A moving target detection method based on optimal fusion of multi-channel ATI-SAR in strong clutter background

Abstract

This paper proposes a moving target detection method using optimal fusion of multi-channel ATI-SAR in a strong clutter background. This method first performs local detection using a set of multi-baseline clutter-free ATI phase tests and AMF amplitude tests. It then designs and optimizes global detection based on the optimal fusion rule. The present invention achieves maximum target detection performance with a constant false alarm probability. To facilitate implementation, the present invention develops a cognitive detection framework that does not require prior target knowledge. Because the present invention fully utilizes spatial degrees of freedom and information from environmental feedback, it effectively improves the detection performance of small, weak targets with low signal-to-noise ratios and low radial velocities in a strong clutter background.

Description

Technical Field

[0001] This invention belongs to the field of radar technology, and more specifically, to a method for detecting moving targets in strong clutter backgrounds using optimal fusion of multi-channel Along-Track Interferometry-Synthetic Aperture Radar (ATI-SAR). This method can be used in ATI-SAR systems to detect small, slow-moving targets with low signal-to-noise ratios in strong clutter backgrounds. Background Art

[0002] The ATI-SAR system is a special form of multi-channel SAR system, which achieves multi-frame SAR image observation of the same scene by configuring two or more SAR imaging channels in the direction of the platform's motion track. Compared with single-channel systems, multi-channel systems increase the spatial degrees of freedom of the signal and can more effectively suppress clutter, thereby improving the detection performance of low signal-to-noise ratio moving targets. However, under the influence of complex scenes and system errors, multi-channel systems will suffer severe signal-to-noise ratio loss during the adaptive clutter suppression process, resulting in a decrease in clutter suppression capability. Therefore, existing multi-channel ATI-SAR systems are still insufficient in terms of spatial degree of freedom utilization. In particular, in the presence of strong clutter, the detection of slow-moving targets with low signal-to-noise ratio still faces great difficulties.

[0003] In their paper "The CFAR detection of ground moving targets based on a joint metric of SAR interferogram's magnitude and phase" (IEEE Transactions on Geoscience and Remote Sensing, vol. 50, no. 9, pp. 3618-3624, 2012), Gao Gui et al. proposed a moving target detection method called IMP (Interferometric Magnitude Phase) that combines the interferometric amplitude and interferometric phase of SAR images along the track. The implementation process of this method is to first perform interferometric processing on the SAR image data of the two channels along the track. Then, the detection statistics are constructed using the interferometric signal amplitude and ATI phase to detect the SAR image and obtain the moving targets. However, the method still has shortcomings. For multi-channel synthetic aperture radar systems, the spatial degree of freedom utilization is low. In addition, the estimation performance of the ATI phase detection statistic is severely degraded under low signal-to-noise ratio conditions and is sensitive to channel errors and system noise. Therefore, these factors make it difficult for this method to detect slow-moving ground moving targets with low signal-to-noise ratio in strong clutter backgrounds.

[0004] In their paper "Two-step detector for RADARSAT-2's experimental GMTI mode" (IEEE Transactions on Geoscience and Remote Sensing, vol. 51, no. 1, pp. 436–454, 2013), CH Gierull, I. Sikaneta, et al. proposed a two-step detection method. This method involves performing preliminary detection on SAR images using signal amplitude information after multi-channel adaptive clutter suppression. Then, a second-step detection is performed using the along-track interferometric (ATI) phase of the two channels. The final detection result is the logical AND of the two detection steps. However, this method still has shortcomings: the ATI phase detection statistic in the second step only utilizes echo data from two channels, which wastes spatial degrees of freedom for multi-channel radar systems. Furthermore, the estimation performance of this ATI phase statistic deteriorates significantly in low signal-to-noise ratio conditions, resulting in unrobust detection performance in strong clutter environments. Summary of the Invention

[0005] The purpose of the present invention is to address the shortcomings of the above-mentioned existing technologies and propose a moving target detection method based on optimal fusion of multi-channel ATI-SAR in a strong clutter background, so as to solve the problem of low spatial degree of freedom and low signal-to-noise ratio in the existing technology in a multi-channel ATI-SAR system with difficulty in detecting slow targets in a strong clutter background.

[0006] The technical solution for achieving the purpose of the present invention is that the present invention utilizes AMF detection and multi-baseline clutter removal ATI phase detection to form a multi-scale optimal fusion target detector, performs high false alarm rate estimation on the initialized target parameters, utilizes the statistical characteristics of each detection quantity, and designs global optimization according to the optimal fusion rule to determine the optimal local detection threshold. Then, the local detection results are optimally fused to perform potential target judgment, and potential targets are detected as much as possible. Then, the maximum likelihood method is used to adaptively estimate the parameters of the potential target, update the target parameters, and finally, adaptively judge the final target based on the parameters of the potential target through the multi-scale optimal fusion target detector and global detection optimization. Because the present invention fully utilizes the spatial degrees of freedom and the information from the environmental feedback, it effectively improves the problem of the difficulty of detecting small targets with low signal-to-noise ratio and slow speed under strong clutter background.

[0007] The specific implementation steps of the present invention include the following:

[0008] S1: Use N SAR images to construct an M+1 scale detector, where N is equal to the number of radar channels, M=N(N-1) / 2;

[0009] S2: Initialize each detector to obtain the target parameter K, perform the global detection optimization method on all detectors, and obtain the optimal local false alarm probability corresponding to each detector in the automatic perception stage;

[0010] S3: By estimating the corresponding statistical characteristics of each detector, the corresponding local detection threshold is obtained according to the optimal local false alarm probability and the corresponding statistical characteristics. The corresponding local detection threshold is used to perform local detection on each detection quantity of each detector to obtain the corresponding binary detection result of each detector. The corresponding binary detection results of each detector are weighted and fused to obtain the global detection quantity of the target; the global detection quantity of the target is globally detected using the optimal fusion rule to determine the potential target;

[0011] S4: Adaptively estimate the detection amount corresponding to each detector for each potential target through the maximum likelihood method, and estimate the target parameters of each potential target; using the target parameters of each potential target and the set global detection false alarm rate, perform the same global detection optimization method as step S2 to obtain the optimal local false alarm probability corresponding to each detector in the enhanced detection stage;

[0012] S5: Using the same method as step S3, obtain the global detection value of each potential target; using the same optimal fusion rule as step S3, perform global detection on each potential target to determine the real target.

[0013] Furthermore, the M+1 scale detectors are composed of one detector for AMF amplitude detection and M detectors for ATI phase detection.

[0014] Furthermore, the moving target detection method of multi-channel ATI-SAR optimal fusion under strong clutter background is characterized in that the AMF amplitude detection value is expressed as:

[0015]

[0016] Where T represents the AMF amplitude detection quantity, n represents the number of views, ∑ represents the summation operation, k represents the ordinal number of the pixel in the SAR image, u represents the normalized filter weight vector, (·) H represents the conjugate transpose operation, z represents the data vector of the complex image point of the pixel to be detected, σ cn Indicates the minimum resolvable noise power.

[0017] Furthermore, the de-cluttered ATI phase detection quantity is expressed as:

[0018]

[0019] in, represents the phase detection amount after the ATI processing of the lth channel and the mth channel, arg[·] represents the phase angle, y l Indicates the n-view DPCA processing results of the lth channel and the 1st channel, represents the conjugate complex number of the n-view DPCA processing results of the m-th channel and the first channel, z l (k) and z1(k) represent the k-th pixel value in the SAR image corresponding to the l-th channel and the 1st channel, respectively.

[0020] Furthermore, the initialization of each detector refers to respectively initializing the amplitude detection amount of the AMF in the target parameter K, the target de-cluttering ATI phase estimation detection amount, and the correlation coefficient between the target channels;

[0021] Initialize the AMF amplitude estimation detection quantity according to the following formula:

[0022]

[0023] in, represents the initialization detection quantity of the AMF amplitude estimation, σ s represents the minimum resolvable signal power, represents the space-time steering vector, σ n represents the minimum resolvable noise power, k scnr =σ s / σ cn Indicates the minimum resolvable signal-to-noise ratio of the array element channel;

[0024] Initialize the target clutter removal ATI phase estimation detection quantity according to the following formula:

[0025]

[0026] in, represents the AIT phase estimation initialization detection quantity corresponding to the lth channel and the mth channel, [·] π represents the phase folding operation, v m represents the maximum unambiguous velocity in the airspace, λ represents the radar wavelength, and d l,m represents the baseline length, V represents the target radial velocity;

[0027] Initialize the space-time steering vector according to the following formula:

[0028]

[0029] in, represents the initialization space-time steering vector, exp(·) represents the exponential operation with the natural constant e as the base, and i represents the imaginary unit symbol;

[0030] Initialize the correlation coefficient between target channels according to the following formula:

[0031]

[0032] in, represents the correlation coefficient initialized between the target lth channel and the mth channel, ρ0 represents a constant, ρ0 = 0.99.

[0033] Furthermore, the steps of the global detection optimization method are as follows:

[0034] In the first step, the following optimization problem is established to obtain a set of optimal local detection false alarm probabilities:

[0035]

[0036] Among them, |·| represents the absolute value operation, P d represents the detection probability of global detection, P f represents the false alarm probability of global detection, r represents the global false alarm probability set according to the receiver performance requirements, P fj represents the false alarm probability of the jth detection unit;

[0037] The second step is to calculate the global detection false alarm probability P corresponding to each set of local detection false alarm probability candidate values. f for:

[0038]

[0039] Among them, L represents the number of candidate values ​​of each group of local detection false alarm probability that exceeds the global detection threshold η G The number of η G =log(P(H0) / P(H1)), log represents the logarithmic operation with base 10, P(H1) and P(H0) represent the prior probability of the existence of the moving target, ∏· represents the symbol of the continuous multiplication operation, D 1,l 、D 0,l They represent the set of identifiers exceeding the local detection threshold and the set of identifiers not exceeding the local detection threshold in group l, l = 1, 2, ..., L, u j Represents the detection result of the local detection unit, u j = +1 indicates that the local detection unit determines that there is a target, u j = -1 indicates that the local detection unit determines that there is no target, 1-P fk represents the correct judgment probability of the kth detection unit;

[0040] The third step is to calculate the global detection probability P corresponding to the candidate value of the local detection false alarm probability. d for:

[0041]

[0042] Among them, P dj represents the detection probability of the jth detection unit, 1-P dk represents the missed detection probability of the kth detection unit;

[0043] In the fourth step, by traversing each candidate value of the local detection false alarm probability, repeating the second and third steps and recording the global false alarm probability and detection probability, until the optimal local detection false alarm probability of the optimization problem in the first step is obtained.

[0044] Furthermore, the step of obtaining the corresponding local detection threshold according to the optimal local false alarm probability and the corresponding statistical characteristics is as follows:

[0045] The first step is to use the following formula to calculate the probability density function of the AMF amplitude detection quantity T under the background of non-uniform clutter:

[0046]

[0047] Among them, f AMF (·) represents the probability density function of the AMF amplitude detection quantity T, t represents the AMF amplitude detection quantity, Represents the equivalent view number, χ represents the non-uniformity of the clutter, χ>0, the larger the χ value, the more uniform the clutter scene is, and the χ value is estimated by the second-order statistic m2 of the AMF amplitude detection quantity. H0 represents the hypothesis without a target, Γ(·) represents the Gamma function;

[0048] When there is a moving target signal, according to the following formula, let its corresponding AMF amplitude detection amount be ω, and the probability density function of the AMF amplitude detection amount T is:

[0049]

[0050] Where t represents the AMF amplitude detection quantity, H1 represents the hypothesis of the existence of the target, and 2F1(·) represents the Gaussian hypergeometric function;

[0051] Given the false alarm probability P f0 , according to the following formula, the corresponding amplitude detection threshold is obtained:

[0052]

[0053] in, Indicates the calculation of the probability density function f from the threshold η0 to positive infinity AMF Cumulative probability of (·);

[0054] The second step is to remove the clutter ATI phase detection quantity according to the following formula under the clutter background. The probability density function of is:

[0055]

[0056] Among them, f p (·) represents the amount of ATI phase detection after removing clutter The probability density function of represents the mean of the denoised ATI phase of the clutter samples, Represents the correlation coefficient between channel l and channel m, and is estimated according to the following formula:

[0057]

[0058] in, Represents the estimated value of the correlation coefficient between channel l and channel m;

[0059] When there is a moving target signal, let the corresponding ATI phase detection value be ATI phase detection quantity after removing clutter The probability density function of is:

[0060]

[0061] Given the false alarm probability P fj , the corresponding detection threshold η j It can be obtained by the following formula

[0062]

[0063] Here, |·| represents the absolute value operation.

[0064] Furthermore, the steps of performing local detection on each detection quantity of each detector using the corresponding local detection threshold to obtain the binary detection result corresponding to each detector are as follows:

[0065] The first step is to perform local detection: for the AMF amplitude detection, determine whether the target exists according to T≥η0, where η0 represents the detection threshold; for any de-cluttered ATI phase detection, determine whether the target exists according to Determine whether the target exists; represents the ATI phase detection value of the jth detection unit, η j represents the detection threshold of the j-th detection unit;

[0066] In the second step, the amount of detection that passes the local limit is recorded as u j = +1, the detection amount that fails to pass the local detection threshold is recorded as u j=-1, j = 0, 1, 2, ... M, then the binary detection result corresponding to each detector is obtained.

[0067] Furthermore, the use of the optimal fusion rule refers to the optimal fusion rule based on Chair and Varshney.

[0068] Furthermore, the global detection amount of the target obtained by weighted fusion of the binary detection results corresponding to each detector is obtained by the following formula:

[0069]

[0070] Among them, a j Represents the weighted fusion coefficient obtained from the detection reliability of the local detection quantity.

[0071] Furthermore, the steps of adaptively estimating the detection amount corresponding to each detector for each potential target and estimating the target parameters of each potential target by the maximum likelihood method are as follows:

[0072] In the first step, the mean of the AMF amplitude estimation detection values ​​of all potential targets is used as the AMF amplitude estimation detection value;

[0073] In the second step, the estimated value of the ATI phase estimation detection quantity of each potential target is estimated according to the following formula:

[0074]

[0075] in, represents the estimated value of the AIT phase detection quantity of the qth potential target decluttering in the lth channel and the mth channel;

[0076] The third step is to estimate the channel correlation coefficient of each potential target according to the following formula:

[0077]

[0078] in, represents the estimated value of the channel correlation coefficient between the lth channel and the qth potential target of the mth channel, E[·] represents the expected operation, z l Represents the complex signal data of the lth channel, z m Represents the complex signal data of the mth channel.

[0079] Compared with the prior art, the present invention has the following advantages:

[0080] First, the present invention uses N SAR images to construct a group of multi-scale detectors, namely, AMF test and multi-baseline clutter removal ATI phase test, and estimates the distribution characteristics of each multi-scale detector, so that the receiver performance of each local test can be correctly predicted. The optimal local detection threshold is determined by global optimization through optimal fusion rule design, which overcomes the problems of excessively high false alarm probability of moving target detection and deterioration of system MDV caused by isolated strong clutter residues in a strong clutter background. The present invention has the ability to detect low signal-to-noise ratio and slow-moving targets, and improves the performance of multi-channel ATI-SAR in detecting small targets with low signal-to-noise ratio and small radial velocity in a strong clutter background.

[0081] Second, the present invention develops a cognitive detection framework that does not require prior target knowledge. Since the information fed back from the environment is more accurate than fixed information, the true target is retained, while the false positive targets are eliminated. This enables the present invention to achieve optimal fusion moving target detection of multi-channel ATI-SAR in a strong clutter background, effectively improving the detection performance of small targets with low signal-to-noise ratio and small radial velocity in a strong clutter background. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 is a flow chart of an embodiment of the present invention;

[0083] Figure 2 1 is a comparison diagram of the estimated error of the conventional ATI phase and the de-cluttered ATI phase according to the input signal-to-noise ratio (SCNR);

[0084] Figure 3 This is a curve showing how the detection probability of the comparison method changes with the false alarm probability when the signal-to-noise ratio of the moving target is 10dB. DETAILED DESCRIPTION

[0085] The present invention will be described in more detail below with reference to the accompanying drawings and embodiments.

[0086] Refer to the attached Figure 1 The specific implementation steps of the present invention are further described in the following embodiments.

[0087] Step 1: Construct AMF detection quantity.

[0088] Under the unknown target-oriented constraint, a robust clutter suppression method is adopted, and its filter weight vector w opt It is obtained by the following formula:

[0089]

[0090] Where R represents the maximum likelihood estimate of the clutter plus noise covariance matrix, a s represents the velocity steering vector of the moving target, ε represents the velocity steering vector set of the target of interest, v m Indicates the maximum unambiguous speed in the airspace.

[0091] The normalized filter weight vector obtained by solving is

[0092] After clutter suppression, the average output power of the background is estimated to be σ cn 2 =u H Such as.

[0093] After normalizing the filtered output power, the amplitude detection value is:

[0094]

[0095] Where z represents the complex image data vector of the pixel to be detected, n represents the number of views, ∑ represents the summation operation, (·) H represents the conjugate transpose operation, H0 represents the hypothesis that there is no target, H1 represents the hypothesis that there is a target, and η1 represents the amplitude detection threshold.

[0096] Step 2: Construct a multi-baseline clutter-free ATI phase detection quantity.

[0097] Step 2.1: For an N-channel ATI-SAR system, take the first channel as the reference channel and perform n-view DPCA processing on the lth channel and the first channel to obtain the signal after cancellation:

[0098]

[0099] Among them, y l represents the n-view DPCA processing results of the lth channel and the 1st channel, z l (k) represents the data corresponding to the kth pixel on the SAR image of the lth channel, and z1(k) represents the data corresponding to the kth pixel on the SAR image of the 1st channel. For N channels, n-1 independent cancellation signals can be obtained by performing n-view DPCA processing results, which are expressed as y l (k), l = 2, 3, ..., N, k represents the pixel number on the SAR image.

[0100] Step 2.2, using the cancellation signals to perform n-view ATI processing on each pair, we can obtain M = (N-2)(N-2) / 2 independent de-cluttered ATI phase detection quantities, with y l (k) and y m (k) Example, the corresponding de-cluttered AIT phase detection quantity is expressed as:

[0101]

[0102] In step 2.3, after the N-1 cancellation signals are processed by S21 and S22, the corresponding M=(N-2)(N-2) / 2 independent de-cluttered ATI phase detection quantities are:

[0103]

[0104] Step 3: Estimate statistical properties.

[0105] Step 3.1, under the background of non-uniform clutter, the probability density function of the S14 amplitude detection quantity T is:

[0106]

[0107] in, represents the equivalent view number, χ represents the degree of clutter non-uniformity, and χ > 0. A larger χ value indicates a more uniform clutter scene. The χ value is estimated from the second-order statistic m2 of the amplitude detection quantity, ν = (2m2n-(n+1)2) / (m2n-(n+1)2), and Γ(·) represents the Gamma function.

[0108] Step 3.2: When there is a moving target signal, let its amplitude detection value be ω, and the probability density function of the S14 amplitude detection value T be:

[0109]

[0110] where 2F1(·) represents the Gaussian hypergeometric function.

[0111] Step 3.3, under the clutter background, remove the clutter ATI phase detection quantity The probability density function of is:

[0112]

[0113] in, represents the mean of the denoised ATI phase of the clutter samples, Represents the correlation coefficient between channel l and channel m.

[0114] Step 3.4, The estimation formula is

[0115] Step 3.5: When there is a moving target signal, let the corresponding ATI phase detection value be ATI phase detection quantity after removing clutter The probability density function of is:

[0116]

[0117] Step 4: Initialize target parameters.

[0118] The estimated detection quantity of AMF is initialized as:

[0119]

[0120] Among them, σ s represents the minimum resolvable signal power, represents the space-time steering vector, σ n represents the minimum resolvable noise power, k scnr =σ s / σ cn Indicates the minimum resolvable signal-to-noise ratio of the array element channel.

[0121] Target clutter removal ATI phase estimation detection quantity Initialized as:

[0122]

[0123] in,[·] π represents the phase folding operation, λ represents the radar wavelength, and d l,m represents the baseline length, and V represents the target radial velocity.

[0124] Space-time steering vector Initialized as:

[0125]

[0126] Here, exp(·) represents the exponential operation with the natural constant e as the base, and i represents the imaginary unit symbol.

[0127] Initialize the correlation coefficient between the target lth channel and the mth channel:

[0128]

[0129] Wherein, ρ0 represents a constant, and in the embodiment of the present invention, ρ0=0.99.

[0130] Step 5: Global detection optimization.

[0131] Global detection optimization is performed using target parameters initialized by S4 and high false alarm.

[0132] For each set of local detection false alarm probabilities, it is not necessarily possible to obtain a good global detection result. Based on maximizing the global detection probability, the following optimization problem is established to obtain a set of optimal local detection false alarm probabilities:

[0133]

[0134] Among them, |·| represents the absolute value operation, P f and P dThey represent the global false alarm probability and detection probability respectively, r represents the false alarm probability given according to the receiver performance requirements, P fj represents the false alarm probability of the jth detection unit.

[0135] Given a set of candidate values ​​for the local detection false alarm probability, the global detection false alarm probability corresponding to S52 is:

[0136]

[0137] Where L represents the number of candidates with a given local false alarm probability that exceeds the global detection threshold, and D 1,l and D 0,l Respectively represent the set of identifiers that exceed the local detection threshold and the set of identifiers that do not exceed the local detection threshold in these L cases, l = 1, 2, ..., L, u j and u k Represents the detection result of the local detection unit, u j = +1 indicates that the local detection unit determines that there is a target, u k = -1 indicates that the local detection unit determines that there is no target, 1-P fk represents the correct judgment probability of the kth detection unit.

[0138] The corresponding global detection probability is:

[0139]

[0140] Among them, P d It represents the detection probability of global detection corresponding to the candidate value of local detection false alarm probability.

[0141] By traversing each candidate value of the local detection false alarm probability, the final optimized detection result can be obtained through S53 and S54.

[0142] Step 6: Identify potential targets.

[0143] For AMF local detection, given the false alarm probability P f0 , the corresponding amplitude detection threshold η0 and false alarm probability are as follows:

[0144]

[0145] Among them, f AMF (·) represents the probability density function of the AMF amplitude detection quantity.

[0146] For AMF local detection, under the detection threshold η1, the corresponding detection probability is:

[0147]

[0148] For any ATI phase detection quantity, Determine whether the target exists, η j Indicates the detection threshold.

[0149] Given the false alarm probability P fj , the corresponding detection threshold η j It can be obtained by the following formula:

[0150]

[0151] At the detection threshold η j The corresponding detection probability is calculated as follows:

[0152]

[0153] The optimal local detection P obtained after S5 global detection optimization fa Next, perform local detection.

[0154] Based on the local detection results, the detection amount passing the local limit is recorded as u j = +1, the detection amount that fails to pass the local detection threshold is recorded as u j =-1, j = 0, 1, 2, ... M, then the global detection quantity of the moving target is expressed as: Y = a0u0 + a1u1 + a2u2 + ... + a M u M ;a j It means that the weighting coefficient is determined by the detection reliability of the local detection quantity:

[0155]

[0156] Global detection is expressed as: Among them, η G represents the global detection threshold, η G =log(P(H0) / P(H1)), P(H1) and P(H0) respectively represent the prior probability of the existence or non-existence of the moving target. Without loss of generality, P(H1)=P(H0)=0.5.

[0157] Perform potential target determination and output potential target results.

[0158] Step 7: Adaptively estimate target parameters.

[0159] According to the potential target results, assuming there are Q potential targets, the target parameters are adaptively estimated for each potential target according to the maximum likelihood method.

[0160] is the average value of the AMF detection of Q potential targets, and the AMF detection of each target is:

[0161] Adaptive Estimation Where q∈{1,2,…,Q}.

[0162] Adaptive Estimation Where q∈{1,2,…,Q}.

[0163] Step 8: Update target parameters and optimize global detection;

[0164] The latent target parameters from step 7 are used to perform global detection optimization.

[0165] For each set of local detection false alarm probabilities, it is not necessarily possible to obtain a good global detection result. Based on maximizing the global detection probability, the following optimization problem is established to obtain a set of optimal local detection false alarm probabilities:

[0166]

[0167] Among them, P f and P d They represent the global false alarm probability and detection probability respectively, and r represents the false alarm probability given according to the receiver performance requirements.

[0168] Given a set of candidate values ​​for the local detection false alarm probability, the global detection false alarm probability corresponding to S82 is:

[0169]

[0170] The corresponding global detection probability is:

[0171]

[0172] Where L represents the number of candidates with a given local false alarm probability that exceeds the global detection threshold, and D 1,l and D 0,l They represent the set of identifiers that exceed the local detection threshold and the set of identifiers that do not exceed the local detection threshold in these L cases, l=1, 2,…, L.

[0173] By traversing each local detection false alarm probability candidate value, the final optimized detection result can be obtained through S84.

[0174] Step 9: Determine the final goal.

[0175] For AMF local detection, given the false alarm probability P f0 The corresponding relationship between the amplitude detection threshold η0 and the false alarm probability is as follows:

[0176]

[0177] For AMF local detection, under the detection threshold η1, the corresponding detection probability is:

[0178]

[0179] For any ATI phase detection quantity, Determine whether the target exists, where η j Indicates the detection threshold.

[0180] Given the false alarm probability P fj , the corresponding detection threshold η j It can be obtained by the following formula:

[0181]

[0182] At the detection threshold η j Next, calculate the corresponding detection probability according to the following formula:

[0183]

[0184] The optimal local detection P obtained after the global detection optimization in step 8 fa Next, perform local detection.

[0185] Based on the local detection results, the detection amount passing the local limit is recorded as u j = +1, the detection amount that fails to pass the local detection threshold is recorded as u j =-1, j = 0, 1, 2, ... M, then the global detection quantity of the moving target is expressed as: Y = a0u0 + a1u1 + a2u2 + ... + a M u M , where a j It means that the weighting coefficient is determined by the detection reliability of the local detection quantity:

[0186]

[0187] S98: Global detection is expressed as:

[0188]

[0189] Among them, η G represents the global detection threshold, η G =log(P(H0) / P(H1)), P(H1) and P(H0) represent the prior probability of the presence or absence of the moving target, respectively. Without loss of generality, P(H1)=P(H0)=0.5;

[0190] S99: Make final target determination and output the final target result.

[0191] The effects of the present invention are further described below in conjunction with simulation experiments:

[0192] 1. Simulation experiment conditions:

[0193] The environment of the simulation experiment of the present invention is: MATLAB R2021a, Intel(R)Core(TM)2Duo CPU 16GHz, and Windows 10 Professional Edition.

[0194] 2. Simulation content and result analysis:

[0195] The effectiveness of the algorithm is verified using simulation data.

[0196] The scene parameters in the simulation experiment are shown in Table 1. When the radar is working, the antenna transmits with full aperture and receives with sub-aperture. After the four channels receive the echo signals, they perform SAR imaging processing, radiation error correction, terrain interferometry phase compensation, and channel registration processing to obtain four SAR images. Then, we used the AMF detection method in the literature (F. Robey, D. Fuhrmann, E. Kelly, and R. Nitzberg, “A CFAR adaptive matched filter detector,” IEEE Transactions on Aerospace and Electronic Systems, vol. 28, no. 1, pp. 208–216, 1992.), the ATI phase detection method in the literature (CH Gierull, “Closed-form expressions for InSAR sample statistics and its application to non-Gaussian data,” IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 5, pp. 3967–3980, 2021.), and the ATI phase detection method in the literature (CH Gierull, I. Sikaneta, and D. Cerutti-Maori, “Two-step detector for RADARSAT-2's experimental GMTI mode,” IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 5, pp. 3967–3980, 2021.). The two-step detection method in Remote Sensing (2013) and the proposed method are used to detect moving targets. The detection probability (Pd) and false alarm probability (Pfa) of the two methods are statistically compared through 1 million Monte Carlo simulation experiments.

[0197] Table 1 Simulation experiment scenario parameters

[0198] name Numerical Radar wavelength 0.02 Platform speed 100m / s Baseline length 1m Number of channels 4 Clutter scene non-uniformity 3 CNR Changes between 20dB and 50dB Equivalent visual number 2 Moving target radial velocity 2m / s

[0199] Figure 2 : is a comparison diagram of the estimated error of the traditional ATI phase and the de-cluttered ATI phase of the present invention as a function of the input signal-to-noise ratio (SCNR) in the simulation experiment of the present invention, wherein: Figure 2 The right vertical axis in represents the output SCNR of the moving target after DPCA processing.

[0200] Figure 3 This is a curve showing how the detection probability of the comparison method changes with the false alarm probability when the signal-to-noise ratio of the moving target in the simulation experiment of the present invention is 10 dB.

[0201] from Figure 2 It can be seen that compared with the traditional ATI phase, the estimation error of the decluttered ATI phase of the present invention is smaller, especially when the moving target input signal-to-noise ratio (SCNR) is low, the accuracy is higher. Figure 3 It can be seen that under the same false alarm probability, the proposed method can achieve a better detection rate than the AMF detection method, the ATI detection method and the AMF+ATI two-step detection method, indicating that the proposed method has better detection performance.

[0202] The foregoing is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art may make various improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are also within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained herein shall, unless otherwise specified or limited, be implemented in accordance with conventional means in the art.

Claims

1. A moving target detection method based on optimal fusion of multi-channel ATI-SAR in a strong clutter background, characterized by: Using multi-channel ATI-SAR data, a multi-scale optimal fusion target detector was generated, and a cognitive detection framework without prior target knowledge was developed; the detection method includes the following steps: the following: S1: Use N SAR images to construct an M+1 scale detector, where N is equal to the number of radar channels, M=N(N-1) / 2; S2: Initialize each detector to obtain the target parameter K, perform the global detection optimization method on all detectors, and obtain the optimal local false alarm probability corresponding to each detector in the automatic perception stage; S3: By estimating the corresponding statistical characteristics of each detector, the corresponding local detection threshold is obtained according to the optimal local false alarm probability and the corresponding statistical characteristics. The corresponding local detection threshold is used to perform local detection on each detection quantity of each detector to obtain the corresponding binary detection result of each detector. The corresponding binary detection results of each detector are weighted and fused to obtain the global detection quantity of the target; the global detection quantity of the target is globally detected using the optimal fusion rule to determine the potential target; S4: Adaptively estimate the detection amount corresponding to each detector for each potential target through the maximum likelihood method, and estimate the target parameters of each potential target; using the target parameters of each potential target and the set global detection false alarm rate, perform the same global detection optimization method as step S2 to obtain the optimal local false alarm probability corresponding to each detector in the enhanced detection stage; S5: Using the same method as step S3, obtain the global detection value of each potential target; using the same optimal fusion rule as step S3, perform global detection on each potential target to determine the real target.

2. The moving target detection method of multi-channel ATI-SAR optimal fusion under strong clutter background according to claim 1, characterized in that: The M+1 scale detectors in step S1 are composed of one detector for AMF amplitude detection and M detectors for ATI phase detection.

3. The moving target detection method of multi-channel ATI-SAR optimal fusion under strong clutter background according to claim 2, characterized in that: The AMF amplitude detection quantity is expressed as: Where T represents the AMF amplitude detection quantity, n represents the number of views, ∑ represents the summation operation, k represents the ordinal number of the pixel in the SAR image, u represents the normalized filter weight vector, (·) H represents the conjugate transpose operation, z represents the data vector of the complex image point of the pixel to be detected, σ cn Indicates the minimum resolvable noise power.

4. The moving target detection method of multi-channel ATI-SAR optimal fusion under strong clutter background according to claim 3, characterized in that: The ATI phase detection quantity for removing clutter is expressed as: in, represents the phase detection amount after the ATI processing of the lth channel and the mth channel, arg[·] represents the phase angle, y l Indicates the n-view DPCA processing results of the lth channel and the 1st channel, represents the conjugate complex number of the n-view DPCA processing results of the m-th channel and the first channel, z l (k) and z1(k) represent the k-th pixel value in the SAR image corresponding to the l-th channel and the 1st channel, respectively.

5. The moving target detection method of multi-channel ATI-SAR optimal fusion under strong clutter background according to claim 4, characterized in that: Initializing each detector in step S2 means respectively initializing the AMF amplitude detection quantity, the target de-cluttering ATI phase estimation detection quantity, and the correlation coefficient between the target channels in the target parameter K; Initialize the AMF amplitude estimation detection quantity according to the following formula: in, represents the initialization detection quantity of the AMF amplitude estimation, σ s represents the minimum resolvable signal power, represents the space-time steering vector, σ n represents the minimum resolvable noise power, k scnr =σ s / σ cn Indicates the minimum resolvable signal-to-noise ratio of the array element channel; Initialize the target clutter removal ATI phase estimation detection quantity according to the following formula: in, represents the AIT phase estimation initialization detection quantity corresponding to the lth channel and the mth channel, [·] π represents the phase folding operation, v m represents the maximum unambiguous velocity in the airspace, λ represents the radar wavelength, and d l,m represents the baseline length, V represents the target radial velocity; Initialize the space-time steering vector according to the following formula: in, represents the initialization space-time steering vector, exp(·) represents the exponential operation with the natural constant e as the base, and i represents the imaginary unit symbol; Initialize the correlation coefficient between target channels according to the following formula: in, represents the correlation coefficient initialized between the target lth channel and the mth channel, ρ0 represents a constant, ρ0 = 0.

99.

6. The method for detecting moving targets by optimal fusion of multi-channel ATI-SAR in a strong clutter background according to claim 5, characterized in that: The steps of the global detection optimization method in step S2 are as follows: In the first step, the following optimization problem is established to obtain a set of optimal local detection false alarm probabilities: Among them, |·| represents the absolute value operation, P d represents the detection probability of global detection, P f represents the false alarm probability of global detection, r represents the global false alarm probability set according to the receiver performance requirements, P fj represents the false alarm probability of the jth detection unit; The second step is to calculate the global detection false alarm probability P corresponding to each set of local detection false alarm probability candidate values. f for: Among them, L represents the number of candidate values ​​of each group of local detection false alarm probability that exceeds the global detection threshold η G The number of η G =log(P(H0) / P(H1)), log represents the logarithmic operation with base 10, P(H1) and P(H0) represent the prior probability of the existence of the moving target, ∏· represents the symbol of the continuous multiplication operation, D 1,l 、D 0,l They represent the set of identifiers exceeding the local detection threshold and the set of identifiers not exceeding the local detection threshold in group l, l = 1, 2, ..., L, u j and u k Represents the detection result of the local detection unit, u j = +1 indicates that the local detection unit determines that there is a target, u k = -1 indicates that the local detection unit determines that there is no target, 1-P fk represents the correct judgment probability of the kth detection unit; The third step is to calculate the global detection probability P corresponding to the candidate value of the local detection false alarm probability. d for: Among them, P dj represents the detection probability of the jth detection unit, 1-P dk represents the missed detection probability of the kth detection unit; In the fourth step, by traversing each candidate value of the local detection false alarm probability, repeating the second and third steps and recording the global false alarm probability and detection probability, until the optimal local detection false alarm probability of the optimization problem in the first step is obtained.

7. The method for detecting moving targets by optimal fusion of multi-channel ATI-SAR in a strong clutter background according to claim 6, characterized in that: The steps of obtaining the corresponding local detection threshold according to the optimal local false alarm probability and the corresponding statistical characteristics in step S3 are as follows: The first step is to use the following formula to calculate the probability density function of the AMF amplitude detection quantity T under the background of non-uniform clutter: Among them, f AMF (·) represents the probability density function of the AMF amplitude detection quantity T, t represents the AMF amplitude detection quantity, Represents the equivalent view number, χ represents the non-uniformity of the clutter, χ>0, the larger the χ value, the more uniform the clutter scene is, and the χ value is estimated by the second-order statistic m2 of the AMF amplitude detection quantity. H0 represents the hypothesis without a target, Γ(·) represents the Gamma function; When there is a moving target signal, according to the following formula, let its corresponding AMF amplitude detection amount be ω, and the probability density function of the AMF amplitude detection amount T is: Where t represents the AMF amplitude detection quantity, H1 represents the hypothesis of the existence of the target, and 2F1(·) represents the Gaussian hypergeometric function; Given the false alarm probability P f0 , according to the following formula, the corresponding amplitude detection threshold is obtained: in, Indicates the calculation of the probability density function f from the threshold η0 to positive infinity AMF Cumulative probability of (·); The second step is to remove the clutter ATI phase detection quantity according to the following formula under the clutter background. The probability density function of is: Among them, f p (·) represents the amount of ATI phase detection after removing clutter The probability density function of represents the mean of the denoised ATI phase of the clutter samples, Represents the correlation coefficient between channel l and channel m, and is estimated according to the following formula: in, Represents the estimated value of the correlation coefficient between channel l and channel m; When there is a moving target signal, let the corresponding ATI phase detection value be ATI phase detection quantity after removing clutter The probability density function of is: Given the false alarm probability P fj , the corresponding detection threshold η j It can be obtained by the following formula: Here, |·| represents the absolute value operation.

8. The method for detecting moving targets by optimal fusion of multi-channel ATI-SAR in a strong clutter background according to claim 7, characterized in that: The steps of performing local detection on each detection quantity of each detector using the corresponding local detection threshold in step S3 to obtain the binary detection result corresponding to each detector are as follows: The first step is to perform local detection: for the AMF amplitude detection, determine whether the target exists according to T≥η0, where η0 represents the detection threshold; for any de-cluttered ATI phase detection, determine whether the target exists according to Determine whether the target exists; represents the ATI phase detection value of the jth detection unit, η j represents the detection threshold of the j-th detection unit; In the second step, the amount of detection that passes the local limit is recorded as u j = +1, the detection amount that fails to pass the local detection threshold is recorded as u j =-1, j = 0, 1, 2, ... M, then the binary detection result corresponding to each detector is obtained.

9. The method for detecting moving targets by optimal fusion of multi-channel ATI-SAR in a strong clutter background according to claim 8, characterized in that: The optimal fusion rule in step S3 refers to the optimal fusion rule based on Chair and Varshney.

10. The method for detecting moving targets by optimal fusion of multi-channel ATI-SAR in a strong clutter background according to claim 9, characterized in that: The global detection amount of the target obtained by weighted fusion of the binary detection results corresponding to each detector in step S3 is obtained by the following formula: Among them, a j Represents the weighted fusion coefficient obtained from the detection reliability of the local detection quantity.

11. The moving target detection method of multi-channel ATI-SAR optimal fusion under strong clutter background according to claim 10, characterized in that: The steps of adaptively estimating the detection amount corresponding to each detector for each potential target and estimating the target parameters of each potential target by the maximum likelihood method in step S4 are as follows: In the first step, the mean of the AMF amplitude estimation detection values ​​of all potential targets is used as the AMF amplitude estimation detection value; In the second step, the estimated value of the ATI phase estimation detection quantity of each potential target is estimated according to the following formula: in, represents the estimated value of the AIT phase detection quantity of the qth potential target decluttering in the lth channel and the mth channel; The third step is to estimate the channel correlation coefficient of each potential target according to the following formula: in, represents the estimated value of the channel correlation coefficient between the lth channel and the qth potential target of the mth channel, E[·] represents the expected operation, z l Represents the complex signal data of the lth channel, z m Represents the complex signal data of the mth channel.

Images