A radar discrimination method for complex structure targets and simple shape targets

Abstract

This invention belongs to the field of radar target recognition technology, specifically relating to a radar identification method for distinguishing between complex-structured targets and simple-shaped targets. The method includes preprocessing the high-resolution range profile of the radar echo using an improved frequency-domain target length feature extraction method; extracting the target echo signal from sea clutter using the spectral differences between the target and clutter; calculating the range profile features and polarization features of each target; substituting the training feature set into a support vector machine (SVM) to obtain a trained SVM classifier; and using the classifier to classify the sample features obtained from the preprocessed radar echo signal to distinguish between complex-structured targets and simple-shaped targets. This invention addresses the situation where the radar echo contains both complex-structured target signals and simple-shaped targets, enabling effective identification between them.

Description

Technical Field

[0001] This invention belongs to the field of radar target recognition technology. Specifically, it relates to the ability to distinguish between complex structural targets and simple shape targets when polarimetric radar detects both at the same time. This is achieved by processing and analyzing the acquired data. Background Technology

[0002] Target identification remains a crucial and complex problem in modern radar applications. Real-world applications involve a wide variety of targets, differing in shape, size, material, and motion patterns, such as ships and buoys. However, the radar echoes of these targets are often difficult to distinguish due to various factors. After effectively detecting targets in complex and changing environments, radar systems need to quickly and accurately identify them. In-depth analysis of target structural characteristics, coupled with data-driven machine learning methods, can enable rapid and accurate target identification.

[0003] Complex targets typically possess multiple scattering centers, irregular shapes, and diverse material compositions, resulting in complex and variable echo signals. In contrast, simple targets have relatively regular geometric shapes, and their radar echo signals are more predictable and regular. Therefore, utilizing structural features for target identification has a theoretical basis. Currently, scholars have analyzed structural characteristics for related identification research; for example, some researchers have used the difference in radial length between complex and simple targets in high-resolution one-dimensional range profiles to distinguish between them. However, this is not applicable to situations where multiple simple targets are cascaded together. Other researchers have used polarization decomposition to utilize the polarization characteristics of complex and simple targets for target identification; however, polarization decomposition requires time-invariant scattering conditions.

[0004] In summary, while there are existing studies on target recognition using structural features, some features have low adaptability and are difficult to extract. Furthermore, many studies focus solely on single-domain features, limiting the improvement of discrimination performance and requiring further robustness enhancement. Therefore, a robust and easily implementable method for distinguishing complex structural targets from simple shaped targets, based on multi-domain feature combination, has yet to be developed and validated by real-world data. Summary of the Invention

[0005] The purpose of this invention is to propose a radar identification method for complex structure targets and simple shape targets, which can effectively identify the two when the radar echo contains both complex structure target signals and simple shape target signals.

[0006] The technical solution of this invention is: a radar identification method for targets with complex structures and targets with simple shapes, specifically including the following steps:

[0007] Step 1: The high-resolution range profile (HRRP) of the radar received echo is preprocessed using an improved frequency domain target length feature extraction method, and the target echo signal is extracted from the sea clutter by utilizing the spectral difference between the target and clutter.

[0008] Step 2: Based on the distance cells occupied by each target obtained in Step 1, calculate the range image features of each target: total half-peak width, average difference amplitude, and polarization features: average position difference of the strongest point.

[0009] Step 3: After processing the training dataset in steps 1-2, obtain the training feature set, and then substitute it into the Support Vector Machine (SVM) for training to obtain the trained SVM classifier.

[0010] Step 4: Use the classifier described above to classify the sample features obtained by the preprocessing of the radar echo signal to distinguish between targets with complex structures and targets with simple shapes.

[0011] Further, the preprocessing method in step 1 is specifically implemented as follows: a sliding window is used to slide from left to right on the high-resolution range image, and Fourier transforms are performed on the echo data within the sliding window in sequence. The product of the mean μ and standard deviation σ of the transformed amplitude spectrum |X(f)| is used as the test statistic T. Then, the range cells with test statistics greater than the threshold T1 are selected as target range cells, and the others are sea clutter cells. Each target region is shifted back by half the sliding window length. Finally, the interval between each target region is tested to exclude the case where a target region is truncated in the middle weak scattering region.

[0012] Furthermore, the sliding window length M is taken as an empirical value of 3.

[0013] Furthermore, the threshold T1 is the sample mean of the test statistic multiplied by the threshold factor α.

[0014] Furthermore, the threshold factor α is adjusted based on the size of the self-calibrated target, radar angle information, and range resolution within the scene.

[0015] Furthermore, the verification process involves determining whether the proportion of units exceeding the detection threshold η between the two regions is less than K. If the above condition is met, it is determined to be a clutter region; otherwise, it is determined to be a truncated region, and the two regions are then merged into the same target region.

[0016] Furthermore, the detection threshold η is determined by the sea clutter amplitude distribution and the false alarm probability;

[0017] Furthermore, K is taken as an empirical value of 0.1;

[0018] Furthermore, step 2 specifically includes the following sub-steps:

[0019] Step 21: Search for all scattering peaks in the target region that are greater than the detection threshold η, calculate their half-width, which is the number of scattering units greater than the peak height, and sum them to obtain the total half-width of the target.

[0020] Step 22: Calculate the difference magnitude within the target region, normalize it using the maximum and minimum values ​​within the region, and take the mean value to obtain the average difference magnitude of the target region.

[0021] Step 23: Search for the number of distance units at the point of strongest scattering of the target in the HH and VV polarization channels, and calculate its absolute difference; randomly select the results obtained from multiple pulse processing, and take the average value to obtain the average position difference of the target's strongest point;

[0022] Furthermore, the training dataset in step 3 is an actual field test dataset.

[0023] The beneficial effects of this invention are:

[0024] I. This invention utilizes high-resolution range image data obtained by radar, which is easy to acquire and store, thus facilitating feature calculation; and as can be seen from the technical solution of this invention, the calculation amount of the implementation steps is small, therefore, this invention is easy to implement in engineering.

[0025] Second, the present invention has been verified by actual test data and is applicable to a variety of scenarios, demonstrating strong adaptability;

[0026] Third, the present invention has been verified by actual test data. Due to the use of multi-domain features, it can still achieve good results even when the target sample type is single and the coverage of application is insufficient. Therefore, the present invention is highly robust. Attached Figure Description

[0027] Figure 1 This is a flowchart of the radar identification method for complex structure targets and simple shape targets in an embodiment of the present invention;

[0028] Figure 2 This refers to a one-dimensional distance image of a pulse HH channel from two sets of measured data used in this embodiment of the invention.

[0029] Figure 3 This is a preprocessed image of the echo extraction result of a complex structural target provided in this embodiment of the invention;

[0030] Figure 4 This is a comparison chart of the total half-peak width of complex structure targets and simple shape targets obtained from the processing of measured data in this embodiment of the invention;

[0031] Figure 5 This is a comparison chart of the average difference magnitude between complex structure targets and simple shape targets obtained from the processing of measured data in this embodiment of the invention;

[0032] Figure 6 This is a comparison chart of the average position difference of the strongest point between complex structure targets and simple shape targets, obtained from the processing of measured data, provided in this embodiment of the invention.

[0033] Figure 7 This is a three-dimensional feature space distribution map of complex structural targets and simple shape targets obtained by processing measured data, provided in this embodiment of the invention. Detailed Implementation

[0034] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0035] Figure 1 This is a flowchart of the radar discrimination method for complex structure targets and simple shape targets in this embodiment, including:

[0036] Step 1: Preprocess the HRRP echo data received by the radar using an improved frequency domain target length feature extraction method to extract the radar target echo signal. This specifically includes the following sub-steps:

[0037] Step 11, Parameter initialization and detection threshold calculation: Target starting unit t s and endpoint unit t e Initialize them as the HRRP start and end cells respectively, with the threshold coefficient α initialized to 1 and the false alarm probability P. F Set to 0.05. Take sea clutter cells, sort them from largest to smallest, and set the false alarm probability P. F The sea clutter amplitude corresponding to the location of the product N of the number of selected units is the threshold η.

[0038] Step 12, Sliding Window Modulus Taking: Using a sliding window of length M and step size 1, slide the window function from left to right, perform FFT on the data within each window, and take the modulus; assuming the m-th sliding window spectrum modulus taking result is represented by |X m | indicates that:

[0039]

[0040] In the formula: x i The HRRP value is set for the i-th distance cell.

[0041] Step 13, determine the starting unit t s : to |X m The product of the mean μ and standard deviation σ is used as the test statistic T for the region to be tested. The corresponding value for each distance unit is calculated as: T = μσ; when the discriminant first satisfies: This indicates entry into the target region, and the starting unit t is updated. s In the formula: To take the mean; α is the threshold coefficient.

[0042] Step 14, determine the endpoint unit t e Each time a starting unit t is determined s Subsequently, when the test statistic is first found to be less than the threshold value, an endpoint cell t is updated. e .

[0043] Step 15, Sliding window compensation to correct the target area: To eliminate the effects of sliding window, the results need to be appropriately corrected: In the formula: [·] represents rounding down.

[0044] Step 16, Truncation Region Inspection: To avoid unreasonable truncation within the target area, an inspection is required; if the proportion of units exceeding the detection threshold η between the two regions is less than K, that is:

[0045]

[0046] If it is determined to be a clutter region, then it is determined to be a truncation region. The two regions are then merged into the same target region, where: v n The magnitude x of HRRP only in the nth distance cell n Satisfy x n When p > η, the threshold is 1, where η is the detection threshold, determined by the sea clutter amplitude distribution and the false alarm probability. s,t p is the starting unit of the next region. e,t-1 This is the endpoint unit of the previous region; in this example, K is taken as an empirical value of 0.1.

[0047] Step 17, Radial Size Estimation and Verification: Based on prior information of the self-calibrated target within the scene: Target length L t Width W t And radar information: azimuth The grazing angle θ can be used to calculate the theoretical range of values ​​for the radial length [C]. min C max ]:

[0048]

[0049] Then, based on the above target area, determine [p] s ,p e Based on the radar range resolution ΔR, the estimated radial length is calculated:

[0050] c e =(p e -p s )ΔR,

[0051] If the estimated value exceeds the theoretical range, the threshold coefficient needs to be adjusted, i.e., return to step 13 to recalculate until the estimated value within the range is obtained. At this time, the corresponding area is the distance unit occupied by the target.

[0052] Step 2 involves calculating the corresponding characteristic values ​​for each radar target echo signal obtained in Step 1, specifically including the following sub-steps:

[0053] Step 21: Calculate the total half-peak width for each target. The specific steps are as follows:

[0054] Step 21a: Search for all peak and valley locations within the target area p v Its quantity is denoted as k, and together with the region boundary, they form a sequence P1, where P1 = {p s ,p v1 ,p v2 ,…,p vk ,p e}, represented as P1={p v0 ,p v1 ,p v2 ,…,p vk ,p v(k+1)};

[0055] Step 21b: Search for the peak p between each adjacent position in sequence P1 p Their quantity is k+1, forming a sequence P2, where P2={p p1 ,p p2 ,…,p p(k+1)}

[0056] Step 21c: Sequentially determine whether each element in P2 is greater than the detection threshold η. If p... pi If the value is greater than η, then its corresponding full width at half maximum (HWHM) is HPB. i for:

[0057]

[0058] Otherwise, it is 0, where:

[0059]

[0060] The total half-peak width (THPB) is obtained by summing all the half-peak widths.

[0061] Step 22: Calculate the average difference magnitude for each objective. First, calculate the normalized difference magnitude for each objective:

[0062]

[0063] The mean difference amplitude (MDA) is obtained by taking the average value.

[0064] Step 23: Calculate the average position difference of the strongest points corresponding to each target; define the position difference of the strongest points as the absolute value of the difference between the strongest point positions in the HH channel and the strongest point positions in the VV channel, denoted as ΔP. M Assume the distance to the strongest point of the target is P units. M Then ΔPM =|P M,HH -P M,VV The average position difference of the strongest point can be obtained by averaging the results of processing multiple random pulses.

[0065]

[0066] In the formula: For the a-th i The pulses correspond to the position difference of the strongest point; L is the number of pulses taken; and 1≤a i ≤N, 1≤i≤L≤N.

[0067] Step 3: Based on the three feature quantities obtained after processing the field measured data in Steps 1-2: total half-peak width, average difference amplitude, and average position difference of the strongest point, construct feature vectors F = [THPB, MDA, MDSPP] for multiple complex and simple targets, and then form a training feature set. Input the set into the preset support vector machine to train the SVM classifier. The kernel function of the support vector machine is selected as the Gaussian radial basis kernel function.

[0068] Step 4: Use the trained classifier to classify the sample feature vectors obtained from the radar echo signals processed in Step 1-2, and output classification labels to distinguish between complex structure targets and simple shape targets.

[0069] To verify the effectiveness and practicality of the radar identification method for complex-structured targets and simple-shaped targets provided in this application, this embodiment further illustrates the method through experimental data processing of a polarimetric coherent radar under two different test scenarios:

[0070] Figure 2 The example shows a single-pulse range image of a pulse echo HH channel from two sets of field test data, in which... Figure 2 (a) A scene corresponding to a single simple object and a complex structure is denoted as Scene 1; Figure 2 (b) The scenario corresponding to two simple shape targets connected together and a complex structure target is denoted as Scenario 2; in this example, the method performance is verified based on the data from Scenario 1 and 2.

[0071] Figure 3 This embodiment describes the echo of a complex target structure (equivalent to the range cell occupied by the target) extracted using a frequency domain length feature extraction method based on radar measured data. Figure 3 (a) shows the result obtained by the traditional frequency domain length feature extraction method. Based on known information, it is known that truncation occurs in the weak scattering region of the target. The improved method of this invention, which incorporates a verification step, can suppress this situation. See the results below. Figure 3 (b)

[0072] Figure 4 This is a comparison of the total half-peak width (FWHM) of complex-structured targets and simple-shaped targets, obtained by processing radar measured data in this embodiment. From... Figure 4 (a) It can be seen that the total half-width of simple-shaped targets is significantly smaller than that of complex-structured targets; from Figure 4 (b) It can be seen that the total half-peak width (HWHM) of the two simple-shaped targets connected together increases with the increase in the number of targets, but it is still significantly smaller than the HWHM of the complex-structured target in Scenario 2, and most likely smaller than the HWHM of the complex-structured target in Scenario 1. This is because the HWHM is a length feature that includes peak value information and is greatly affected by the actual size of the target. However, the selection of scattering peaks during feature calculation weakens the interference effect of the distance between the two simple-shaped targets connected together in length, so the HWHM is more advantageous than the radial dimension.

[0073] Figure 5 This is a comparison chart of the average differential amplitude between complex-structured targets and simple-shaped targets, obtained by processing radar field measurement data in this example. From Figure 5 (a) It can be seen that the average difference amplitude of complex structure targets is greater than 0.1, while that of single simple shape targets is less than 0.1; from Figure 5 (b) It can be seen that the value ranges of the complex structure target and the tandem of two simple shapes overlap, but the complex structure target is still more likely to have a value greater than 0.1, while the tandem of two simple shapes is more concentrated in the region less than 0.1. Generally, the average differential amplitude of the complex structure target is greater than that of the simple shape target. This is because simple shapes usually have a large radar cross-section and strong scattered echo. However, their size is much smaller than that of the complex structure target, and their HRRP shows a sharp decreasing trend from the peak to both sides within a small range cell. Therefore, the volatility of a single simple shape target is greater than that of the complex structure target. However, since the symmetry center point of the simple shape target has a much larger scattering intensity than that of the complex structure target, its rate of change after normalization is actually smaller than that of the complex structure target. Due to the presence of clutter cells between the targets, the value range of the tandem of two simple shapes partially overlaps with that of the complex structure target.

[0074] Figure 6 This is a comparison chart of the average position difference of the strongest points between complex-structured targets and simple-shaped targets, obtained by processing radar field measurement data in this example. From Figure 6 As can be seen from (a) and 6(b), whether it is a single simple-shaped target or two simple-shaped targets connected in series, the average position difference of the strongest point is likely smaller than that of a complex-structure target. This is because the range image of a complex-structure target is formed by the superposition of echo signals from multiple scattering centers. Once the polarization mode is changed, each sub-echo changes accordingly, easily causing the strongest scattering point of the superimposed echo to appear in different positions. In contrast, simple-shaped targets have extremely high symmetry, and the scattering mechanisms of each scattering point are similar, with little difference depending on the polarization mode. The peak point usually appears at the target scattering center.

[0075] Figure 7 This is a spatial distribution map of the three-dimensional features of complex-structured targets and simple-shaped targets, obtained by processing radar field measurement data in this example. From Figure 7 It can be seen that in the three-dimensional feature space, the data of complex structure targets and simple shape targets become linearly separable.

[0076] Table 1 shows the identification results of the radar field measurement data processed and tested using the radar identification method for complex structure targets and simple shape targets provided in this example. It can be seen that the proposed method is applicable not only to the identification of single shape targets and complex structure targets, but also to the identification of two shape targets in series and complex structure targets. Furthermore, it can still achieve good results even when the training sample coverage is insufficient, thus enabling effective identification of complex structure targets and simple shape targets.

[0077] Table 1 Comparison of Identification Results

[0078]

Claims

1. A radar method for distinguishing between targets with complex structures and targets with simple shapes, characterized in that, Specifically, the steps include the following: Step 1: The high-resolution range profile of the radar echo is preprocessed using an improved frequency domain target length feature extraction method, and the target echo signal is extracted from the sea clutter by utilizing the spectral difference between the target and clutter. Step 2: Based on the distance cells occupied by each target obtained in Step 1, calculate the range image features of each target: total half-peak width, average difference amplitude, and polarization features: average position difference of the strongest point. Specifically, it includes the following sub-steps: Step 21: Search the target region for all objects larger than the detection threshold. The scattering peaks are identified, and their half-widths (WHMs) are calculated. The WHMs are greater than the number of scattering units at the peak midpoint. The sum of these WHMs is then used to obtain the total WHM of the target. Step 22: Calculate the difference magnitude within the target region, normalize it using the maximum and minimum values ​​within the region, and take the mean value to obtain the average difference magnitude of the target region. Step 23: Search for the number of units at which the target's scattering is strongest in the HH and VV polarization channels, and calculate the absolute difference between them; The average position difference of the target's strongest point is obtained by randomly selecting and averaging the results from multiple pulse processing. Step 3: After processing the training dataset in steps 1-2, obtain the training feature set, and then substitute it into the support vector machine (SVM) for training to obtain the trained SVM classifier. Step 4: Use the classifier described above to classify the sample features obtained by the preprocessing of the radar echo signal to distinguish between targets with complex structures and targets with simple shapes.

2. The method according to claim 1, characterized in that, The preprocessing method in step 1 is specifically implemented as follows: Using a sliding window, slide it from left to right on the high-resolution range image, and perform Fourier transforms on the echo data within the window sequentially. The amplitude spectrum after the transform is then used. mean and standard deviation Product as a test statistic ; Then, the test statistic is greater than the threshold. The range cell is selected as the target range cell, and the others are sea clutter cells. The target regions are shifted back by half a sliding window length. Finally, the intervals between each target area are checked to rule out the possibility that a target area is truncated in the middle weak scattering region.

3. The method according to claim 2, characterized in that, Sliding window length Take 3 experience points.

4. The method according to claim 2, characterized in that, Threshold To test the sample mean multiplied by the threshold factor .

5. The method according to claim 4, characterized in that, Threshold factor Adjustments are made based on the size of the target being calibrated, radar angle information, and range resolution within the scene.

6. The method according to claim 2, characterized in that, The process of checking the interval between each target region is to determine whether the interval between two consecutive regions exceeds the detection threshold. The proportion of units is less than ; If the above conditions are met, it is determined to be a clutter region; Otherwise, it is determined to be a truncated area, and the two areas are then merged into the same target area.

7. The method according to claim 6, characterized in that, Detection threshold It is determined by the amplitude distribution of sea clutter and the false alarm probability.

8. The method according to claim 6, characterized in that, Take the empirical value of 0.

1.

9. The method according to claim 1, characterized in that, The training dataset in step 3 is the field test dataset.

Images