Robust target detection method and system in CG context based on model-data fusion

Abstract

The application relates to a target robust detection method and system based on model-data fusion in a CG background, and relates to the technical field of radar adaptive detection.The existing detector is based on a single assumed signal model, and when a real echo signal model is mismatched with the assumed signal model, the signal cannot be effectively and accurately estimated, and the detection performance is reduced.Compared with the existing detector method based on a single assumed signal model, the application has stronger robustness for different types of signal models, and the detection performance is better than that of the existing detector method in the case of signal model mismatch.The technical scheme of the application has CFAR characteristics for texture components and robustness for speckle components in a compound Gaussian background.A DMF-ANMF detector effectively solves the problem of detection performance reduction of the ANMF detector in the case of target signal model mismatch, and improves the robustness of the method for a target echo signal model.

Description

Technical Field

[0001] This invention relates to the field of radar adaptive detection technology, specifically to a robust target detection method and system based on model-data fusion in a CG background. Background Technology

[0002] To improve radar detection capabilities, coherent detection is widely used in radar target detection. In real-world radar detection environments, multiple scenarios are typically involved. For each scenario, an optimal detector can be designed based on the statistical model of clutter and the target echo signal model. However, since the type of scenario is often unknown beforehand, the optimal detector designed for a single scenario faces the risk of model mismatch.

[0003] Currently, most coherent detectors typically assume that background clutter follows a Gaussian distribution, such as Kelly's GLRT, AMF, ACE, and MSD detectors. However, as radar resolution increases, the statistical characteristics of clutter change, and the probability density function of the clutter envelope exhibits a tailing phenomenon. The compound Gaussian distribution is a typical non-Rayleigh distribution model. The aforementioned detectors cannot guarantee a specified false alarm probability in compound Gaussian clutter and may even experience a decline in detection performance.

[0004] To address the issue of detecting signals in a rank-1 subspace against a complex Gaussian clutter background, Conte et al. proposed a Novel Magnetic Model (NMF) detector. This detector achieves good detection performance even with small shape parameters or large pulse numbers. Gini et al. proposed a subspace-based detection method to improve the detection performance of detectors in a complex Gaussian background when the signal echo model is unknown. De Maio proposed the RAMF method based on second-order cone constraints, considering the mismatch in the target echo signal steering vector against a Gaussian clutter background. This method can characterize the degree of mismatch in the target signal steering vector by the angle of the cone. Dai Xin et al. extended De Maio's method to a complex Gaussian background, proposing the RANMF method. However, these methods all require prior information about the true target echo signal. The selection of the subspace or cone needs to ensure that the true echo signal vector falls within it, but this prior information is difficult to obtain in practice. Furthermore, once the subspace or cone is selected, the range of the signal model used by the method is fixed and cannot be adjusted according to the true target echo signal. When the actual target echo signal does not conform to the model assumptions of the above method, the detection performance will degrade. Therefore, for the target detection problem under the background of composite Gaussian clutter, it is of great significance to carry out the design of a robust target coherent detector, improve the detector's robustness to the target echo signal model under the background of composite Gaussian clutter, and improve the detector's detection performance under the condition of target echo signal model mismatch. Summary of the Invention

[0005] The purpose of this invention is to address the problem that existing detectors, which are based on a single hypothetical signal model, cannot effectively and accurately estimate the signal when there is a mismatch between the actual echo signal model and the hypothetical signal model, thus leading to a decline in detection performance. The invention proposes a robust target detection method and system based on model-data fusion in a CG background.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0007] A robust target detection method based on model-data fusion in CG backgrounds includes the following steps:

[0008] Step 1: Acquire the radar echo complex data after A / D sampling. The radar echo complex data includes the data {z1,…,z...} in the leading reference window. K / 2}, Data {z} in the trailing edge reference window K / 2+1 ,…,z K} and the data z of the unit to be detected;

[0009] Step 2: Estimate the covariance matrix using the data from the leading and trailing reference windows to obtain the estimated value.

[0010] Step 3: Based on the estimated value obtained in Step 2 By combining the data from the unit to be detected and the rank-1 subspace signal model, a model-based signal estimate is obtained.

[0011] Step 4: Based on model-based signal estimation The fusion weights χ are specified, thus obtaining the estimation based on model-data fusion.

[0012] Step 5: Utilize model-data fusion-based estimation estimated value The data z of the unit to be detected is used to obtain the detection statistic l(z) of DMF-ANMF;

[0013] Step 6: Based on the false alarm probability, obtain the threshold value corresponding to the false alarm probability. Then compare the detection statistic l(z) of DMF-ANMF with the corresponding threshold value. If the detection statistic l(z) of DMF-ANMF is not less than the corresponding threshold value, it is determined that there is a target; otherwise, it is determined that there is no target.

[0014] Furthermore, the estimated value Represented as:

[0015]

[0016] Where K is the length of the reference window, N is the number of coherent accumulation pulses, and z i For the i-th reference data, For z i The conjugate transpose of .

[0017] Furthermore, the rank-1 subspace signal model is represented as:

[0018] s∈Ω={s∈C N |s=αp,α∈C}

[0019] Where p = [1, e 2πf ,…,e 2πf(N-1) [ ] is the signal steering vector of the rank-1 subspace signal model, f is the normalized Doppler frequency, α is the unknown signal amplitude, s is the signal conforming to the rank-1 subspace signal model, Ω is the set of possible signal values ​​corresponding to the rank-1 subspace signal model, C is the complex field, C N Let be an N-dimensional complex vector space.

[0020] Furthermore, the model-based signal estimation Represented as:

[0021]

[0022]

[0023] Where, p H This is the conjugate transpose of p. The signal estimate is based on the rank-1 subspace signal model. This is an estimate of the signal amplitude based on the subspace model. express The inverse matrix.

[0024] Furthermore, the estimation based on model-data fusion Represented as:

[0025]

[0026] Where P = diag(p) ​​is the diagonal matrix composed of the signal steering vectors of the rank-1 subspace signal model, and I is the identity matrix. It is a vector composed of signal amplitude estimates based on the rank-1 subspace signal model.

[0027] Furthermore, the detection statistic l(z) of the DMF-ANMF is expressed as:

[0028]

[0029] Among them, z H Let z be the conjugate transpose of the data z of the unit to be detected. for The conjugate transpose of .

[0030] A robust target detection system based on model-data fusion in CG background includes: a radar echo complex data acquisition module, a covariance matrix estimation module, a signal estimation module, a DMF-ANMF detection statistics acquisition module, and a decision module.

[0031] The radar echo complex data acquisition module is used to acquire the radar echo complex data after A / D sampling. The radar echo complex data includes the data {z1,…,z...} in the leading reference window. K / 2}, Data {z} in the trailing edge reference window K / 2+1 ,…,z K} and the data z of the unit to be detected;

[0032] The covariance matrix estimation module is used to estimate the covariance matrix using data from the leading edge reference window and the trailing edge reference window, and obtain the estimated value.

[0033] The signal estimation module is used to estimate the value obtained in step two. By combining the data from the unit to be detected and the rank-1 subspace signal model, a model-based signal estimate is obtained. And based on model-based signal estimation The fusion weights χ are specified, thus obtaining the estimation based on model-data fusion.

[0034] The detection statistics acquisition module of the DMF-ANMF is used to utilize model-data fusion-based estimation. estimated value The data z of the unit to be detected is used to obtain the detection statistic l(z) of DMF-ANMF;

[0035] The determination module is used to obtain the threshold value corresponding to the false alarm probability based on the false alarm probability, and then compare the detection statistic l(z) of DMF-ANMF with the corresponding threshold value. If the detection statistic l(z) of DMF-ANMF is not less than the corresponding threshold value, it is determined that there is a target; otherwise, it is determined that there is no target.

[0036] Furthermore, the estimated value Represented as:

[0037]

[0038] Where K is the length of the reference window, N is the number of coherent accumulation pulses, and z i For the i-th reference data, For z i The conjugate transpose of;

[0039] The rank-1 subspace signal model is represented as follows:

[0040] s∈Ω={s∈C N |s=αp,α∈C}

[0041] Where p = [1, e 2πf ,…,e 2πf(N-1) [ ] is the signal steering vector of the rank-1 subspace signal model, f is the normalized Doppler frequency, α is the unknown signal amplitude, s is the signal conforming to the rank-1 subspace signal model, Ω is the set of possible signal values ​​corresponding to the rank-1 subspace signal model, C is the complex field, C N Let be an N-dimensional complex vector space.

[0042] Furthermore, the model-based signal estimation Represented as:

[0043]

[0044]

[0045] Where, pH This is the conjugate transpose of p. The signal estimate is based on the rank-1 subspace signal model. This is an estimate of the signal amplitude based on the subspace model. express The inverse matrix;

[0046] The estimation based on model-data fusion Represented as:

[0047]

[0048] Where P = diag(p) ​​is the diagonal matrix composed of the signal steering vectors of the rank-1 subspace signal model, and I is the identity matrix. It is a vector composed of signal amplitude estimates based on the rank-1 subspace signal model.

[0049] Furthermore, the detection statistic l(z) of the DMF-ANMF is expressed as:

[0050]

[0051] Among them, z H Let z be the conjugate transpose of the data z of the unit to be detected. for The conjugate transpose of .

[0052] The beneficial effects of this invention are:

[0053] This application mitigates signal model mismatch by fusing actual measurement data with information from a hypothetical signal model, enabling more accurate signal estimation. Compared to existing detector methods based on a single hypothetical signal model, it exhibits stronger robustness to different types of signal models and outperforms existing detector methods even under signal model mismatch conditions. The technical solution of this application demonstrates CFAR characteristics for texture components against a composite Gaussian background and exhibits robustness to speckle components. The DMF-ANMF detector effectively addresses the performance degradation problem of ANMF detectors under target signal model mismatch conditions by designing a model-data fusion method, improving the robustness of this method to target echo signal models. Attached Figure Description

[0054] Figure 1 DMF-ANMF detection flowchart;

[0055] Figure 2 Verify the texture component map for the CFAR properties of DMF-ANMF;

[0056] Figure 3To verify the speckle component plot for the CFAR characteristics of DMF-ANMF;

[0057] Figure 4 The detection probability curve of the Model 1 target in the composite Gaussian clutter is shown.

[0058] Figure 5 The detection probability curve of the Model 2 target in the composite Gaussian clutter is shown.

[0059] Figure 6 This is a graph showing the detection probability of the Model 3 target in a composite Gaussian clutter. Detailed Implementation

[0060] It should be noted that, where there is no conflict, the various embodiments disclosed in this application can be combined with each other.

[0061] Specific implementation method one: Refer to Figure 1 This embodiment describes a robust target detection method for CG backgrounds based on model-data fusion, comprising the following steps:

[0062] Step 1: Obtain the complex radar echo data after A / D sampling; the radar echo data includes clutter and targets, the clutter follows a composite Gaussian distribution, and there are reference cell data that only contain clutter;

[0063] Step 2: Estimate the covariance matrix of clutter based on the reference cell data; estimate the signal echo based on the rank-1 subspace model of the cell to be detected;

[0064] Step 3: Combine the signal echo estimate based on the rank-1 subspace model obtained in the previous step with the measured data of the unit to be detected, calculate the detection statistics, and make a decision based on the pre-calculated threshold to determine whether the target exists.

[0065] Figure 1 A flowchart of the DMF-ANMF detector is shown. The method includes the following steps:

[0066] Step 1, assume that the complex data in the reference window obtained after A / D sampling of the radar coherent pulse train echo is {z1,…,z...} K / 2 ,z K / 2+1 ,…,z K}, where {z1,…,z K / 2} represents the data in the preceding reference window, {z K / 2+1 ,…,z K} represents the data in the trailing edge reference window, and the data of the cell to be detected is z.

[0067] Step 2: Use the data within the reference window to estimate the covariance matrix. The estimated value is:

[0068]

[0069] Where K is the length of the reference window, N is the number of coherent accumulation pulses, and z i This represents the i-th reference data. Indicate z i The conjugate transpose of .

[0070] Step 3: Calculate the signal estimate based on the rank-1 subspace model. The rank-1 subspace model assumes that the signal echo vector lies within a known one-dimensional subspace, i.e.:

[0071] s∈Ω={s∈C N |s=αp,α∈C}

[0072] Where p = [1, e] 2πf ,…,e 2πf(N-1) [ ] is the steering vector of the rank-1 subspace model, f is the normalized Doppler frequency, N is the number of coherent accumulation pulses, α is the unknown signal amplitude, Ω represents the set of possible signal values ​​corresponding to the rank-1 subspace model, C represents the complex field, C N Let represent an N-dimensional complex vector space.

[0073] If the data of the unit to be detected is z, then the signal estimation based on the rank-1 subspace model is:

[0074]

[0075]

[0076] in For signal amplitude estimation based on the rank-1 model, For signal estimation based on the rank-1 model, p H This represents the conjugate transpose of p. express The inverse matrix.

[0077] Step 4, specify the fusion weights χ, where χ∈[0,1), and calculate the signal estimate based on model-data fusion:

[0078]

[0079]

[0080] in For signal amplitude estimation based on model-data fusion, For signal estimation based on model-data fusion, where P = diag(p) ​​is the diagonal matrix formed by the steering vectors of the rank-1 subspace model, P HLet P be the conjugate transpose, and I be the identity matrix.

[0081] Step 5, calculate the detection statistic. The detection statistic for DMF-ANMF is:

[0082]

[0083] in in express The conjugate transpose of .

[0084] Step 6: Based on the specified false alarm rate, look up the corresponding threshold value in the table according to the pre-calculated threshold.

[0085] Step 7: Compare whether the detection statistic exceeds the threshold. If it exceeds the threshold, it is determined that there is a target; otherwise, it is determined that there is no target.

[0086] Finally, the effectiveness of the algorithm was verified using Monte Carlo simulation experiments. The number of coherent pulses was set to N = 16, the reference window length to K = 40, and the false alarm rate to P. fa =10 -4 The threshold Monte Carlo simulation iterations are calculated to be 100 / P. fa The Monte Carlo simulation for calculating the detection probability was performed 10 times. 4 Next, the shape and scale parameters of the composite Gaussian distribution are set to b = v = 1. This invention names the detector with the optimal fusion weights as DMF-ANMF opt-χ. To verify the effectiveness of this invention, DMF-ANMF is compared with existing methods, where ANMF represents the adaptive normalized matched filter method, AMSD represents the adaptive subspace detector method, AMSD2 represents the AMSD method with the target signal steering vector matrix H = [p2, p3], AMSD3 represents the AMSD method with the target signal steering vector matrix H = [p1, p2, p3], SOC represents the second-order cone constraint method, and φ represents the cone angle.

[0087] Table 1 lists the target signal models tested. Model 1 is a rank-1 subspace signal model with a normalized Doppler frequency of 0.3 for the steering vector. Model 2 is a special three-dimensional subspace signal model with the same amplitude for each component and normalized Doppler frequencies of 0.3, 0.2, and 0.4 for the three components, respectively. Model 3 is a linear frequency modulated signal model with a normalized center Doppler frequency of 0.3 and a normalized Doppler frequency change rate of 0.1.

[0088] Table 1 Target Signal Model Settings

[0089] Model formula parameter Model 1 <![CDATA[s=αp0]]> <![CDATA[f0=0.3]]> Model 2 <![CDATA[s=α(p1+p2+p3)]]> <![CDATA[[f1,f2,f3]=[0.3,0.2,0.4]]]> Model 3 <![CDATA[s=αP chirp ]]> [f, k] = [0.3, 0.1]

[0090] Figure 4-6 The detection performance curves of the detector for different model targets are shown in the background of a composite Gaussian distribution. Table 2 shows the ranking of the detection performance of different methods under different models. The results show that in Model 1, the proposed method DMF-ANMF achieves the best detection performance by selecting appropriate fusion weights; in Model 2, the proposed method DMF-ANMF achieves suboptimal detection performance, although it does not achieve the best performance, it is a significant improvement compared to the unusable performance of ANMF; in Model 3, only our method can detect the target.

[0091] Table 2 Comparison of detection performance of different methods

[0092] method Model 1 Model 2 Model 3 DMF-ANMF opt-χ 1 2 1 DMF-ANMF χ=0.3 3 4 3 DMF-ANMF χ=0.5 4 3 2 ANMF 1 (Match) - - AMSD2 - 5 - AMSD3 2 1 (Match) - SOCφ=10° 1 - - SOCφ=60° 3 - -

[0093] Note: The smaller the number, the better the detection performance; "-" indicates that the method cannot detect the target of this model.

[0094] The detection results show that only our proposed method can detect the targets of all three models, verifying the robustness of our proposed method.

[0095] It should be noted that the specific embodiments are merely explanations and illustrations of the technical solution of the present invention and should not be used to limit the scope of protection. Any modifications made in accordance with the claims and specification of the present invention that are only partial should still fall within the protection scope of the present invention.

Claims

1. A robust target detection method in CG background based on model-data fusion, characterized in that... Includes the following steps: Step 1: Acquire the radar echo complex data after A / D sampling. The radar echo complex data includes the data in the leading edge reference window. Data in the trailing reference window and the data of the unit to be detected ; Step 2: Estimate the covariance matrix using the data from the leading and trailing reference windows to obtain the estimated value. ; Step 3: Based on the estimated value obtained in Step 2 By combining the data of the unit to be detected and the rank-1 subspace signal model, a model-based signal estimate is obtained. ; Step 4: Based on model-based signal estimation and specify the fusion weights. This leads to an estimate based on model-data fusion. ; Step 5: Utilize model-data fusion-based estimation Estimated value Data of the unit to be detected The detection statistic of DMF-ANMF was obtained. ; Step Six: Based on the false alarm probability, obtain the threshold value corresponding to the false alarm probability, and then calculate the detection statistic of DMF-ANMF. Compared with the corresponding threshold value, if the detection statistic of DMF-ANMF is... If the value is not less than the corresponding threshold, then a target is determined to exist; otherwise, no target is determined to exist. The estimation based on model-data fusion Represented as: in, This is a diagonal matrix consisting of the signal steering vectors of the rank-1 subspace signal model. It is the identity matrix. This is a vector composed of signal amplitude estimates based on the rank-1 subspace signal model. express The inverse matrix, It is the signal steering vector of the rank-1 subspace signal model.

2. The robust target detection method based on model-data fusion in CG background according to claim 1, characterized in that... The estimated value Represented as: Where K is the length of the reference window, and N is the number of coherent accumulation pulses. For the first One reference data point, for The conjugate transpose of .

3. The robust target detection method based on model-data fusion in CG background according to claim 2, characterized in that... The rank-1 subspace signal model is represented as follows: in, For the signal steering vector of the rank-1 subspace signal model. The normalized Doppler frequency, For unknown signal amplitude, For signals that conform to the rank-1 subspace signal model, Let C be the set of possible signal values ​​corresponding to the rank-1 subspace signal model, and let C be the complex field. Let be an N-dimensional complex vector space.

4. The robust target detection method based on model-data fusion in CG background according to claim 3, characterized in that... The model-based signal estimation Represented as: in, for The conjugate transpose of . The signal estimate is based on the rank-1 subspace signal model. This is an estimate of the signal amplitude based on the subspace model. express The inverse matrix.

5. The robust target detection method based on model-data fusion in CG background according to claim 4, characterized in that... The detection statistics of DMF-ANMF Represented as: in, Data of the unit to be detected The conjugate transpose of . for The conjugate transpose of .

6. A robust target detection system in a CG background based on model-data fusion, characterized in that... include: The system includes a radar echo complex data acquisition module, a covariance matrix estimation module, a signal estimation module, a DMF-ANMF detection statistics acquisition module, and a judgment module. The radar echo complex data acquisition module is used to acquire the radar echo complex data after A / D sampling, wherein the radar echo complex data includes the data in the leading edge reference window. Data in the trailing reference window and the data of the unit to be detected ; The covariance matrix estimation module is used to estimate the covariance matrix using data from the leading edge reference window and the trailing edge reference window, and obtain the estimated value. The signal estimation module is used to estimate the value obtained in step two. By combining the data of the unit to be detected and the rank-1 subspace signal model, a model-based signal estimate is obtained. And based on model-based signal estimation and specify the fusion weights. This leads to an estimate based on model-data fusion. ; The detection statistics acquisition module of the DMF-ANMF is used to utilize model-data fusion-based estimation. Estimated value Data of the unit to be detected The detection statistic of DMF-ANMF was obtained. ; The determination module is used to obtain the threshold value corresponding to the false alarm probability based on the false alarm probability, and then calculate the detection statistics of DMF-ANMF. Compared with the corresponding threshold value, if the detection statistic of DMF-ANMF is... If the value is not less than the corresponding threshold, then a target is determined to exist; otherwise, no target is determined to exist. The estimation based on model-data fusion Represented as: in, This is a diagonal matrix consisting of the signal steering vectors of the rank-1 subspace signal model. It is the identity matrix. This is a vector composed of signal amplitude estimates based on the rank-1 subspace signal model. express The inverse matrix, It is the signal steering vector of the rank-1 subspace signal model.

7. The robust target detection system based on model-data fusion in CG background according to claim 6, characterized in that... The estimated value Represented as: Where K is the length of the reference window, and N is the number of coherent accumulation pulses. For the first One reference data point, for The conjugate transpose of; The rank-1 subspace signal model is represented as follows: in, For the signal steering vector of the rank-1 subspace signal model. The normalized Doppler frequency, For unknown signal amplitude, For signals that conform to the rank-1 subspace signal model, Let C be the set of possible signal values ​​corresponding to the rank-1 subspace signal model, and let C be the complex field. Let be an N-dimensional complex vector space.

8. The robust target detection system based on model-data fusion in CG background according to claim 7, characterized in that... The model-based signal estimation Represented as: in, for The conjugate transpose of . The signal estimate is based on the rank-1 subspace signal model. This is an estimate of the signal amplitude based on the subspace model. express The inverse matrix.

9. The robust target detection system based on model-data fusion in CG background according to claim 8, characterized in that... The detection statistics of DMF-ANMF Represented as: in, Data of the unit to be detected The conjugate transpose of . for The conjugate transpose of .

Images