Space-borne high resolution SAR imaging method and system

Abstract

The application provides a kind of spaceborne high-resolution SAR imaging method and system, comprising: input high-resolution SAR original echo data;The obtained original echo is divided into sub-aperture imaging to obtain the low-resolution SAR image of target area;Based on low-resolution SAR image, the position, class and attitude of land target are identified by deep learning target detection network, and the three-dimensional geometric model of target is constructed;Based on the three-dimensional geometric model of the target, the imaging reference elevation of each pixel is determined, and the target is imaged with high precision using back projection algorithm;Output target image with focus depth optimization.The application uses deep learning method to identify targets through low-resolution images, and introduces three-dimensional models of targets into imaging processing, thereby improving the focusing ability of high-resolution SAR systems on targets.

Description

Technical Field

[0001] This invention relates to the technical field of synthetic aperture radar (SAR) signal processing, and more specifically, to a spaceborne high-resolution SAR imaging method and system. Background Technology

[0002] Synthetic Aperture Radar (SAR) has been widely used in civilian fields such as marine environmental monitoring, agricultural resource surveys, and high-precision topographic mapping due to its ability to acquire high-resolution Earth observation images 24 / 7 and in all weather conditions. However, with the continuous improvement of the imaging resolution of spaceborne SAR systems, especially when the resolution reaches the decimeter level, the assumptions commonly used in traditional imaging processing, such as the walk-stop model, the standard hyperbolic distance model under the assumption of a flat Earth surface, and the spatial invariance of atmospheric propagation delay within the imaging scene, are no longer applicable. The impact of target elevation on the satellite's equivalent velocity becomes significant. Existing imaging methods for high-resolution SAR systems only consider ground elevation, neglecting the influence of the height of the observed object (such as power line towers) on the imaging, which leads to the problem of target defocusing in high-resolution SAR images. Therefore, how to accurately acquire detailed target elevation data within the imaging area and perform precise geometric and motion compensation for the propagation path and phase history of radar signals is the core challenge currently facing the improvement of imaging quality of spaceborne high-resolution SAR.

[0003] The paper "Echo Model Analyses and Imaging Algorithm for High-ResolutionSAR on High-Speed ​​Platform" (Y. Liu et al. IEEE Transactions on Geoscience and Remote Sensing, vol. 50, no. 3, pp. 933-950, March 2012) analyzes the accurate echo model when the walk-stop model is not applicable and proposes a spotlight imaging algorithm based on this model. It introduces errors caused by the walk-stop model approximation by analyzing the motion between the transmitted and received pulses, including quadratic phase error in the range direction, range migration error, cubic phase error in the azimuth direction, and target position deviation. Then, it introduces a SAR secondary focusing method and uses a range-frequency-dependent reference function in the azimuth preprocessing, proposing a spotlight imaging algorithm based on accurate echoes. The effectiveness of the proposed imaging algorithm is verified through point target imaging simulations. However, this method only provides a way to handle high-resolution SAR echoes from high-speed platforms when the walk-stop model is not applicable. For actual systems such as spaceborne SAR, the imaging geometry is different from the geometry used in this imaging algorithm.

[0004] The paper "Processing of Ultrahigh-Resolution spaceborne sliding spotlight SAR data on curved orbit" (F. He, Q. Chen, Z. Dong and Z. Sun. IEEE Transactions on Aerospace and Electronic Systems, vol. 49, no. 2, pp. 819-839, APRIL 2013) analyzes the impact of high-order orbit models on imaging and proposes an imaging processing algorithm based on these models. This method first calculates the equivalent parameters of the local hyperbolic curves for each sub-aperture of the sliding spotlight SAR data, proposes a piecewise hyperbolic range history model, and processes high-resolution sliding spotlight spaceborne SAR data based on this model. This method is easy to implement and can be effectively applied to imaging of high-resolution spaceborne SAR systems. However, this paper does not consider the influence of ground elevation on the equivalent velocity of the spaceborne radar.

[0005] The paper "A High-Resolution Spaceborne SAR Imaging Method for Large Scenes Based on Acceleration Slant Range Model" (Wu Yuan, Sun Guangcai, Yang Jun, et al. Journal of Electronics and Information Technology, 2015, 37(05):1116-1121.) describes a new acceleration slant range model. This model takes into account the azimuth spatial variation of velocity. First, it eliminates the azimuth spatial variation of velocity through azimuth time-domain resampling. Then, it removes the cubic and quartic errors in the two-dimensional frequency domain. Finally, it uses the range migration algorithm for imaging. Simulation results demonstrate the effectiveness of the acceleration slant range model and imaging method. However, this method only focuses on analyzing the influence of the orbital slant range model on imaging and does not consider the range ambiguity or azimuth ambiguity present in large-scene high-resolution spaceborne SAR.

[0006] Patent document CN102788972A discloses a self-focusing method suitable for high-resolution SAR imaging. This method converts the data format from polar coordinates to rectangular coordinates through two one-dimensional interpolations for range and azimuth. After spectral truncation of the data in the range direction, coarse-resolution imaging is performed in the range direction. Then, a conventional self-focusing algorithm is used to estimate the azimuth phase error, which is compensated for based on the analytical relationship between residual range migration and azimuth phase error before precise focusing imaging. However, this method does not consider the impact of atmospheric propagation delay on imaging and cannot be directly applied to spaceborne high-resolution SAR systems.

[0007] Patent document CN104237885A discloses a method for secondary focusing of synthetic aperture radar (SAR) images in terms of azimuth. This method first establishes a polynomial model relating radar equivalent velocity, target azimuth, range, and elevation. Then, it uses this polynomial model and target location results to quickly calculate the radar equivalent velocity, and constructs a Doppler-domain equivalent velocity spatially varying error compensation function to perform secondary focusing of traditional SAR imaging results in terms of azimuth. However, this method only considers ground elevation and does not account for the influence of target height on imaging.

[0008] In summary, existing high-resolution SAR imaging methods do not consider the impact of the target's height on imaging. Therefore, there is a market need for a spaceborne high-resolution SAR imaging method and system that identifies and confirms the target's height information and then compensates for the target's elevation to achieve effective refocusing of the target. Summary of the Invention

[0009] In view of the deficiencies in the prior art, the purpose of this invention is to provide a spaceborne high-resolution SAR imaging method and system.

[0010] According to the present invention, a spaceborne high-resolution SAR imaging method includes the following steps: Step S1: Input high-resolution SAR raw echo data; Step S2: Divide the acquired raw echo into sub-aperture imaging to obtain a low-resolution SAR image of the target area; Step S3: Based on the low-resolution SAR image, identify the location, category, and attitude of land targets using a deep learning target detection network, and construct a three-dimensional geometric model of the targets; Step S4: Determine the imaging reference elevation of each pixel based on the three-dimensional geometric model of the target, and use the back projection algorithm to perform high-precision imaging of the target; Step S5: Output the target image with optimized focus depth.

[0011] Preferably, step S2 includes: Step S2.1: Determine the sub-aperture image resolution requirements based on the target type; Step S2.2: Calculate the sub-aperture azimuth pulse number, divide the original echo into sub-apertures, and ensure that the number of sub-aperture azimuth sampling points satisfies the preset formula:

[0012] In the formula, PRF is the pulse repetition frequency, and Vg is the beam ground velocity. The sub-aperture image azimuth resolution. This is the beamwidth factor.

[0013] Preferably, step S3 includes: Step S3.1: Identify the target category using a pre-trained deep learning algorithm and output the probability of each category; Step S3.2: Determine the target attitude based on the target scattering information database corresponding to the identified category; Step S3.3: Construct a 3D geometric model of the target ground parking area. The height model of each scattering point is as follows:

[0014] In the formula, and The coordinates are respectively The ground elevation and target height at the location.

[0015] Preferably, step S4 includes: Step S4.1: Project the 3D geometric model onto the orientation-distance plane to determine the 3D spatial position information of each pixel of the target. The projection method can be represented as follows:

[0016] In the formula, and These are the azimuth and distance positions, respectively. The target horizontal position, The height of the target scattering point. The projection matrix; Step S4.2: Based on the spatial location information of each scattering point, complete the target focusing imaging based on the back projection algorithm.

[0017] Preferably, in step S3, the deep learning object detection network is pre-trained based on a preset dataset; The dataset contains various SAR images of land targets and their corresponding categories and attitude labels.

[0018] According to the present invention, a spaceborne high-resolution SAR imaging system includes: The data input module is used to input high-resolution SAR raw echo data; The sub-aperture imaging module is used to divide the raw echo into sub-apertures and image them, outputting a low-resolution SAR image of the target area; The target modeling module is used to identify the location, category, and attitude of land targets through a deep learning target detection network and construct a three-dimensional geometric model of the target. The high-precision imaging module is used to determine the imaging reference elevation of each pixel based on a three-dimensional geometric model, and to complete the high-precision imaging of the target through a back projection algorithm. The image output module is used to output a target image with optimized focus depth.

[0019] Preferably, the sub-aperture imaging module is further configured to adaptively adjust the number of sub-aperture divisions based on the target area size, the amount of raw echo data, and imaging efficiency requirements.

[0020] Preferably, the target modeling module has a built-in target scattering information database, which is used to match the identified target categories and assist in determining the target attitude.

[0021] Preferably, the high-precision imaging module has a built-in projection matrix for projecting the target three-dimensional geometric model onto the azimuth-distance plane.

[0022] Preferably, the system is adapted to X-band pure focus imaging mode, and the synthesis aperture time and transmission signal bandwidth can be preset according to the imaging resolution requirements.

[0023] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention is the first to propose introducing deep learning target recognition and three-dimensional geometric models into the SAR imaging process. It accurately identifies the target's location, category, and attitude through low-resolution images and constructs a three-dimensional model. Based on this, the imaging reference elevation of each pixel is determined. This effectively solves the technical problem of inconsistent focusing effects on targets at different heights and deterioration of focusing on high-altitude targets in traditional methods that rely on external DEM compensation. It achieves optimal focusing effects for targets at both high and low altitudes, significantly improves the focusing capability of high-resolution SAR systems for complex land targets, and provides high-quality image support for refined target identification and analysis.

[0024] 2. This invention adopts a step-by-step strategy of "first coarse imaging with sub-apertures, then target recognition and modeling, and finally precise focusing". First, the original echo is imaged at low resolution by dividing the sub-apertures. Target recognition and modeling are completed based solely on the low-resolution image, eliminating the need for complex calculations on the full-aperture high-resolution data. This significantly reduces the computational load of deep learning target detection and subsequent imaging processing. At the same time, the sub-aperture division can be adaptively adjusted according to the target area size, data volume, and efficiency requirements, further optimizing the allocation of computational resources. Compared with traditional full-aperture direct imaging algorithms, the computational efficiency is significantly improved, making it more suitable for the limited computing resources and real-time imaging requirements of spaceborne platforms. Attached Figure Description

[0025] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the spaceborne high-resolution SAR imaging method of the present invention; Figure 2 The image results are obtained using the traditional DEM compensation method. Figure 3 These are the results of focusing on targets at heights of 0 meters and 24 meters using traditional methods; Figure 4 The imaging results are obtained using the spaceborne high-resolution SAR imaging method of this invention; Figure 5 These are the focusing results of the present invention's method on targets at heights of 0 meters and 24 meters. Detailed Implementation

[0026] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0027] According to the present invention, a spaceborne high-resolution SAR imaging method is provided, such as... Figure 1 As shown, the method includes the following steps: Step S1: Input high-resolution SAR raw echo data.

[0028] Step S2: Divide the acquired raw echo into sub-aperture imaging to obtain a low-resolution SAR image of the target area. Step S2 includes the following steps: Step S2.1: Determine the sub-aperture image resolution based on the target type. Requirements, for example, the azimuth resolution of sub-aperture images of building targets must meet certain standards. .

[0029] Step S2.2: Calculate the sub-aperture azimuth pulse number and divide the original echo into sub-apertures. Number of sub-aperture azimuth sampling points. for:

[0030] In the formula, The pulse repetition frequency, For beam ground velocity, The sub-aperture image azimuth resolution. The beamwidth factor has a range of values. .

[0031] Step S3: Based on low-resolution SAR images, identify the location, category, and orientation of land targets using a deep learning target detection network, and construct a three-dimensional geometric model of the targets. Step S3 includes the following steps: Step S3.1: Identify the target category using a pre-trained deep learning algorithm and output the category probability. The pre-trained deep learning algorithm used in step S3.1 preferably employs a target detection architecture of YOLOv8 or higher. The neural network architecture is trained based on a SAR image dataset labeled with target categories. The neural network with the optimal weights is selected to implement the pre-trained deep learning algorithm to identify the target category and output the category probability index.

[0032] Step S3.2: Determine the target attitude based on the target scattering information database under this category. The radar target scattering information database in Step S3.2 is based on the simulation analysis results of the radar target scattering model and consists of the radar target scattering model, imaging information of each radar target attitude, and the center angle of the corresponding attitude. Based on the target type identified in Step S3.1, the target information corresponding to the radar target scattering information database is retrieved. The center angle of the corresponding target is matched according to the image information of the target imaging area, and the target elevation and azimuth angles are output.

[0033] Step S3.3: Construct a 3D geometric model of the target resting on the ground. (Height model of each scattering point) for

[0034] In the formula, and The coordinates are respectively The ground elevation and target height at the location.

[0035] Step S4: Based on the three-dimensional geometric model of the target, determine the imaging reference elevation of each pixel, and use a back projection algorithm to perform high-precision imaging of the target. Step S4 includes the following steps: Step S4.1: Project the target's three-dimensional geometric model onto the azimuth-range plane to determine the three-dimensional spatial position information of each pixel of the target in the SAR image. The projection method can be expressed as:

[0036] In the formula, and These are the azimuth and distance positions, respectively. The target horizontal position, The height of the target scattering point. The projection matrix; Step S4.2: Based on the spatial location information of the pixels where each scattering point of the target is located, focus the image of the target using a back projection algorithm. Compared with the traditional DEM compensation algorithm, which only determines the reference elevation based on the DEM and ignores the height of the target itself, resulting in defocused target imaging, this invention constructs a three-dimensional geometric model of the target and compensates for the phase error caused by the target height during back projection imaging, thereby optimizing the target imaging focus and improving imaging accuracy.

[0037] Step S5: Output the target image with optimized focus depth.

[0038] The method of this invention first divides the target into sub-apertures based on the resolution required for target identification and then performs coarse imaging. The coarse imaging results are then used to identify the target through deep learning. The three-dimensional model of the target is then introduced into the imaging process. Based on the target's shape information, the target is refocused and imaged, which greatly improves the focusing capability of the high-resolution SAR system. This method is the first of its kind both domestically and internationally.

[0039] The following simulation utilizes the TerraSAR-X satellite orbital parameters (orbital altitude 514.8 km, orbital inclination 97.44°, ascending node local time 18:00) and X-band pure spot-beam imaging mode. The synthetic aperture time is set to 9 s, the transmit signal bandwidth to 2 GHz, the pulse repetition frequency to 3000 Hz, the beam ground velocity to 7100 m / s, and the sub-aperture image azimuth resolution to 0.5 m. The number of sub-aperture azimuth sampling points is calculated to be 9000. The simulated observation target is 24 meters high, with a point of identical scattering characteristics spaced 1 meter apart. Figure 2 The imaging results using the traditional DEM compensation method are presented. Figure 3 The focusing results of the target at heights of 0 meters and 24 meters are given respectively. As can be seen from the figure, the focusing result at 24 meters deteriorates when only external DEM compensation is used. Figure 4 The results using the imaging method of this patent are presented. Figure 5 The focusing results for targets at heights of 0 meters and 24 meters are presented respectively. As can be seen from the figure, the focusing results for the target at 24 meters are basically consistent with those for the target at 0 meters. This simulation experiment fully demonstrates the effectiveness of the method of this invention.

[0040] The present invention also provides a spaceborne high-resolution SAR imaging system, which can be implemented by executing the process steps of the spaceborne high-resolution SAR imaging method. That is, those skilled in the art can understand the spaceborne high-resolution SAR imaging method as a preferred embodiment of the spaceborne high-resolution SAR imaging system.

[0041] The present invention also provides a spaceborne high-resolution SAR imaging system, the system comprising the following modules: Module M1: Input high-resolution SAR raw echo data.

[0042] Module M2: Divides the acquired raw echo into sub-aperture images to obtain a low-resolution SAR image of the target area. Module M2 includes the following steps: Module M2.1: Determines the sub-aperture image resolution requirements based on the target type; Module M2.2: Calculates the sub-aperture azimuth pulse number and divides the raw echo into sub-apertures.

[0043] Module M3: Based on low-resolution SAR images, a deep learning target detection network is used to identify the location, category, and attitude of land targets, and to construct a three-dimensional geometric model of the targets. Module M3 includes the following steps: Module M3.1: Identify the target category using a pre-trained deep learning algorithm and output the category probability; Module M3.2: Determine the target attitude based on a target scattering information database for that category; Module M3.3: Construct a three-dimensional geometric model of the target resting on the ground.

[0044] Module M4: Based on the target's three-dimensional geometric model, determine the imaging reference elevation for each pixel, and use a back projection algorithm to perform high-precision imaging of the target. Module M4 includes the following steps: Module M4.1: Project the target's three-dimensional geometric model onto the azimuth-range plane to determine the three-dimensional spatial position information of each pixel of the target in the SAR image; Module M4.2: Based on the spatial position information of the pixels where each scattering point of the target is located, perform focused imaging of the target using a back projection algorithm.

[0045] Module M5: Outputs a target image with optimized focus depth.

[0046] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0047] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A spaceborne high-resolution SAR imaging method, characterized in that, Includes the following steps: Step S1: Input high-resolution SAR raw echo data; Step S2: Divide the acquired raw echo into sub-aperture imaging to obtain a low-resolution SAR image of the target area; Step S3: Based on the low-resolution SAR image, identify the location, category, and attitude of land targets using a deep learning target detection network, and construct a three-dimensional geometric model of the targets; Step S4: Determine the imaging reference elevation of each pixel based on the three-dimensional geometric model of the target, and use the back projection algorithm to perform high-precision imaging of the target; Step S5: Output the target image with optimized focus depth; Step S2 includes: Step S2.1: Determine the sub-aperture image resolution requirements based on the target type; Step S2.2: Calculate the sub-aperture azimuth pulse number, divide the original echo into sub-apertures, and ensure that the number of sub-aperture azimuth sampling points satisfies the preset formula: In the formula, PRF is the pulse repetition frequency, and Vg is the beam ground velocity. The sub-aperture image azimuth resolution. Beam widening factor; Step S3 includes: Step S3.1: Identify the target category using a pre-trained deep learning algorithm and output the probability of each category; Step S3.2: Determine the target attitude based on the target scattering information database corresponding to the identified category; Step S3.3: Construct a 3D geometric model of the target ground parking area. The height model of each scattering point is as follows: In the formula, and The coordinates are respectively Ground elevation and target height at the location; Step S4 includes: Step S4.1: Project the 3D geometric model onto the orientation-distance plane to determine the 3D spatial position information of each pixel of the target. The projection method can be represented as follows: In the formula, and These are the azimuth and distance positions, respectively. The target horizontal position, The height of the target scattering point. The projection matrix; Step S4.2: Based on the spatial location information of each scattering point, complete the target focusing imaging based on the back projection algorithm; Deep learning object detection networks are pre-trained on a pre-defined dataset; The dataset contains various SAR images of land targets and their corresponding categories and attitude labels.

2. A spaceborne high-resolution SAR imaging system, characterized in that, The spaceborne high-resolution SAR imaging method according to claim 1 includes: The data input module is used to input high-resolution SAR raw echo data; The sub-aperture imaging module is used to divide the raw echo into sub-apertures and image them, outputting a low-resolution SAR image of the target area; The target modeling module is used to identify the location, category, and attitude of land targets through a deep learning target detection network and construct a three-dimensional geometric model of the target. The high-precision imaging module is used to determine the imaging reference elevation of each pixel based on a three-dimensional geometric model, and to complete the high-precision imaging of the target through a back projection algorithm. The image output module is used to output a target image with optimized focus depth.

3. The spaceborne high-resolution SAR imaging system according to claim 2, characterized in that, The sub-aperture imaging module is also used to adaptively adjust the number of sub-aperture divisions based on the target area size, the amount of raw echo data, and imaging efficiency requirements.

4. The spaceborne high-resolution SAR imaging system according to claim 2, characterized in that, The target modeling module has a built-in target scattering information database, which is used to match the identified target categories and assist in determining the target attitude.

5. The spaceborne high-resolution SAR imaging system according to claim 2, characterized in that, The high-precision imaging module has a built-in projection matrix, which is used to realize the projection transformation of the target three-dimensional geometric model to the azimuth-distance plane.

6. The spaceborne high-resolution SAR imaging system according to claim 2, characterized in that, The system is adapted to X-band pure beam imaging mode, and the synthesis aperture time and transmission signal bandwidth can be preset according to the imaging resolution requirements.

Images