Frequency agility radar target detection method

Abstract

A frequency agility radar target detection method comprises the following steps: firstly, designing a self-adaptive iteration step length based on SCCS, and solving the problem that the iteration step length cannot be adjusted in time when the residual error change of a traditional SAMP algorithm is small, so that the signal recovery efficiency is reduced; and then, a grid correction method combining a regularization staged backtracking idea and gradient descent optimization is adopted, so that stray sidelobe energy can be effectively suppressed under the condition of lacking priori information of a sparseness level. According to the method, the problem that phase-coherent accumulation gain loss is increased due to the fact that the step length is fixed in a traditional SAMP method is solved, and the target detection performance can be improved.

Description

Technical Field

[0001] The present invention relates to the field of radar signal processing technology, and in particular to a frequency agile radar target detection method based on an improved sparsity adaptive matching pursuit algorithm (SAMP). Background Art

[0002] The coherent accumulation method can improve the target detection and clutter suppression of coherent frequency-agile radar. However, if the radiating source radar transmits a waveform parameter agile signal (such as radar carrier frequency agility), the consistency of different echo phases will be destroyed, resulting in an increase in the loss of coherent accumulation gain, which will bring great challenges to target detection. Commonly used coherent accumulation methods, such as Keystone transform (KT), adjacent cross-correlation function (ACCF), axial rotation moving target detection (MTD), etc., are only applicable to traditional radar signal processing. Many scholars have achieved a lot of results in the study of coherent processing methods for frequency-agile radar. For example, the stretch method with Doppler filter group is used to design a high-resolution range-velocity two-dimensional matched filter. The stretch method is based on the idea of ​​matched filtering. Its advantage is that the entire calculation process is a linear operation, the calculation complexity is low, and the calculation results are relatively stable. However, for frequency-agile radar, due to the irregular change of carrier frequency, there will be a high level of spurious sidelobes on the range-velocity plane, which may lead to a decrease in detection performance. In order to compress the energy of spurious sidelobes, more commonly used methods such as Radon-Fourier transform (RFT) are used. According to the target motion information, this method compensates for phase fluctuations through a two-dimensional range-velocity iterative search method to obtain echo coherent accumulation results, which can effectively suppress spurious sidelobes, but this method is computationally intensive. In order to solve this problem, in 2012, the IEEE Transaction on Aerospace and Electronic Systems journal open document "Randon-Fourier transform for radar target detection III: optimality and fast implementations" proposed two fast RFT implementation methods based on fast Chirp-Z transform: band RFT and sub-band RFT methods. In the same year, the Electronics letters journal open document "Fast implementation of generalized Randon-Fourier transform for maneuvering radar target detection" proposed an RFT method (RFT-WOA) based on particle swarm optimization (PSO). In 2018, the IEEE Transaction on Aerospace and Electronic Systems journal open document "Adaptive Randon-Fourier transform for weak radar target detection" proposed a sub-aperture adaptive RFT method to achieve fast detection of weak targets.However, these methods usually require complex preprocessing, such as subband search, appropriate multi-parameter initialization, sub-aperture partitioning, etc.

[0003] In recent years, sparse representation has made great progress in the fields of radar detection and radar three-dimensional imaging, and has also provided new ideas for signal processing of frequency-agile radar. Sparse reconstruction algorithm is an important part of sparse representation algorithm and has a crucial impact on the results of sparse representation algorithm. Existing reconstruction algorithms can be summarized into two categories: convex optimization method and matching pursuit method. Convex optimization method requires less measurement, but the computational complexity is high, which is not commensurate with actual large-scale applications, such as BP algorithm. Matching pursuit (MP) algorithm based on iterative greedy pursuit principle can achieve sparse reconstruction with low computational complexity. Typical algorithms of this type include orthogonal matching pursuit (OMP), compressed sampling MP (CoSaMP) and regularized OMP (ROMP) algorithm. Specifically, OMP optimizes the traditional MP method by orthogonalizing the support set atoms. The CoSaMP algorithm is developed from the OMP algorithm, and multiple atoms are selected in each iteration of this method. Both algorithms are proposed based on the idea of ​​backtracking. The biggest difference between them is that the atoms selected in each iteration of the ROMP method are always retained and used again in the next iteration, while the atoms selected in each iteration of the CoSaMP method may be discarded in the next iteration. Although the reconstruction complexity of these methods is close to O(KMN), which is much lower than the BP method, they all require prior information on the sparsity level of the target signal, which may not be met in practical applications.

[0004] The traditional sparse adaptive matching pursuit algorithm (SAMP) can achieve target signal reconstruction without sparsity level restrictions. However, the iteration step size of the traditional SAMP algorithm is difficult to determine and is not adaptive, which seriously affects the reconstruction performance. Specifically, if the step size of SAMP is set large during the iteration process, overestimation will occur, resulting in a decrease in reconstruction accuracy; if the step size is set small, more iterations are required, resulting in a decrease in reconstruction efficiency. The more commonly used improved SAMP algorithms include: the reduced iteration orthogonal matching pursuit (RIOMP) algorithm (setting a threshold based on the backtracking idea to reduce the number of iterations), the binary variable step size sparse adaptive matching pursuit (VSSAMP) algorithm based on energy residual theory (the step size is adaptive), the adaptive step size matching pursuit (AStMP) algorithm (using an adaptive step size to estimate sparsity in each iteration), and the correlation coefficient based SAMP algorithm (CCSAMP) (there is no need to compare the current iteration result with the previous iteration result in each iteration to improve the algorithm speed). Among them, the RIOMP and AStMP algorithms can effectively improve the accuracy of the sparse reconstruction algorithm, but the number of iterations of these algorithms is still quite high. The VSSAMP algorithm based on iteration residuals designs an adaptive iteration step size, but it may not be able to adjust the iteration step size in time when the iteration residual does not change much. Although the CCSAMP algorithm solves the problem that the iteration step size is insensitive to the iteration residual, in actual applications, the grid mismatch problem may cause non-stationary spurious sidelobes to be generated, resulting in a significant decrease in target detection performance in a low signal-to-clutter ratio (SCR) environment. The main disadvantage of the traditional SAMP algorithm is the high energy of the spurious sidelobes.

[0005] The statements herein merely provide background art related to the present invention and do not necessarily constitute prior art. Summary of the invention

[0006] The object of the present invention is to provide a frequency agile radar target detection method, which can effectively suppress stray sidelobe energy and improve target detection probability and reconstruction efficiency.

[0007] To achieve the above-mentioned purpose, the present invention provides a frequency agile radar target detection method, comprising:

[0008] Step S1, initialize the algorithm input parameters: initial residual r0=y, iteration step size s0, s0≥1, index set size L=s0, iteration number t, t≥1, correlation coefficient threshold η1, 0<η1≤1, polynomial fitting error convergence threshold Maximum number of iterations T max =250;

[0009] Step S2, regularizing the signal y to be detected;

[0010] |St (i)|≤2||S t (j)|| (i,j∈J0)

[0011] Among them, [S t ,J t ]=supp(|Φ T r t-1 |,L)=supp(|u|,L)

[0012] Step S3: Update candidate set Among them, C t represents the candidate set for the tth iteration, The L largest non-zero coordinates of the observation vector u or all non-zero coordinates are used to obtain the atomic set, F t-1 represents the candidate set corresponding to the t-1th iteration;

[0013] Step S4: Obtain the grid mismatch coefficient using the gradient descent method and t Corresponding sensor matrix Make updates;

[0014]

[0015] subject to|δr l |≤Δr / 2,|δf l |≤Δf / 2

[0016] in, and They represent the distance and Doppler resolution after grid division, x is the sparse coefficient vector, y is the signal to be detected, and Φ is the sensor matrix;

[0017] Ψ(r l +δr l ,f l +δf l ) to expand the first-order Taylor formula:

[0018]

[0019] Calculate the mismatch coefficients corresponding to the non-zero atoms in the reconstruction vector x:

[0020]

[0021] in, represents the residual signal before correction, represents the weighted partial derivative matrix of the mismatched atom, P∈C M×2L . w=[δr1,…,δr L ,δv1,…,δv L ]T represents the target distance and Doppler mismatch coefficient, w∈C 2L ;

[0022] Step S5: Calculate the vector using the least squares estimation method And update the support set F t =max(x t |,L) and the residual signal in, Denotes the support set F t The corresponding sensing matrix;

[0023] Step S6: Calculate the correlation coefficient ρ t =SCCS(y t ,y tL ), and update the iteration step size s according to the following conditions;

[0024] If t ≤η1, then s=s0+(10*(1-ρ t ));

[0025] If t ≤1-10 -6 , then s=s0;

[0026] If t ≤1, then s=1;

[0027] If ||r t ||≤||r end || or t ≥ T max , r end Represents the corresponding residual at the end of the iteration, then the output is x t ;

[0028] If ||r t ||2≥||r t-1 ||2 or Then L = L + s;

[0029] Otherwise, t = t + 1, until the iteration stop condition || r is met t ||≤||r end || or t>T max ;

[0030] Step S7: Output the final estimated sparse signal x t :

[0031] Step S8: sparse signal x t Perform rearrangement and obtain echo coherent accumulation results.

[0032] The present invention has the following beneficial effects:

[0033] 1. Improve reconstruction efficiency: An adaptive step size method based on the Spearman correlation coefficient (SCCS) is proposed to solve the problem of large number of iterations of the traditional SAMP method. In addition, the signal reconstruction efficiency is further improved by introducing a convergence criterion based on polynomial fitting error, which helps to improve the efficiency of target detection.

[0034] 2. Improve target detection probability: A grid mismatch correction method combining regularization principle and gradient descent optimization is proposed, which can effectively suppress spurious sidelobe energy, reduce coherent accumulation gain loss, and help improve target detection probability without requiring prior information on sparsity level. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 The present invention provides a flow chart of a frequency agile radar target detection method.

[0036] Figure 2 It is a schematic diagram of the principle of the grid mismatch correction method.

[0037] Figure 3 It is based on the matched filtering coherent accumulation result.

[0038] Figure 4 It is based on ROMP coherent accumulation and reconstruction results (assuming the sparsity is known).

[0039] Figure 5 It is based on ROMP coherent accumulation and reconstruction results (sparseness = 14).

[0040] Figure 6 It is based on the SAMP coherent accumulation and reconstruction results.

[0041] Figure 7 It is based on the coherent accumulation and reconstruction results of CCSAMP.

[0042] Figure 8 It is the coherent accumulation and reconstruction result of SAMP-OC method.

[0043] Fig. 9 It is the coherent accumulation and reconstruction result of the method of the present invention. DETAILED DESCRIPTION

[0044] The following is based on Figures 1 to 9 , specifically describe the preferred embodiments of the present invention.

[0045] Assuming that the frequency agile radar transmits a linear frequency modulation (LFM) signal and the carrier frequency of each pulse changes randomly, the transmitted signal can be expressed as:

[0046]

[0047] in, t f Indicates fast time, T p represents the radar pulse width, f m =f0+d m Δf(m=0,1,2,...,M-1) represents random frequency, where f0 is the carrier frequency, d m is a random integer whose value range is [0, B / Δf], Δf is the frequency step value, B is the radar baseband synthesis bandwidth, m is the pulse number, M is the total number of pulses during the coherent integration period, and k is the frequency modulation slope.

[0048] For targets with extended range, the signal finally received by the radar can be approximately regarded as the linear superposition of the echoes of all scattering points of the target with the same moving speed in the observation scene. For the convenience of explanation, assuming that the number of scattering points of the extended target is 1, the echo signal can be expressed as follows after modulation and pulse compression:

[0049]

[0050] Where A0 represents the target echo amplitude, sin c(x) = sinπx / (πx) represents the sin c equation. Assuming the target moves in a uniform straight line, τ = 2R(t m ) / c represents the delay, where R(t m )=R0+v0mT r is the instantaneous radial distance between the target and the radar, t m represents slow time, c is the speed of light, R0 and v0 represent the initial radial distance and speed of the target, T r is the pulse repetition interval.

[0051] Extracting the echo signal at the distance unit where the target is located along the slow time dimension can be expressed as:

[0052]

[0053] It can be seen from formula (3) that as the number of pulses m increases, the phase of the slow-time sampling signal changes non-uniformly. After coherent integration, the level of spurious side lobes on the range-velocity plane is high, which may lead to a decrease in target detection performance or even detection failure. The traditional coherent integration method based on fast Fourier transform (FFT) will no longer be applicable.

[0054] To solve this problem, the target echo range-Doppler two-dimensional plane can be discretized according to formula (3). Specifically, the range dimension and velocity dimension data where the target may exist are discretized into P and Q grid points respectively, and formula (3) can be rewritten as:

[0055]

[0056] Among them, Ωp and Ω q are all possible value sets of target high-resolution distance and speed, α(p,q) is the target scattering coefficient corresponding to the grid point (p,q) on the two-dimensional range-Doppler plane, where and represent the high-resolution range phase factor and velocity phase factor respectively.

[0057] The main task of signal processing is to reconstruct the target scene, that is, to estimate the target echo scattering coefficient α(p,q) on the high-resolution range-Doppler two-dimensional plane based on the echo data of formula (4). In addition, the grid should cover all the range and velocity units of interest, and the number of grid points determines the fineness of the reconstructed scene. For example, the higher the reconstruction fineness, the denser the grid and the more grid points. However, due to the range-velocity resolution capability of the radar system, the number of grid points cannot be increased indefinitely.

[0058] After discretization, the target echo scattering coefficient α(p,q) can be expressed by a complex matrix Γ∈P×Q, that is:

[0059]

[0060] Among them, (p l ,q l ) represents the distance-velocity pair corresponding to a grid point after grid discretization, l = 0, 1, 2, ..., PQ-1.

[0061] Vectorize the columns of matrix Γ to form vector x, which can be expressed as:

[0062] x=[a(p0,q0),...,a(p P-1 ,q0),...,a(p P-1 ,q Q-1 )] T (6)

[0063] Assume that the distance and phase factor corresponding to the lth element in vector x are recorded as l p =0,1,...,P-1,l q =0,1,2,...,Q-1. Therefore, formula (4) can be rewritten as:

[0064]

[0065] According to formula (7), the radar observation signal can be written in matrix form:

[0066] y=Φx (8)

[0067] Among them, y=[y(0),y(1),...,y(M-1)] T ∈C M represents the detection signal, Φ=[φ(0),φ(1),...,φ PQ-1 ]∈C M×PQ represents the sensing matrix.

[0068] The reconstruction of the radar observation scene can be modeled as estimating the vector x from formula (8). The atoms with an absolute value of zero in the vector x indicate that there are no significant target scattering points on the corresponding high-resolution range-velocity unit. In addition, based on the position of the non-zero atoms in the vector x, the high-resolution range and velocity information of the target at the corresponding position of the atom can be inferred. In general, formula (8) can be understood as a sparse representation of the echo signal based on the compressed sensing (CS) theory. When the sensing matrix satisfies the matrix constrained isocapacitance (RIP), the reconstruction result of the original signal can be obtained using the traditional SAMP method without prior information on the sparsity level.

[0069] Aiming at the shortcomings of the traditional SAMP algorithm, the present invention proposes a new greedy algorithm, which can effectively suppress the spurious sidelobe energy and improve the detection performance in the absence of prior information on the sparsity level, and has low computational cost, providing more effective solutions for radar target detection in low SCR environments.

[0070] like Figure 1 As shown, the method of the present invention mainly includes two steps: adaptive step size design based on Spearman correlation coefficient (SCCS) and grid mismatch correction.

[0071] 1. Adaptive step length design;

[0072] In the traditional SAMP algorithm, the vector y of the tth iteration is t and tL The relationship between can be expressed by the following formula:

[0073]

[0074] From the above formula, we can see that as the number of iterations t increases, the vector r rt gradually approaches the zero vector, vector y t and tL Linearly related and y tL Gradually tending to y t Therefore, we can use the vector y t and tL The correlation coefficient between them completes the signal reconstruction. Since the vector y t and tLThe correlation coefficient between them is variable, which can be used to convert the change of iteration step size, thereby solving the problem that the iteration step size is insensitive to the change of signals with similar residuals in two adjacent iterations.

[0075] The present invention uses the SCCS method to evaluate the vector y t and tL Similarity:

[0076] ρ t =SCCS(y t ,y tL ) (10)

[0077] Note that in practical applications, the signal to be detected may contain noise or clutter, and the echo data may not necessarily meet the normal distribution condition. Therefore, the present invention uses the SCCS method instead of the Pearson correlation coefficient (PCCS) method to describe the correlation between vectors.

[0078] In addition, if ||r t ||2<||r t-1 ||2, indicating that the estimated sparsity level can be close to the original signal sparsity level. In this case, for the traditional SAMP algorithm, the index set size L may remain unchanged in the next few iterations, resulting in an increase in the number of iterations, thereby reducing the reconstruction efficiency. In order to solve this problem, the present invention introduces a convergence criterion based on polynomial fitting error to adjust the size of the index set L, further accelerating the signal reconstruction speed. The convergence criterion can be expressed as:

[0079]

[0080] Among them, r t 'Represented as iterative residual r t The fitting results can be obtained by t =[r0,r1,...,r t-1 ] was obtained by polynomial fitting method. is the convergence threshold, used to adjust the size of L. Note: when ||r t ||2<||r t-1 ||2 and It indicates that the estimated signal sparsity level is close to the true value, but the reconstruction accuracy cannot meet our acceptable value (or is not yet the ideal value, then adjust L=L+s(estimated sparsity level) to further approach the true sparsity level of the original signal and further improve the reconstruction efficiency.

[0081] 2. Grid mismatch correction;

[0082] Although the OMP method can achieve high-precision reconstruction of echo signals, it requires prior information on the sparsity level. The traditional SAMP method can achieve echo signal reconstruction in the absence of prior information on the echo sparsity level, but it does not solve the grid mismatch problem, and the algorithm reconstruction performance is greatly reduced in a low SCR environment. In order to overcome this problem, the present invention proposes a grid mismatch correction algorithm that combines a regularized atom selection strategy and a gradient descent optimization (GDO) method to improve signal reconstruction accuracy, effectively suppress the spurious sidelobe energy after coherent accumulation, and improve target detection performance.

[0083] The grid mismatch correction method mainly includes two steps: regularization constraint and mismatch coefficient calculation.

[0084] 2.1, Regularization constraints;

[0085] In each iteration, the regularization idea is used to constrain the elements in the candidate list to ensure that the selected elements have greater energy than the unselected elements.

[0086] The idea of ​​regularization can be expressed as:

[0087] |S t (i)|≤2||S t (j)|| (i,j∈J0) (12)

[0088] Among them, [S t ,J t ]=supp(|Φ T r t-1 |,L)=supp(|u|,L).

[0089] According to formula (12), the purpose of regularization is to select J0∈J elements with the largest average energy from the candidate set. After regularization, the probability of selecting elements with lower importance will be significantly reduced, which improves the reconstruction accuracy and reduces the computational complexity. In addition, the set J can be obtained from the L largest non-zero value coordinates of the observation vector u or all non-zero value coordinates (selecting the coordinate set with less data), and then the next iteration is performed and the support set and the current residual are updated, as shown in the following formula:

[0090]

[0091] Among them, F t-1 represents the (t-1)th iteration support set, Denotes the support set F t The corresponding sensor matrix.

[0092] 2.2. Calculate the mismatch coefficient;

[0093] In actual detection scenarios, when the target parameters deviate from the selected grid, off-grid phenomena may occur, which may reduce the algorithm reconstruction performance. To solve this problem, Figure 2 As shown, the present invention uses a gradient descent optimization method to obtain the grid mismatch coefficient and adjust the sensor matrix, thereby reducing the offset between the actual position of the target scattering point and the grid position corresponding to the discrete sensor matrix, thereby suppressing the echo spurious sidelobe energy after coherent integration and improving the target detection performance.

[0094] To directly calculate the distance mismatch coefficient δr (δr = [δr1,…,δr K ], K is the sparsity level) and the Doppler frequency mismatch coefficient δf (δf=[δf1,…,δf K ]), we need to solve the following joint optimization problem:

[0095]

[0096] in, and They represent the distance and Doppler resolution after grid division, x is the sparse coefficient vector, y is the signal to be detected, and Φ is the sensing matrix.

[0097] Formula (14) is a non-convex equation solution problem with high computational complexity. The present invention uses the gradient descent method to convert the above joint optimization problem into a two-iteration optimization solution problem, thereby converting the above formula into a convex function solution problem.

[0098] First, Ψ(r l +δr l ,f l +δf l ) to expand the first-order Taylor formula:

[0099]

[0100] Then, the joint optimization problem in equation (14) is decomposed into two least squares solution problems. The mismatch coefficients corresponding to the non-zero atoms in the reconstruction vector x are calculated according to equation (16), as shown in the following equation:

[0101]

[0102] in, represents the residual signal before correction, represents the weighted partial derivative matrix of the mismatched atom, P∈C M×2L . w=[δr1,…,δr L ,δv1,…,δv L ] T represents the target distance and Doppler mismatch coefficient, w∈C 2L .

[0103] Finally, simulation data is used to verify the effectiveness of the proposed method. ROMP, SAMP, CCSAMP, RFT-WOA and SAMP-OC are used as comparison methods, and the target detection probability (p d ), running time (T') and target signal energy ratio r main (r main =S0 / S, S0 and S represent the reconstructed target signal energy and echo total energy respectively) and other indicators to evaluate the performance of each method. Specifically, the running time (T') and the target signal energy ratio r main They are used to compare the reconstruction efficiency and reconstruction accuracy of each method, and the detection probability (p d ) is used to evaluate the detection performance of each method. The methods of the present invention are all carried out on Matlab 2022a software, the experimental computer uses Windows 10 operating system, the CPU is Intel i5-12500 (2.5GHz), and the memory is 16.0GB.

[0104] In the simulation, it is assumed that the frequency agile radar transmits a linear frequency modulation signal, and the radar simulation parameters are shown in Table 1.

[0105] Table 1

[0106] parameter Numeric Signal bandwidth 100MHz Pulse Width 50us Sampling frequency 2.5GHz Pulse repetition time 300us Radar system loss 4dB Carrier frequency step interval 30MHz Total number of pulses during coherent processing 128

[0107] To simplify the processing, the problem of radar echo moving across distance units during coherent processing is ignored. Assume that the number of target scattering points is set to 2, and the RCSs of each scattering point is 10m 2 The relative distances between each scattering point and the radar are 6000m and 6020m respectively. The target moves at a uniform speed with an average speed of 10m / s. The clutter obeys the Weibull distribution, the clutter shape parameter is 1.5, the scale parameter is 1, and the SCR of the target echo after pulse compression is 16.76dB. The target detection probability is obtained through 100 Monte Carlo tests, and the false alarm rate is 10 -5 , in each Monte Carlo experiment, the maximum number of iterations is 256.

[0108] like Figure 3 As shown in Figure 1, due to the radar carrier frequency agility effect, the matched filtering method cannot obtain satisfactory coherent accumulation gain, resulting in target detection failure. The coherent accumulation results of ROMP, SAMP, CCSAMP, RFT-WOA (the maximum number of iterations is set to 50, and the number of search individuals is set to 20), SAMP-OC and the method of the present invention are shown in Figure 1. Figure 4-Figure 9 As shown. Figure 4 As shown in the figure, the ROMP method based on sparsity level prior information can effectively suppress the grid mismatch phenomenon, and the reconstructed non-target range-Doppler position mapping coefficients are all zero.

[0109] Table 2

[0110]

[0111]

[0112] As shown in Table 2, compared with other methods, ROMP can achieve better reconstruction performance, with the target signal energy ratio and detection probability both being 1, and the calculation time is only 0.0047s. However, the reconstruction accuracy of the ROMP method is greatly affected by the sparsity level and has poor robustness. Figure 5 As shown in the figure, in the absence of prior information on the sparsity level, there are a large number of spurious side lobes in the coherent processing results based on the ROMP method, the grid mismatch phenomenon is serious, and the target detection performance is reduced.

[0113] like Figure 6-Figure 8 It is shown that through the qualitative comparison of the coherent processing results of each method, it can be seen that the SAMP and CCSAMP methods have similar capabilities in target energy accumulation and spurious sidelobe suppression. For example, the target signal energy ratios of these two methods are 0.66 and 0.68, and the detection probabilities are 0.99 and 1, respectively, as shown in Table 2. By comparing the coherent accumulation of these three methods, it can be seen that the reconstruction performance of the RFT-WOA method is better, for example, the target signal energy ratio and detection probability are 0.99 and 1, respectively. However, the calculation time of RFT-WOA is 9.01s, which is much higher than other methods. For example, the calculation time of SAMP and CCSAMP methods is 0.22s and 0.068s, respectively. However, when the real target scattering point is not at the predetermined discrete grid position, it leads to a grid mismatch problem. At this time, the coherent accumulation results based on SAMP, CCSAMP and RFT-WOA methods all have certain spurious sidelobes, the target signal energy ratio is reduced, and the reconstruction accuracy is reduced, which is not conducive to improving the target detection performance.

[0114] like Figure 8 It can be seen that the SAMP-OC method can effectively accumulate target energy and suppress spurious sidelobe energy, suppress and solve the grid mismatch phenomenon, and improve reconstruction accuracy. For example, the target signal energy ratio and detection probability are 0.98 and 1 respectively, which are better than ROMP (lack of prior information on sparsity level), SAMP and CCSAMP methods. However, the SAMP-OC method is still based on the traditional SAMP method, which means that the iterative step size of this method is fixed. From the previous analysis, it can be seen that if the step size is not appropriate, it may affect the reconstruction performance of the algorithm and lead to a decrease in detection performance.

[0115] like Fig. 9As shown in Table 2, the target signal energy ratio, detection probability and calculation time of the method of the present invention are 1, 1 and 0.38s respectively, which are better than the SAMP-OC method (the target signal energy ratio, detection probability and calculation time of this method are 0.98, 1 and 0.5s respectively), which also proves that the adaptive step size design method proposed in the present invention is helpful to improve the algorithm reconstruction efficiency. Therefore, the quantitative and qualitative results show that in the absence of prior information on the sparsity level, compared with ROMP, SAMP and CCSAMP, the method of the present invention has better performance in terms of target signal energy ratio and detection probability, and the calculation time is better than the SAMP-OC method, and the reconstruction efficiency is higher.

[0116] It should be noted that, in the embodiments of the present invention, the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the embodiments, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0117] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0118] It should be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0119] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0120] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0121] As used in this specification and the appended claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0122] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as limiting the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention. After reading the above content, it will be obvious for those skilled in the art to make various modifications and substitutions to the present invention. Therefore, the scope of protection of the present invention should be defined by the attached claims.

Claims

1. A frequency agile radar target detection method, characterized in that: Include: Step S1, initialize the algorithm input parameters: initial residual r0=y, iteration step size s0, s0≥1, index set size L=s0, iteration number t, t≥1, correlation coefficient threshold η1, 0<η1≤1, polynomial fitting error convergence threshold Maximum number of iterations T max =250; Step S2, regularizing the signal y to be detected; |S t (i)|≤2||S t (j)|| (i,j∈J0) Among them, [S t ,J t ]=supp(|Φ T r t-1 |,L)=supp(|u|,L) Step S3: Update candidate set Among them, C t represents the candidate set for the tth iteration, The L largest non-zero coordinates of the observation vector u or all non-zero coordinates are used to obtain the atomic set, F t-1 represents the candidate set corresponding to the t-1th iteration; Step S4: Obtain the grid mismatch coefficient using the gradient descent method and t Corresponding sensor matrix Make updates; in, and They represent the distance and Doppler resolution after grid division, x is the sparse coefficient vector, y is the signal to be detected, and Φ is the sensor matrix; Ψ(r l +δr l ,f l +δf l ) to expand the first-order Taylor formula: Calculate the mismatch coefficients corresponding to the non-zero atoms in the reconstruction vector x: in, represents the residual signal before correction, represents the weighted partial derivative matrix of the mismatched atoms, P∈C M×2L w=[δr1,…,δr L ,δv1,…,δv L ] T represents the target distance and Doppler mismatch coefficient, w∈C 2L ; Step S5: Calculate the vector using the least squares estimation method And update the support set F t =max(|x t |,L) and the residual signal in, Denotes the support set F t The corresponding sensing matrix; Step S6: Calculate the correlation coefficient ρ t =SCCS(y t ,y tL ), and update the iteration step size s according to the following conditions; If t ≤η1, then s=s0+(10*(1-ρ t )); If t ≤1-10 -6 , then s=s0; If t ≤1, then s=1; If ||r t ||≤||r end || or t ≥ T max , r end Represents the corresponding residual at the end of the iteration, then the output is x t ; If ||r t ||2≥||r t-1 ||2 or Then L = L + s; Otherwise, t = t + 1, until the iteration stop condition || r is met t ||≤||r end || or t>T max ; Step S7: Output the final estimated sparse signal Step S8: sparse signal x t Perform rearrangement and obtain echo coherent accumulation results.

Images