Through-the-wall radar ghosting recognition and suppression method based on template matching

Abstract

The invention discloses a through-the-wall radar ghosting identification and suppression method based on template matching, and the method comprises the steps: firstly, carrying out the matching of candidate targets at different positions of a radar image through the imaging templates of the targets at different positions, and identifying the ghosting and the targets according to the matching degree; furthermore, a ghosting mask is generated according to an identification result, so that the suppression of the ghosting is realized. The effectiveness of the algorithm is verified through simulation and actual measurement analysis. The method does not need any geometric prior information, and is a universal image domain ghosting identification and suppression method.

Description

Technical Field

[0001] The present invention belongs to the field of through-wall radar imaging detection, and relates to a method for identifying and suppressing ghost images of through-wall radar based on template matching. Background Art

[0002] In recent years, there has been a great demand for indoor human detection and positioning in both military and civilian fields, such as smart city detection, intelligent logistics, etc. Through-wall radar penetrates walls by emitting electromagnetic waves in the L / S band, and is a method for indoor human detection and positioning that can be used in non-cooperative environments.

[0003] During the propagation of electromagnetic waves in the indoor environment, reflections will occur on the surface of the inner wall. Since the wavelength of the electromagnetic waves emitted by the through-wall radar is in the decimeter range, which is much larger than the roughness of the wall surface, the electromagnetic waves are approximately specularly reflected on the wall surface, resulting in multipath propagation. Multipath propagation will cause false targets to appear in non-target positions in the radar image, which are called ghost images. The multipath ghost images seriously distort the image, increase the false alarm rate of target detection, and reduce the target positioning accuracy.

[0004] Many scholars have proposed methods for suppressing multipath ghost images. P. Setlur deduced the positions of multipath ghost images of SAR through-wall radar, and enhanced the target echo using multipath information, realizing the utilization of multipath. In addition, some scholars have used the method of sparse recovery to suppress multipath. M. Leigsnering proposed to reconstruct the image using the characteristics of group sparsity to achieve multipath suppression. S. Guo et al. imaged the target and multipath using different imaging dictionaries, and achieved ghost image suppression through multi-image fusion. In addition, S. Guo et al. proposed a similarity method, which can effectively suppress ghost images when the target and ghost images overlap. However, the above methods all require known building layout information. In the actual radar detection environment, it is difficult to obtain the prior information of the building layout, so the above methods are no longer applicable.

[0005] A class of methods that do not require prior information of the building layout is to identify and suppress ghost images by using the various parameter differences between the target and the ghost image. R. Feng analyzed the differences between the ghost image and the target in the range and velocity dimensions, and used the Hough transform to detect the target-first-order multipath-second-order multipath straight line in the R-V domain to realize the identification of the ghost image and the target. However, this method requires multiple cycle echo signals and relies on the existence of weak second-order ghost images. J. Park proposed to identify ghost images by using the differences between the target and the ghost image in the angle-velocity domain, but still requires multi-cycle information, and the real-time performance of the algorithm is limited. H. Luo and J. Kim have successively proposed to identify the target and the ghost image by using two-dimensional angle information. However, the methods are all used for millimeter-wave radar. When the radar band is in the L / S band, additional problems will occur in the near-field two-dimensional angle estimation.

[0006] Another method that does not require building geometric layout is the azimuth-dependent ghost suppression method. A. Abdalla proposed the azimuth-dependent ghost suppression method, which utilizes the characteristic that the ghost positions are different in different sub-apertures to suppress ghosts. However, this method requires a large array aperture. M. Wang analyzed the positions of the first-order ghosts in MIMO radar and proposed the method of relevant sub-apertures to achieve ghost suppression. J. Liu used the PCF method to achieve ghost suppression, but this method has an upper limit on ghost suppression. In MIMO radar, it is difficult to form a large aperture like SAR radar. Therefore, the azimuth-dependent ghost suppression method has limitations. Summary of the Invention

[0007] In view of this, the present invention provides a method for identifying and suppressing ghosts in a through-wall radar based on template matching, which is applied to through-wall radar imaging detection. First, potential target points are detected in the radar image. Subsequently, the candidate target area is matched with the template. The matching results are clustered to obtain real targets. Further, a mask is generated based on the matching results to suppress ghosts. The template used for matching is priori known when the radar is determined. This method does not require any information on building layout and is a general method for identifying and suppressing ghosts in radar images.

[0008] To achieve the above object, the method for identifying and suppressing ghosts in a through-wall radar based on template matching of the present invention includes the following steps:

[0009] Step 1, calculate the radar imaging result according to the radar echo;

[0010] The signal received by the radar is expressed as

[0011]

[0012] The received signal is down-converted using the de-chirp method, and the obtained signal is expressed as

[0013]

[0014] The detection scene is imaged using the time-domain coherent BP algorithm, and the imaging result at the imaging grid point C is expressed as

[0015]

[0016] Step 2, generate templates for different imaging grid points;

[0017] For the imaging grid point C, its coordinates are expressed as (θ C , r C ) in the theta-range image. The template corresponding to this grid point is expressed as

[0018]

[0019] Among them, ρ θ (θ C ) represents the angular resolution at grid point C, and ρ r represents the distance resolution at grid point C. Ω C represents the range of grid points occupied by the template at grid point C, which is expressed as

[0020]

[0021] Step 3, extract the candidate target area;

[0022] First, extract the local maximum of the radar image, and record the response size of the preliminarily screened candidate target points in the image. Among them, the amplitude of this point is commonly used to represent its response size, so the response at grid point C at this time is expressed as

[0023]

[0024] Process it using the non-maximum suppression method to obtain the set of candidate target point coordinates as follows

[0025]

[0026] For the nth candidate target, its area is expressed as

[0027]

[0028] Step 4, template matching;

[0029] Five methods for calculating the difference between matrices are considered to measure the matching degree between the candidate target area and the template, namely the normalized Manhattan distance, structural similarity, Pearson correlation coefficient, Chatterjee correlation coefficient, and KL divergence.

[0030]

[0031] To better fuse the calculation results of each index, two new indexes are defined as

[0032] M a = M nm ·M KL

[0033] M b = M SSMI ·M pcc ·M ccc

[0034] Step 5, ghost identification and suppression;

[0035] After calculating the metrics between each target candidate region and the template, the identification of the target and the ghost can be achieved using a clustering method, such as the K-mean algorithm. That is, assuming that the calculation result of the metrics corresponding to the nth target is M a (n) and M b (n), then the condition for the nth candidate region to be a target is

[0036] (M a (n),M b (n)) ∈ KM t

[0037] where KM t represents the cluster corresponding to the target, and its characteristic is that the center M a of this category is the smallest and M b is the largest. Therefore, the ghost mask is generated according to the position of the target as

[0038]

[0039] Therefore, the image after ghost suppression is

[0040] I o (C) = I(C) · I mask (C).

[0041] Beneficial effects:

[0042] 1. The present invention derives the expression of radar imaging and analyzes the differences between the target and the ghost; it can be obtained that both the target and the ghost can be modeled as ellipses in the radar image, and the major axes and rotation angles of the target ellipse and the ghost ellipse are different;

[0043] 2. The present invention proposes a ghost suppression method based on template matching. In this method, two metrics are used to measure the similarity between the candidate target and the template, and they are used in the clustering algorithm to identify the ghost;

[0044] 3. Compared with the existing methods, the present invention can achieve ghost suppression without any prior information on the geometric layout and can achieve good ghost suppression effects in the image domain. Description of the drawings

[0045] Figure 1 Flowchart of the method for identifying and suppressing ghosts in through-wall radar based on template matching of the present invention;

[0046] Figure 2 Scene diagram of through-wall radar detection;

[0047] Figure 3 Direct path and multipath propagation model;

[0048] Figure 4 Schematic diagram of echo path delay;

[0049] Figure 5 Index M a and M b Calculation result;

[0050] Figure 6 Simulation scenario diagram and radar array layout;

[0051] Figure 7 Ghost suppression result. (a) Original BP imaging result (b) PCF method (c) Sub-aperture method (d) Proposed method;

[0052] Figure 8 Measured data processing result (a) Schematic diagram of measured scenario (b) Original imaging result of the first group of measured data (c) Processing result of the proposed algorithm for the first group of measured data (d) Original imaging result of the second group of measured data (e) Processing result of the proposed algorithm for the second group of measured data. Detailed implementation manner

[0053] The present invention discloses a method for identifying and suppressing ghosts in a through-wall radar based on template matching. First, using the imaging templates of the target at different positions, the candidate target regions at different positions in the radar image are matched, and ghosts and targets are identified according to the matching degree. Further, a ghost mask is generated based on the identification result to achieve the suppression of ghosts. Simulation and measured analysis verify the effectiveness of the proposed algorithm. This method does not require any geometric prior information and is a general method for ghost identification and suppression in the image domain. The present invention will be described in detail below with reference to the accompanying drawings and by way of examples.

[0054] A method for identifying and suppressing ghosts in a through-wall radar based on template matching according to the present invention has a processing flow chart as Figure 1 shown, and the specific steps are as follows:

[0055] Step 1, calculate the radar imaging result according to the radar echo;

[0056] Consider a two-dimensional MIMO through-wall radar with Q transmitters and P receivers placed against the wall, and the detection scenario is as Figure 1 shown. Taking the radar center as the coordinate origin, the coordinates of the q-th (q = 1,..., Q) transmit antenna are (x q , 0, z q ), and the coordinates of the p-th receive antenna are (x p , 0, z q ), and this antenna pair is considered the k-th (k = 1,..., QP) channel. The radar transmits a linear frequency modulated continuous wave signal to detect the target behind the wall, which is expressed as

[0057] s t (t) = exp(j2πf0t + jπμt 2 )

[0058] Among them, f0 is the starting frequency of the transmitted signal, and μ = B / T p is the frequency modulation slope of the signal. B is the signal bandwidth, and T p is the pulse repetition period. There is a moving target behind the wall located at (x T , y T , 0). The present invention focuses on the identification and suppression of side wall ghosts. Therefore, the z coordinate will be ignored in the subsequent discussion. For the ghosts caused by the ceiling and the floor, their mechanisms and suppression methods are the same as those of the side wall, and the method proposed in the present invention is still applicable to the suppression of such multipath ghosts. The signal received by the radar is expressed as

[0059]

[0060] Among them, a rw represents the energy of the wall echo s rw (t). The wall energy is suppressed by the MTI method. Therefore, this component will no longer be considered in the subsequent discussion of this article. τ i = R i / c + Δτ i represents the time delay of the i-th (i = 00, 01, 10,...) direct path or multipath echo. R i represents the distance traveled by the electromagnetic wave in this path. c represents the speed of the electromagnetic wave propagating in a vacuum. Δτ i is the additional time delay of the electromagnetic wave caused by the wall and is removed by the compensation method. a ti represents the echo energy of the i-th path. N(t) represents the received noise. The received signal is down-converted using the de-chirp method, and the resulting signal is expressed as

[0061]

[0062] In the formula, (·) * represents the conjugate operation, and N'(t) represents the noise after the de-chirp operation. The detection scene is imaged using the time-domain coherent BP algorithm, and the imaging result at the imaging grid point C is expressed as

[0063]

[0064] In the formula, τ C represents the time delay between the q-th transmitting antenna, the p-th receiving antenna and the imaging grid point C, and its calculation method is similar to τ i .

[0065] The present invention uses Figure 3 to more intuitively show the propagation time delay relationship of different paths. Without loss of generality, Figure 3The reflective wall in it can be placed arbitrarily. There are three main paths in the scene. The first is the direct path of the radar irradiating the target, and its path number is 00:P q →P qa →P T →P pa →P p 。 Its path time delay τ 00 is calculated as

[0066]

[0067] In the formula, P q P T represents the path length from point P q to point P T , Δτ 00 represents the additional path brought by the wall for this path, and its approximate expression is

[0068]

[0069] where d fw and ε fw represent the thickness and dielectric constant of the front wall respectively. Assuming the target is isotropic and the radar cross-section is σ T , then the echo energy of this path is expressed as where A t represents the energy attenuation of the electromagnetic wave penetrating the wall, which is related to the transmission coefficient and is approximately a constant. The second path is that the radar emits electromagnetic waves, which are reflected by the wall and then irradiate the target, and then the reflected wave of the target is received by the receiver. The path number is 10:P q →P qb →P rq →P T →P pa →P p 。 Its path delay is calculated as τ 10 =(P q P rq +P rq P T +P T P p ) / c + Δτ 10 。 And there is Δτ 10 ≈Δτ 00 。 The echo energy of this path is expressed as where A r <1 represents the energy attenuation of the electromagnetic wave reflected on the wall surface, which is related to the wall reflection coefficient and is approximately a constant. Similarly, the third path is that the radar emits electromagnetic waves to irradiate the target, after being reflected by the target, it is reflected by the rear wall once, and then is received by the receiver. The path number is 01:P q →Pqa →P T →P rp →P pb →P p 。Its path delay is calculated as τ 01 =(P q P T +P T P rp +P rp P p ) / c + Δτ 01 。And Δτ 01 ≈Δτ 00 。The echo energy of this path is expressed as Actually, there are also multiple reflected waves in the detection scenario. Assume that the electromagnetic wave emitted by the radar has passed through r q times of wall reflections before reaching the target, and after being reflected by the target, it has passed through r p times of wall reflections and is received by the receiving antenna. Then the energy of this multipath Since A r < 1, so when r q +r p ≥ 2, Therefore, the influence of multiple reflected waves on the echo can be ignored.

[0070] In the literature, the positions of different ghosts have been deduced. The position of the first-order multipath depends on the position of the transmitting antenna. That is, when there are Q transmitting antennas in the radar, theoretically there are 2Q first-order ghosts. Among them, Q ghosts are located on the ellipse with the intersection of the line connecting the center of the receiving array and the target and the transmitting antenna and the target as the intersection points, and Q ghosts are located on the hyperbola with the intersection of the line connecting the center of the receiving array and the second-order ghost and the transmitting antenna and the second-order ghost as the intersection points. The position of the second-order ghost is axisymmetric with the target position about the reflecting surface. In this paper, it is considered too weak to be detected. Through the simulation of different paths, we have obtained the positions of the target and the ghosts, as Figure 4 shown.

[0071] In Figure 4 , the blue triangles represent the left transmitting antennas, the red triangles represent the right transmitting antennas, and the black circles represent the receiving antennas. The black asterisk in the figure represents the position of the target, the blue asterisks represent the two multipath ghosts corresponding to the left transmitting antenna, and the red asterisks represent the two multipath ghosts corresponding to the right transmitting antenna. The solid line segments in the figure represent the direct paths, that is, Figure 3 the path 00 in Figure 3 ; the dotted lines represent the transmitted direct paths and the received single-reflection paths, that is, Figure 3Path 10 in it. It can be seen that for the target, it involves all transmitting antennas and all receiving antennas, and for the ghost image, it involves a single transmitting antenna and all receiving antennas. We can draw the following conclusions.

[0072] ● For the target, it presents an elliptical shape in the radar image. The major axis of the target ellipse is the angular resolution of the entire array where is the wavelength corresponding to the center frequency. The minor axis of the target ellipse is the range resolution of the radar waveform The rotation angle of the ellipse is 0.

[0073] ● For the ghost image ellipse, the major axis of the ellipse is the angular resolution of the entire receiving array The minor axis of the target ellipse is the range resolution of the radar waveform The rotation angle of the target is

[0074] Step 2, generate templates for different imaging grid points;

[0075] For the imaging grid point C, its coordinates are represented as (θ C , r C ) in the theta-range image. The template corresponding to this grid point is represented as

[0076]

[0077] where ρ θ (θ C ) represents the angular resolution at grid point C, and ρ r represents the range resolution at grid point C. Ω C represents the range of grid points occupied by the template at grid point C, which is represented as

[0078]

[0079] In theta-range imaging, the range resolution is the same in different grids, but the azimuth resolution varies with the angle. Therefore, for grids with the same angle and different ranges, their corresponding templates are the same.

[0080] Step 3, extract the candidate target area;

[0081] Template matching is carried out in two steps. First, local maxima are extracted from the radar image, and the response magnitudes of the preliminarily screened candidate target points are recorded in the image. Among them, the magnitude of this point is commonly used to represent its response magnitude, and the response at grid point C at this time is represented as

[0082]

[0083] However, after this operation, many peaks will be generated, including the peak changes in the image caused by noise and some pseudo-peaks caused by the discontinuity of grid points. In order to further remove such peaks and further reduce the computational amount, the method proposed in the present invention uses the non-maximum suppression method to process them, so as to obtain the set of candidate target point coordinates as follows

[0084]

[0085] where C ct-n represents the nth candidate target, and N represents the number of candidate targets. For the nth candidate target, its region is expressed as

[0086]

[0087] After non-maximum suppression, the number of candidate target points is greatly reduced, further reducing the number of coordinates that need to be template-matched, and thus reducing the computational amount.

[0088] Step 4, template matching;

[0089] Through non-maximum suppression, the positions of candidate target points can be obtained. For the case where the candidate target is a real target, its candidate target region is similar to the template; while for the case where the candidate target point is a ghost, its region is different from the template. Five methods for calculating the difference between matrices are considered to measure the matching degree between the candidate target region and the template, namely the normalized Manhattan distance, structural similarity, Pearson correlation coefficient, Chatterjee correlation coefficient, and KL divergence.

[0090] · Normalized Manhattan distance

[0091]

[0092] In the formula, num represents the number of imaging grid points in the candidate region. The normalized Manhattan distance is used to measure the average difference between the imaging results of each imaging grid point and the template.

[0093] · Structural similarity

[0094]

[0095] In the formula, is the pixel mean value of the candidate region, is the pixel mean value within the template. is the pixel standard deviation of the candidate region, is the pixel standard deviation within the template, is the pixel covariance between the template and the candidate region. c1 and c2 represent two constants, and for the normalized template and candidate region images, their values are often taken as 0.0001 and 0.0009. Structural similarity is used to measure the similarity between two images, which conforms to the human visual system and is especially suitable for measuring the information of two images. When the candidate target region is consistent with the template, its value is close to 1. When the candidate target region is a ghost, the elliptical angle of its imaging amplitude is different from that of the template, so its structural consistency decreases.

[0096] · Pearson correlation coefficient

[0097]

[0098] The Pearson correlation coefficient is a relatively classical statistical index used to measure the linear correlation between two variables. This coefficient is insensitive to the size of the imaging range. For the candidate region corresponding to the target, it has a strong correlation with the template, while for the candidate region corresponding to the ghost, its correlation with the template is poor.

[0099] · Chatterjee correlation coefficient

[0100]

[0101] In the formula, R i is the order statistic of the candidate target region, that is, the sorting of the imaging result values within the candidate target region. D i represents the rank difference of the nearest neighbor value of R corresponding in the template i This correlation coefficient was newly proposed by Chatterjee of Stanford University. This correlation coefficient is not limited to the linear relationship between two variables, but can measure various relationships between two variables, and thus can more comprehensively measure the correlation between the candidate target region and the template.

[0102] · KL divergence

[0103]

[0104] In the formula, ln(·) represents the natural logarithm operation. When using KL divergence to measure the similarity between the candidate target region and the template, the two are regarded as two probability distributions, and the difference between the two probabilities is measured. In order to better integrate the calculation results of each index, two new indexes are defined as

[0105] M a = M nm · M KL

[0106] M b = M SSMI · M pcc · M ccc

[0107] Fifty Monte Carlo experiments were conducted to observe the calculation results of the fusion index between the template and the target candidate regions. In each experiment, the positions of the target and the wall were randomly generated. The experimental results are as Figure 5 shown, and the regions corresponding to the target are concentrated in the smaller M a value and the larger M b value.

[0108] Step 5, ghost identification and suppression;

[0109] After calculating the index between each target candidate region and the template, the clustering method, such as the K-mean algorithm, can be used to identify the target and the ghost. That is, assuming that the calculation results of the index corresponding to the nth target are M a (n) and M b (n), then the condition for the nth candidate region to be the target is

[0110] (M a (n), M b (n)) ∈ KM t

[0111] where KM t represents the cluster corresponding to the target, and its characteristic is that the center M a of this category is the smallest and M b is the largest. Therefore, the ghost mask is generated according to the position of the target as

[0112]

[0113] Therefore, the image after ghost suppression is

[0114] I o (C) = I(C) · I mask (C).

[0115] Example 1

[0116] A two-dimensional MIMO planar array is placed against the wall to verify the effectiveness of the proposed algorithm. The simulation scenario is as Figure 6 (a) shown. To simplify the discussion, it is assumed that only the multipath echoes of one side wall in the room can be received, and the energies of the remaining echoes are weak and thus cannot be received. Also, the echoes from the front wall have been removed by the preprocessing algorithm, such as the ACF method. The MIMO radar consists of four transmitting antennas and eight receiving antennas, and the radar array is as Figure 6(as shown in (b)). The transmitting antenna emits a frequency-modulated continuous wave to detect the target behind the wall. The starting frequency is 2.5 GHz, the bandwidth is 1 GHz, and the pulse repetition period is 1 ms. The side wall is a brick wall with an air layer of 0.24 m thickness, and when the center of the array is the origin, the equation corresponding to the position of the side wall is x = 5. The target is located behind the wall, and its position parameters are (θ t , r t ) = (5°, 5 m). The simulation experiment parameter table is shown in Table 1.

[0117] Table 1 Simulation Parameter Table

[0118]

[0119] To verify the effectiveness of the algorithm, we compare the proposed algorithm with the PCF method and the sub-aperture-based method. The images after ghost suppression by different methods are as Figure 7 shown. As Figure 7 (a) shows, since the array can be equivalent to two transmitting antennas on the left and right in the imaging area, there are 4 ghost targets in the original BP image, which is the same as the theoretical analysis. The ghosts in the figure will affect the detection and positioning of the target. As Figure 7 (b) shows, although the PCF method suppresses the ghosts to a certain extent, the suppression degree is limited, and the remaining ghosts will still affect the detection and positioning of the target. Figure 7 In (c), although the method using sub-apertures can also suppress the ghosts to a certain extent, due to the difficulty in comparing the separation degree between sub-apertures in the MIMO array with that in the SAR mode, the performance of this method is limited in MIMO radar. In Figure 7 (d), the proposed method correctly identifies the ghosts through template matching. Therefore, through the ghost mask, the ghosts can be completely removed, and only the target image is retained. It can be seen from the image that the ghost suppression effect of the proposed method is better than that of the existing methods.

[0120] Example 2

[0121] To verify the effectiveness of the proposed algorithm in measured data, a carefully designed MIMO is used to detect the target, and the test scenario is as Figure 8 (a) shows. Among them, the radar is set up at a position of 1.3 m, which is level with the chest position of an adult. The radar parameters are consistent with the simulation. The position of the wall in the x-y coordinate system is x = 2 m. Two groups of experiments are carried out to illustrate the effectiveness of the proposed algorithm. In the first group of experiments, the human target moves to the position of about (0°, 4 m), and in the second group of experiments, the human moves to the position of about (18.43°, 3.16 m). The algorithm processing results are as Figure 8 shown. It can be seen that the proposed algorithm can correctly identify and suppress the multipath ghosts in the presence of multipath ghosts, and then correctly detect the target.

[0122] In summary, the above are only the preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for identifying and suppressing ghost images of a through-wall radar based on template matching, characterized in that, It includes the following steps: Step 1: Calculate the radar imaging result based on the radar echo; Step 2: Generate templates for different imaging grid points; Step 3: Extract the candidate target area; Step 4: Template matching; Step 5: Ghost identification and suppression.

2. The method according to claim 1, wherein, In Step 2, the method for generating templates for different imaging grid points is as follows: For the imaging grid point C, whose coordinates are represented as (θ C , r C ) in the theta-range image; the corresponding template of this grid point is represented as where ρ θ (θ C ) represents the angular resolution at grid point C, and ρ r represents the distance resolution at grid point C; Ω C represents the range of grid points occupied by the template at grid point C, expressed as 3. The method according to claim 1, characterized in that, In Step 3, the method for extracting the candidate target area is as follows: Extract the local maximum of the radar image, and record the response size of the initially screened candidate target points in the image; usually, the amplitude of this point is used to represent its response size, so the response at grid point C at this time is expressed as where I(C) is the original radar image; process it using the non-maximum suppression method to obtain the set of candidate target point coordinates as follows Among them, C ct-n represents the nth candidate target, and (θ ct-n , r ct-n ) are the coordinates of this candidate target; for the nth candidate target, its area is expressed as 4. The method according to claim 1, characterized in that The template matching method in Step 4 is as follows: Five methods for calculating the difference between matrices are considered to measure the matching degree between the candidate target area and the template, namely the normalized Manhattan distance, structural similarity, Pearson correlation coefficient, Chatterjee correlation coefficient, and KL divergence: where num represents the number of imaging grid points within the candidate region; / num is the pixel mean of the candidate region, / num is the pixel mean within the template; is the pixel standard deviation of the candidate region, is the pixel standard deviation within the template, is the pixel covariance between the template and the candidate region; c1 and c2 represent two constants, and for the normalized template and candidate region images, their values are often taken as 0.0001 and 0.0009. Structural similarity is used to measure the similarity between two images, which conforms to the human visual system and is particularly suitable for measuring the information of two images; R i The order statistic of the candidate target region, that is, the sorting of the imaging result values within the candidate target region; D i represents the corresponding R in the template i The rank difference of the nearest neighbor value; ln(·) represents the natural logarithm operation.

5. The method according to claim 1, characterized in that, Two new indicators in Step 4 are defined as:

6. The method according to claim 1, wherein In step 5, the condition for the n-th candidate region to be the target is (M a (n), M b (n)) ∈ K M t Among them, M a (n) and M b (n) correspond to the settlement results of two new indicators defined in step 4, KM t represents the cluster corresponding to the target, and its feature is that the center M a of this category is the smallest and M b is the largest.

7. The method according to claim 1, wherein In Step 5, the ghost mask can be generated according to the position of the target as:

8. The method according to claim 1, wherein In Step 5, the image after ghost suppression is: I o (C) = I(C)·I mask (C).

Images