Target detection method based on target spatiotemporal stability

Abstract

The present application relates to the radar signal processing neighborhood, specifically to a kind of target detection method based on target space-time stability.The method includes: based on the time-domain signal representation of multiple frames sliding, generate subframe by dynamic matching filter, obtain subframe peak point information, including the energy, distance, doppler information of peak point on each subframe RD spectrum;Based on the spatial signal processing of single snap angle, generate multi-subframe peak point RDA spectrum, obtain the azimuth information of peak point on each subframe RD spectrum, utilize the feature information of each subframe peak point, remove false alarm point using false alarm suppression method based on mahalanobis distance;After false alarm suppression, the result is further removed discrete false alarm point using target track based on space-time stationarity, the method proposed in the present application improves the problem of too many false alarms in the RD spectrum after traditional clutter suppression algorithm;It can effectively reduce false alarm rate while retaining weak target signal, improve target detection capability;It is suitable for multiple system radars, and the application range is wide.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing, and more specifically to a target detection method based on the spatiotemporal stability of targets that can achieve better target detection results. Background Technology

[0002] Target detection is a crucial research topic and ultimate goal in the radar field. However, in complex clutter environments, the performance of traditional clutter suppression techniques is limited, failing to completely eliminate the influence of clutter. Residual clutter leads to a large number of false alarms on the RD spectrum. The presence of numerous false alarms severely impacts radar target detection performance. On one hand, during target track initiation, if the extracted data contains a large number of false alarms, these points will participate in track establishment, resulting in false tracks. This not only wastes system resources but also increases the computational load and complexity of computer data processing. Therefore, the presence of false alarms directly restricts the quality of automatic radar track initiation. On the other hand, false alarms may be associated with existing correct tracks, causing track deviations and preventing the radar tracking module from effectively completing its tracking task. Therefore, researching methods to reduce false alarms is of great significance.

[0003] False alarms are primarily caused by residual clutter after strong clutter suppression, exhibiting non-stationary characteristics across multiple dimensions such as time and space. In contrast, target points are generated from echoes reflected from real targets, exhibiting stationary characteristics across the same time and space. Therefore, a solution is urgently needed to reduce the false alarm rate and improve target detection capabilities. Summary of the Invention

[0004] This invention provides a target detection method based on the spatiotemporal stability of the target, which improves the radar's target detection capability in complex electromagnetic environments by utilizing the stationarity difference between the target and the false alarms, thereby significantly reducing the number of false alarms while retaining the target.

[0005] This invention is achieved through the following technical solution:

[0006] A target detection method based on target spatiotemporal stability, characterized by the following steps:

[0007] Step 1: Based on the time-domain signal representation of multi-frame sliding, sub-frames are generated through dynamic matched filtering to obtain the peak point information of the sub-frames, including the energy, distance, and Doppler information of the peak points on the RD spectrum of each sub-frame;

[0008] Step 2: Based on the spatial signal processing of single-shot angle measurement, generate multi-subframe peak point RDA spectrum and obtain the azimuth information of the peak points on the RD spectrum of each subframe;

[0009] Step 3: Using the feature information of the peak points of each subframe, remove false alarm points using a false alarm suppression method based on Mahalanobis distance;

[0010] Step 4: The results after false alarm suppression are further processed by removing discrete false alarm points using the target track based on spatiotemporal stationarity.

[0011] In step 1 of this invention, the target is stationary for a short period of time. To ensure the short-term stationarity of the target, this invention proposes a time-domain signal representation based on multi-frame sliding. The radar target echo signal can be regarded as a sample of the reference signal after time delay and Doppler frequency shift, which is a linear superposition of multiple target echoes, clutter, and noise. The echo signal is passed through a range-Doppler two-dimensional matched filter at a sampling rate f. s The sampled received signal, assuming the baseband signal transmitted by the radar is s. ref (n), n = 0, 1, ..., N-1; the echo signal after clutter suppression is s surv (n), then the two-dimensional matched filtering process of the radar echo signal can be expressed as:

[0012]

[0013] Two-dimensional matched filtering is achieved using a "distance correlation + Doppler transform" processing method. The above equation can be rewritten as follows:

[0014]

[0015] Where N b n is the length of the sub-segment. b Let N be the number of sub-segments, and N = n. b N b .like The above equation is then simplified to:

[0016]

[0017] To reduce false alarms, the stationarity difference between the target signal and the false alarm is utilized. To ensure short time intervals, a sliding matched filtering method is employed. This involves concatenating adjacent frames and then performing matched filtering using a shorter time window to generate RD information. Continuous frames are then concatenated pairwise, with αt (Δt < T) as the time interval. frame By performing matched filtering with time interval sliding, multiple subframe RDs can be obtained. i (i = 1, 2, ..., m), where m = T frame / αt,RD i Represented as:

[0018]

[0019] In the formula ΔN b =Δt×f sCompared to matched filtering per frame, the sliding subframes RDi have a smaller time interval to ensure the stability of the target signal.

[0020] In step 2 of this invention, after obtaining the distance R, Doppler velocity v, and energy E information on the subframe RD spectrum, it is also necessary to obtain the angle information of these peaks in order to more effectively distinguish between targets and false alarms. This invention selects the DOA estimation method based on sparse representation to estimate the angle. It has the characteristics of high precision and high resolution, and at the same time, it does not require phase deconvolution processing and has no requirement for the number of snapshots of the data.

[0021] Consider a spatial domain divided at equal angles as {θ1, θ2, ..., θ L}, and assume every possible angle θ n (n = 1, 2, ..., L) each correspond to a potential target signal s(θ) n If the array manifold matrix Φ = {a(θ1), a(θ1), ..., a(θ)}, then the array manifold matrix Φ = {a(θ1), a(θ1), ..., a(θ)} L Each column of the array corresponds to the azimuth information of a potential target signal. To reflect sparsity, the number of potential targets L is much larger than the number of actual targets N, thus constructing an L×1 dimensional sparse signal S = [s1, s2, ..., s]. L ] T , where s n =s(θ) n (n = 1, 2, ..., L), there are only N positions θ where the target actually exists. n The corresponding s n It is a non-zero value, while the other LN s n All are zero. The model for the sparse representation of the signal is:

[0022] Y = ΦS + E,

[0023] Where Y = Y j It consists of single snapshot data from the j-th distance gate after pulse compression; E represents the noise component of the data; matrix Φ is a redundant dictionary composed of guide vectors for all possible angles in space, i.e.:

[0024] Φ = {a(θ1), a(θ1), ..., a(θ)} L )},

[0025] A single snapshot data Y can be sparsely represented using a redundant dictionary Φ as the transformation basis. Its representation coefficients are the sparse signal S. By solving for the sparse signal S and finding the positions of its non-zero elements, the target's DOA angle can be estimated. For solving this problem, minimizing l... pThe constraint criterion of the norm (0 < p < 1), and at the same time, the idea of regularization is introduced into the algorithm to reduce the influence of noise on the reconstruction algorithm, so as to improve the estimation accuracy of the angle. The DOA estimation problem can be equivalently solved as the following problem:

[0026]

[0027] where 0 < p < 1 is the l p norm,

[0028] Under the maximum a posteriori probability (MAP) criterion, the solution of the above formula can be expressed as

[0029]

[0030] where the cost function The regularization coefficient η = σ 2 / β p , σ 2 is the noise variance, To obtain the optimal solution S * , it is necessary to satisfy its necessary condition:

[0031]

[0032] where the matrix ∑ = diag{|S(1)| P-2 , …, |S(L)| P-2}, the parameter From the above formula, we can get

[0033] (Φ H ΦS + 2λ∑)S * = Φ H Y,

[0034] Borrowing the idea of weighted minimization iterative solution of the FOCUSS algorithm, a weighted matrix is introduced:

[0035]

[0036] Substituting it in, we can get

[0037] S * = W((ΦW) H ΦW + λI) -1 (ΦW) H Y,

[0038] Introducing the iterative process, let W k+1 = diag{|S k (1)| 1-(P / 2) , …, |S k (L)| 1-(P / 2)} and U k+1 = Φk W k The optimal solution can be obtained iteratively using the following formula.

[0039]

[0040] The obtained sparse solution Find The non-zero elements or the N largest elements Location, and then based on s n ~θ n By establishing the correspondence, the estimated DOA of the target can be obtained.

[0041]

[0042] Where P CS This is the normalized spatial spectrum. From this, the angle estimate of any (τ,f) peak can be obtained.

[0043] In step 3 of this invention, the RD values ​​of each subframe are statistically analyzed. i Peak information matrix in (i = 1, 2, ..., m) Each subframe RD i have One peak, among which The peak information vector x obtained from the RD spectrum of the i-th subframe i,n =[R i,n ,v i,n E i,n ,θ i,n ], R i,n ,v i,n E i,n ,θ i,n These represent the distance cell, Doppler cell, energy, and azimuth information corresponding to the peak, respectively. Assume there are N... t One target, peak information matrix X i It can be rewritten as

[0044]

[0045] Where X t , Subframe RD i The target peak matrix and the false peak matrix, and

[0046] For the spike information matrix X of subframe RDi i The arbitrary spike information vector x in i,n From the distance-Doppler information obtained solely from the RD spectrum, it can be known that the peak information matrix X of adjacent subframes RDi+1 is... i+1 The vast majority of peaks and xi,n They are not matched. To reduce computational complexity and improve algorithm efficiency, we can first perform preliminary pairing based on distance-Doppler information. Here, we use Euclidean distance to do this, that is, for any spike x in subframe RDi... i,n Distance element - Doppler element (R i,n ,v i,n ), defined with (R i,n ,v i,n The matching range is a circular region centered at r0 with radius r0, and the peaks in adjacent subframes RDi+1 are the matching range. satisfy:

[0047] x can be considered i,n With x i+1,k Match, retain x i,n With x i+1,k Unmatched spikes are considered false spikes and are eliminated. The matching range r0 is typically chosen to be 2-3 cell sizes to ensure that the target is not lost while avoiding the introduction of more false alarms. This yields the matching spike information matrix X for adjacent subframes. i ′ and X′ i+1 The peak information vector is less, which reduces the computational load for subsequent steps while completing the initial reduction of false alarms;

[0048] To further reduce false alarms, the range cell, Doppler cell, energy, and angle information of the spike are fully utilized to calculate the spike x′ in the matched subframe RDi. i,n Match the spike x′ in subframe RDi+1 it1 The Mahalanobis distance is used, with the minimum Mahalanobis distance as the peak x′. i,n Distance d to subframe RDi+1 i,n For d i,n have

[0049]

[0050] in For x′ i,n to x′ i+1,k The Mahalanobis distance. ∑ -1 For x′ i,n to x′ i+1,k The inverse of the covariance matrix:

[0051]

[0052] If the covariance matrix is ​​the identity matrix, the Mahalanobis distance simplifies to the Euclidean distance; through the above operations, we can obtain the subframe RD. i All the peaks x′ i,n To adjacent subframe RDi+1 Mahalanobis distance Because the target signal is stationary relative to the false alarm signal within a short time interval, the Mahalanobis distance of the target is smaller than the Mahalanobis distance of the false alarm. Therefore, in order to detect N... t One goal becomes statistical D i,n N t The peak corresponding to the minimum Mahalanobis distance is sufficient;

[0053] However, the exact target number cannot be determined in actual data testing, therefore this invention sets a reference threshold d. thr When subframe RD i All peaks x′ i,n To adjacent subframe RD i+1 Mahalanobis distance D i,n The value in is less than the reference threshold d thr When the peak point is reached, it can be considered a suspected target. Reference threshold d thr The setting references the Mahalanobis distance when the target spike information vector may change the most within the interval between adjacent subframes. Assume that the target information of a subframe is T. i =[R i ,v i E i ,θ i If the value changes most significantly in adjacent subframes, then T is the value of T. i+1 =[R i +ΔR,v i +Δv,E i +ΔE,θ i +Δθ], based on practical engineering experience, ΔR and Δv are usually no more than 2 elements, ΔE is no more than 5dB, and Δθ is 1 / 2 beamwidth, therefore the reference threshold d thr for

[0054] d thr =MD(T i ,T i+1 ),

[0055] The final target detection result should be the intersection of the detection results from multiple subframes.

[0056]

[0057] Through the above steps, a large number of false alarms were eliminated.

[0058] In step 4 of this invention, after removing a large number of false alarms, there are still sporadically distributed false alarms in the RD spectrum. However, these false alarms cannot form stable tracks. Therefore, tracks can be generated based on the spatiotemporal stability of the target to further remove scattered points. The specific implementation method is as follows: First, all points in the first frame of data are used as track starters. Then, the Mahalanobis distance and SNR between points in adjacent frames are used as the criteria for target association. If the actual target track is not generated in the first frame, the newly generated points in each frame that are not associated with the previous frame will be used as the new track starters. In order to reduce false tracks and reduce the amount of computation, tracks that are not associated with each frame are cut off. The track update process is performed after each frame is detected as follows:

[0059] (1) Prediction: Assume that the uncut trace at time k-1 is {T} k-1 The state of (i)} is {x} k-1 (i)=[R k-1 (i),D k-1 (i),θ k-1 (i)]}, where the two parameters represent distance, velocity, and azimuth, respectively. Assuming the aircraft moves at a constant linear velocity within adjacent frames, the predicted state of the point at time k is:

[0060]

[0061] The above formula assumes that after time T, the target distance state will change according to the Doppler state, while the azimuth angle remains stable. The Doppler state changes slightly due to the change in position relative to the receiving station during flight, so w is the Doppler state adjustment parameter.

[0062] (2) Association: The set of points {x} detected at time k k (j)=[R k (j),D k (j),θ k (j)]} and the unfinished trace {T k-1 (i)} are associated according to the following rules:

[0063] First, filter to obtain the point T. k-1 (i) Predicted state p k (i) Two surrounding distance cells and four Doppler cells (assuming the range is U) i Neighborhood point set within ) If the neighborhood point set {x k If ′(m)} is an empty set, then continue updating the point state directly. If the neighborhood point set {x k If there is at least one point in ′(m)}, then the point with the strongest SNR is used to select the associated point;

[0064] (3) Evaluate the trace {T} k(i)} quality and update the point status: For points that are not associated with new points, the process ends after 3 consecutive frames of no association; for associated points, in order to reduce the impact of point offset caused by measurement error and process error, the point status is updated with a certain smoothing process according to the following formula:

[0065] x k (i)=x k-1 (i)+p(x k ′(m)-x k-1 (i)),

[0066] p is a smoothing coefficient, ranging from 0 to 1. Points not associated with any other point are used as new starting points. By tracking the trajectory, target points that form stable trajectories are retained to obtain the target trajectory, while scattered points that cannot form trajectories are eliminated, thereby reducing false alarms and improving target detection capabilities.

[0067] Compared with existing technologies, this invention utilizes the stationarity difference between the target and the false alarm to significantly reduce the number of false alarms while retaining the target, thereby improving the radar's target detection capability in complex electromagnetic environments. Attached Figure Description

[0068] Figure 1 This is a schematic diagram of the structure of the present invention.

[0069] Figure 2 This is a schematic diagram of the time-domain signal representation process based on multi-frame sliding in this invention.

[0070] Figure 3 This invention relates to a time-domain signal characterization method based on multi-frame sliding, specifically the multi-subframe RD spectrum.

[0071] Figure 4 This invention relates to the spatial signal characterization of multi-subframe RDA spectra based on single-shot angle measurement.

[0072] Figure 5 This is a schematic diagram of the false alarm suppression method based on Mahalanobis distance in this invention.

[0073] Figure 6 This is a comparison of the RD spectra before and after processing by the false alarm suppression method based on Mahalanobis distance in this embodiment of the invention. Figure 6 (a) is the original RD spectrum. Figure 6 (b) is the RD spectrum after clutter suppression without suppressing false alarms. Figure 6 (c) is the RD spectrum after false alarm suppression.

[0074] Figure 7 This is a schematic diagram of the target trajectory generation process based on spatiotemporal stability in an embodiment of the present invention.

[0075] Figure 8This is a comparison of the track generation results before and after suppressing false alarms in an embodiment of the present invention, wherein... Figure 8 (a) shows the track tracking results before false alarm suppression. Figure 8 (b) shows the track tracking results after false alarm suppression. Detailed Implementation

[0076] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0077] Combination Figure 1 This example provides a target detection method based on the spatiotemporal stability of the target, which includes the following steps;

[0078] Step 1: Temporal signal representation based on multi-frame sliding;

[0079] Step 2: Spatial signal processing based on single-shot angle measurement;

[0080] Step 3: False alarm suppression method based on Mahalanobis distance;

[0081] Step 4: Target trajectory generation based on spatiotemporal stationarity.

[0082] Since the target signal is stationary over a short period of time, in order to suppress false alarms by utilizing the difference in stationarity between the target signal and clutter, this invention first concatenates data from adjacent frames and then performs matched filtering using a time window with a smaller time interval. Figure 2 As shown.

[0083] The suppressed consecutive frame data are concatenated pairwise, with Δt (Δt < T) frame By performing matched filtering with time interval sliding, multiple subframe RDs can be obtained. i (i = 1, 2, ..., m), such as Figure 3 As shown, where m = T frame / Δt, RD i It can be represented as

[0084]

[0085] In the formula ΔN b =Δt×f s Compared to matched filtering per frame, sliding-generated subframe RD iSmaller time intervals are used to ensure the stability of the target signal. After multi-frame sliding matched filtering, the RD spectrum of the subframe is obtained, and all peaks on the spectrum are counted to obtain the distance R, Doppler velocity v, and energy E of these peaks. It can be determined that the target peak is contained among these numerous peaks, but due to the large number of false alarms, the target peak is difficult to distinguish. After obtaining the distance R, Doppler velocity v, and energy E information on the RD spectrum of the subframe, the angle information of these peaks is also needed to more effectively distinguish between the target and false alarms. This invention selects a DOA estimation method based on sparse representation to estimate the angle, which has the characteristics of high accuracy and high resolution, while requiring no phase deconvolution processing and having no requirement for the number of snapshots.

[0086] The target's azimuth information is obtained by estimating the angles of the peak points extracted from the RD spectrum of each subframe. Combining the distance and Doppler information of each peak point, the angle spectrum (RDA) of each subframe's peak point is obtained. The angles of each peak point are represented by different colors, as shown in the figure. Figure 4 As shown.

[0087] After the above operations, the RD values ​​for each subframe can be statistically analyzed. i Peak information matrix in (i = 1, 2, ..., m) Each subframe RD i have One peak, among which The peak information vector x obtained from the RD spectrum of the i-th subframe i,n =[R i,n ,v i,n E i,n ,θ i,n ], R i,n ,v i,n E i,n ,θ i,n These represent the range cell, Doppler cell, energy, and azimuth information corresponding to the peak. Assume there are N... t One target, peak information matrix X i It can be rewritten as

[0088]

[0089] Where X t , Subframe RD i The target peak matrix and the false peak matrix, and

[0090] For the spike information matrix X of subframe RDi i The arbitrary spike information vector x in i,n From the distance-Doppler information obtained solely from the RD spectrum, it can be known that the peak information matrix X of adjacent subframes RDi+1 is...i+1 The vast majority of peaks and x i,n Since they are not matched, to reduce computational complexity and improve algorithm efficiency, preliminary pairing can be performed based on distance-Doppler information. Here, Euclidean distance is used, that is, for any spike x in subframe RDi... i,n Distance element - Doppler element (R i,n ,v i,n ), defined with (R i,n ,v i,n The matching range is a circular region centered at r0 with radius r0, and the peaks in adjacent subframes RDi+1 are the matching range. satisfy:

[0091] x can be considered i,n With x i+1,k Match, retain x i,n With x i+1,k False spikes are eliminated by identifying spikes that do not match. The matching range r0 is typically chosen to be 2-3 cell sizes, ensuring that no target is lost while minimizing the introduction of false alarms. This yields the matching spike information matrix X for adjacent subframes. i ′ and X′ i+1 The peak information vector is less, which reduces the computational load for subsequent steps while completing the initial reduction of false alarms.

[0092] To further reduce false alarms, the range cell, Doppler cell, energy, and angle information of the spike are fully utilized to calculate the spike x′ in the matched subframe RDi. i,n Match the spike x′ in subframe RDi+1 i+1 The Mahalanobis distance is used, with the minimum Mahalanobis distance as the peak x′. i,n Distance d to subframe RDi+1 i,n The process is as follows Figure 5 As shown, for d i,n have

[0093]

[0094] in For x′ i,n to x′ i+1,k The Mahalanobis distance. ∑ -1 For x′ i,n to x′ i+1,k The inverse of the covariance matrix:

[0095]

[0096] To verify the effectiveness of the algorithm in practical applications, we conducted a field experiment. The radar beam of this system was directed towards the ocean and was mainly used to detect maritime targets. Real ADS-B track information was used to compare the track tracking results of the algorithm.

[0097] pass Figure 6 (a) As can be seen, since the radar main lobe faces the ocean, the clutter in the original data RD spectrum is mainly sea clutter. The presence of this clutter increases the floor energy of the original data RD spectrum, causing the target to be buried under the clutter energy. Therefore, clutter suppression is needed to reduce the floor energy and expose the target spike. Traditional time-domain suppression results are as follows: Figure 6 As shown in (b), after clutter suppression, the substrate energy decreased by approximately 25 dB, which is sufficient to reveal the target. However, due to the influence of clutter residue, a large number of false alarms exist in the suppressed RD spectrum, which adversely affects target detection. To improve target detection performance, false alarm reduction processing was performed on the suppressed data. The results are shown in (b). Figure 6 As shown in (c), after the false alarm reduction processing, a large number of false alarm peaks in the RD spectrum were eliminated. Using the same method to process consecutive multi-frame data in batches, the average processing effect of the multi-frame data can achieve the elimination of 91% of false alarm points.

[0098] Even after removing a large number of false alarms, sporadic false alarms still exist in the RD spectrum. However, these false alarms cannot form stable tracks. Therefore, track generation can be performed based on the spatiotemporal stability of the target to further remove scattered points. The specific implementation method is as follows: first, all points in the first frame of data are used as track starters; then, the Mahalanobis distance and SNR between points in adjacent frames are used as the criteria for target association. If an actual target track is not generated in the first frame, newly generated points in each frame that are not associated with the previous frame are used as new track starters. To reduce false tracks and computational load, tracks that are not associated across multiple frames are terminated. The track tracking process is as follows: Figure 7 As shown.

[0099] To compare the impact of reducing false alarms on tracking, a tracking algorithm based on target spatiotemporal stability was used to track the target trajectory of multiple peak points detected before and after the false alarm reduction. The tracking results were then compared with the actual ADS-B. Figure 8 (a) and Figure 8(b) The figures show the tracking results before and after reducing false alarms. The lines connecting the dots represent the tracked points, the scattered points represent the actual UAV tracks, and the color of the dots represents the azimuth angle. Due to the large amount of interference in the measured data, if false alarm suppression is not performed after temporal clutter suppression, false alarms will be mixed into the track during the tracking process. False alarms are unstable and will cause track interruptions, making it difficult to track points similar to the actual UAV tracks. After using the algorithm proposed in this invention to suppress false alarms, the obtained tracking results are very close to the actual UAV tracks, fully demonstrating the effectiveness of the target detection method based on the spatiotemporal stability of the target.

Claims

1. A target detection method based on the spatiotemporal stability of the target, characterized in that, Includes the following steps: Step 1: Based on the time-domain signal representation of multi-frame sliding, sub-frames are generated through dynamic matched filtering to obtain the peak point information of the sub-frames, including the energy, distance, and Doppler information of the peak points on the RD spectrum of each sub-frame; Step 2: Based on the spatial signal processing of single-shot angle measurement, generate multi-subframe peak point RDA spectrum and obtain the azimuth information of the peak points on the RD spectrum of each subframe; Step 3: Using the feature information of the peak points of each subframe, remove false alarm points using a false alarm suppression method based on Mahalanobis distance; Step 4: After suppressing false alarms, further remove discrete false alarm points using the target trajectory based on spatiotemporal stationarity; In step 3, the RD values ​​of each subframe are statistically analyzed. i Peak information matrix in (i = 1, 2, ..., m) Each subframe RD i have One peak, among which The spike information vector obtained from the RD spectrum of the i-th subframe , These represent the distance cell, Doppler cell, energy, and azimuth information corresponding to the peak, respectively. Assume there are... One target, peak information matrix Rewritten as ， in Subframe RD i The target peak matrix and the false peak matrix, and , First, preliminary pairing is performed based on range-Doppler information, using Euclidean distance, i.e., for any spike in subframe RDi. Distance element - Doppler element ( ), defined by ( Centered on ) The matching range is a circular region with a radius of 1, and the peaks in adjacent subframes RDi+1 are the matching range. satisfy: , think and Match, retain The method eliminates spikes that do not match, considering them false spikes. The matching range is selected to be 2-3 cell sizes, thus obtaining the matching spike information matrix of adjacent subframes. and ; To further reduce false alarms, the spike in RDi of the matched subframe is calculated. Match the spike in subframe RDi+1 The Mahalanobis distance is used as the minimum Mahalanobis distance as the peak. Distance to subframe RDi+1 ,for have , in for arrive Mahalanobis distance, for arrive The inverse of the covariance matrix: ， If the covariance matrix is ​​the identity matrix, the Mahalanobis distance simplifies to the Euclidean distance, thus obtaining the subframe RD. i All the spikes To adjacent subframe RD i+1 Mahalanobis distance Because the target signal is a stationary signal relative to the false alarm within a short time interval, the Mahalanobis distance of the target is smaller than that of the false alarm, thus enabling detection. One goal becomes statistics middle The peak corresponding to the minimum Mahalanobis distance: Set a reference threshold d thr When subframe RD i All peaks To adjacent subframe RD i+1 Mahalanobis distance The value in is less than the reference threshold. d thr At that time, the corresponding peak point was considered a suspected target. Reference threshold d thr The setting references the Mahalanobis distance when the target spike information vector may change the most within the interval between adjacent subframes. Assume that the target information of one subframe is... Then, the time when its change is greatest in adjacent subframes is , No more than 2 units No more than 5dB 1 / 2 beamwidth, reference threshold d thr for: ， The final target detection result is the intersection of the detection results from multiple subframes: ， Complete the elimination of false alarms.

2. The target detection method based on the spatiotemporal stability of the target according to claim 1, characterized in that, In step 1, the radar target echo signal is passed through a range-Doppler two-dimensional matched filter at a sampling rate of The sampled received signal, assuming the baseband signal transmitted by the radar is... The echo signal after clutter suppression is The two-dimensional matched filtering process of the radar echo signal is expressed as: ， Two-dimensional matched filtering is achieved by using a method combining distance correlation and Doppler transform. The above equation can be rewritten as follows: ， in The length of the sub-segment. Let be the number of sub-segments, and ,like Then the above formula is simplified to: ， To reduce false alarms, the stationarity difference between the target signal and the false alarm signal is utilized. To ensure short time intervals, a sliding matched filtering method is used. This involves concatenating adjacent frames and then applying matched filtering using a smaller time window to generate RD information. Continuous frames are then concatenated pairwise. , Matched filtering is performed by sliding the time interval to obtain multiple subframe RDs. i , i =1,2,...,m, where RD i Represented as: ， In the formula Compared to matched filtering per frame, the sliding subframes RDi have a smaller time interval to ensure the stability of the target signal.

3. The target detection method based on the spatiotemporal stability of the target according to claim 2, characterized in that, In step 2, a DOA estimation method based on sparse representation is selected to estimate the angle, considering the spatial domain partitioning with equal angles. And assume every possible angle Each of the n=1,2,...,L corresponds to a potential target signal. Then the array manifold matrix Each column corresponds to the azimuth information of a potential target signal. To reflect sparsity, the number of potential targets L is much larger than the actual number of targets N. Therefore, a... sparse signals of dimension ,in n=1,2,...,L, representing only N locations where the target actually exists. corresponding It is a non-zero value, while the other L−N are... If all zeros are present, the sparse representation model of the signal is: ， in It consists of single snapshot data from the j-th distance gate after pulse compression; E represents the noise component of the data; matrix Φ is a redundant dictionary composed of guide vectors for all possible angles in space, i.e.: ， The DOA estimation problem is equivalent to solving the following problem: ， where , 0 < p < 1 is a norm Under the maximum a posteriori (MAP) criterion, the solution to the above equation is expressed as: ， Where the cost function Regularization coefficient , For noise variance, To obtain the optimal solution It needs to meet its necessary conditions: ， Where the matrix ,parameter From the above equation, we get: ， Borrowing the idea of ​​weighted minimization iterative solution from the FOCUSS algorithm, a weighting matrix is ​​introduced: ， Substituting, we get: ， Introducing an iterative process, let and The optimal solution is then obtained iteratively using the following formula. ， The obtained sparse solution Find The non-zero elements or the N largest elements Location, and then according to Based on the correspondence, the estimated DOA of the target is obtained. , ， in To normalize the spatial spectrum, we obtain arbitrary Peak angle estimation.

4. The target detection method based on target spatiotemporal stability according to claim 3, characterized in that, In step 4, after removing a large number of false alarms, a trajectory is generated based on the spatiotemporal stability of the target, further removing scattered points. Specifically, all points in the first frame of data are used as the starting point for the trajectory. Then, the Mahalanobis distance and SNR between points in adjacent frames are used as the criteria for target association. If an actual target trajectory is not generated in the first frame, newly generated points in each frame that are not associated with the previous frame are used as the starting point for the new trajectory. To reduce false trajectories and computational load, unassociated trajectories across multiple frames are cut off. The trajectory update process after each frame's detection is as follows: Step 5-1: Prediction, assuming no cutoff point at time k-1. The status is: The two parameters represent distance, velocity, and azimuth, respectively. Assuming the aircraft moves at a constant velocity in a straight line within adjacent frames, the predicted state of the point at time k is: ， The above formula assumes that after time T, the target's range state will change according to the Doppler state, while the azimuth angle remains stable. The Doppler state undergoes subtle changes due to the shift in position relative to the receiving station during flight. w Adjust parameters for Doppler state; Step 5-2: Association: Assemble the set of points detected at time k With unfinished dots Associations should be made according to the following rules: First, filter to obtain the dots. Predicted state The set of neighboring points within a range of 2 distance cells and 4 Doppler cells, assuming this range is... , If the neighborhood point set If the set is empty, continue updating the point state directly. If the neighborhood point set If there is at least one point, the point with the strongest SNR is selected as the associated point; Step 5-3: Evaluate the dots To improve the quality and update the point status: For points that are not associated with new points, the process ends after 3 consecutive frames of no association; for associated points, to reduce the impact of point offset caused by measurement errors and process errors, the point status is smoothed using the following formula when updating: ， p is a smoothing coefficient, with a value between 0 and 1. For points that are not associated with any track, they are used as new starting tracks. Through track tracking, target points that form stable tracks are retained to obtain target tracks, while scattered points that cannot form tracks are eliminated, thereby reducing false alarms and improving target detection capabilities.

Images