CSAR dense artificial target detection method combined with digital elevation information

Abstract

The invention relates to the field of target detection, and more specifically relates to a CSAR dense artificial target detection method combined with digital elevation information, and the method comprises the steps: obtaining observation scene echo data through an airborne CSAR system; digital elevation information extraction is carried out according to observation scene echo data, and a high-precision CSAR image is generated in combination with the digital elevation information; generating a maximum extremal region according to the high-precision CSAR image; and obtaining vehicle average length and width information, and obtaining a detection result according to the vehicle average length and width information and the maximum extremal region. According to the method, firstly, digital elevation information extraction is carried out according to airborne CSAR observation scene echo data, and then a high-precision CSAR image is generated in combination with the digital elevation information; through the high-precision CSAR image combined with digital elevation information, the problem that adjacent vehicles interfere with each other due to broadening of vehicle contours when vehicle targets are dense can be effectively solved, so that the detection effect is improved.

Description

Technical Field

[0001] The present invention relates to the field of target detection, and more specifically, to a CSAR dense artificial target detection method combined with digital elevation information. Background Art

[0002] Synthetic Aperture Radar (SAR) imaging is a high-resolution earth observation technology that has the characteristics of working all day and all weather. It has been rapidly developed and widely concerned in recent years. Circular SAR (CSAR), as a new SAR mode, can obtain all-round high-resolution images of scene targets through 360° circular trajectory observation. The advantage of CSAR imaging mode is that the CSAR platform moves in a circle at a certain height, and the radar beam always points to the observation scene, so that all-round information and feature data of the observed target can be obtained. For man-made vehicle targets, the imaging plane is generally lower than the height of the vehicle target. Due to the overlap effect and other phenomena in CSAR images, the contour of the vehicle target will be widened. When the man-made vehicle targets are arranged more dispersedly, even if the contour of the vehicle target is widened, the vehicle target can generally be detected and identified well. However, for the dense arrangement of man-made vehicle targets in the parking lot, the widening of the vehicle contour will cause interference between adjacent vehicles, which easily makes it difficult to distinguish multiple dense vehicle targets, resulting in a high false alarm rate for dense man-made vehicle target detection.

[0003] The prior art discloses a SAR vehicle small target rapid detection and processing method, comprising step 1: constructing an optical remote sensing vehicle small target data set and a SAR ship small target data set based on an open source optical remote sensing data set and a SAR ship data set; step 2: constructing a SAR vehicle artificial data set according to a SAR image; step 3: designing a lightweight small target detection network structure based on a residual module network structure; step 4: using the data sets obtained in step 1 and step 2 to perform transfer transfer learning on the lightweight small target detection network, and training the lightweight small target detection network; step 5: eliminating false alarms in the detection results based on a post-processing false alarm elimination algorithm based on prior-data fusion; on the one hand, the method can efficiently and quickly detect SAR vehicle small targets, and on the other hand, it can effectively eliminate a large number of false alarms in the detection results, and has the characteristics of good detection and false alarm elimination effects and high detection result accuracy. However, the method still has the problem that the widening of the vehicle contour will cause interference between adjacent vehicles, which easily causes multiple dense vehicle targets to be indistinguishable, thereby making the detection of dense artificial vehicle targets have a high false alarm rate. Summary of the invention

[0004] The purpose of the present invention is to disclose a CSAR dense artificial target detection method combined with digital elevation information with better detection effect.

[0005] In order to achieve the above object, the present invention provides a CSAR dense artificial target detection method combined with digital elevation information, comprising:

[0006] S1: Obtain observation scene echo data through the airborne CSAR system;

[0007] S2: Extract digital elevation information based on the observed scene echo data and generate high-precision CSAR images based on the digital elevation information;

[0008] S3: Generate the maximum extreme value region based on the high-precision CSAR image;

[0009] S4: Obtain average length and width information of the vehicle, and obtain a detection result based on the average length and width information of the vehicle and the maximum extreme value area.

[0010] Furthermore, in step S1, it includes:

[0011] The airborne CSAR system performs uniform circular motion in a plane at a height of H from the horizontal plane, with a motion radius of R and a tangential velocity of V. During the motion observation process, the beam center of the airborne CSAR system always points to the center of the observation scene O. Therefore, when the airborne CSAR system performs a complete circular motion, a circular observation scene area will be formed with a radius of R. s , the distance between the airborne radar and the center of the observation scene is:

[0012]

[0013] When the airborne CSAR system moves to the positive half-axis position of X, the slow time η is zero, and the azimuth angle φ of the radar is:

[0014]

[0015] The airborne CSAR system moves along a circular trajectory to a position coordinate r at any azimuth angle φ. l =(x l ,y l ,z l )=(Rcosφ,Rsinφ,H), where φ∈[0,2π); Assume that the position coordinates of any point target P in the observation area are r p =(x p ,y p ,z p ), then the instantaneous distance from the radar to the point target P is:

[0016]

[0017] The radar signal emitted by the airborne CSAR system is a linear frequency modulation signal, namely:

[0018]

[0019] Among them, rect(·) is the rectangular window function, f c is the center frequency of the transmitted signal, τ is the fast time variable, T p is the duration of a single pulse signal, γ is the modulation frequency of the linear frequency modulation signal;

[0020] When the transmitted signal reaches the point target P, after backscattering, the signal received by the radar antenna of the airborne CSAR system is:

[0021]

[0022] Among them, σ lp is the backscattering coefficient of the point target P, c is the speed of light, w lp is the modulation factor of the radar antenna, is the time delay from the radar signal being transmitted from the antenna to the point target P and then received by the antenna after reflection. For the continuous pulse signal transmitted by the radar system, the total signal time is t, and the echo signal received by the radar antenna is:

[0023]

[0024] Right now:

[0025]

[0026] Among them, nT is the motion time variable of the radar antenna, recorded as slow time η, and slow time η = nT is used as the time quantity to describe the motion of the antenna, and the relationship between fast and slow time variables is τ = t-η = t-nT;

[0027] For the echo signal S r (τ) is orthogonally demodulated and the carrier frequency is removed, and the point target echo signal received by the airborne CSAR system becomes:

[0028]

[0029] The echo data of the observed scene are obtained based on all the received point target echo signals.

[0030] Further, in step S2, digital elevation information is extracted according to the observed scene echo data, including:

[0031] S2.1: First, the circular full-aperture echo data is divided into n sub-apertures to obtain sub-aperture echo data, where the azimuth angle of each sub-aperture is Δθ, and the azimuth angle between adjacent sub-apertures is Δφ. The relationship expression is:

[0032] L = Δθ + (n-1) × Δφ

[0033] Where L is the angle corresponding to the circular full-aperture echo data;

[0034] S2.2: Using the back projection (BP) algorithm to process the sub-aperture echo data, a coarsely focused sub-aperture image is obtained;

[0035] S2.3: The coarsely focused sub-aperture image is subjected to autofocus processing based on autofocus backprojection (ABP) to obtain a finely focused sub-aperture image, and the error phase is estimated using the phase estimation criterion, which is expressed as:

[0036]

[0037] in, is the phase error to be estimated, It is a function for evaluating the sharpness of an image, namely, the energy evaluation function;

[0038] S2.4: Project the finely focused sub-aperture image in the height direction, estimate the elevation range of the observed scene, and establish the height coordinate axis. Then, set the height interval, and project the sub-aperture image to each height value according to the calculated coordinate offset, thereby forming a sub-aperture image group.

[0039] S2.5: In the sub-aperture image group, two adjacent sub-aperture image groups are selected to perform correlation calculation pixel by pixel. The correlation between the corresponding height layers between the sub-aperture image groups after height projection is used to calculate each height layer. The height layer with the strongest correlation is taken as the height result at that location. The correlation calculation formula is:

[0040]

[0041] Among them, I A and I B are the pixel amplitude values ​​of the slider windows of two adjacent sub-aperture images A and B, and are the average values ​​of the pixel amplitudes in the slider window, K is the parameter for adjusting the slider window, and the size of the slider window is (2×K+1)×(2×K+1) pixels. The correlation coefficient ρ of each height layer is calculated layer by layer according to the method of each height layer, and the maximum correlation coefficient ρ is taken.max The corresponding height layer is taken as the measured height h of the pixel (i, j), and its calculation formula is:

[0042]

[0043] S2.6: According to the principle of correlation calculation of adjacent sub-aperture image groups, all sub-aperture image groups are traversed, and the digital elevation model (DEM) information of all adjacent sub-aperture images is superimposed and averaged to obtain digital elevation information.

[0044] Furthermore, in step S2, a high-precision CSAR image is generated in combination with the digital elevation information, including: adding digital elevation information to the grid points of the imaging area, compensating for the error Δh in the height direction, and generating a high-precision CSAR image based on the compensated imaging area grid.

[0045] Further, in step S2, generating a high-precision CSAR image according to the compensated imaging area grid includes: generating a high-precision CSAR image by using a BP algorithm.

[0046] Furthermore, in step S3, it includes: generating a maximum extreme region according to the high-precision CSAR image by using a Maximum Stable Extreme Region (MSER) ​​algorithm.

[0047] Further, in step S3, generating the maximum extreme value region according to the high-precision CSAR image by using the MSER algorithm includes:

[0048] Based on the prior information of the pixel area occupied by the vehicle target in the CSAR image, the area of ​​all detected extreme value areas is limited to improve the accuracy and efficiency of detection. The area limitation range is:

[0049]

[0050] Among them, S min and S max are the minimum and maximum values ​​of the limited extreme area respectively, It is the extreme value area, and the extreme value area whose area exceeds this limit will be discarded; thus the extreme value area is obtained.

[0051] Furthermore, in step S3, it further includes: determining whether the extreme value region is stable:

[0052]

[0053] in, represents the area change rate, ε is the limit value of the area change rate, and the corresponding like The corresponding extreme value region is a stable region and is retained as the maximum extreme value region; otherwise, it is an unstable region and is discarded.

[0054] Furthermore, in step S4, it includes:

[0055] In the maximum extreme value area, the center of each extreme value area is extracted, and the distance between the centers of different extreme value areas is calculated. Then, a comprehensive judgment is made based on the length and width information of the vehicle. If the distance between the centers of two extreme value areas is greater than the average vehicle length and width, they are judged to be two vehicle targets. Otherwise, they are judged to be the same vehicle target, and the detection result is obtained.

[0056] Further, in step S4, obtaining the average length and width information of the vehicle includes: collecting statistics on the length and width information of multiple types of vehicles, and obtaining the average length and width information of the vehicle according to the length and width information of the multiple types of vehicles.

[0057] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0058] The present invention first extracts digital elevation information based on the airborne CSAR observation scene echo data, and then generates a high-precision CSAR image in combination with the digital elevation information; the high-precision CSAR image combined with the digital elevation information can effectively solve the problem of interference between adjacent vehicles due to the widening of the vehicle outline when the vehicle targets are dense, thereby improving the detection effect. In addition, the present invention also uses the vehicle target length and width information to perform extreme value area fusion screening to obtain high-precision detection results of dense artificial vehicle targets in the observation scene. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A flow chart of a CSAR dense man-made target detection method combined with digital elevation information described in Example 1;

[0060] Figure 2 A schematic diagram of the geometric configuration of the airborne CSAR system imaging described in Example 2;

[0061] Figure 3 This is a result diagram after the ABP-based self-focusing process described in Example 2;

[0062] Figure 4 A three-dimensional schematic diagram of the overlapping effect of CSAR imaging described in Example 3;

[0063] Figure 5 A top view schematic diagram of the overlapping effect of CSAR imaging described in Example 3;

[0064] Figure 6 A schematic diagram of a vehicle outline in an imaging result when the imaging reference plane is highly consistent with the vehicle as described in the third embodiment;

[0065] Figure 7 A schematic diagram of a vehicle outline in an imaging result when the imaging reference plane is lower than the height of the vehicle as described in the third embodiment;

[0066] Figure 8 The actual vehicle photo picture described in Example 3;

[0067] Fig. 9 The vehicle outline image when the imaging reference plane height is 0m as described in Example 3;

[0068] Fig.10 The vehicle outline image when the imaging reference plane described in the third embodiment is consistent with the vehicle height;

[0069] Fig.11 It is an optical image of the observation scene of the Gotcha data described in Example 3;

[0070] Fig.12 The radar chart of the observation scene of the Gotcha data described in Example 3;

[0071] Fig.13 The vehicle target annotation map for evaluating DEM information described in Example 3;

[0072] Fig.14 This is a DEM information result diagram obtained by the method of the present invention described in Example 3;

[0073] Fig.15 This is the imaging result diagram without DEM information compensation described in Example 3;

[0074] Fig.16 This is the imaging result diagram combined with DEM information compensation described in Example 3;

[0075] Fig.17 A schematic diagram for generating an extreme value region as described in the third embodiment;

[0076] Fig.18 This is the extreme value region fusion result graph using only vehicle length information as described in Example 3;

[0077] Fig.19 This is the extreme value region fusion result map using vehicle length and width information as described in Example 3;

[0078] Fig. 20 This is a man-made vehicle target detection result diagram obtained by the multi-height fusion method described in Example 3;

[0079] Fig.21 This is a graph of the artificial vehicle target detection result obtained by the present invention described in Example 3;

[0080] Fig. 22 A local magnified image of a densely populated man-made vehicle target area obtained by the multi-height fusion method described in Example 3;

[0081] Fig.23 A partial enlarged view of a densely populated man-made vehicle target area obtained by the present invention as described in Example 3; DETAILED DESCRIPTION

[0082] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0083] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0084] Embodiment 1:

[0085] This embodiment provides Figure 1 A CSAR dense man-made target detection method combined with digital elevation information is shown, comprising:

[0086] The present invention provides a CSAR dense artificial target detection method combined with digital elevation information, comprising:

[0087] S1: Obtain observation scene echo data through the airborne CSAR system;

[0088] S2: Extract digital elevation information based on the observed scene echo data and generate high-precision CSAR images based on the digital elevation information;

[0089] S3: Generate the maximum extreme value region based on the high-precision CSAR image;

[0090] S4: Obtain average length and width information of the vehicle, and obtain a detection result based on the average length and width information of the vehicle and the maximum extreme value area.

[0091] This embodiment first extracts digital elevation information based on the airborne CSAR observation scene echo data, and then generates a high-precision CSAR image in combination with the digital elevation information; the high-precision CSAR image combined with the digital elevation information can effectively solve the problem of interference between adjacent vehicles due to the widening of the vehicle outline when the vehicle targets are dense, thereby improving the detection effect. In addition, this embodiment also uses the vehicle target length and width information to perform extreme value area fusion screening to obtain high-precision detection results for dense artificial vehicle targets in the observation scene.

[0092] Embodiment 2:

[0093] This embodiment further discloses the following on the basis of the first embodiment:

[0094] Furthermore, in step S1, it includes:

[0095] The geometric configuration of airborne CSAR system imaging is shown below: Figure 2 As shown in the figure, the airborne CSAR system performs uniform circular motion in a plane at a height of H from the horizontal plane, with a motion radius of R and a tangential velocity of V. During the motion observation process, the beam center of the airborne CSAR system always points to the center of the observation scene O. Therefore, when the airborne CSAR system performs a complete circular motion, a circular observation scene area will be formed with a radius of R. s , the distance between the airborne radar and the center of the observation scene is:

[0096]

[0097] When the airborne CSAR system moves to the positive half-axis position of X, the slow time η is zero, and the azimuth angle φ of the radar is:

[0098]

[0099] The airborne CSAR system moves along a circular trajectory to a position coordinate r at any azimuth angle φ. l =(x l ,y l ,z l )=(Rcosφ,Rsinφ,H), where φ∈[0,2π); Assume that the position coordinates of any point target P in the observation area are r p =(x p ,y p ,z p ), then the instantaneous distance from the radar to the point target P is:

[0100]

[0101] The radar signal emitted by the airborne CSAR system is a linear frequency modulation signal, namely:

[0102]

[0103] Among them, rect(·) is the rectangular window function, f c is the center frequency of the transmitted signal, τ is the fast time variable, T p is the duration of a single pulse signal, γ is the modulation frequency of the linear frequency modulation signal;

[0104] When the transmitted signal reaches the point target P, after backscattering, the signal received by the radar antenna of the airborne CSAR system is:

[0105]

[0106] Among them, σ lp is the backscattering coefficient of the point target P, c is the speed of light, w lp is the modulation factor of the radar antenna, is the time delay from the radar signal being transmitted from the antenna to the point target P and then received by the antenna after reflection. For the continuous pulse signal transmitted by the radar system, the total signal time is t, and the echo signal received by the radar antenna is:

[0107]

[0108] Right now:

[0109]

[0110] Among them, nT is the motion time variable of the radar antenna, recorded as slow time η, and slow time η = nT is used as the time quantity to describe the motion of the antenna, and the relationship between fast and slow time variables is τ = t-η = t-nT;

[0111] For the echo signal S r (τ) is orthogonally demodulated and the carrier frequency is removed, and the point target echo signal received by the airborne CSAR system becomes:

[0112]

[0113] The echo data of the observed scene are obtained based on all the received point target echo signals.

[0114] Further, in step S2, digital elevation information is extracted according to the observed scene echo data, including:

[0115] S2.1: First, the circular full-aperture echo data is divided into n sub-apertures to obtain sub-aperture echo data, where the azimuth angle of each sub-aperture is Δθ, and the azimuth angle between adjacent sub-apertures is Δφ. The relationship expression is:

[0116] L = Δθ + (n-1) × Δφ

[0117] Where L is the angle corresponding to the circular full-aperture echo data;

[0118] S2.2: Use BP algorithm to process the sub-aperture echo data to obtain a coarsely focused sub-aperture image;

[0119] S2.3: The coarsely focused sub-aperture image is subjected to ABP-based autofocusing processing to obtain a finely focused sub-aperture image, and the error phase is estimated using the phase estimation criterion, which is expressed as:

[0120]

[0121] in, is the phase error to be estimated, It is a function for evaluating the sharpness of an image, namely, the energy evaluation function;

[0122] The results after ABP-based self-focusing processing are as follows Figure 3 As shown;

[0123] S2.4: Project the finely focused sub-aperture image in the height direction, estimate the elevation range of the observed scene, and establish the height coordinate axis. Then, set the height interval, and project the sub-aperture image to each height value according to the calculated coordinate offset, thereby forming a sub-aperture image group.

[0124] S2.5: In the sub-aperture image group, two adjacent sub-aperture image groups are selected to perform correlation calculation pixel by pixel. The correlation between the corresponding height layers between the sub-aperture image groups after height projection is used to calculate each height layer. The height layer with the strongest correlation is taken as the height result at that location. The correlation calculation formula is:

[0125]

[0126] Among them, I A and I B are the pixel amplitude values ​​of the slider windows of two adjacent sub-aperture images A and B, and are the average values ​​of the pixel amplitudes in the slider window, K is the parameter for adjusting the slider window, and the size of the slider window is (2×K+1)×(2×K+1) pixels. The correlation coefficient ρ of each height layer is calculated layer by layer according to the method of each height layer, and the maximum correlation coefficient ρ is taken. max The corresponding height layer is taken as the measured height h of the pixel (i, j), and its calculation formula is:

[0127]

[0128] S2.6: According to the principle of correlation calculation of adjacent sub-aperture image groups, all sub-aperture image groups are traversed, the DEM information of all adjacent sub-aperture images obtained are superimposed and the average is taken to obtain digital elevation information.

[0129] Furthermore, in step S2, a high-precision CSAR image is generated in combination with the digital elevation information, including: adding digital elevation information to the grid points of the imaging area, compensating for the error Δh in the height direction, and generating a high-precision CSAR image based on the compensated imaging area grid.

[0130] Further, in step S2, generating a high-precision CSAR image according to the compensated imaging area grid includes: generating a high-precision CSAR image by using a BP algorithm.

[0131] Furthermore, in step S3, it includes: generating a maximum extreme value region according to the high-precision CSAR image by using the MSER algorithm.

[0132] Further, in step S3, generating the maximum extreme value region according to the high-precision CSAR image by using the MSER algorithm includes:

[0133] Based on the prior information of the pixel area occupied by the vehicle target in the CSAR image, the area of ​​all detected extreme value areas is limited to improve the accuracy and efficiency of detection. The area limitation range is:

[0134]

[0135] Among them, S min and S max are the minimum and maximum values ​​of the limited extreme area respectively, It is the extreme value area, and the extreme value area whose area exceeds this limit will be discarded; thus the extreme value area is obtained.

[0136] Furthermore, in step S3, it further includes: determining whether the extreme value region is stable:

[0137]

[0138] in, represents the area change rate, ε is the limit value of the area change rate, and the corresponding like The corresponding extreme value region is a stable region and is retained as the maximum extreme value region; otherwise, it is an unstable region and is discarded.

[0139] Furthermore, in step S4, it includes:

[0140] In the maximum extreme value area, the center of each extreme value area is extracted, and the distance between the centers of different extreme value areas is calculated. Then, a comprehensive judgment is made based on the length and width information of the vehicle. If the distance between the centers of two extreme value areas is greater than the average vehicle length and width, they are judged to be two vehicle targets. Otherwise, they are judged to be the same vehicle target, and the detection result is obtained.

[0141] Further, in step S4, obtaining the average length and width information of the vehicle includes: collecting statistics on the length and width information of multiple types of vehicles, and obtaining the average length and width information of the vehicle according to the length and width information of the multiple types of vehicles.

[0142] This embodiment first extracts digital elevation information based on the airborne CSAR observation scene echo data, and then generates a high-precision CSAR image in combination with the digital elevation information; the high-precision CSAR image combined with the digital elevation information can effectively solve the problem of interference between adjacent vehicles due to the widening of the vehicle outline when the vehicle targets are dense, thereby improving the detection effect. In addition, this embodiment also uses the vehicle target length and width information to perform extreme value area fusion screening to obtain high-precision detection results for dense artificial vehicle targets in the observation scene.

[0143] Embodiment three:

[0144] When performing airborne CSAR imaging, the three-dimensional diagram of the overlapping effect of CSAR imaging is as follows: Figure 4 As shown in the figure, for any point target P in the observation scene, the height difference between its height and the imaging reference plane is Δh. Assume that the height of the imaging reference plane is 0m, and the point target P (x PA ,y PA ,Δh) is imaged and projected at the sub-aperture A, then the coordinates of the point target P on the imaging reference plane are A (x PA′ ,y PA′ ,0) is:

[0145]

[0146] Where φA is the azimuth at aperture A, and θA is the downward viewing angle at aperture A. In the CSAR system, full-aperture imaging is performed on a point target P. The point target P will appear defocused and projected into a ring shape on the imaging reference plane. The overlapping effect of CSAR imaging is shown in the following figure. Figure 5 As shown. Therefore, the imaging result of the point target P will produce geometric deformation, that is, it will cause the overlapping effect of the CSAR image. It can be found that the geometric deformation of the CSAR image is related to the height difference between the point target location and the imaging reference plane. The greater the height difference between the height plane where the point target is located and the imaging reference plane, the greater the geometric deformation of the CSAR image. At the same time, the beam pitch angle θ and azimuth angle φ will also affect the degree of geometric deformation of the CSAR image.

[0147] When there is a height difference Δh between the target height and the imaging reference plane, the target imaging result will have a position offset. For man-made vehicle targets, when the imaging reference plane is consistent with the vehicle target height, the imaging result can better reflect the vehicle contour information. When the imaging reference plane is consistent with the vehicle height, the vehicle contour in the imaging result is shown as follows: Figure 6 When the imaging reference plane is lower than the target height of the vehicle, the vehicle outline in the imaging result will be widened. When the imaging reference plane is lower than the vehicle height, the vehicle outline in the imaging result is shown as follows: Figure 7 shown.

[0148] For man-made vehicle targets, the imaging reference plane is generally lower than the height of the vehicle target. Due to the existence of phenomena such as the overlap effect, the outline of the vehicle target will be widened. When the man-made vehicle targets are arranged more dispersedly, even if the outline of the vehicle target is widened, the vehicle target can generally be detected and identified well. However, for the dense arrangement of man-made vehicle targets in the parking lot, the widening of the vehicle outline will cause interference between adjacent vehicles, which can easily cause multiple dense vehicle targets to be indistinguishable, resulting in a high missed alarm rate for dense man-made vehicle target detection. Actual vehicle photos such as Figure 8 As shown in the figure, a full-aperture image of an actual vehicle with a height of 1.67m and a width of 1.77m is taken. When the imaging reference plane height is 0m, the width of the two sides of the vehicle relative to the center is L1≈L2=1.82m, and the width of the entire vehicle is about 3.64m. When the imaging reference plane height is 0m, the vehicle outline is as follows Fig. 9 When the height of the imaging reference plane is 1.67m (i.e., consistent with the height of the vehicle), the width of both sides of the vehicle relative to the center is L1≈L2=0.95m, and the width of the entire vehicle is about 1.9m, which is consistent with the actual width of the vehicle. When the imaging reference plane is consistent with the height of the vehicle, the vehicle outline is as follows: Fig.10 Compared with the imaging results without DEM information compensation, the energy of the vehicle target in the imaging results combined with DEM information compensation is more concentrated in the central area, which is convenient for subsequent detection and identification of vehicle targets.

[0149] In order to verify the correctness and effectiveness of the method proposed in this embodiment, experimental verification is performed using CSAR measured data (Gotcha data set), which is CSAR experimental data released by the US Air Force Research Laboratory in 2006. This embodiment uses the CSAR data collected by the first baseline for experimental verification. The scene observed in the experiment is a parking lot with several civilian vehicles parked. The optical image of the observation scene of the Gotcha data is as follows: Fig.11 As shown in the figure, the radar chart of the observation scene of Gotcha data is as follows Fig.12 The CSAR system parameters of the Gotcha dataset are shown in Table 1.

[0150] Table 1 CSAR system parameters of Gotcha dataset

[0151] System Parameters Numeric Carrier frequency 9.6GHz Signal bandwidth 640MHz Radar platform flight altitude 7200m Radar platform flight radius 7200m 2D resolution 0.1m

[0152] In order to verify the effectiveness of the improved chain correlation method to extract DEM information, the vehicle target annotation used to evaluate DEM information is as follows: Fig.13As shown in the figure, 7 vehicle targets (corresponding to numbers A to G) used to evaluate the measurement accuracy of DEM information are marked, and the direction of the arrow is the direction of the front of the vehicle target. The DEM information extraction results of vehicle targets A to G are evaluated below. The DEM information extraction results of the two methods are quantitatively analyzed, that is, the average value and root mean square error of the vehicle target height are calculated based on the DEM information. In this embodiment, the average height of the top area of ​​the vehicle target is set as the measured height, and the calculation formula of the root mean square error is:

[0153]

[0154] Where Q represents the number of pixels used for accuracy evaluation, h q represents the actual elevation value of the qth pixel, represents the estimated elevation value of the qth pixel. The DEM information result obtained by the method proposed in this embodiment is as follows Fig.14 The calculation results are shown in Table 2. The experimental results show that the method proposed in this embodiment can effectively extract the DEM information of the scene.

[0155] Table 2 DEM information extraction results

[0156]

[0157]

[0158] The Gotcha dataset was imaged and the imaging results without DEM information compensation are as follows: Fig.15 As shown in Figure 2, the imaging results combined with DEM information compensation are shown in Figure 2. Fig.16 As shown in the figure. Man-made vehicle targets in dense areas are marked with red boxes. It can be found that compared with the imaging results without DEM information compensation, the compensation processing combined with DEM information can obtain CSAR images with higher accuracy. In addition, the mutual interference between the contours of dense vehicle targets is significantly reduced, the overlapping effect of vehicle targets is significantly improved, and the distinguishability of man-made vehicle targets in dense areas is significantly improved.

[0159] The generation of extreme value regions using the MSER algorithm is shown below: Fig.17 To illustrate the advantage of using the average vehicle length and width of the vehicle for comprehensive judgment in MSER fusion, compared with using only the average vehicle length information as the judgment condition, the fusion result of the extreme value area using only the vehicle length information is shown in FIG. Fig.18 As shown, the extreme area fusion result using vehicle length and width information is as follows Fig.19 As shown in Figure 2, since extreme value areas may be generated between dense vehicle targets, using only vehicle length information will lead to false alarms. On the contrary, using both vehicle length and width information can effectively reduce the false alarm rate.

[0160] In order to verify the effectiveness of the CSAR dense man-made target detection method combined with digital elevation information proposed in this embodiment, the multi-height fusion method is used as a comparison method for dense vehicle target detection. The multi-height fusion method accumulates the CSAR full-aperture images obtained at different imaging heights and only uses the vehicle target length information to detect the scene man-made vehicle targets. The man-made vehicle target detection results obtained by the multi-height fusion method are shown in Figure 2. Fig. 20 As shown, the artificial vehicle target detection result obtained in this embodiment is as follows Fig.21 As shown, the red circle represents the detected vehicle target. Fig. 22 This is a local magnified image of the densely populated man-made vehicle target area obtained by the multi-height fusion method. Fig.23 This is a partial enlarged view of the dense man-made vehicle target area obtained in this embodiment, where area A is part of the dense vehicle area and area B is the street lamp area.

[0161] Table 3 Comparison of two man-made vehicle target detection algorithms

[0162]

[0163]

[0164] The results of artificial vehicle target detection in the observation scene are statistically analyzed, and Table 3 gives the quantitative evaluation indicators. The experimental results show that the detection rate of dense vehicle targets obtained by the multi-height fusion method is 94.12%, but the false alarm rate reaches 4.71%. While the method proposed in this embodiment ensures a low false alarm rate of 2.35%, the detection rate of dense vehicle targets reaches 97.65%. Compared with the multi-height fusion method, the detection rate of the method proposed in this embodiment is significantly improved, thereby verifying the effectiveness and practicality of the method proposed in this embodiment. In addition, there are still false alarms for very individual vehicle targets, which may be due to the presence of street lamps next to the artificial vehicle targets. During CSAR imaging detection, due to the high scattering energy of street lamps, it is difficult to avoid the generation of extreme value areas at the street lamps, which leads to false alarms in vehicle target detection.

[0165] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A CSAR dense man-made target detection method combined with digital elevation information, characterized in that: include: S1: Obtain observation scene echo data through the airborne CSAR system; S2: Extract digital elevation information based on the observed scene echo data and generate high-precision CSAR images based on the digital elevation information; S3: Generate the maximum extreme value region based on the high-precision CSAR image; S4: Obtain average length and width information of the vehicle, and obtain a detection result based on the average length and width information of the vehicle and the maximum extreme value area.

2. The CSAR dense man-made target detection method combined with digital elevation information according to claim 1, characterized in that: In step S1, it includes: The airborne CSAR system performs uniform circular motion in a plane at a height of H from the horizontal plane, with a motion radius of R and a tangential velocity of V. During the motion observation process, the beam center of the airborne CSAR system always points to the center of the observation scene O. Therefore, when the airborne CSAR system performs a complete circular motion, a circular observation scene area will be formed with a radius of R. s , the distance between the airborne radar and the center of the observation scene is: When the airborne CSAR system moves to the positive half-axis position of X, the slow time η is zero, and the azimuth angle φ of the radar is: The airborne CSAR system moves along a circular trajectory to a position coordinate r at any azimuth angle φ. l =(x l ,y l ,z l )=(Rcosφ,Rsinφ,H), where φ∈[0,2π); Assume that the position coordinates of any point target P in the observation area are r p =(x p ,y p ,z p ), then the instantaneous distance from the radar to the point target P is: The radar signal emitted by the airborne CSAR system is a linear frequency modulation signal, namely: Among them, rect(·) is the rectangular window function, f c is the center frequency of the transmitted signal, τ is the fast time variable, T p is the duration of a single pulse signal, γ is the modulation frequency of the linear frequency modulation signal; When the transmitted signal reaches the point target P, after backscattering, the signal received by the radar antenna of the airborne CSAR system is: Among them, σ lp is the backscattering coefficient of the point target P, c is the speed of light, w lp is the modulation factor of the radar antenna, is the time delay from the radar signal being transmitted from the antenna to the point target P and then received by the antenna after reflection. For the continuous pulse signal transmitted by the radar system, the total signal time is t, and the echo signal received by the radar antenna is: Right now: Among them, nT is the motion time variable of the radar antenna, recorded as slow time η, and slow time η = nT is used as the time quantity to describe the motion of the antenna, and the relationship between fast and slow time variables is τ = t-η = t-nT; For the echo signal S r (τ) is orthogonally demodulated and the carrier frequency is removed, and the point target echo signal received by the airborne CSAR system becomes: The echo data of the observed scene are obtained based on all the received point target echo signals.

3. The CSAR dense man-made target detection method combined with digital elevation information according to claim 1 is characterized in that: In step S2, digital elevation information is extracted according to the observed scene echo data, including: S2.1: First, the circular full-aperture echo data is divided into n sub-apertures to obtain sub-aperture echo data, where the azimuth angle of each sub-aperture is Δθ, and the azimuth angle between adjacent sub-apertures is Δφ. The relationship expression is: L = Δθ + (n-1) × Δφ Where L is the angle corresponding to the circular full-aperture echo data; S2.2: Using BP algorithm to process sub-aperture echo data to obtain a coarsely focused sub-aperture image; S2.3: The coarsely focused sub-aperture image is subjected to ABP-based autofocusing processing to obtain a finely focused sub-aperture image, and the error phase is estimated using the phase estimation criterion, which is expressed as: in, is the phase error to be estimated, It is a function for evaluating the sharpness of an image, namely, the energy evaluation function; S2.4: Project the finely focused sub-aperture image in the height direction, estimate the elevation range of the observed scene, and establish the height coordinate axis. Then, set the height interval, and project the sub-aperture image to each height value according to the calculated coordinate offset, thereby forming a sub-aperture image group. S2.5: In the sub-aperture image group, two adjacent sub-aperture image groups are selected to perform correlation calculation pixel by pixel. The correlation between the corresponding height layers between the sub-aperture image groups after height projection is used to calculate each height layer. The height layer with the strongest correlation is taken as the height result at that location. The correlation calculation formula is: Among them, I A and I B are the pixel amplitude values ​​of the slider windows of two adjacent sub-aperture images A and B, and are the average values ​​of the pixel amplitudes in the slider window, K is the parameter for adjusting the slider window, and the size of the slider window is (2×K+1)×(2×K+1) pixels. The correlation coefficient ρ of each height layer is calculated layer by layer according to the method of each height layer, and the maximum correlation coefficient ρ is taken. max The corresponding height layer is taken as the measured height h of the pixel (i, j), and its calculation formula is: S2.6: According to the principle of correlation calculation of adjacent sub-aperture image groups, all sub-aperture image groups are traversed, the DEM information of all adjacent sub-aperture images obtained are superimposed and the average is taken to obtain digital elevation information.

4. The CSAR dense man-made target detection method combined with digital elevation information according to claim 1, characterized in that: In step S2, a high-precision CSAR image is generated in combination with the digital elevation information, including: adding digital elevation information to the grid points of the imaging area, compensating for the error Δh in the height direction, and generating a high-precision CSAR image based on the compensated imaging area grid.

5. The CSAR dense man-made target detection method combined with digital elevation information according to claim 4 is characterized in that: In step S2, generating a high-precision CSAR image according to the compensated imaging area grid includes: generating a high-precision CSAR image by using a BP algorithm.

6. The CSAR dense man-made target detection method combined with digital elevation information according to claim 1, characterized in that: In step S3, it includes: generating a maximum extreme value region according to the high-precision CSAR image by using the MSER algorithm.

7. The CSAR dense man-made target detection method combined with digital elevation information according to claim 6, characterized in that: In step S3, the maximum extreme value region is generated according to the high-precision CSAR image by using the MSER algorithm, including: Based on the prior information of the pixel area occupied by the vehicle target in the CSAR image, the area of ​​all detected extreme value areas is limited to improve the accuracy and efficiency of detection. The area limitation range is: Among them, S min and S max are the minimum and maximum values ​​of the limited extreme area respectively, It is the extreme value area, and the extreme value area whose area exceeds this limit will be discarded; thus the extreme value area is obtained.

8. The CSAR dense man-made target detection method combined with digital elevation information according to claim 7, characterized in that: In step S3, it also includes: judging whether the extreme value region is stable: in, represents the area change rate, ε is the limit value of the area change rate, and the corresponding like The corresponding extreme value region is a stable region and is retained as the maximum extreme value region; otherwise, it is an unstable region and is discarded.

9. The CSAR dense man-made target detection method combined with digital elevation information according to claim 1, characterized in that: In step S4, it includes: In the maximum extreme value area, the center of each extreme value area is extracted, and the distance between the centers of different extreme value areas is calculated. Then, a comprehensive judgment is made based on the length and width information of the vehicle. If the distance between the centers of two extreme value areas is greater than the average vehicle length and width, they are judged to be two vehicle targets. Otherwise, they are judged to be the same vehicle target, and the detection result is obtained.

10. The CSAR dense man-made target detection method combined with digital elevation information according to claim 1, characterized in that: In step S4, obtaining the average length and width information of the vehicles includes: collecting statistics on the length and width information of multiple types of vehicles, and obtaining the average length and width information of the vehicles according to the length and width information of multiple types of vehicles.

Images