sea environment based on multi-feature association.

[0008] The technical solution adopted by the present invention is as follows:

[0009] A radar target stable tracking method based on multi-feature association in a complex sea environment aims at the problem of unstable tracking caused by strong sea clutter or small weak targets in a complex sea environment. It uses multi-features to construct a feature template and compare the feature matrix of the suspicious target area with similar distances, and selects the target corresponding to the closest feature matrix for tracking. In the tracking task, the present invention mainly includes the following steps:

[0010] The tracking result at time k-1 is used in combination with the filter to predict the position and state of the target at time k. At the same time, the feature matrix judged as the target at time k-1 is used as the feature template at time k.

[0011] Using the echo data at time k, low-threshold CFAR detection is carried out around the estimated position area to obtain N suspicious target areas.

[0012] The suspicious target area is incoherently accumulated to obtain a one-dimensional range image, which reduces the target's posture sensitivity. The threshold is calculated using the amplitude of the surrounding sea area, and the peak point on the one-dimensional range image that exceeds the threshold is taken as the scattering point to obtain the scattering point of each suspicious target area. The calculation formula for the scattering point threshold is:

[0013] y=max(x(n))+3std(x(n))

[0014] Where x(n) is the sea clutter around the target at time k-1.

[0015] After performing max-min normalization on the one-dimensional range image of each suspicious target area, multiple features corresponding to the distance channel of each scattering point are extracted to obtain the (C*R) of N suspicious target areas. i Feature matrix, i is the i-th suspicious target area, C is the type corresponding to the feature, R is the scattering point distance channel corresponding to each suspicious target area, and the R of each matrix is ​​different. Furthermore, this method extracts four features: time domain-fractal, frequency domain-Doppler entropy, energy domain-relative average amplitude, and modulation domain-time-frequency ridge, that is, C=4.

[0016] The echo data at time k-1 is used to calculate the difference between each feature. First, the difference between each feature target and the clutter is obtained. The formula is as follows:

[0017]

[0018] Among them, M is the number of scattering points of the target at time k-1, j is the jth feature, is the jth eigenvalue corresponding to the i-th scattering point of the target at time k-1, is the average value of the jth feature in the sea clutter region at time k-1.

[0019] Then calculate the feature difference of each feature, the formula is as follows:

[0020]

[0021] The extracted features are combined with the corresponding feature differences to ensure the consistency of the distinguishing contribution of each element in the matrix.

[0022] Carry out matrix feature dimensionality reduction based on scattering point intensity. The scattering point intensity reflects the extent of the impact of sea clutter at that location to a certain extent. Calculate the probability of scattering intensity proportion

[0023]

[0024] Among them, r is the distance channel of each scattering point, i is the i-th suspicious target area, P r,i is the energy intensity corresponding to the rth scattering point in the i-th suspicious target area.

[0025] The eigenvalue corresponding to each scattering point is weighted by the probability of scattering intensity proportion, and the feature matrix is ​​reduced in dimension (C*R) i →(C*1) i .

[0026] Find the minimum correlation distance between the scattering matrix of each suspicious area and the feature template, and the correlation coefficient is

[0027]

[0028] The correlation distance is dist = 1-ρ XY , the value range is [0,2], which satisfies non-negativity. The smaller the correlation distance is, the more similar the two features are, and vice versa. Therefore, the suspicious target area corresponding to the minimum correlation distance is selected as the tracking target.

[0029] In the first three times, the feature matrix with the smallest correlation distance is directly used as the feature template. From the fourth time, the current correlation distance is compared with the target judgment area calculated by the previous N times. The target judgment area is:

[0030]

[0031] where dist N It is the set of relevant distances between the target and the feature template selected for the previous N times. When the relevant distance is not in the target determination area, the feature template is not updated. When the relevant distance is in the target determination area, the feature template is updated.

[0032] As mentioned above, in view of the influence of feature difference and value range on the tracking process, this scheme first performs max-min normalization on the data to be extracted, so that the feature magnitude is maintained in the same range as much as possible, and the selection result will not be biased towards features with larger magnitude due to the magnitude difference of the feature value; at the same time, feature difference is introduced when constructing the feature matrix, so that the contribution of each value in the matrix to the entire judgment of the target and clutter difference is relatively consistent, further improving the effectiveness of using the correlation distance as the target judgment.

[0033] In view of the problem of attitude sensitivity and flickering of scattering points, it is easy to produce the phenomenon that the position of the scattering points corresponding to the target is not fixed, and the number of scattering points may be inconsistent before and after. Previous methods often use the strongest point feature as the representative feature of the target, but the strongest point may correspond to different positions at different times, resulting in large differences in the extracted features themselves; some methods also select an area larger than the target length to extract the corresponding features, but due to the introduction of a large amount of sea clutter, the results have certain deviations; for the phenomenon that the number of scattering points is inconsistent before and after, some methods directly use PCA to force dimensionality reduction, but this method assumes that there is no difference between the scattering points. On the basis of retaining the selection of all scattering point features, this scheme considers that the intensity of each scattering point can reflect the prominence of the target at that location and the degree of interference by sea clutter, introduces the probability of the proportion of scattering point intensity, and combines multiple scattering point features into one by combining their scattering intensity proportion probability weighted. No matter how the number of scattering points changes before and after, the dimension of the feature matrix for comparison is not affected.

[0034] The feature template needs to be updated regularly to ensure that the feature template at the current moment is the most suitable for the target state. In the past, algorithms often used the original template all the time, or updated the feature template immediately after each processing. If the original template is used all the time, it is impossible to accurately describe the feature changes in the target's motion state due to the mobility of the target; if the feature template is updated at every moment, once a wrong choice is made at a certain moment, it is likely to cause the wrong target to be selected in subsequent tracking. In order to use the feature template to describe the target state relatively accurately and prevent the selection of clutter features as feature templates due to mismatching at a certain moment without setting limits, the matching distance is considered each time and the target interval is set. Only when the matching distance falls within the target interval will the feature template be updated to obtain the optimal tracking result.

[0035] From the above description, it can be seen that the present invention aims at the problems existing in the prior art and proposes a radar target stable tracking method in a complex sea environment based on multi-feature association. The current position of the target is estimated at the previous moment, and the four types of features of the target, namely, time domain-fractal, frequency domain-Doppler entropy, energy domain-relative average amplitude and modulation domain-time-frequency ridge, are retained at the previous moment as the feature template at the current moment. The above-mentioned feature extraction is carried out on all scattering points in the suspicious area at the current moment to form a feature matrix, and the feature matrix is ​​adjusted in combination with the difference in the distinguishing ability of each feature, and the feature matrix is ​​reduced in dimension based on the amplitude characteristics of the scattering points. The target at the current moment is selected for tracking by comparing the correlation distance between the reduced feature matrix and the feature template. In order to ensure the stability of the target feature template, the target judgment area is introduced to update the feature template, so as to obtain a more stable tracking effect. The experimental results show that the algorithm proposed in this paper flexibly uses the feature correlation of adjacent time data and the difference of different features on the same comparison object to achieve stable target tracking under conditions such as strong clutter and small targets, and performs well on multiple test data, and has a more obvious effect in tracking high sea conditions or small targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of a method for stable tracking of radar targets in a complex sea environment based on multi-feature association according to an embodiment of the present invention.

[0037] Figure 2 This is a diagram showing the effect of target-clutter differentiation in selecting features in an embodiment of the present invention.

[0038] Figure 3 This is a scattering point distribution diagram of the same target at different times according to an embodiment of the present invention.

[0039] Figure 4 It is a comparison diagram of the results of the embodiment of the present invention under traditional energy-based tracking and multi-feature association-based tracking. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0041] The main idea of ​​the radar target stable tracking method in complex sea environment based on multi-feature association is: on the basis of traditional tracking algorithm methods, such as using the strongest energy point as the tracking point, multiple features, feature differences and scattering point intensity ratio probability are integrated to prevent target tracking errors caused by strong clutter environment or the target's own low energy and easy loss. By setting the target area at the relevant distance to achieve effective update of the feature template, the stability of target tracking is further improved.

[0042] Please see Figure 1 The present invention provides a radar target stable tracking method in a complex sea environment based on multi-feature association. The processing flow can be described as follows: based on the target determined at the previous moment, the target position at the current moment is predicted, the feature matrix of the selected target is retained as a feature template, and the difference of each eigenvalue at the previous moment is calculated. The low threshold CFAR algorithm is used to obtain the suspicious target area around the target prediction position of the echo data at the current moment. After calculating the scattering points of each suspicious target area, the four features of time domain-fractal, frequency domain-Doppler entropy, energy domain-relative average amplitude and modulation domain-time-frequency ridge are used in sequence to extract the features of each scattering point in the suspicious target area. After all four features are processed, each eigenvalue is trimmed in combination with the feature difference to obtain the feature matrix of each suspicious target area. In order to reduce the dimension of the feature matrix, the probability of the scattering point intensity ratio of each scattering point is calculated, and the processing results of each feature are superimposed in a weighted manner to reduce the multi-dimensional matrix to a one-dimensional matrix. The relationship between the feature matrix of each current suspicious target area and the feature template at the previous moment is calculated using the correlation distance, and the target corresponding to the feature matrix with the smallest correlation distance is selected as the tracking result. In order to further improve the tracking stability, the target area is calculated using the relevant distance of the selected target at each moment, and the feature template is selectively updated so that the tracked target is always the expected target. The specific steps include:

[0043] Step 1. Use the tracking results at time k-1 and combine the filter to predict the position and state of the target at time k. At the same time, use the feature matrix of the selected target at time k-1 as the feature template at time k.

[0044] Step 2. Using the echo data at time k, perform low-threshold CFAR detection around the estimated location area to obtain N suspicious target areas.

[0045] Step 3. Perform incoherent accumulation on several suspicious target areas to obtain a one-dimensional range image, reduce the target's attitude sensitivity, and use the amplitude of the surrounding sea area to calculate the threshold. The peak point exceeding the threshold is the scattering point, and the scattering point of each suspicious target area is obtained. The calculation formula of the scattering point threshold is:

[0046] y=max(x(n))+3std(x(n))

[0047] Where x(n) is the sea clutter around the target at time k-1.

[0048] Step 4. Perform max-min normalization on the one-dimensional range image of each suspicious target area and extract the features corresponding to the range channel of each scattering point to obtain the (C*R) of N suspicious target areas. iFeature matrix, i is the i-th suspicious target area, C is the type corresponding to the feature, R is the scattering point distance channel corresponding to each suspicious target area, and the R of each matrix is ​​different; further, this algorithm extracts four features: time domain-fractal, frequency domain-Doppler entropy, energy domain-relative average amplitude and modulation domain-time-frequency ridge, that is, C=4.

[0049] Step 5. Use the echo data at time k-1 to calculate the discrimination difference of each feature. First, find the difference between each feature target and the clutter. The formula is as follows:

[0050]

[0051] Among them, M is the number of scattering points of the target at time k-1, j is the jth feature, is the jth eigenvalue corresponding to the i-th scattering point of the target at time k-1, is the average value of the jth feature in the sea clutter region at time k-1.

[0052] Then calculate the feature difference of each feature, the formula is as follows:

[0053]

[0054] Step 6. Combine the extracted features with the corresponding feature differences to ensure the consistency of the distinguishing contribution of each element in the matrix.

[0055] Step 7. Carry out matrix feature dimensionality reduction based on scattering point intensity. The scattering point intensity reflects the extent of the impact of sea clutter at that location to a certain extent. Calculate the probability of scattering intensity proportion.

[0056]

[0057] Among them, r is the distance channel of each scattering point, i is the i-th suspicious target area, P r,i is the energy intensity corresponding to the rth scattering point in the i-th suspicious target area.

[0058] The eigenvalue corresponding to each scattering point is weighted by the probability of scattering intensity proportion, and the feature matrix is ​​reduced in dimension (C*R) i →(C*1) i .

[0059] Step 8. Find the minimum correlation distance between the scattering matrix of each suspicious area and the feature template. The correlation coefficient is

[0060]

[0061] The correlation distance is dist = 1-ρ XY, the value range is [0,2], which satisfies non-negativity. The smaller the correlation distance is, the more similar the two features are, and vice versa. Therefore, the suspicious target area corresponding to the minimum correlation distance is selected as the tracking target.

[0062] Step 9. For the first three times, the feature matrix with the smallest correlation distance is directly used as the feature template. From the fourth time, the current correlation distance is compared with the target judgment area calculated in the previous times. The judgment area is:

[0063]

[0064] where dist N It is the set of relevant distances between the target and the feature template selected for the previous N times. When the relevant distance is not in the target determination area, the feature template is not updated. When the relevant distance is in the target determination area, the feature template is updated.

[0065] The extracted classification features include four features: time domain-fractal, frequency domain-Doppler entropy, energy domain-relative average amplitude, and modulation domain-time-frequency ridge:

[0066] (1) Modulation domain - time-frequency ridge characteristics

[0067] Assuming that the analysis window function γ(t) is stable (pseudo-stationary) within a short time interval, the window function is moved to achieve windowing and translation near the pulse pressure signal z(t), so that z(t)γ(t) is a stable signal within different finite time widths, thereby calculating the power spectrum at different moments, and arranging the obtained series of Fourier transform results to obtain a two-dimensional representation, that is, the time-frequency distribution diagram.

[0068] The short-time Fourier transform formula of the pulse pressure signal z(t) is STFT z (t,f):

[0069]

[0070] Where z(t) is the pulse pressure signal, γ(t) is the window function, γ*(t) is the conjugate function of γ(t), t is time, and f is frequency.

[0071] Assume that the time-frequency distribution of the signal is an M×N matrix with an amplitude of f(i,j) (j∈[1,M], j∈[1,N]). The Gaussian operator is used to smooth the time-frequency image amplitude f(i,j). The Gaussian operator is G(i,j):

[0072]

[0073] Where i is the number of rows in the time-frequency distribution matrix, j is the number of columns in the time-frequency distribution matrix, and σ is the standard deviation. The smoothing process can be controlled by adjusting σ. Smoothing the time-frequency distribution graph is to convolve the original distribution graph with the Gaussian function, that is, L(i,j)=G(i,j)*f(i,j), where L(i,j) is the image after convolution.

[0074] Since the ridge features of the image are directional, in order to better extract the ridge features of the image, all operators are made non-directional and rotation-translation invariant in a certain way. The two-dimensional Laplacian operator can well meet these conditions. Define w as the gradient of the time-frequency image L(i,j), its size is the gradient modulus ||w||, and its direction is the direction with the largest gradient change, that is,

[0075]

[0076] Where L x With L y are the first-order derivatives of the image in the x and y directions, respectively.

[0077] When the point is on both sides of the image ridge, the direction of w always points to the ridge, and when the point is on the ridge, its direction is along the ridge. Define the vector ν perpendicular to the direction of w, that is

[0078]

[0079] where ||ν|| is the magnitude of the vector ν, θ ν is the direction of vector ν. For points on the ridge line, the direction of ν is perpendicular to the ridge. When ν is on the ridge, its direction changes the most. Its second-order derivative L y There will be a relatively large amplitude. Therefore, as long as a suitable threshold is selected, the ridge features of the image can be extracted. The second-order derivative of Laplacian of ν The operation formula is:

[0080]

[0081] In the formula and are the second-order derivatives of the image in the x and y directions, L xy Calculate the partial derivatives of the image in the x and y directions respectively.

[0082] For time-frequency distribution images with obvious ridge features, after selecting appropriate standard deviations σ and When the value is , the ridge line feature of the image can be extracted by using the above algorithm, and the ridge line of the time-frequency image can be extracted well.

[0083] (2) Time domain - fractal characteristics:

[0084] A set of acquired radar sea clutter time domain pulse pressure signals are processed and the Yule-Walker equation method is used to process the radar sea clutter time domain pulse pressure signals to obtain the sea clutter time domain AR spectrum sequence.

[0085] The structure function F(m) and scale m of the sea clutter time-domain AR spectrum sequence are calculated according to the "random walk" model, and the log-log curve is used to check whether there is a linear relationship between them. If the curve has a linear relationship, it is determined that the time-domain AR spectrum sequence of the sea clutter data has fractal characteristics; otherwise, it is determined that the time-domain AR spectrum sequence does not have fractal characteristics.

[0086] If it is determined that the time domain AR spectrum sequence of sea clutter has fractal characteristics, the Hurst index H of the AR spectrum of sea clutter can be further obtained. AR :

[0087]

[0088] Where F(m) is the structure function of the sea clutter time domain AR spectrum sequence, and m is the scale. The Hurst exponent of the AR spectrum can be used as a characteristic parameter to identify targets in the sea clutter background.

[0089] (3) Frequency domain - Doppler spectral entropy

[0090] The Doppler amplitude of the pure clutter unit is distributed in a wider bandwidth, while the Doppler amplitude of the unit where the target is located is distributed in a narrower range. Entropy can describe the degree of chaos of the system or data. Similarly, the Doppler spectrum vector entropy can be used to determine whether the target exists, which is defined as

[0091]

[0092] in It is the frequency domain accumulated data obtained by coherently integrating the time domain pulse pressure data.

[0093] (4) Energy domain - relative average amplitude

[0094] For a signal of length N, the average amplitude is defined as follows

[0095]

[0096] Where x represents the echo time domain pulse pressure signal. Differentiating the magnitude of echo intensity is the main basis for traditional radar to identify targets. RAA calculates the relative average amplitude by calculating the echo intensity around the unit to be estimated. and Represent the average echo of the unit to be estimated and the average echo of the surrounding reference units respectively. The relative average amplitude is calculated as follows

[0097]

[0098] Please see Figure 2 The present invention provides a radar target stable tracking method in a complex sea environment based on multi-feature association. For the four features of time domain-fractal, frequency domain-Doppler entropy, energy domain-relative average amplitude and modulation domain-time-frequency ridge, sea clutter and target have obvious differences in various feature domains, but the difference degree in each feature domain is not the same, and the feature changes with the movement of the target. Considering that the traditional CFAR can screen out suspicious target areas, what this paper needs to do is to select the suspicious area closest to the target at the previous moment around the target prediction position as the target at this moment for tracking. Generally speaking, judgment based on a single feature has certain limitations, and it is impossible to find a certain feature that can distinguish the target from the clutter. At the same time, the traditional multi-feature method assumes that each feature has the same distinguishing ability, and sometimes even most features fail. Only a few features can effectively distinguish the target from the clutter, but due to the misjudgment of most features, the final judgment is wrong. In order to effectively use multiple features to achieve the purpose of target discrimination, this paper introduces the feature difference degree to solve the problem of how to ensure the authenticity and effectiveness of the final judgment result when there are differences in the judgment results of different features.

[0099] Please see Figure 3 The present invention provides a radar target stable tracking method in a complex sea environment based on multi-feature association, which solves the problem of inconsistent number of target scattering points at different times. The algorithm in this paper takes into account that each scattering point can provide effective feature information, so feature extraction is performed on each scattering point. However, the number of target scattering points varies greatly, and the inconsistent dimensions of the feature template and the feature matrix at each moment will make it difficult to calculate the relevant distance. This paper introduces the probability of scattering point intensity proportion and weighting mechanism to solve the problem that the number of target scattering points at different times leads to inconsistent dimension of feature matrix and is difficult to effectively compare.

[0100] Please see Figure 4 The present invention provides a comparison of the tracking results of a radar target stable tracking method in a complex sea environment based on multi-feature association and the traditional energy-based tracking results. After multi-feature processing, the target that needs to be tracked at the current moment can be effectively determined, and the tracking accuracy can be effectively improved compared to single feature tracking. It should be understood that the above embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. In addition, it should be understood that after reading the contents taught by the present invention, those skilled in the art can make various changes or modifications to the present invention without departing from the spirit and scope of the technical solution of the present invention, which should all be covered in the scope of the claims of the present invention.

Claims

1. A radar target stable tracking method in a complex sea environment based on multi-feature association, characterized in that: The following steps are included: Using the tracking results at time k-1, combined with the filter, the position and state of the target at time k are predicted, and the feature matrix determined as the target at time k-1 is used as the feature template at time k; Using the echo data at time k, low-threshold CFAR detection is performed around the estimated location area to obtain N suspicious target areas; The suspicious target area is incoherently accumulated to obtain a one-dimensional range image. The threshold is calculated using the amplitude of the surrounding sea area. The peak value exceeding the threshold on the one-dimensional range image is taken as the scattering point to obtain the scattering point of each suspicious target area. After the one-dimensional range image of each suspicious target area is normalized by max-min, multiple features corresponding to the distance channel of each scattering point are extracted to obtain the (4*R) i Feature matrix, i is the i-th suspicious target area, C is the type corresponding to the feature, R is the scattering point distance channel corresponding to each suspicious target area, and R of each matrix is ​​different; The echo data at time k-1 is used to calculate the discrimination difference of each feature. The difference between the target and the clutter of each feature is first obtained, and then the feature difference of each feature is obtained. Combine the extracted features with the corresponding feature differences to ensure the consistency of the contribution of each element in the matrix; Carry out matrix feature dimensionality reduction based on scattering point intensity, calculate the probability of scattering intensity proportion, weight the eigenvalue corresponding to each scattering point with the probability of scattering intensity proportion, and reduce the feature matrix dimension (C*R) i →(C*1) i ; Obtain the correlation distance between the feature matrix and feature template of each suspicious target area, and select the area corresponding to the minimum correlation distance as the target. For the first three times, the feature matrix with the minimum correlation distance is directly used as the feature template. From the fourth time, the current correlation distance is compared with the target judgment area obtained by the previous N calculations. The target determination area is: where dist N It is the set of relevant distances between the first N selected targets and feature templates.

2. The radar target stable tracking method in a complex sea environment as claimed in claim 1, characterized in that: The multiple features are time domain-fractal, frequency domain-Doppler entropy, energy domain-relative average amplitude and modulation domain-time-frequency ridge; C=4.

3. The radar target stable tracking method in a complex sea environment as claimed in claim 2, characterized in that: The calculation formula of the scattering point threshold is as follows: y=max(x(n))+3std(x(n)) Where x(n) is the intensity set of sea clutter around the target at time k-1.

4. The radar target stable tracking method in a complex sea environment as claimed in claim 2, characterized in that: The difference formula between the target and clutter is as follows: Among them, M is the number of scattering points of the target at time k-1, j is the jth feature, is the jth eigenvalue corresponding to the i-th scattering point of the target at time k-1, is the average value of the jth feature in the sea clutter region at time k-1; The formula of the feature difference is as follows:

5. The radar target stable tracking method in a complex sea environment as claimed in claim 2, characterized in that: The probability formula of the scattering point intensity ratio is as follows Among them, r is the distance channel of each scattering point, i is the i-th suspicious target area, P r,i is the energy intensity corresponding to the rth scattering point in the i-th suspicious target area.

6. The radar target stable tracking method in a complex sea environment as claimed in claim 2, characterized in that: The formula for the correlation distance is as follows, dist=1-ρ XY , the value range is [0,2], where ρ XY is the correlation coefficient, The formula for the correlation coefficient is as follows, Where X and Y are the feature matrix and feature template to be compared.

Images