CN-122023656-A - Space target three-dimensional reconstruction method based on radar multi-view observation under low signal-to-noise ratio

CN122023656ACN 122023656 ACN122023656 ACN 122023656ACN-122023656-A

Abstract

The invention discloses a space target three-dimensional reconstruction method based on radar multi-view observation under low signal-to-noise ratio. Firstly, preprocessing such as imaging and diversity is carried out on multi-angle observation data, so that a three-dimensional reconstruction architecture based on self-supervision and geometric structure consistency is constructed, and secondly, a self-adaptive three-dimensional reconstruction network is designed based on multi-angle observation information, so that three-dimensional reconstruction is realized adaptively and robustly. Finally, the joint constraint and optimization of the three-dimensional reconstruction network parameters are realized by designing a loss function based on J-invariance denoising and a loss function based on geometric structure consistency, so that the three-dimensional reconstruction precision is ensured. The invention aims to provide a three-dimensional intelligent reconstruction method based on multi-angle observation, which is suitable for observing scenes and unknown target morphology, and can be expected to be applied to three-dimensional imaging of space targets by a foundation radar.

Inventors

DING ZEGANG
LI LINGHAO
Zheng Pengnan
WANG GUANXING

Assignees

北京理工大学

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (6)

1. A space target three-dimensional reconstruction method based on radar multi-view observation under low signal-to-noise ratio is characterized by comprising the following steps: firstly, quantitatively analyzing a projection relation between a target three-dimensional structure and a radar two-dimensional image, and establishing a radar three-dimensional imaging model; Step two, constructing a mapping conforming to J-invariance by utilizing mutual independence of radar images acquired by each angle through a multi-angle observation process, and further designing a self-supervision three-dimensional reconstruction architecture by utilizing multi-angle observation information, thereby realizing end-to-end mapping from a low signal-to-noise ratio two-dimensional image to a high-precision three-dimensional structure; And thirdly, designing a self-adaptive three-dimensional reconstruction network based on multi-angle observation information, and restraining the network through J-invariance and geometric structure consistency, so that the accuracy and the robustness of three-dimensional reconstruction are improved.
2. The method for three-dimensional reconstruction of a spatial target based on radar multi-view observation at a low signal-to-noise ratio according to claim 1, wherein in the first step, the two-dimensional radar image of the target is a projection of a three-dimensional structure thereof on an imaging plane, and the method is modeled by mathematical deduction: (1); Where kr represents a distance projection vector, kd represents a Doppler projection vector, r represents a target distance dimension coordinate, d represents a target Doppler dimension coordinate, and q represents a target three-dimensional coordinate.
3. The method for three-dimensional reconstruction of a spatial target based on radar multi-view observation at a low signal-to-noise ratio according to claim 1, wherein in the second step, a self-supervision three-dimensional reconstruction architecture using multi-view observation information is designed, so that the training process based on the self-supervision architecture is equivalent to a supervised training process: Inputting a noisy image, a distance and a Doppler projection vector; and outputting a three-dimensional reconstruction result.
4. The method for three-dimensional reconstruction of a spatial target based on radar multi-view observation at a low signal-to-noise ratio as defined in claim 1, wherein in the second step, a map g conforming to J invariance is constructed, namely ; The optimal mapping g is required to minimize the following: (2); Wherein, the , The method is characterized by comprising the following steps of: (3); Is that Is an unbiased estimate of (1), Is a clean image.
5. The method for three-dimensional reconstruction of a spatial target based on radar multi-view observation at a low signal-to-noise ratio as defined in claim 1, wherein in step three, a loss function conforming to J invariance is constructed Optimal parameters Expressed as: (4); Wherein FC represents the complement of F; Output of network The method comprises the following steps: (5); Wherein, the And N fix represents a fixed noise floor.
6. The method for three-dimensional reconstruction of a spatial target based on radar multi-view observation at a low signal-to-noise ratio according to claim 1, wherein in the third step, a three-dimensional geometric consistency loss function is introduced assuming that v three-dimensional reconstruction results share a consistent three-dimensional geometric structure The geometric consistency of the three-dimensional reconstruction result under different view angles is restrained and enhanced, and the specific form is as follows: (6); wherein S is a three-dimensional reconstruction result, and further, the method can be used for reconstructing And Combining, thereby realizing self-adaptive and robust low signal-to-noise ratio high precision three-dimensional reconstruction, as shown in the following formula: 。

Description

Space target three-dimensional reconstruction method based on radar multi-view observation under low signal-to-noise ratio Technical Field The invention relates to a space target three-dimensional reconstruction method based on radar multi-view observation under a low signal-to-noise ratio, and belongs to the technical field of radars. Background The radar imaging technology can acquire high-resolution images of targets all the day and all the weather, so that the radar imaging technology is widely applied to the fields of remote sensing detection, target identification, topographic mapping and the like. Conventional radar imaging is mainly two-dimensional imaging, such as Synthetic Aperture Radar (SAR) and Inverse Synthetic Aperture Radar (ISAR) imaging methods, which are mainly output as range-doppler two-dimensional images. Compared with two-dimensional imaging, the radar three-dimensional imaging can describe the spatial structure information of the target more completely, and has important application value in complex target analysis and fine modeling. However, in practical application, the radar imaging process is inevitably affected by factors such as system noise, environmental clutter and the like, so that the signal-to-noise ratio of an imaging result is low, and the accuracy and the robustness of the three-dimensional reconstruction of the radar are seriously affected. Existing radar three-dimensional imaging and reconstruction methods typically rely on high quality observations or object prior models, such as known object scattering characteristics, sparsity assumptions, or geometric constraints. When the observation environment is complex and the target morphology characteristics are unknown, the traditional methods are difficult to obtain high-precision three-dimensional reconstruction results. In addition, multi-view radar observation can provide information of targets under different observation geometries, but the existing method often fails to fully utilize the geometric consistency relationship between multi-angle observations, and manual annotation or a real three-dimensional model is usually required to serve as supervision information, so that the application of the multi-view radar observation in an actual scene is limited. Therefore, research on a method capable of fully utilizing multi-view radar observation data to realize high-precision three-dimensional reconstruction under the condition of lacking target and scene prior information is needed. Based on the method, the self-supervision radar three-dimensional reconstruction framework combining J-invariance and geometric structure consistency is provided, self-supervision constraint is constructed through multi-angle observation information, effective noise suppression is achieved, and the target three-dimensional reconstruction precision is improved under the condition that a real three-dimensional morphology is not needed. Disclosure of Invention The three-dimensional reconstruction method of the space target based on radar multi-view observation under the low signal-to-noise ratio is provided for solving the problem that the traditional radar three-dimensional reconstruction method fails when the prior information of the target scene is unknown and the observation signal-to-noise ratio is low. The invention is realized by the following technical scheme: firstly, quantitatively analyzing a projection relation between a target three-dimensional structure and a radar two-dimensional image, and establishing a radar three-dimensional imaging model; Step two, constructing a mapping conforming to J-invariance by utilizing mutual independence of radar images acquired by each angle through a multi-angle observation process, and further designing a self-supervision three-dimensional reconstruction architecture by utilizing multi-angle observation information, thereby realizing end-to-end mapping from a low signal-to-noise ratio two-dimensional image to a high-precision three-dimensional structure; And thirdly, designing a self-adaptive three-dimensional reconstruction network based on multi-angle observation information, and restraining the network through J-invariance and geometric structure consistency, so that the accuracy and the robustness of three-dimensional reconstruction are improved. Drawings FIG. 1, a flow chart of the proposed method; FIG. 2, schematic illustration of CST electromagnetic simulation of a spatial target; FIG. 3 is a three-dimensional reconstruction result under different methods based on CST electromagnetic simulation data; fig. 4 is a quantitative evaluation result of three-dimensional reconstruction results obtained under different methods based on CST electromagnetic simulation data. Advantageous effects 1. According to the method, under the condition that the three-dimensional structure of the target and the prior information of the scene are unknown, the high-precision three-dimensional reconstruction of th