Radar target detection method based on morphology

Abstract

The invention discloses a radar target detection method based on morphology, and the method comprises the steps: 1, carrying out the two-dimensional FFT processing of a two-dimensional echo data matrix collected by a radar, and obtaining an energy matrix after the two-dimensional FFT processing; step 2, performing two-dimensional CA-CFAR preliminary screening on the energy matrix after the two-dimensional FFT processing to obtain data after preliminary screening; 3, performing morphological false alarm suppression on the data subjected to the two-dimensional CA-CFAR preliminary screening to obtain a region subjected to false alarm suppression; and step 4, parameterized GLRT fine detection is carried out on the area after false alarm suppression, and whether a target exists is finally determined. The method can further improve the detection probability, improves the radar detection precision, reduces the false alarm number, provides a high-robustness solution for the real-time detection of a low observable target, and effectively solves a problem that the FFT speed measurement precision is insufficient in the prior art.

Description

Technical Field

[0001] The invention belongs to the technical field of radar detection and relates to a radar target detection method based on morphology. Background Art

[0002] As modern radars face increasingly complex target environments (such as low-altitude penetration drones and stealth fighters), small target detection has become one of the core challenges in radar signal processing. These targets typically have low radar cross section (RCS), low signal-to-noise ratio (SNR < -20dB), and strong ground and sea clutter interference, causing the performance of traditional detection algorithms to degrade dramatically. Constant False Alarm (CFAR) detection [1][2] As a key technology for radar target recognition, it aims to maintain a constant false alarm rate in a complex noise environment by setting an adaptive threshold. However, traditional CFAR algorithms (such as unit average CA-CFAR) [3] , ordered statistics OS-CFAR [4] ) faces an inherent contradiction under extremely low signal-to-noise ratio conditions: if the false alarm rate threshold is lowered to suppress noise false alarms, weak target signals are easily missed; if the threshold is relaxed to increase the detection probability, the number of false alarms will increase sharply, significantly increasing the back-end data processing burden. In addition, problems such as multi-target masking effects and non-uniform distribution of clutter further aggravate the degradation of detection performance. [5] Taking CA-CFAR as an example, its noise estimation relies on fixed-size reference cells. When the target energy diffuses in the range-Doppler domain, the reference cells are susceptible to contamination by the target's sidelobes, leading to biased noise estimation and, in turn, threshold inaccuracy. This limitation is particularly prominent in complex low-altitude environments, necessitating breakthroughs in novel detection architectures.

[0003] In recent years, scholars at home and abroad have proposed a variety of improvement schemes for CFAR detection optimization. For example, the variable size protection unit (VI-CFAR) improves the robustness of noise estimation by dynamically adjusting the reference window. In 2020, Liu Jia et al. [6] A multi-strategy CFAR detection algorithm based on adaptive threshold selection is adopted, combining the advantages of detection algorithms such as CA, GO and ACCA. By judging the current clutter background, the corresponding strategy is reasonably selected, but its computational complexity is relatively high. Morphological filtering suppresses clutter through structural element design. [7] , but it relies heavily on prior knowledge of the target morphology. In 2023, Wang Shenghua [8] et al. proposed an inference matrix to judge the interactive behavior between group targets, and to complete the target association and stable tracking under the cross, split and merge behaviors of group targets; a detector based on machine learning [9]Although it can extract features adaptively, it faces the bottleneck of lack of training data and real-time performance. In terms of refined discrimination, the generalized likelihood ratio test (GLRT) optimizes the test statistic through hypothesis testing.

[10]

[11] However, when used alone, it has high computational complexity and cannot meet the needs of real-time processing. Existing methods often focus on optimizing a single link and fail to systematically solve the problem of synergizing "wide threshold capture" and "high-precision discrimination", which restricts engineering applications in complex scenarios.

[0004] The public documents involved in the present invention are as follows:

[0005] [1]Finn HM, Johnson R S.Adaptive detection mode with thresholdcontrol as a function of spatially sampled clutter-level estimates[J].RCAreview,1968,29:414-464.

[0006] [2] Ding Lufei, Geng Fulu, Chen Jianchun. Principles of Radar[M]. Beijing: Electronic Industry Press, 2014.

[0007] [3]Hansen V G.Constant false alarm rate processing in search radars[C] / / IEEE International Radar Conference,London,1973:325-332.

[0008] [4]Rickard JT, Dillard G M. Adaptive detection algorithms for multiple-target situations[J]. IEEE Transactions on Aerospace and Electronic Systems, 1977, 13(4): 338-343.

[0009] [5] Zhang Yuxin. Research on adaptive constant false alarm rate detection technology under complex background[D]. Xidian University, 2021.DOI:10.27389 / d.cnki.gxadu.2021.002110.

[0010] [6] Liu Jia, Xiao Pengbin, Yuan Youhong. Distributed multi-strategy CFAR detection algorithm based on adaptive threshold selection[J]. Electro-Optics & Control, 2020, 27(09): 60-65.

[0011] [7] Zhang Bin, Hu Qingrong, Wei Lideng, et al. Improved morphological interference pattern filtering method [J]. Systems Engineering and Electronics, 2018, 40(10): 2230-2236.

[0012] [8] Wang Shenghua, Deng Yukun, Zhao Chenbo, et al. Group target tracking based on morphological filtering and inference matrix[J]. Modern Radar, 2023, 45(10): 94-99. DOI: 10.16592 / j.cnki.1004-7859.2023.10.012.

[0013] [9] Wang Zhifei, Yu Junpeng, Sun Jingming, et al. Detection of extended targets on sea surface based on machine learning[J]. Modern Radar, 2022, 44(03): 18-23. DOI: 10.16592 / j.cnki.1004-7859.2022.03.003.

[0014]

[10] Zhao Junxiang, Liang Xingdong, Li Yanlei. A SAR coherent change detection method based on likelihood ratio statistics[J]. Journal of Radars, 2017, 6(02): 186-194.

[0015]

[11] BANDIERA F, ORLANDO D, RICCI G. CFAR detection strat egies for distributed targets under conic constraints [J]. IEEE Transactions on SignalProcessing, 2009, 57(9): 3305-3316. Summary of the Invention

[0016] The purpose of the present invention is to provide a radar target detection method based on morphology to solve the problem of insufficient FFT velocity measurement accuracy in the prior art.

[0017] In order to achieve the above object, the present invention adopts the following technical solutions:

[0018] A radar target detection method based on morphology is characterized by comprising the following steps:

[0019] Step 1: Perform two-dimensional FFT processing on the two-dimensional echo data matrix collected by the radar to obtain the energy matrix after two-dimensional FFT processing;

[0020] Step 2: Perform a two-dimensional CA-CFAR preliminary screening on the energy matrix after two-dimensional FFT processing to obtain the data after preliminary screening;

[0021] Step 3: Perform morphological false alarm suppression on the data after the initial screening of 2D CA-CFAR to obtain the area after false alarm suppression;

[0022] Step 4: Perform parameterized GLRT precision detection on the area after false alarm suppression to finally determine whether the target exists.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] The method of the present invention adopts a cascade detection architecture that integrates morphological operations and statistical models. Its core innovation lies in achieving a balance between performance and complexity through a three-stage processing strategy:

[0025] (1) A two-dimensional CA-CFAR initial screening module captures potential targets with a high false alarm rate through reference units, protection units, and noise estimation;

[0026] (2) Morphological false alarm suppression module, which uses 4-neighborhood connectivity analysis and area screening to filter out discrete noise false alarms;

[0027] (3) The parameterized GLRT precision inspection module combines dynamic template construction with multi-dimensional feature fusion to improve the weak target identification capability.

[0028] In summary, the present invention breaks through the performance boundary of traditional single-threshold detection through the cascade architecture of "coarse screening-filtration-fine judgment". Simulation results show that in a strong noise environment with a signal-to-noise ratio of -20dB, compared with the traditional CA-CFAR algorithm, the present invention can further improve the detection probability, improve the radar detection accuracy, and reduce the number of false alarms. It provides a highly robust solution for the real-time detection of low-observable targets, and the system solves the problem of coordination between wide-threshold capture and high-precision discrimination. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of the method of the present invention;

[0030] Figure 2 is the target image after two-dimensional FFT;

[0031] Figure 3 is the detection probability under different signal-to-noise ratios; where (a) target 1, (b) target 2.

[0032] Figure 4 are the detection results of different false alarm rates; among them, (a) the false alarm rate is 1e -5 , (b) false alarm rate is 1e -4 .

[0033] Figure 5 It is the result of morphological processing;

[0034] Figure 6It is the result of GLRT processing. DETAILED DESCRIPTION

[0035] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] like Figure 1 As shown, the radar target detection method based on morphology provided by the present invention specifically includes the following steps:

[0037] Step 1: Perform two-dimensional FFT processing on the two-dimensional echo data matrix collected by the radar to obtain the energy matrix after two-dimensional FFT processing.

[0038] Step 1 specifically includes the following operations:

[0039] Assume that the pulse Doppler radar transmits a linear frequency modulation (LFM) signal, and the transmission signal of the mth pulse is s m (t):

[0040]

[0041] in:

[0042] s m (t)—transmitted signal of the mth pulse;

[0043] A—amplitude;

[0044] —Rectangular window function;

[0045] t—time;

[0046] T p — pulse width;

[0047] j—imaginary number symbol;

[0048] f c - carrier frequency;

[0049] k—frequency modulation;

[0050] The collected echo matrix is ​​then r(m,n), where m is the number of pulses and n is the number of sampling points of an echo; the matched filter impulse response is the conjugate flip of the transmitted signal:

[0051]

[0052] in:

[0053] h(t): matched filter;

[0054] s m (-t) * : conjugate flip of the transmitted signal;

[0055] Pulse compression output:

[0056] y(m,n)=ifft(fft(h(t))×fft(r(m,n))) (3)

[0057] in:

[0058] y(m,n)—pulse compression output result;

[0059] fft—Fast Fourier transform function;

[0060] ifft—inverse Fourier transform function;

[0061] Perform FFT processing on the same distance unit of m pulses to obtain the result after two-dimensional FFT processing modulo, which is the energy matrix after two-dimensional FFT processing:

[0062] S(m,n)=fftshift[fft(y(m,n))] (4)

[0063] in:

[0064] S(m,n)—the result after two-dimensional FFT modulo;

[0065] abs—modulo function;

[0066] fftshift — Spectral centering function.

[0067] Step 2: Perform two-dimensional CA-CFAR preliminary screening on the energy matrix after two-dimensional FFT processing to obtain the data after preliminary screening.

[0068] Step 2 specifically includes the following sub-steps:

[0069] Step 21: Data preprocessing, converting the complex data S(m,n) into power data:

[0070] P(m,n)=|S(m,n)| 2 (5);

[0071] Step 22, set parameters.

[0072] According to the target situation, set the number of protection units G in the distance and Doppler direction n and G m To prevent energy leakage, set the number of reference units R outside the protection unit n and R m , used to estimate noise. Set the false alarm rate P according to system requirements fa , use formula (6) to calculate the total number of reference units N ref .

[0073] Nref =(2G n +2G m +1)×(2R n +2R m +1)-(2R n +1)×(2R m +1) (6)

[0074] The threshold factor is calculated using formula (7):

[0075]

[0076] Where: P fa —Constant false alarm rate.

[0077] Step 23, traverse all range-Doppler units and perform the following operations on each detection unit (i, j):

[0078] (d) Determine the scope of the protected area

[0079] Doppler direction range: [iG m ,i+G m ], the range of distance direction: [jG n ,j+G n ];

[0080] (e) Collect reference units

[0081] Outside the protection area, traverse within the reference window:

[0082] Doppler direction: from iG m -R m To i+G m +R m , skip the protection unit.

[0083] Distance direction: from jG n -R n to j+G n +R n , skip the protection unit, add the power values ​​in all reference units to get the power sum P ref .

[0084] (f) Calculate the power mean of all reference units:

[0085]

[0086] Step 24, calculate the threshold T ca The formula is as follows:

[0087] T ca =αE avg (9);

[0088] Step 25: Compare the energy matrix after two-dimensional FFT processing with the threshold T ca Compare, mark it as 1 when it is greater than the threshold, mark it as 0 when it is less than the threshold, and get the data Ω after initial screening k .

[0089] Step 3: Perform morphological false alarm suppression on the data after the initial screening of the two-dimensional CA-CFAR to obtain the area after false alarm suppression.

[0090] Specifically, after step 2, the real target usually occupies a large continuous area, while the noise false alarm is mostly a small sporadic area. Count the number of pixels A in each connected area k By eliminating small false alarms, a more accurate target area can be obtained. The original data is subjected to a two-dimensional FFT, and the sinc function is presented in each dimension. Therefore, the 4-domain connected area is selected to filter the data after the initial screening in step 2, and the 4-domain connected filtered data is obtained, thereby obtaining the area after false alarm suppression. The formula is as follows:

[0091]

[0092] in:

[0093] D final (a, b)—Data after 4-domain connectivity filtering;

[0094] Ω k —Data after initial screening in step 2;

[0095] A k — Each connected region pixel;

[0096] γ—Statistically connected area pixels, the minimum target occupied area γ is estimated based on prior information such as radar resolution and target reflection cross-section (RCS).

[0097] Step 4: Perform parameterized GLRT precision detection on the area after false alarm suppression to finally determine whether the target exists.

[0098] Step 41: Dynamic Template Construction

[0099] Generate a signal template that matches the actual target parameters based on the area after false alarm suppression to improve detection accuracy. The specific operations are as follows:

[0100] (c) Region D after false alarm suppression based on step 3 final (a, b), extract the local peak points of each region (i p ,j p ), and estimate the corresponding delay and Doppler frequency

[0101]

[0102] Where: PRF—Pulse Repetition Frequency;

[0103] B—bandwidth;

[0104] m—number of pulses;

[0105] c—Speed ​​of light in vacuum.

[0106] (d) Template construction: based on refined parameters Constructing theoretical signal templates

[0107]

[0108] Step 42: Calculate the adaptive GLRT statistics.

[0109] Construct a target existence hypothesis model:

[0110]

[0111] Estimating the complex amplitude by maximum likelihood Compute the generalized likelihood ratio statistic:

[0112]

[0113] Step 43: Calculate the threshold η based on the false alarm probability. If Λ≥η, the target is considered to exist; otherwise, the target is considered not to exist. The calculation formula of the threshold η is:

[0114] η=-2lnP fa (15).

[0115] In order to prove the feasibility and effectiveness of the present invention, the following simulation experiment is carried out.

[0116] 1 Simulation parameters

[0117] Table 1 Simulation experiment parameters

[0118] Parameter name Value / Expression unit illustrate Number of sampling points N＝4096 Points Sampling frequency <![CDATA[f s =100e6]]> Hz ADC sampling rate Target 1 coordinates R=3000, V=500 m, m / s Target 1 distance Target 2 coordinates R=4000, V=300 m, m / s Target 1 speed Number of pulses M＝128 Number of pulses Center frequency f0=3e9 Hz Radar transmit center frequency Pulse duration <![CDATA[T p =8e-6]]> s Pulse width Pulse repetition time PRT=4.096e-5 s Pulse repetition interval bandwidth B=30e6 Hz FM signal bandwidth Speed ​​of Light c=3e8 m / s The speed of electromagnetic waves in a vacuum Signal-to-noise ratio SNR = -20 dB Receiver FM slope k=3.75e12 Hz / s Pulse width False alarm rate <![CDATA[P fa1 =1e -4 ,P fa2 =1e -5 ]]> False alarm probability Reference Unit 6*6 indivual Protection Unit 3*3 indivual

[0119] 2 Simulation Experiment Results

[0120] According to the above parameters, the results after two-dimensional FFT are shown in the figure. At this time, the SNRs corresponding to targets 1 and 2 are -30dB and -42dB respectively. The energy of target 1 is spread in a rectangular box of about 80*100, while the energy of target 2 is low and not obvious. Figure 2 shown.

[0121] 2.1 Comparison of detection performance under different signal-to-noise ratios

[0122] To further verify the effectiveness of this algorithm, the following simulation was conducted. The signal-to-noise ratio of target 1 was set from -30dB to -20dB, and the signal-to-noise ratio of target 2 was set from -52dB to -32dB. The CA_CFAR detection results were compared. The false alarm of CA_CFAR was set to 10-6, and the CFAR coarse screening constant false alarm of this method was set to 10-4. 500 Monte Carlo simulations were performed to compare the detection probabilities of the two targets. Figure 6 As shown in Figure 2, target 1 can be successfully detected. The detection probability of target 2 by this method is better than that of CFAR detection, which verifies that the collaboration of "wide threshold capture" and "high-precision discrimination" is feasible. Figure 3 shown.

[0123] 2.2 Changes in the number of false alarms

[0124] The experimental results of two-dimensional CFAR detection with two false alarm rates are as follows: Figure 4 As shown, the false alarm probability of (a) is 1e -5 , Figure 4 The false alarm probability of (b) is 1e -4 , after 1000 simulations, the number of false alarms is 98, and the average number of false alarms is 108. The subsequent process is based on Figure 4 (b) is processed. After the morphological false alarm suppression processing module, Figure 5 As shown, the results after GLRT processing are as follows Figure 6 shown.

[0125] Through Matlab simulation verification, it is shown that after morphological processing in a strong noise environment, small-area false alarms (such as isolated noise points) are filtered out, and the real target is retained due to the continuous area formed by energy aggregation. Through the fusion of multi-dimensional features (energy distribution, residual statistics), the weak target identification ability is improved, and the remaining false alarms caused by noise fluctuations are suppressed, verifying the performance of the proposed cascade detection architecture.

[0126] 3 Conclusion

[0127] In response to the inherent contradiction between false alarm rate and detection probability faced by traditional constant false alarm rate (CFAR) algorithms in radar weak target detection, this paper proposes a cascade detection architecture based on morphological operations and generalized likelihood ratio test (GLRT). Through the three-stage processing strategy of "coarse screening-filtering-fine judgment", the detection performance is improved in an extremely low signal-to-noise ratio (SNR=-30dB) environment, verifying the synergistic advantages of wide threshold capture and statistical fine judgment. In terms of engineering applications, this architecture can be compatible with hardware acceleration through modular design, providing a highly robust solution for real-time detection of typical low-observable targets such as low-altitude UAVs and stealth fighters. Future work will focus on the optimization of morphological parallel computing in multi-target dense scenarios, and explore the fusion framework of deep learning models and statistical tests to meet the challenges of target identification in complex clutter environments.

Claims

1. A radar target detection method based on morphology, characterized in that: The specific steps include: Step 1: Perform two-dimensional FFT processing on the two-dimensional echo data matrix collected by the radar to obtain the energy matrix after two-dimensional FFT processing; Step 2: Perform a two-dimensional CA-CFAR preliminary screening on the energy matrix after two-dimensional FFT processing to obtain the data after preliminary screening; Step 3: Perform morphological false alarm suppression on the data after the initial screening of 2D CA-CFAR to obtain the area after false alarm suppression; Step 4: Perform parameterized GLRT precision detection on the area after false alarm suppression to finally determine whether the target exists.

2. The radar target detection method based on morphology according to claim 1, wherein: Step 1 specifically includes the following operations: Assume that the pulse Doppler radar transmits a linear frequency modulation signal, and the transmission signal of the mth pulse is s m (t): in: s m (t)—transmitted signal of the mth pulse; A—amplitude; —Rectangular window function; t—time; T p — pulse width; j—imaginary number symbol; f c - carrier frequency; k—frequency modulation; The collected echo matrix is ​​then r(m,n), where m is the number of pulses and n is the number of sampling points of an echo; the matched filter impulse response is the conjugate flip of the transmitted signal: in: h(t): matched filter; s m (-t) * : conjugate flip of the transmitted signal; Pulse compression output: y(m,n)=ifft(fft(h(t))×fft(r(m,n))) in: y(m,n)—pulse compression output result; fft—Fast Fourier transform function; ifft—inverse Fourier transform function; Perform FFT processing on the same distance unit of m pulses to obtain the result after two-dimensional FFT processing modulo, which is the energy matrix after two-dimensional FFT processing: S(m,n)=fftshift[fft(y(m,n))] in: S(m,n)—the result after two-dimensional FFT modulo; abs—modulo function; fftshift — Spectral centering function.

3. The radar target detection method based on morphology according to claim 2, wherein: Step 2 specifically includes the following sub-steps: Step 21: Data preprocessing, converting the complex data S(m,n) into power data: P(m,n)=|S(m,n)| 2 ; Step 22, setting parameters; According to the target situation, set the number of protection units G in the distance and Doppler direction n and G m , set the number of reference units R outside the protection unit n and R m , set the false alarm rate P according to system needs fa , calculate the total number of reference cells N ref : N ref =(2G n +2G m +1)×(2R n +2R m +1)-(2R n +1)×(2R m +1) The threshold factor is calculated using the following formula: Where: P fa —Constant false alarm rate; Step 23, traverse all range-Doppler units and perform the following operations on each detection unit (i, j): (a) Determine the scope of the protected area Doppler direction range: [iG m ,i+G m ], the range of distance direction: [jG n ,j+G n ]; (b) Collect reference units Outside the protection area, traverse within the reference window: Doppler direction: from iG m -R m To i+G m +R m , skip the protection unit; Distance direction: from jG n -R n to j+G n +R n , skip the protection unit, add the power values ​​in all reference units to get the power sum P ref ; (c) Calculate the power mean of all reference units: Step 24, calculate the threshold T ca , the formula is as follows: T ca =αE avg Step 25: Compare the energy matrix after two-dimensional FFT processing with the threshold T ca Compare, mark it as 1 when it is greater than the threshold, mark it as 0 when it is less than the threshold, and get the data Ω after initial screening k .

4. The radar target detection method based on morphology according to claim 3, wherein: Step 4 specifically includes the following sub-steps: Step 41: Dynamic Template Construction Generate a signal template that matches the actual target parameters based on the area after false alarm suppression. The specific operations are as follows: (a) Region D after false alarm suppression based on step 3 final (a, b), extract the local peak points of each region (i p ,j p ), and estimate the corresponding delay and Doppler frequency Where: PRF—Pulse Repetition Frequency; B—bandwidth; m—number of pulses; c—the speed of light in vacuum; (b) Template construction: based on refined parameters Constructing theoretical signal templates Step 42: Adaptive GLRT statistics calculation Construct a target existence hypothesis model: Estimating the complex amplitude by maximum likelihood Compute the generalized likelihood ratio statistic: Step 43: Calculate the threshold η based on the false alarm probability. If Λ≥η, the target is considered to exist; otherwise, the target is considered not to exist. Threshold η calculation formula: η=-2lnP fa 。

Images