Track deviation estimation method for UAV-mounted through-wall radar based on image domain autofocus

Abstract

The present invention belongs to the field of through-wall radar, and more particularly relates to a method for estimating track deviations for unmanned aerial vehicle-mounted through-wall radars based on through-wall imaging image domain autofocusing. First, the Radon transform is used to integrate wall energy, and in the transform domain, image autofocusing is used to estimate track deviations in the line of sight. Secondly, maximum contrast autofocusing is performed on multiple point target areas distributed behind the wall to estimate track deviations along the course. This method can reduce the dimensionality of optimization variables, improve estimation efficiency, and fully utilize the space-varying slant range error information provided by targets at different distances and azimuths.

Description

Technical Field

[0001] The present invention belongs to the field of through-wall radar, and in particular relates to a method for estimating track deviation of a through-wall radar carried by an unmanned aerial vehicle based on self-focusing in a through-wall imaging image domain. Background Art

[0002] Small unmanned aerial vehicles (UAVs) offer excellent maneuverability and stealth, providing a new platform for battlefield reconnaissance and suitable for through-the-wall detection within high-rise buildings. However, small UAVs are more susceptible to interference from factors such as manual control and external wind gusts, resulting in track and attitude deviations, defocused imaging, and positioning errors. The low accuracy of their onboard global positioning systems (GPS) and inertial navigation systems (INS) does not meet the requirements for motion compensation. Therefore, research is urgently needed on track deviation estimation algorithms for UAV-mounted through-the-wall radars to provide technical support for the subsequent realization of high-resolution target imaging and positioning tracking.

[0003] Currently, airborne radar motion error estimation methods are primarily divided into parametric algorithms, such as image offset autofocusing, maximum image contrast autofocusing, and multi-aperture correlation, and nonparametric methods, such as phase gradient autofocusing and weighted least squares. These algorithms all estimate and compensate for space-invariant phase errors. To address the problem of estimating space-variant errors, researchers have proposed wide-beam motion error estimation and compensation algorithms. These algorithms use time-frequency segmentation to ignore the space-variance of the error within a sub-block and then concatenate the estimated results from multiple sub-blocks. However, this can introduce errors at the sub-block junctions, and the number of sub-blocks increases dramatically in close-range airborne through-wall radar detection scenarios, making it difficult to meet practical application requirements. Small UAV radars have small antennas and wide beamwidths, resulting in the two-dimensional space-variance of slant range error that cannot be ignored. Furthermore, their complex flight paths make traditional polynomial phase error models too high-order. Furthermore, the UAV's track deviation can be significantly greater than the range resolution, significantly impacting the envelope error. Traditional SAR imaging methods struggle to decouple range migration and phase error estimation. The main problem is that existing techniques cannot effectively estimate the motion error of UAV through-wall radars. Summary of the Invention

[0004] To solve the above problems, the present invention proposes a track deviation estimation method for UAV-mounted through-wall radar based on image domain autofocusing. When the measurement trajectory obtained by the GPS system has deviations, the track deviation value can be accurately estimated and a focused image can be obtained when the error space-varying characteristics are significant.

[0005] A method for estimating track deviation of a UAV-mounted through-wall radar based on image domain autofocusing includes the following steps:

[0006] In step S1, the drone-mounted radar moves toward the building wall while detecting and acquiring radar echoes. This echo is then processed into a range-compression signal. The intensity of the range-compression signal is then compensated based on the distance to obtain an energy-compensated range-compression signal. The range-compression signal includes a wall component and a target component behind the wall.

[0007] S2: Based on the refraction theorem and geometric relationship, the path length of the electromagnetic wave after refraction through the wall is approximately obtained;

[0008] S3: BP imaging of the building area is performed using the energy-compensated distance compression signal and GPS coarse positioning data obtained in S1, as well as the path length of the electromagnetic wave after refraction through the wall obtained in S2. The BP imaging results include the wall and the distributed targets behind the wall.

[0009] S4: Perform Radon transform on the wall imaging result obtained in S3 to obtain R f Transform domain image;

[0010] S5: R obtained according to S4 f Transform domain image and GPS coarse positioning data, with image entropy as the optimization objective function, the particle swarm optimization algorithm is used to obtain the line of sight track deviation vector Δy=[Δy1,Δy2,…Δy l ] and the BP imaging results of the wall part with restored straight line characteristics;

[0011] S6: Using the mean cancellation method on the BP imaging result of the wall portion with restored linear characteristics obtained in S5 to suppress wall clutter, obtaining an imaging result of the target portion behind the wall after clutter suppression;

[0012] S7: The corrected radar positioning data is obtained by the line-of-sight direction track deviation vector obtained in S5. Based on this, the image contrast is used as the optimization objective function. The particle swarm optimization algorithm is used to iteratively obtain the track deviation vector Δx=[Δx1, Δx2,…Δx l ] and the focused image of the target area;

[0013] In step 1, the range compression signal obtained at each azimuth sampling point is used to compensate the echo energy of the target at a longer distance by using the compensation factor:

[0014] Azimuth sampling point P t,a The FMCW range compression echo signal at is:

[0015]

[0016] Where T p is the pulse width, f r =-2kr a / c is the distance frequency, k is the frequency modulation slope, λ is the wavelength, r a is the target slant range. The range compression echo signal at each azimuth sampling point is multiplied by the compensation factor β:

[0017]

[0018] Where n is the range unit, N is the number of FFT points in the range direction, and α is a hyperparameter. The criterion for selecting α is that the energy of the target at different distances is as close as possible after compensation.

[0019] In step 2, the path length of the electromagnetic wave after refraction through the wall is approximately obtained based on the refraction theorem and geometric relationship:

[0020] The path of an electromagnetic wave emitted by radar T1 (X1, Y1), through the free space in front of the wall and refracted by the wall, to the target T2 (X2, Y2) behind the wall is equivalent to the case where the radar is directly adjacent to the wall, simplifying the calculation into two paths. Assuming the relative dielectric constant of the wall is ε and the thickness is d, and assuming the distance between the target and the antenna is much greater than the wall thickness, the slant distance after refraction from the wall is:

[0021]

[0022]

[0023] Where ‖T1T2‖ is the straight-line distance between the radar and the target, θ i is the incident angle on the free space side, θ t is the refraction angle on the inner side of the wall, θ q is the angle between the line between the radar and the target and the normal.

[0024] In step 3, the method for performing preliminary imaging of the building area is:

[0025] The imaging area is divided into grid points, and the grid point size is slightly smaller than the radar range and azimuth resolution. t,a The pixel value of the obtained range compression signal projected onto the grid point where the target T2 is located is:

[0026]

[0027] The BP imaging result at the grid point where the target T2 is located is obtained by coherently superposing multiple azimuth sampling points:

[0028]

[0029] where N a is the number of sampling points in the azimuth direction. The same operation is performed on each grid point in the imaging area to obtain the BP imaging result of the building area.

[0030] In the fourth step, the method of performing Radon transformation on the imaging result of the wall portion is:

[0031] Each pixel point of the wall part of the BP imaging result is projected to R according to ρ and θ f domain, where ρ is the distance from the pixel to the reference point, and θ is the rotation angle, ranging from 0° to 180°.

[0032] The straight line in the wall imaging result is represented by the function f(x,y), and its corresponding Radon transform R f (ρ,θ) is obtained by performing a line integral on the straight line defined by ρ and θ.

[0033] R f (ρ,θ)=∫f(x,y)dl

[0034] R f The extreme point is obtained by integrating the distance ρ and angle θ corresponding to the wall line in the transform domain.

[0035] In step 5, the particle swarm optimization algorithm is used to iteratively obtain the line of sight direction track deviation vector Δy=[Δy1, Δy2, ... Δy l ] and a partially focused image of the wall:

[0036] The optimization objective function is R f The image entropy S of the domain.

[0037] argminS=-∑ω(ρ,θ)logω(ρ,θ)

[0038]

[0039] Take the track deviation vector Δy at L azimuth sampling points = [Δy1, Δy2, ... Δy l ] as a particle, which is considered to represent the changing trend of the track deviation curve, and is iterated. In each iteration, through Y a =Y m +Δy to get the true position of the radar in the direction of sight at each sampling point, where Y a and Y m They are the real position of the radar in the line of sight and the GPS rough positioning position, and then an R is obtained through the BP imaging algorithm. f Domain image.

[0040] In each iteration, the particle tracks the global optimal solution p gd and the individual optimal solution p id Two extreme values ​​to update their own speed and position:

[0041]

[0042]

[0043] When the difference between the objective function values ​​obtained in consecutive iterations is small enough, the image is considered to be focused.

[0044] In step six, a mean cancellation method is used to suppress wall clutter.

[0045] Based on the BP imaging result of the wall portion with restored linear characteristics obtained in step 5, all pixels are averaged along the azimuth direction to obtain the average pixel, and then the average pixel is subtracted from each azimuth pixel to suppress the wall signal.

[0046]

[0047] Among them, I(x i ,y j ), I′(x i ,y j ) respectively represent (x i ,y j ) before and after mean cancellation. N represents the total number of pixels in the azimuth direction.

[0048] In step 7, the particle swarm optimization algorithm is used to iteratively obtain the track deviation vector Δx=[Δx1, Δx2,…Δx l ] and the target area focused image.

[0049] The image contrast of the imaging area where multiple distributed targets are located is used as the optimization objective function:

[0050]

[0051]

[0052] Where I(x,y) is the pixel value at position (x,y) in the BP imaging result, and the deviation vector Δx along the heading track is taken as [Δx1, Δx2,…Δx l ] is iterated as a particle, and in each iteration, X a =X m +Δx to obtain the true position of the radar along the course at each sampling point, where X a and X m The radar's true position along the course and the GPS coarse positioning position are respectively used to re-obtain the BP imaging results of a target area. The image is considered to be focused until the difference between the objective function values ​​obtained in consecutive iterations is less than the set minimum value ε.

[0053] Beneficial effects:

[0054] 1. This invention provides an autofocus optimization method based on through-wall BP imaging results. The optimization independent variables are the track offsets Δx and Δy at each radar sampling point. This method transforms the space-variable phase error estimation problem into the absolute magnitude estimation problem of the track offset, eliminating the need for block splicing and prior information on the error distribution. Furthermore, the optimization process simultaneously considers the effects of both envelope and phase. Simulation and field experiments demonstrate that this method can adapt to situations where the track deviation of small UAVs is significantly greater than the range resolution. Compared with traditional algorithms that only consider phase error, this method is more accurate, achieving track deviation estimation accuracy at the range resolution level.

[0055] 2. The present invention provides a step-by-step track deviation estimation method based on the linear characteristics of a wall and the distributed targets behind it. First, the Radon transform is used to integrate the wall energy. In the transform domain, the line-of-sight track deviation is estimated using image autofocusing. Secondly, maximum contrast autofocusing is performed on multiple point target areas distributed behind the wall to estimate the track deviation along the course. This method reduces the dimensionality of the optimization variables, improves estimation efficiency, and fully utilizes the space-varying slant range error information provided by targets at different distances and azimuths. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is the through-wall radar scene signal model on which the method of the present invention is based;

[0057] Figure 2 is a signal processing flow chart of an embodiment of the present invention;

[0058] Figure 3 is a schematic diagram of the wall echo on which the method of the present invention is based;

[0059] Figure 4 This is a schematic diagram of the refraction path of electromagnetic waves through walls, which is the basis of the method of the present invention;

[0060] Figure 5 Schematic diagram of target echo behind the wall based on the method of the present invention;

[0061] Figure 6 is a scene graph simulated by the method of the present invention;

[0062] Figure 7 : These are the result diagrams of the wall portion obtained by processing the simulation data using the method of the present invention, wherein (a) is the wall imaging result under ideal conditions, (b) is the wall imaging result when there is a track deviation, (c) is the wall imaging result after compensation using the method of the present invention, (d) is the Radon transform under ideal conditions, (e) is the Radon transform when there is a track deviation, and (f) is the Radon transform after compensation using the method of the present invention.

[0063] Figure 8It is the line-of-sight direction track deviation estimation result of processing simulation data by the method of the present invention, wherein (a) is the line-of-sight direction track deviation estimation value, and (b) is the residual line-of-sight direction track deviation.

[0064] Figure 9 1 is a diagram of the target part of the result of processing the simulation data using the method of the present invention, wherein (a) is the target RD imaging result under ideal conditions, (b) is the target RD imaging result when there is a track deviation, (c) is the target imaging result after compensation by the comparison algorithm (Improved Minimum-Entropy Autofocus Algorithm), (d) is the target BP imaging result under ideal conditions, (e) is the target BP imaging result when there is a track deviation, and (f) is the target imaging result after compensation by the method of the present invention.

[0065] Figure 10 It is the along-track deviation estimation result of processing the simulation data by the method of the present invention, wherein (a) is the along-track deviation estimation value, and (b) is the residual along-track deviation.

[0066] Figure 11 It is a scene diagram taken by the method of the present invention;

[0067] Figure 12 Figure 3 is a partial result diagram of the wall obtained by processing the measured data using the method of the present invention, wherein (a) is the wall imaging result under ideal conditions, (b) is the wall imaging result when there is a track deviation, (c) is the wall imaging result after compensation using the method of the present invention, (d) is the Radon transform under ideal conditions, (e) is the Radon transform when there is a track deviation, and (f) is the Radon transform after compensation using the method of the present invention.

[0068] Figure 13 It is the line-of-sight direction track deviation estimation result of processing the measured data by the method of the present invention, wherein (a) is the line-of-sight direction track deviation estimation value, and (b) is the residual line-of-sight direction track deviation.

[0069] Figure 14 : It is the target part result diagram of the simulation data processed by the method of the present invention, wherein (a) is the target imaging result under ideal conditions, (b) is the target BP imaging result when there is a track deviation, and (c) is the target imaging result after compensation by the method of the present invention.

[0070] Figure 15 It is the along-track deviation estimation result of processing the measured data by the method of the present invention, wherein (a) is the along-track deviation estimation value and (b) is the residual along-track deviation. DETAILED DESCRIPTION

[0071] This invention aims to address the inaccurate positioning of small drones and the inaccurate motion error estimation of existing methods for small drone-mounted through-wall radars under close-range detection conditions. By proposing a method for estimating the track deviation of drone-mounted through-wall radars based on image domain autofocus, this method is applicable to scenarios where multiple targets are located behind a straight, uniform wall with known thickness and material, providing technical support for subsequent high-resolution imaging and positioning tracking of targets.

[0072] Figure 1 This is the wall-penetrating radar signal model based on the present invention. The drone-mounted radar moves along the wall in the x-axis direction, and the beam illumination direction is the line of sight in the y-axis direction. The solid curve is the actual track of the drone-mounted radar, and the dotted curve is the measurement track obtained by GPS. The sampling point P at a certain azimuth is t,a Since the error is located at P t,m , there are track deviations Δx, Δy. For the detected target T2, the actual slant range is r a The solid line shown, the measured slope distance obtained according to the measurement position is r m Dotted line shown. Figure 2 This is a signal processing flow chart of an embodiment of the present invention. The present invention is based on the case of a single-transmit and single-receive through-wall radar. After acquiring the echo, the present invention is implemented through the following steps:

[0073] Step 1: Compensate for weak target echo energy;

[0074] The drone-mounted radar moves along the building wall while detecting and acquiring radar echoes. Considering that the farther the target is from the wall, the smaller its echo energy is, and it may be ignored in the optimization process, the echo of each channel is multiplied by the compensation factor:

[0075]

[0076] Where k is the distance unit, and α is a hyperparameter. The target amplification factor increases with distance. The criterion for selecting α is that the energy of the target at different distances is as close as possible after compensation.

[0077] Step 2: Calculate the refraction path of electromagnetic waves through the wall;

[0078] like Figure 4 As shown in the figure, the electromagnetic wave emitted by radar T1, refracted by the wall, and reaching target T2 behind the wall has three paths: in front of the wall, inside the wall, and behind the wall. It can be deduced that the slant range between the radar and the target is independent of the distance between the radar and the wall. This is equivalent to the radar being in close contact with the wall, simplifying the calculation into two paths. Assuming the relative dielectric constant of the wall is ε and the thickness is d, the one-way propagation path of the electromagnetic wave is:

[0079]

[0080] θi and θ t Satisfies Snell's law:

[0081]

[0082] Since the distance between the target and the antenna is much greater than the wall thickness, the following applies:

[0083]

[0084] Also because

[0085]

[0086] The final slope distance R is:

[0087]

[0088]

[0089] Step 3: BP imaging of building area;

[0090] The imaging area is divided into grid points, the slant range from the radar to the grid points is calculated, and the signals at the corresponding distances of the one-dimensional range echoes at multiple azimuth sampling points are coherently superimposed to obtain the two-dimensional imaging results.

[0091] Azimuth sampling point P t,a The FMCW range compression echo signal at is:

[0092]

[0093] Where T p is the pulse width, f r is the distance frequency, k is the frequency modulation slope, and λ is the wavelength. t,a The pixel value of the resulting echo projected onto the target T2 is:

[0094]

[0095] It can be seen that due to the existence of track deviation, the phase compensation function H cannot completely cancel the phase of the echo, and the envelope is also affected.

[0096] The BP imaging results are obtained by coherently superimposing multiple azimuth sampling points:

[0097]

[0098] where N a is the number of sampling points in azimuth.

[0099] Step 4: Radon transform of the wall part of the imaging result;

[0100] Radon transform R corresponding to f(x,y) f (ρ,θ) is obtained by performing a line integral along the line defined by ρ and θ.

[0101] R f (ρ,θ)=∫f(x,y)dl

[0102] Since the wall has a straight geometric characteristic, its imaging result is transformed into R f The extreme points are obtained by integrating the corresponding positions in the transform domain.

[0103] Step 5: Particle swarm optimization algorithm is iterated to achieve wall imaging self-focusing;

[0104] The optimization objective function is R f The image entropy S of the domain.

[0105] argminS=-Σω(ρ,θ)logω(ρ,θ)

[0106]

[0107] Since the wall echo depends only on the nearest slant distance, the line of sight track deviation Δy affects the wall imaging result, as shown in Figure 3 As shown. Take the track deviation vector Δy at L azimuth sampling points = [Δy1, Δy2, ... Δy l ] as a particle, which is considered to represent the changing trend of the track deviation curve, and is iterated. In each iteration, through Y a =Y m +Δy to get the actual position of the azimuth sampling point, where Y a and Y m are the true Y-axis position and measured position of the radar respectively.

[0108] In each iteration, the particle tracks the global optimal solution p gd and the individual optimal solution p id Two extreme values ​​to update their own speed and position:

[0109]

[0110]

[0111] When the difference between the objective function values ​​obtained from the continuous iterations is small enough, the image is considered to be focused. At the same time, this step obtains the line of sight track deviation vector Δy=[Δy1,Δy2,…Δy l ].

[0112] Step 6: Mean cancellation to suppress wall clutter;

[0113] Track deviation along the course requires estimation using targets behind walls. Wall clutter can affect targets close to the wall. After line-of-sight track deviation estimation and compensation, the wall imaging result essentially restores a straight-line geometry. All pixels along the azimuth are averaged to obtain the average pixel, which is then subtracted from each azimuth pixel to suppress the wall signal.

[0114]

[0115] Among them, I(x i ,y j ), I′(x i ,y j ) respectively represent (x i ,y j ) before and after mean cancellation. N represents the total number of pixels in the azimuth direction.

[0116] Step 7: Particle swarm optimization algorithm is iterated to achieve self-focusing of target area imaging;

[0117] like Figure 5 As shown in Figure 1, after the line-of-sight direction track deviation is compensated, the residual along-track deviation is estimated by multiple targets distributed behind the wall. Similar to step 3, the along-track deviation vector Δx = [Δx1, Δx2, ... Δx l ] as particles for iteration. The wider the target distribution, the more sufficient the space-variant information provided, and the more accurate the estimated along-course track deviation. This completes a track deviation estimation method for UAV-mounted through-wall radar based on image domain autofocusing.

[0118] In order to verify the track deviation estimation method of UAV-borne through-wall radar based on image domain autofocusing proposed in this invention, simulation and actual measurement experiments were designed for analysis and verification.

[0119] Simulation scenario such as Figure 6 The simulation parameters are shown in Table 1:

[0120] Table 1 Simulation parameter settings

[0121]

[0122] The drone-mounted radar moves along the wall, and the track deviation is set to a sinusoidal curve with a maximum deviation value A of 0.4m:

[0123]

[0124] Figure 7 This is the imaging result of the wall part estimated and compensated by the method of the present invention. After compensation, the linear characteristics of the wall are basically restored, corresponding to R f The transform domain peak is also focused. Figure 8This is the estimation result of the line-of-sight track deviation. It can be seen that the estimated curve is basically consistent with the actual deviation curve. The maximum residual error is 0.14m, the compensation accuracy is at the range resolution level, and the variance is 0.0013. Figure 9 The image shows the target area imaging result. The comparison algorithm is an improved Minimum-Entropy Autofocus Algorithm. Based on this algorithm, distance segmentation is used to improve the accuracy of distance space-varying error estimation. It can be seen that after compensation, the comparison algorithm produces false targets around the target, and residual linear phase causes deviation in the target position. After compensation using the method of the present invention, the target pixels are focused and the target position is correct. The weak arc-shaped shadows around the target are caused by coherent superposition of the BP algorithm. Figure 10 The estimated results of the track deviation along the course are shown in Figure 2. The obtained curve is basically consistent with the actual deviation curve, with a maximum residual error of 0.05m, a compensation accuracy of the range resolution level, and a variance of 2.898×10 -4 The above results show that the method of the present invention can effectively self-focus the imaging results when the track deviation is much larger than the range resolution, and the estimated track deviation accuracy reaches the range resolution level.

[0125] The measured scene is as follows Figure 11 As shown in Table 2, the experiment uses a stepped frequency signal, and the parameters are shown in Table 2:

[0126] Table 2 Measured parameter settings

[0127]

[0128] The track deviation is also a sinusoidal curve, with the maximum deviation value being 0.5m. Figure 12 The imaging result of the wall part estimated and compensated by the method of the present invention is consistent with the simulation effect. After compensation, the linear characteristics of the wall are restored. f The domain peak is focused. Figure 13 The estimated results of the track deviation in the line of sight direction show that the estimated curve is basically consistent with the actual deviation curve. The maximum residual error is 0.07m, the compensation accuracy is at the range resolution level, and the variance is 2.3404×10 -4 . Figure 14 This is the imaging result of the target area. After compensation using the method of the present invention, the target pixels are focused and the target position is correct. Figure 15 The curve obtained is the result of the estimated along-course track deviation. It is consistent with the actual deviation curve, with a maximum residual error of 0.1m, a compensation accuracy of the range resolution level, and a variance of 0.0057. These results show that in the actual measurement experiment, the method of the present invention can still autofocus the imaging results when the track deviation is much larger than the range resolution, and the estimated track deviation accuracy reaches the range resolution level.

[0129] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for estimating track deviation of UAV-mounted through-wall radar based on image domain autofocus, characterized in that: The following steps are involved: In step S1, the drone-mounted radar moves toward the building wall while detecting and acquiring radar echoes. The detected radar echoes are processed into range-compression signals. The intensity of the range-compression signals is compensated according to the distance to obtain an energy-compensated range-compression signal. The range-compression signal includes a wall component and a target component behind the wall. S2, obtains the path length of the electromagnetic wave after refraction through the wall; S3, performing BP imaging of the building area based on the GPS coarse positioning data, the energy-compensated distance compression signal obtained in step S1, and the path length of the electromagnetic wave after refraction through the wall obtained in step S2, wherein the BP imaging result includes the wall and the distributed targets behind the wall; S4, performing Radon transformation on the wall portion imaging result obtained in step S3 to obtain R f Transform domain image; S5, based on the GPS coarse positioning data and the R obtained in step S4 f Transform domain image, with image entropy as optimization objective function, adopt particle swarm optimization algorithm to obtain the sight direction track deviation vector Δy=[Δy1,Δy2,…Δy l ] and the BP imaging results of the wall part with restored straight line characteristics; S6, using the mean cancellation method to suppress the wall clutter on the BP imaging result of the wall portion with restored linear characteristics obtained in S5, and obtaining the imaging result of the target portion behind the wall after clutter suppression; S7, obtain the corrected radar positioning data through the line of sight direction track deviation vector obtained in step S5, and use the image contrast as the optimization objective function to iteratively use the particle swarm optimization algorithm to obtain the tracking deviation vector Δx=[Δx1, Δx2, ... Δx] for the imaging result of the target behind the wall after clutter suppression obtained in step S6. l ] and the target area focused image.

2. The method for estimating track deviation of a UAV-mounted through-wall radar based on image domain autofocusing according to claim 1, characterized in that: In step 1, the range compression signal obtained at each azimuth sampling point is used to compensate the echo energy of the target at a longer distance by using the compensation factor. t，a The FMCW range compression echo signal at is: Among them, T p is the pulse width, f r =-2kr a / c is the distance frequency, k is the frequency modulation slope, λ is the wavelength, r a To be the target slant range, multiply the range compression echo signal at each azimuth sampling point by the compensation factor β: Where n is the distance unit, N is the number of FFT points in the range direction, and α is a hyperparameter. The selection criterion of α is that the energy of targets at different distances after compensation is as close as possible.

3. The method for estimating track deviation of a UAV-mounted through-wall radar based on image domain autofocusing according to claim 1 is characterized by: In step 2, the method for approximately obtaining the path length of the electromagnetic wave after refraction through the wall according to the refraction theorem and geometric relationship is: The path of the electromagnetic wave emitted by radar T1 (X1, Y1) through the free space in front of the wall and refracted by the wall to reach the target T2 (X2, Y2) behind the wall is equivalent to the case where the radar is close to the wall. It is simplified into a two-segment path calculation. Assuming that the relative dielectric constant of the wall is ε and the thickness is d, and assuming that the distance between the target and the antenna is much greater than the wall thickness, the slant distance after refraction from the wall is: Among them, ||T1T2|| is the straight-line distance between the radar and the target, θ i is the incident angle on the free space side, θ t is the refraction angle on the inner side of the wall, θ q is the angle between the line between the radar and the target and the normal.

4. The method for estimating track deviation of a UAV-mounted through-wall radar based on image domain autofocusing according to claim 1, characterized in that: In step 3, the method for performing BP imaging on the building area is: The imaging area is divided into grid points. The grid point size is slightly smaller than the radar range and azimuth resolution. The azimuth sampling point P t，a The pixel value of the obtained range compression signal projected onto the grid point where the target T2 is located is: The BP imaging result at the grid point where the target T2 is located is obtained by coherently superposing multiple azimuth sampling points: where N a is the number of sampling points in the azimuth direction. The same operation is performed on each grid point in the imaging area to obtain the BP imaging result of the building area.

5. The method for estimating track deviation of a UAV-mounted through-wall radar based on image domain autofocusing according to claim 1, characterized in that: In the fourth step, the method of performing Radon transformation on the imaging result of the wall portion is: Each pixel point of the wall part of the BP imaging result is projected to R according to ρ and θ f Domain, where ρ is the distance from the pixel to the reference point, and θ is the rotation angle, ranging from 0° to 180°; The straight line in the wall imaging result is represented by the function f(x, y), and its corresponding Radon transform R f (ρ, θ) is obtained by performing a line integral on the straight line defined by ρ and θ; R f (ρ,θ)=∫f(x,y)dl R f The extreme point is obtained by integrating the distance ρ and angle θ corresponding to the wall line in the transform domain.

6. The method for estimating track deviation of a UAV-mounted through-wall radar based on image domain autofocusing according to claim 1, characterized in that: In step 5, the particle swarm optimization algorithm is used to iteratively obtain the line of sight direction track deviation vector Δy = [Δy1, Δy2, ... Δy l ] and a partially focused image of the wall: The optimization objective function is R f The image entropy S of the domain; argmin S=-∑ω(ρ,θ)logω(ρ,θ) Take the track deviation vector Δy at L azimuth sampling points = [Δy1, Δy2, ... Δy l ] as a particle, which is considered to represent the changing trend of the track deviation curve, and is iterated. In each iteration, Y a =Y m +Δy to get the true position of the radar in the direction of sight at each sampling point, where Y a and Y m They are the real position of the radar in the line of sight and the GPS rough positioning position, and then an R is obtained through the BP imaging algorithm. f images of the domain; In each iteration, the particle tracks the global optimal solution p gd and the individual optimal solution p id Two extreme values ​​to update their own speed and position: When the difference between the objective function values ​​obtained from consecutive iterations is less than a set threshold, the image is considered to be focused.

7. The method for estimating track deviation of a UAV-mounted through-wall radar based on image domain autofocusing according to claim 1, characterized in that: In step 6, a mean cancellation method is used to suppress wall clutter; Based on the BP imaging results of the wall part that restores the straight line characteristics, all pixels are averaged along the azimuth direction to obtain the average pixel, and then the average pixel is subtracted from each azimuth pixel to suppress the wall signal; Among them, I(x i ,y j ), I′(x i ,y j ) respectively represent (x i ,y j ) is the imaging pixel value before and after mean cancellation, and N represents the total number of pixels in the azimuth direction.

8. The method for estimating track deviation of a UAV-mounted through-wall radar based on image domain autofocusing according to claim 1 is characterized by: In step 7, the particle swarm optimization algorithm is used to iteratively obtain the course deviation vector Δx = [Δx1, Δx2, ... Δx l ] and the focused image of the target area; The image contrast of the imaging area where multiple distributed targets are located is used as the optimization objective function: Where I(x, y) is the pixel value at position (x, y) in the BP imaging result, and the deviation vector Δx along the heading track is taken as [Δx1, Δx2, ... Δx l ] is iterated as a particle, and in each iteration, X a =X m +Δx to obtain the true position of the radar along the course at each sampling point, where X a and X m They are the true position of the radar along the course and the GPS coarse positioning position, and the BP imaging results of a target area are re-obtained. Until the difference between the objective function values ​​obtained by continuous iteration is less than the set minimum value ε, the image is considered to be focused.

Images