A physical structure guided method for extracting key scattering points of aircraft targets in SAR images
By establishing a physical semantic structure model and topological constraints for the aircraft, combined with masking adaptive design, and adopting a coarse-fine combined positioning strategy, the stability and adaptability problems of scattering point extraction under complex observation conditions were solved, achieving high-precision extraction of key scattering points and improving the accuracy of SAR aircraft target recognition and attitude estimation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING INSTITUTE OF TECHNOLOGY ANHUI INSTITUTE OF AEROSPACE INFORMATION
- Filing Date
- 2026-05-07
- Publication Date
- 2026-07-14
AI Technical Summary
Existing technologies struggle to reliably extract key scattering points from aircraft targets under complex observation conditions, and lack effective semantic completion and deduction mechanisms, resulting in poor accuracy and stability of SAR aircraft target identification.
By establishing a physical semantic structure model of the aircraft, combining topological constraints and occlusion adaptive design, and adopting a coarse-fine positioning strategy and consistency constraints, we can achieve high robustness and high accuracy extraction of key scattering points.
It significantly improves the stability and adaptability of scattering point extraction under complex observation conditions, and enhances the accuracy and reliability of SAR aircraft target identification and attitude estimation.
Smart Images
Figure CN122391620A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of SAR image processing and automatic target recognition technology, and in particular to a method for extracting key scattering points of aircraft targets in SAR images guided by physical structure. Background Technology
[0002] Synthetic Aperture Radar (SAR) is an active microwave imaging system with the core advantages of all-weather, all-day, and long-range imaging. It is unaffected by weather conditions such as sunlight, clouds, fog, rain, or snow, making it irreplaceable in aerospace target monitoring. Unlike optical images, SAR images are essentially formed by the scattering interaction between electromagnetic waves and the target structure. The grayscale intensity of the image is determined by the electromagnetic scattering center generated by the target structure. Typical rigid structural parts of aircraft targets, such as the nose, wings, vertical tail, and engines, produce stable and strongly responsive electromagnetic scattering, which appears as high-brightness localized energy regions—the scattering centers—in SAR images.
[0003] Because aircraft targets exhibit strong geometric regularity, their scattering centers have a fixed structural correlation in space. Accurate extraction of key scattering points is the core foundation for achieving SAR aircraft target identification, orientation determination, and attitude estimation. However, in actual SAR imaging, the same aircraft target can appear vastly different under different observation conditions due to multiple factors, including the occlusion effect caused by side-looking imaging geometry, variations in imaging azimuth and elevation angles, speckle noise, and multipath scattering. Problems such as missing scattering points, weak response, and false scattering interference also exist.
[0004] In existing technologies, traditional extraction methods based on the overall shape and texture features of the target are difficult to stably describe the structural features of the aircraft under complex observation conditions. Most scattering point extraction methods do not explicitly model the physical semantic structure of the aircraft target, and cannot establish a mapping relationship between the inherent structure of the aircraft and the electromagnetic scattering center. It is also difficult to use the rigid structural laws of the aircraft to constrain the scattering point extraction process. At the same time, in response to the problem of missing scattering points in the occluded area caused by side-view imaging, existing methods lack effective semantic completion and inference mechanisms, resulting in poor robustness of extraction under different observation conditions, which seriously restricts the accuracy and stability of downstream tasks such as SAR aircraft target recognition. Summary of the Invention
[0005] The purpose of this invention is to propose a physical structure-guided method for extracting key scattering points of aircraft targets in SAR images. By deeply integrating the physical semantic structure of the aircraft with the electromagnetic scattering mechanism, and combining topological constraints and masking adaptive design, the method achieves high robustness and high precision extraction of key scattering points.
[0006] To achieve the above objectives, this invention provides a method for extracting key scattering points of aircraft targets from SAR images guided by physical structure, comprising the following steps: Step S1: Acquire a synthetic aperture radar (SAR) image containing the aircraft target, determine the area where the aircraft target is located, and generate the corresponding target bounding box; Step S2: Analyze the electromagnetic scattering mechanism of typical aircraft structures, establish a physical semantic structure model of aircraft targets, and define the set of key scattering points of aircraft targets; Step S3: Based on the inherent structural composition of the aircraft, construct the topological relationship of the scattering structure between key scattering points to form a topological map of the scattering structure of the aircraft target; Step S4: To address the geometric occlusion phenomenon in SAR side-looking imaging, a visibility labeling attribute for key scattering points is introduced, and a coarse-fine combined strategy is adopted to locate the key scattering points. Step S5: Incorporate scattering structure consistency constraints, optimize the localization results of key scattering points, and output the final key scattering point extraction results.
[0007] Preferably, in step S2, the physical semantic structure model of the aircraft target decomposes the aircraft into five major functional components: nose, cockpit connection section, wings, vertical tail and engine, and establishes a mapping relationship between the aircraft structure and SAR scattering mechanism: the nose and tail correspond to multi-faceted scattering, the wing tip corresponds to edge diffraction, and the engine corresponds to cavity or wall scattering.
[0008] Preferably, in step S2, the set of key scattering points includes ten scattering points with fixed semantics, which are, in order: nose scattering point, cockpit connection scattering point, left wing root scattering point, right wing root scattering point, left wing tip scattering point, right wing tip scattering point, vertical tail tip scattering point, vertical tail root scattering point, left engine scattering point, and right engine scattering point.
[0009] Preferably, in step S3, the scattering structure topological relationships are divided into three categories of core topological constraints, specifically including: The main axis structure topology of the fuselage is as follows: the nose scattering point, the cockpit connection scattering point, the vertical tail root scattering point and the vertical tail tip scattering point are connected in sequence to form the main axis structure of the aircraft fuselage. Wing structure topology: The scattering point at the root of the left wing and the scattering point at the root of the right wing are connected to form the wing-fuselage connection structure. The scattering point at the root of the left wing and the scattering point at the tip of the left wing are connected to form the left wing structure. The scattering point at the root of the right wing and the scattering point at the tip of the right wing are connected to form the right wing structure. Engine structural topology: The left wing root scattering point is connected to the left engine scattering point to form the left engine mounting structure, and the right wing root scattering point is connected to the right engine scattering point to form the right engine mounting structure.
[0010] Preferably, in step S4, the visibility annotation attribute includes three states: visible, semantically present but not visible, and non-existent.
[0011] Preferably, in step S4, the strategy of combining coarse and fine processing is as follows: Coarse localization stage: Local intensity analysis is performed on the target region of the SAR image to detect the location of local maxima in the image, determine the set of candidate scattering centers, and use the coordinates of candidate points that conform to the physical semantic structure model as the initial values for regression; Precise localization stage: Construct a key point detection network, input the SAR image of the target area, and output the target probability heatmap corresponding to each key scattering point. Determine the precise coordinates of visible key scattering points by the maximum response position of the heatmap; Combine the geometric prior of the aircraft physical semantic structure model to deduce the coordinates of semantically existing but invisible occluded scattering points.
[0012] Preferably, in step S5, the scattering structure consistency constraint includes three types of constraint terms: left-right symmetry constraint, fuselage main axis collinearity constraint, and wingspan constraint.
[0013] Preferred, left-right symmetry constraint: satisfies The formula for the symmetric loss function used to constrain the dimensional consistency of the left and right wing structures is as follows: ; in, The Euclidean distance between the two points is... The scattering point is at the root of the left wing. The scattering point is at the tip of the left wing. The scattering point is at the root of the right wing. The scattering point is at the tip of the right wing; Collinearity constraint of fuselage main axes: Constructed based on the linear structural characteristics of the aircraft fuselage, used to constrain the collinearity of key scattering points along the fuselage main axes, specifically: nose scattering points. Cockpit connection scattering point Vertical tail fin root scattering point and vertical tail tip scattering point Located on the same straight line; Wingspan constraint: satisfies the formula ,in The Euclidean distance between the two points is... The scattering point is at the tip of the left wing. The scattering point is at the tip of the right wing. The scattering point is at the root of the left wing. This is the scattering point at the root of the right wing.
[0014] Preferably, the three types of constraint terms are quantified and the localization results are optimized using a unified topological distance penalty constraint loss function. For any two key points... By calculating the predicted relative topological distance Relative to actual physical distance An exponential mapping kernel is used as the distance penalty constraint, and the distance penalty constraint loss function formula is as follows: ; in, This is the preset distance penalty coefficient.
[0015] Therefore, the present invention employs the above-described method for extracting key scattering points of aircraft targets in SAR images guided by physical structure, which has the following advantages: (1) This invention establishes a one-to-one mapping relationship between typical aircraft structures and electromagnetic scattering mechanisms by explicitly constructing a physical semantic structure model of aircraft targets. This breaks through the limitation of traditional methods that rely solely on image grayscale features, enabling the scattering point extraction process to have clear physical meaning and interpretability, and greatly improving the stability of feature extraction under complex observation conditions. (2) This invention introduces three types of visibility annotation attributes for the geometric occlusion characteristics of SAR side-view imaging. Combined with topological geometric priors, it realizes the semantic inference of scattering points in the occlusion area, solves the problem that traditional methods cannot effectively extract missing scattering points, and significantly improves the adaptability and robustness of the method to different imaging perspectives. (3) The present invention adopts a two-stage positioning strategy that combines coarse and fine positioning. It achieves rapid coarse positioning of scattering points through local intensity analysis and sub-pixel level precise positioning of visible scattering points through a deep learning key point detection network, thus taking into account both extraction efficiency and positioning accuracy. (4) This invention designs three types of core structural consistency constraints and realizes the quantitative fusion of multiple constraints through a unified exponential mapping kernel topological distance penalty loss function. It deeply embeds the rigid structural law of the aircraft into the scattering point optimization process, effectively suppresses the interference of pseudo scattering points, and ensures the consistency between the extracted scattering point topology and the real physical structure of the aircraft. It provides highly reliable structured features for downstream tasks such as SAR aircraft target recognition and attitude estimation.
[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0017] Figure 1 This is a flowchart of a method for extracting key scattering points of aircraft targets in SAR images guided by physical structure, as described in an embodiment of the present invention. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0019] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.
[0020] Example like Figure 1 As shown in the figure, this embodiment proposes a method for extracting key scattering points of aircraft targets in SAR images guided by physical structure. The specific implementation steps are as follows: Step S1: SAR image acquisition and target region determination: This embodiment uses an airborne SAR image with a spatial resolution of 0.5 meters, in the X-band, and in strip imaging mode. First, the original SAR image is preprocessed, including radiometric calibration, Lee filter speckle noise suppression, and geometric correction, to obtain a preprocessed SAR image. Then, the Faster R-CNN target detection algorithm is used to detect aircraft targets in the preprocessed SAR image, outputting axis-aligned bounding boxes for the aircraft targets. The resulting image region containing only a single aircraft target is cropped, and the image size is uniformly scaled to 512×512 pixels for subsequent processing.
[0021] Step S2: Establishment of the aircraft physical semantic structure model and definition of key scattering points: Based on the electromagnetic scattering mechanism analysis of aircraft targets, civil airliner targets are decomposed into five major functional components: nose, cockpit connection section, wing, vertical tail, and engine. A structure-scattering mapping relationship is established: the conical structure of the nose and the angular structure of the tail correspond to multi-faceted angular scattering, which are strong scattering sources in the direction of radar wave incident; the edge structures of the leading and trailing edges and wingtips of the wings correspond to edge diffraction, with the wingtips being stable strong scattering points; the cavity structure of the engine air intake corresponds to cavity scattering, which has a strong and stable backscattering response.
[0022] Based on the above mapping relationship, 10 key scattering points with fixed semantics are defined and numbered sequentially as follows: - : The nose scattering point is located at the very front of the aircraft's nose, corresponding to the center of strong multi-faceted scattering. The cockpit connection scattering point is located at the connection between the cockpit and the main fuselage, corresponding to the composite angular scattering at the front of the fuselage. The scattering point at the root of the left wing is located at the junction of the left wing and the fuselage. The scattering point at the root of the right wing is located at the junction of the right wing and the fuselage. The scattering point at the tip of the left wing is located at the wingtip of the left wing and corresponds to edge diffraction scattering. The right wingtip scattering point is located at the wingtip of the right wing, corresponding to edge diffraction scattering. The vertical tail tip scattering point is located at the very top of the aircraft's vertical tail. The vertical tail root scattering point is located at the junction of the vertical tail and the tail of the fuselage. The left engine scattering point is located at the air intake of the left-side suspended engine, corresponding to the cavity scattering. The right engine scattering point is located at the intake duct of the right-side suspended engine, corresponding to the cavity scattering.
[0023] Step S3: Constructing the topological relationships of the scattering structure: Based on the inherent structural connections of civil aircraft, the topological connections of 10 key scattering points are established to form a scattering structure topology map, which is specifically divided into three categories: Fuselage main shaft structure topology: Establishment , , The connection relationship forms the longitudinal main axis of the aircraft fuselage, representing the aircraft's orientation and fuselage length; Wing structure topology: Establishment The connection relationship characterizes the lateral width of the connection between the wing and the fuselage; establishing , The connection relationships represent the spanwise structure of the left and right wings, respectively; Engine structure topology: Establishment , The connection relationships represent the installation positions of the left and right engines and the wing root, respectively.
[0024] Step S4: Coarse-to-fine combined scattering point localization based on visibility annotations: To address the issue in SAR side-looking imaging where the aircraft's wings and engines, on the side furthest from the radar, easily enter shadow areas, resulting in missing scattered signals, a visibility labeling attribute is assigned to each key scattering point. In this embodiment, the right side of the aircraft is furthest from the radar, therefore... , , Points marked as "semantically present but not visible" are labeled as "visible," while other scattering points are labeled as "visible." A two-stage localization strategy combining coarse and fine mapping is employed. Coarse localization stage: A 5×5 sliding window is used to traverse the target area image, and the local gray-level maxima in each window are detected. Candidate scattering points with gray-level values higher than 3 times the average gray-level of the image are selected, resulting in 28 candidate scattering centers. Combined with the physical semantic structure model in step S2, 10 candidate points that conform to the aircraft structure position distribution are selected, and their pixel coordinates are used as the initial values for regression. In the fine localization stage: a second-order stacked hourglass network is used as the key point detection network. The network input is a 512×512 SAR image of the target area, and the output is 10 single-channel probability heatmaps, each corresponding to a key scattering point; the maximum response location of each heatmap is extracted to obtain 7 visible scattering points. , , , , , , The precise pixel coordinates of the scattering points are obtained; based on the topological geometric priors of the fuselage main axis and wing structure, and using the visible scattering points as a reference, the three invisible scattering points are derived through the proportional relationship of the rigid structure. , , The coordinates of ).
[0025] Step S5: Optimization of scattering points based on structural consistency constraints: Three types of scattering structure consistency constraints are introduced, and the localization results are optimized using a unified topological distance penalty loss function: 1. Left-right symmetry constraint: The left and right wings are constrained to have the same length, and the symmetry loss function is: ; in, The Euclidean distance in pixel coordinates; 2. Focal shaft collinearity constraint: constraint , , , The four scattering points are approximately collinear. The optimal straight line of the four points is fitted by the least squares method, and the sum of the squares of the perpendicular distances from each point to the straight line is calculated as the collinearity loss. 3. Wingspan constraint: The wingspan is constrained to be greater than the distance between the wing roots, i.e. A penalty loss is incurred when this constraint is not met.
[0026] Any two key points corresponding to the three types of constraints Construct a unified topological distance penalty loss function: ; Among them, the distance penalty coefficient Set to 0.5. The prior value is the standard physical distance mapping of the key scattering point pairs of this type of civil aircraft to the pixel coordinate system.
[0027] By using the above structural constraints, the detection results of key scattering points can be optimized, thereby improving the stability and accuracy of scattering point extraction.
[0028] Using the above method, key scattering points of aircraft targets can be stably extracted from SAR images, and the topological relationship of the aircraft scattering structure can be established, realizing the semantic expression of the structural features of aircraft targets, and providing reliable structural features for subsequent SAR aircraft target recognition, structural analysis and attitude estimation.
[0029] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for extracting key scattering points of aircraft targets from SAR images guided by physical structure, characterized in that, Includes the following steps: Step S1: Acquire a synthetic aperture radar (SAR) image containing the aircraft target, determine the area where the aircraft target is located, and generate the corresponding target bounding box; Step S2: Analyze the electromagnetic scattering mechanism of typical aircraft structures, establish a physical semantic structure model of aircraft targets, and define the set of key scattering points of aircraft targets; Step S3: Based on the inherent structural composition of the aircraft, construct the topological relationship of the scattering structure between key scattering points to form a topological map of the scattering structure of the aircraft target; Step S4: To address the geometric occlusion phenomenon in SAR side-looking imaging, a visibility labeling attribute for key scattering points is introduced, and a coarse-fine combined strategy is adopted to locate the key scattering points. Step S5: Incorporate scattering structure consistency constraints, optimize the localization results of key scattering points, and output the final key scattering point extraction results.
2. The method for extracting key scattering points of aircraft targets in SAR images guided by physical structure according to claim 1, characterized in that: In step S2, the physical semantic structure model of the aircraft target decomposes the aircraft into five major functional components: nose, cockpit connection section, wings, vertical tail and engine, and establishes the mapping relationship between the aircraft structure and SAR scattering mechanism: the nose and tail correspond to multi-faceted scattering, the wing tip corresponds to edge diffraction, and the engine corresponds to cavity or wall scattering.
3. The method for extracting key scattering points of aircraft targets in SAR images guided by physical structure according to claim 1, characterized in that: In step S2, the key scattering point set contains ten scattering points with fixed semantics, namely: nose scattering point, cockpit connection scattering point, left wing root scattering point, right wing root scattering point, left wing tip scattering point, right wing tip scattering point, vertical tail tip scattering point, vertical tail root scattering point, left engine scattering point, and right engine scattering point.
4. The method for extracting key scattering points of aircraft targets in SAR images guided by physical structure according to claim 1, characterized in that: In step S3, the scattering structure topological relationships are divided into three types of core topological constraints, specifically including: The main axis structure topology of the fuselage is as follows: the nose scattering point, the cockpit connection scattering point, the vertical tail root scattering point and the vertical tail tip scattering point are connected in sequence to form the main axis structure of the aircraft fuselage. Wing structure topology: The scattering point at the root of the left wing and the scattering point at the root of the right wing are connected to form the wing-fuselage connection structure. The scattering point at the root of the left wing and the scattering point at the tip of the left wing are connected to form the left wing structure. The scattering point at the root of the right wing and the scattering point at the tip of the right wing are connected to form the right wing structure. Engine structural topology: The left wing root scattering point is connected to the left engine scattering point to form the left engine mounting structure, and the right wing root scattering point is connected to the right engine scattering point to form the right engine mounting structure.
5. The method for extracting key scattering points of aircraft targets in SAR images guided by physical structure according to claim 1, characterized in that: In step S4, the visibility annotation attribute includes three states: visible, semantically present but not visible, and non-existent.
6. The method for extracting key scattering points of aircraft targets in SAR images guided by physical structure according to claim 1, characterized in that: In step S4, the strategy of combining coarse and fine processing is as follows: Coarse localization stage: Local intensity analysis is performed on the target region of the SAR image to detect the location of local maxima in the image, determine the set of candidate scattering centers, and use the coordinates of candidate points that conform to the physical semantic structure model as the initial values for regression; Precise localization stage: Construct a key point detection network, input a SAR image of the target area, output a target probability heatmap corresponding to each key scattering point, and determine the precise coordinates of visible key scattering points by the location of the maximum response in the heatmap; By combining the geometric priors of the aircraft physical semantic structure model, the coordinates of the occultation scattering points that exist semantically but are not visible are deduced.
7. The method for extracting key scattering points of aircraft targets in SAR images guided by physical structure according to claim 1, characterized in that: In step S5, the scattering structure consistency constraint includes three types of constraint terms: left-right symmetry constraint, fuselage main axis collinearity constraint, and wingspan constraint.
8. The method for extracting key scattering points of aircraft targets in SAR images guided by physical structure according to claim 7, characterized in that: Left-right symmetry constraint: satisfies The formula for the symmetric loss function used to constrain the dimensional consistency of the left and right wing structures is as follows: ； in, The Euclidean distance between the two points is... The scattering point is at the root of the left wing. The scattering point is at the tip of the left wing. The scattering point is at the root of the right wing. The scattering point is at the tip of the right wing; Collinearity constraint of fuselage main axes: Constructed based on the linear structural characteristics of the aircraft fuselage, used to constrain the collinearity of key scattering points along the fuselage main axes, specifically: nose scattering points. Cockpit connection scattering point Vertical tail fin root scattering point and vertical tail tip scattering point Located on the same straight line; Wingspan constraint: satisfies the formula ,in The Euclidean distance between the two points is... The scattering point is at the tip of the left wing. The scattering point is at the tip of the right wing. The scattering point is at the root of the left wing. This is the scattering point at the root of the right wing.
9. The method for extracting key scattering points of aircraft targets in SAR images guided by physical structure according to claim 8, characterized in that: The three types of constraints are quantified and the localization results are optimized using a unified topological distance penalty constraint loss function for any two key points. By calculating the predicted relative topological distance Relative to actual physical distance An exponential mapping kernel is used as the distance penalty constraint, and the distance penalty constraint loss function formula is as follows: ； in, This is the preset distance penalty coefficient.