A method for segmenting complex star spots based on photometric-geometric hybrid terrain for star sensors with large aberrations

By constructing a photometric-geometric hybrid potential energy surface and performing a watershed immersion transformation on it, the problem of unstable star spot segmentation under large aberration conditions was solved, and a more stable star spot segmentation effect was achieved.

CN123118664APending Publication Date: 2026-11-17CHANGCHUN INST OF OPTICS FINE MECHANICS & PHYSICS CHINESE ACAD OF SCI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611182527.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-05
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Under conditions of large aberrations, existing star spot segmentation methods struggle to maintain the integrity of star spot geometric boundary information and lack joint modeling of luminosity continuity and geometric topology, leading to unstable segmentation results, especially prone to misjudgment in overlapping star points and complex backgrounds.

Method used

By constructing a potential energy surface based on a photometric-geometric hybrid terrain, and combining the geometric distance field and the physical light intensity field for weighted fusion, a hybrid potential energy terrain is constructed, and a watershed immersion transformation is performed on it to achieve robust segmentation of star spots.

Benefits of technology

It improves the stability and accuracy of star spot segmentation under large aberration conditions, effectively suppresses false maxima and saddle points, ensures that segmentation ridges are formed along true energy troughs, and improves the stability and symmetry of segmentation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118664A_ABST
    Figure CN123118664A_ABST
Patent Text Reader

Abstract

This invention relates to a complex star spot segmentation method based on a photometric-geometric hybrid terrain for star sensors with large aberrations. It falls within the field of star sensor image processing and target segmentation technology. Existing star spot segmentation methods do not address the technical problem of topological distortion under large aberration conditions from a physical mechanism perspective. This method includes the following steps: acquiring the star image to be segmented and foreground extraction; constructing a geometric distance field; extracting a physical light intensity field; dimensionless normalization processing; constructing a hybrid potential energy terrain; hybrid watershed segmentation; and outputting the results. This invention addresses the non-convex boundary and topological breakage problems of star spots with large aberrations by explicitly introducing photometric continuity information within the star spot, enabling the segmentation potential energy surface to simultaneously possess geometric and physical energy constraints, avoiding pseudo-topology caused by relying solely on morphological contours.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of star sensor image processing and target segmentation technology, and in particular to a complex star spot segmentation method based on photometric-geometric hybrid terrain for star sensors with large aberrations. Background Technology

[0002] Star sensors measure the attitude of a spacecraft by observing the positions of stars, which relies on the stable extraction and high-precision positioning of star points. As the requirements for lightweight, miniaturized, and low-power payloads in microsatellites, near-space vehicles, and high-altitude platforms continue to increase, the optical systems at the front end of star sensors sometimes have to be simplified. Reducing the number of mirrors, compressing structural dimensions, and decreasing the degrees of freedom for aberration correction can lead to significant asymmetric stretching, tailing, and local contraction of the point spread function in the off-axis field of view. When two star points are spatially close or locally adhere in a high-density star field, traditional segmentation methods based on binary contours or geometric centers are prone to over-segmentation, under-segmentation, and ridge offset, further affecting the stability of subsequent centroid estimation and attitude calculation.

[0003] The existing technologies most similar to this invention fall into three main categories. The first category includes traditional thresholding, connected component extraction, and gray-scale centroid methods. These methods are simple to implement and computationally inexpensive, but typically assume that the star point outline is approximately centrally symmetric or locally convex. When large aberrations cause star spots to exhibit tails, crescent shapes, comet tails, and non-convex boundaries, the geometric center of the binary outline often deviates from the true physical center, and adhered regions are difficult to segment correctly. The second category is the traditional watershed algorithm based on Euclidean distance transformation. This method maps the binary foreground region to a distance field and then finds the segmentation ridge line through an immersion process. For regular, convex targets, their potential terrain is usually a single-peak structure, and the watershed can perform segmentation relatively stably; however, for star spots with large aberrations, non-convex boundaries can induce multiple false local maxima and false saddle points in the distance field, leading to incorrect catchment basins and oversegmentation. The third category includes methods for small target detection and star point separation in complex backgrounds, such as local contrast enhancement, one-dimensional Gaussian correlation thresholding, region growing, and morphological filtering. These methods are effective in background suppression, but their main focus is on improving target saliency rather than correcting topological distortion under large aberration conditions from a physical mechanism perspective.

[0004] Therefore, although existing schemes can achieve basic segmentation in regular star spots or weakly degenerate scenes, when the star spot morphology is affected by strong aberrations and deviates from the convex topology, especially when binary star adhesion, high-density star map and complex background work together, there is still a lack of a robust segmentation method that can preserve the geometric boundary information of star spots and use physical photometric continuity to correct pseudo-topological structures.

[0005] The existing technology has at least the following shortcomings:

[0006] (1) Traditional grayscale thresholding, region growing and centroid methods mainly rely on the geometric integrity of the target contour. When star spots have obvious tailing, bending and local concavity at the edge of the field of view, the topology of the extracted region is no longer consistent with the real energy distribution, and the segmentation results are prone to misjudgment.

[0007] (2) Traditional watershed methods based on Euclidean distance transformation only use geometric distance information to construct the potential energy surface. For non-convex star spots caused by large aberrations, the distance field may produce multiple false peaks and false ridges, causing the same physically continuous star point to be incorrectly cut into multiple fragments, or pushing the real binary star dividing line away from the actual energy trough.

[0008] (3) Most existing complex background suppression algorithms focus on noise suppression or saliency enhancement, lacking explicit modeling of the interaction mechanism between the luminosity continuity and geometric topology inside the star spot, making it difficult to alleviate the topological ambiguity caused by the simplified optical system from the source.

[0009] (4) For overlapping star point scenes, if only traditional distance field or simple morphological operations are used, the position of the segmented ridge line will be highly sensitive to background noise, edge burrs and local threshold fluctuations, resulting in insufficient separation limit and the output area does not have good symmetry and stability. Summary of the Invention

[0010] The present invention aims to address the problem that the main focus of existing star spot segmentation methods is on improving target saliency rather than correcting topological distortion under large aberration conditions from a physical mechanism perspective. It provides a complex star spot segmentation method based on photometric-geometric hybrid terrain for star sensors with large aberrations.

[0011] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0012] A complex star spot segmentation method based on photometric-geometric hybrid terrain for star sensors with large aberrations includes the following steps:

[0013] Step 1: Obtain the star map to be segmented and extract the foreground: Process the observed image to obtain the effective foreground region of the target. and its corresponding background area ;

[0014] Step 2: Construct the geometric distance field: for the foreground region Each valid pixel within Calculate its distance to the nearest background region The geometric distances of the boundaries form a geometric distance field characterizing the topological structure. ;

[0015] Step 3: Extract the physical light intensity field: Extract the true physical light intensity distribution of the corresponding foreground pixels from the original observed image. , as a physical constraint on photometric continuity;

[0016] Step 4: Dimensionless normalization: for the geometric distance field With physical light intensity field Perform range normalization separately within the foreground region to obtain a dimensionless normalized geometric field. With normalized photometric field ;

[0017] Step 5: Constructing Hybrid Potential Energy Terrain: Introducing Regularization Weighting Coefficients Normalized geometric field With normalized photometric field A linear weighted fusion is performed, and its negative mapping is taken to construct a hybrid potential surface oriented towards the watershed immersion transformation. ;

[0018] Step 6: Mixed watershed segmentation: at the mixed potential energy surface Perform watershed submersion transformation to find potential energy minimum points and extract catchment basins and their dividing ridges to achieve robust separation of overlapping star points or complex trailing star points.

[0019] Step 7: Output Results: The segmentation results are labeled with connected components, numbered, and output as candidate sub-regions to obtain a set of target regions for subsequent detection, matching, or localization processing.

[0020] In the above technical solution, step 1 involves processing the observed image, specifically as follows:

[0021] Local adaptive thresholding, or

[0022] Equivalent prospect extraction, or

[0023] Local background estimation subtraction, or

[0024] Regional growth, or

[0025] Significance detection.

[0026] In the above technical solution, step 2 targets the foreground region. Each valid pixel within Calculate its distance to the nearest background region The geometric distance of the boundary is specifically:

[0027] Minimum Euclidean distance, or

[0028] Weighted distance, or

[0029] Geodetic distance, or

[0030] Segmented distance constraints.

[0031] In the above technical solution, in step 5, the normalized geometric field With normalized photometric field Weighted fusion is performed, specifically as follows:

[0032] Linear weighted fusion, or

[0033] Segmented weighted fusion, or

[0034] Adaptive weight fusion.

[0035] In the above technical solution, step 3 specifically includes:

[0036] For any pixel belonging to the foreground region The minimum Euclidean distance to the nearest background boundary is defined as:

[0037]

[0038] In the formula, Background area arbitrary pixel coordinates in; For geometric distance field.

[0039] In the above technical solution, step 4 specifically includes:

[0040] For geometric distance field With physical light intensity field Each is uniformly mapped to In the dimensionless probability space, the dimensionless normalized geometric fields are obtained respectively. With normalized photometric field :

[0041]

[0042]

[0043] In the formula, and These are the minimum and maximum values ​​within the connected region, respectively.

[0044] In the above technical solution, step 5 specifically includes:

[0045] Normalized geometric field With normalized photometric field A linear weighted fusion is performed, and its negative mapping is taken to construct a hybrid potential surface oriented towards the watershed immersion transformation. :

[0046]

[0047] In the formula, To adjust the normalized geometric field With normalized photometric field The regularization coefficients of the weights of both.

[0048] In the above technical solution, step 6 specifically includes:

[0049] Mixed potential energy surface The topological surface is input into the watershed immersion algorithm; during the immersion process, the water flow will converge along the region that has both real light intensity trough and geometric contraction characteristics; the final generated segmentation ridge is anchored at the boundary of real physical energy.

[0050] Compared with existing methods that use a single geometric distance field watershed, simple threshold connected component extraction, or traditional centroid segmentation, this invention has at least the following advantages:

[0051] (1) To address the non-convex boundary and topological breakage issues of star spots with large aberrations, the luminosity continuity information inside the star spot is explicitly introduced, so that the segmented potential energy surface has both geometric and physical energy constraints, thus avoiding pseudo-topology caused by relying solely on morphological contours.

[0052] (2) By constructing a hybrid potential energy surface and performing a watershed immersion transformation, it is possible to effectively suppress false maxima and false saddle points in the range field, making the segmentation ridge line closer to the energy trough region between targets.

[0053] (3) Compared with the traditional distance transformation watershed method, the present invention has a smaller separable spacing and segmentation stability in the overlapping binary star scenario, especially under large off-axis conditions; when the limit resolution distance is similar, it shows better segmentation stability and smaller error fluctuation, which has been verified in simulation and semi-physical experiments. Attached Figure Description

[0054] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0055] Figure 1 This is a schematic diagram of the overall process for a robust segmentation method for complex star spots. It illustrates the complete processing chain, including foreground extraction, distance field construction, luminosity normalization, mixed potential surface generation, and watershed output.

[0056] Figure 2 This is a typical input sample image for simulating overlapping binary stars. It is used to demonstrate the foreground morphology of a bonded binary star system formed under conditions of large aberrations and background noise.

[0057] Figure 3This is a schematic diagram comparing the pure geometric distance field and the mixed potential energy surface. It is used to illustrate that after introducing luminosity continuity, pseudo-local peaks and pseudo-ridges are suppressed, and true energy troughs are more likely to form stable boundaries.

[0058] Figure 4 This is a schematic diagram of the vertical cross-section and normalized one-dimensional profile curve along the principal physical axis of the binary star system. It is used to illustrate the local potential energy difference at the true boundary.

[0059] Figure 5 This is a schematic diagram of the error comparison curve under the two-star spacing scan. It is used to compare the minimum separable spacing and error variation trend of the method of this invention with several common processing methods in the case of overlapping star points.

[0060] Figure 6 This is a comparison chart of the segmentation ridges and output sub-regions of a traditional watershed (top) and a hybrid watershed (bottom). It is used to illustrate the differences between the two methods in terms of ridge location, regional symmetry, and boundary regularity.

[0061] Figure 7 The following are performance comparison charts of two segmentation methods under different DMD micromirror spacings: (a) Segmentation success rate comparison chart; (b) Region area uniformity comparison chart; (c) Double PSF fitting NCC comparison chart; (d) Spacing error comparison chart. Detailed Implementation

[0062] The inventive concept of this invention is as follows:

[0063] The present invention provides a complex star spot segmentation method based on photometric-geometric hybrid terrain for star sensors with large aberrations. By introducing photometric continuity information inside the star spot into the traditional geometric distance field, a hybrid potential energy surface is constructed and a watershed transformation is performed on the potential energy surface. This suppresses pseudo-maxima and pseudo-ridges induced by non-convex boundaries without significantly increasing computational complexity, and improves the segmentation stability under conditions of overlapping star points, trailing star points, and complex backgrounds. This provides reliable candidate regions for subsequent star point detection, matching, or localization processing.

[0064] The core mechanism of the method of this invention is that, in order to address the topological breakage problem caused by the non-convex degradation of star spots with large aberrations, it no longer relies solely on a single geometric distance metric, but constructs a hybrid potential energy terrain by combining geometric centripetal constraints and photometric continuity constraints, enabling the watershed algorithm to form stable segmentation ridges along the valleys of real photon energy.

[0065] The present invention provides a method for segmenting complex star spots based on photometric-geometric hybrid terrain for star sensors with large aberrations, such as... Figure 1 As shown, the basic process is as follows:

[0066] Step 1: Obtain the star map to be segmented and extract the foreground: Perform local adaptive thresholding or equivalent foreground extraction on the observed image to obtain the effective foreground region of the target. and its corresponding background area ;

[0067] Step 2: Construct the geometric distance field: for the foreground region Each valid pixel within Calculate its distance to the nearest background region The minimum Euclidean distance of the boundary forms the geometric distance field characterizing the topological structure. ;

[0068] Step 3: Extract the physical light intensity field: Extract the true physical light intensity distribution of the corresponding foreground pixels from the original observed image. , as a physical constraint on photometric continuity;

[0069] Step 4: Dimensionless normalization: for the geometric distance field With physical light intensity field Perform range normalization separately within the foreground region to obtain a dimensionless normalized geometric field. With normalized photometric field ;

[0070] Step 5: Constructing Hybrid Potential Energy Terrain: Introducing Regularization Weighting Coefficients By linearly weighting and fusing the two types of normalized fields and taking their negative mapping, a hybrid potential surface oriented towards the watershed immersion transformation is constructed. ;

[0071] Step 6: Mixed watershed segmentation: on the mixed potential energy surface Perform watershed submersion transformation to find the potential energy minimum point (valley bottom) and extract the catchment basin and its dividing ridge line, thereby achieving robust separation of overlapping star points or complex trailing star points;

[0072] Step 7: Output Results: The segmentation results are labeled with connected components, numbered, and output as candidate sub-regions to obtain a set of target regions that can be used for subsequent detection, matching, or localization processing.

[0073] The present invention will now be described in detail with reference to the accompanying drawings.

[0074] The technical solution of this invention will be further explained below in conjunction with the foreground region extraction, geometric topological ambiguity analysis, hybrid potential energy construction and watershed segmentation process.

[0075] i. Foreground region extraction for star spots with large aberrations

[0076] (Step 1) In the original star map, target stars are typically superimposed on sky background, readout noise, and shot noise. To limit computational redundancy and noise interference in subsequent watershed operations, this invention first performs local adaptive thresholding and local background estimation subtraction on the observed image. The set of effective target foreground pixels in the extracted full image is defined as follows: The corresponding set of background pixels is This step provides an initial target topological support domain, which strictly constrains the subsequent construction of the potential energy surface. Internally, it reduces the interference of complex background undulations on the overall terrain.

[0077] ii. Topological ambiguity caused by geometric nonconvexity

[0078] In the classic watershed segmentation framework, binary foreground regions are typically relied upon. The constructed range transformation field. For an ideal rotationally symmetric star spot, its range potential field is unimodal; however, in a simple large aberration optical system, the star spot boundary will exhibit severe asymmetric stretching, bending, and local contraction, causing the target geometry to degenerate from convex to non-convex.

[0079] Such non-convex boundaries generate multiple spurious local maxima and pseudo-saddle points connecting them in a pure range field. The process of water inundation accumulates incorrectly along these pseudo-saddle points, producing redundant dividing ridges that fragment what was originally a physically continuous star point into multiple pieces. Therefore, a pure geometric range field cannot fully represent the physical continuity of star spots with large aberrations.

[0080] iii. Methods for constructing photometric-geometric hybrid potential energy terrain

[0081] To overcome the topological ambiguity caused by a single geometric metric, this invention proposes fusing luminosity continuity information within star spots within a geometric distance field. The specific computational model is as follows:

[0082] a) Calculate the geometric distance field (step 2):

[0083] For any pixel belonging to the foreground region The minimum Euclidean distance to the nearest background boundary is defined as:

[0084]

[0085] In the formula, Background area The coordinates of any pixel in the array. Distance value. The larger the value, the closer the point is to the center of the target in terms of geometric topology; conversely, the smaller the value, the closer it is to the edge of the boundary.

[0086] b) Extract the physical light intensity field (step 3):

[0087] Simultaneous extraction of the foreground spatial domain Original light intensity distribution within .

[0088] c) Independent range normalization (step 4):

[0089] To avoid fusion failure caused by significant differences in the numerical range between geometric dimensions (pixel distance) and radiation dimensions (ADU or electron number), both must be uniformly mapped to... In the dimensionless probability space:

[0090]

[0091]

[0092] In the formula, and These are the minimum and maximum values ​​within the connected region, respectively. For normalized geometric fields; This is the normalized luminosity field.

[0093] d) Constructing mixed potential energy terrain (step 5):

[0094] Normalized geometric field With normalized photometric field By performing linear weighted fusion and taking its negative mapping, the mixed potential energy surface required by the watershed algorithm to find the "bottom" is constructed. :

[0095]

[0096] In the formula, To adjust the regularization coefficients for the weights of both, in the star spot energy core and high-brightness tail regions, the luminosity term provides physical support for the geometrically contracted areas through its continuous and monotonic energy distribution, effectively smoothing out spurious maxima induced by the range field. In the edge regions with low signal-to-noise ratios, the geometric term provides centripetal constraints, suppressing boundary noise and contour spikes. In this way, the constructed potential energy surface no longer simply corresponds to the target binary contour, but becomes a hybrid terrain that balances physical energy continuity and geometric topological constraints.

[0097] iv. Watershed Immersion Segmentation Based on Hybrid Potential Surface

[0098] (Step 6) Obtaining a continuous mixed potential energy surface Then, this topological surface is input into the watershed immersion algorithm. Due to the introduction of photometric regularization, the water flow will converge along regions that combine real light intensity troughs with geometric contraction characteristics during immersion. The resulting segmentation ridge is anchored at the boundary of real physical energy, rather than at pseudo-saddle points induced by geometric non-convexity. This step enables stable separation of binary stars with large aberrations and extremely close-knit stars.

[0099] (Step 7) The segmentation results are labeled with connected components, numbered and output as candidate sub-regions to obtain a set of target regions that can be used for subsequent detection, matching or localization processing.

[0100] This invention constructs a limit-overlapping binary star sample in a simulated environment with significant off-axis aberrations. The center-to-center distance between the two stars is set close to the critical segmentation range, and background noise is superimposed to simulate the real sensor environment, resulting in the following... Figure 2 The image shown is of a fused binary star system.

[0101] On this input sample, traditional geometric distance terrain and the hybrid potential energy terrain proposed in this invention are constructed respectively. A three-dimensional potential energy surface is, for example... Figure 3 As shown: Traditional potential energy topography exhibits multiple local peaks, easily leading to erroneous segmentation lines; the hybrid potential energy topography constructed in this invention forms a clearer valley boundary between binary stars. To further illustrate the local differences between the two types of potential energy surfaces at the actual binary star boundary, a vertical cross-section is introduced along the line connecting the physical centers of the binary stars, and a one-dimensional profile curve is extracted, as shown... Figure 4 As shown, traditional distance fields tend to degenerate into flat regions with insufficient gradients in the real boundary region, while the hybrid potential surface constructed in this invention maintains a more continuous valley structure, providing a stable descent path for the watershed algorithm.

[0102] For a binary star system with a fixed field of view radius, overlapping binary star images with spacings of 1 to 10 pixels are constructed using a step size of 0.5 pixels. Several common methods in the processing chain are selected as controls, including the gray-scale centroid method, pure PSF matching method, traditional range field watershed method, and the method of this invention, to evaluate the minimum separable spacing and segmentation stability in overlapping binary star scenarios. The results are as follows: Figure 5 As shown, the present invention exhibits higher stability when the distance between the two stars is close to the limit resolution threshold. Under typical edge field of view conditions, the present invention can achieve effective separation of the two stars at around 4.5 pixels; the traditional distance field watershed requires around 5.0 pixels to stably cut the stars under some large off-axis conditions; and when the limit resolution distances of the two stars are the same at certain field of view positions, the error curve corresponding to the present invention is smoother and the error fluctuation is smaller.

[0103] Figure 6The image shows a comparison of ridge lines segmented by traditional watershed and hybrid watershed methods. Compared to traditional watershed methods, the ridge lines output by the method of this invention are more regular and closer to the actual physical energy troughs between binary stars. The resulting sub-region boundaries are more regular and have better symmetry, providing a more stable input region for subsequent detection, matching, or localization steps.

[0104] Furthermore, the practicality of the method of the present invention was verified through a hardware-in-the-loop experiment based on a DMD star simulator. This hardware-in-the-loop experimental platform based on a DMD star simulator mainly consists of a DMD, a collimator, a three-connected TIR prism, a PSF lens with field-of-view variability, a detector, and auxiliary adjustment mechanisms. The DMD is used to load a controllable star pattern, the collimator is used to convert the pattern on the focal plane of the DMD into quasi-parallel incident light, the lens under test is responsible for forming a star image with field-of-view related degradation characteristics on the detector, and the three-connected TIR prism, composed of three prisms, is used to achieve optical path folding and coupling.

[0105] By controlling the center-to-center spacing of two luminescent micromirror units on a DMD, binary star targets with varying degrees of overlap were generated. For each micromirror spacing, five sets of overlapping star spot images were acquired, and segmentation and local double PSF fitting were performed using both traditional range field watershed and the proposed hybrid terrain watershed. Experimental results are as follows: Figure 7 As shown, when the micromirror spacing is 20–22, both methods struggle to stably separate binary stars. When the spacing increases to 23, the mixed terrain watershed begins to achieve effective segmentation, while the traditional watershed remains largely ineffective. Within the critical overlap range of 24–27 micromirrors, the mixed terrain watershed achieves effective segmentation in all five sample sets, with a success rate of 1.00%. The success rates of the traditional distance field watershed in this range are 0.40, 0.40, 0.60, and 0.80, respectively, with an average of only 0.55, indicating that the proposed method has a more stable separation capability under strong overlap conditions.

[0106] The above results demonstrate that the hybrid terrain watershed proposed in this invention can not only stably form more reasonable dividing ridges in the simulated potential energy field, but also has good applicability under the measured complex star spot conditions.

[0107] In other specific embodiments of the present invention:

[0108] (1) The acquisition of the foreground region is not limited to the local adaptive threshold. It can also be achieved by local background estimation subtraction, region growth, saliency detection or other methods that can output candidate foreground regions.

[0109] (2) In addition to Euclidean distance transformation, the geometric distance term can also be constrained by weighted distance, geodesic distance or piecewise distance, as long as it can characterize the degree of geometric embedding of the pixel relative to the target boundary.

[0110] (3) In addition to directly using the original grayscale, the photometric item can also use the intensity distribution after background correction, smoothing and noise reduction or local contrast enhancement.

[0111] (4) In addition to linear weighting, the fusion of photometric and geometric terms can also be achieved by piecewise weighting, adaptive weighting, or other fusion methods that maintain the synergistic effect of the two types of constraints.

[0112] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for segmenting complex star spots based on photometric-geometric hybrid terrain for star sensors with large aberrations, characterized in that, Includes the following steps: Step 1: Obtain the star map to be segmented and extract the foreground: Process the observed image to obtain the effective foreground region of the target. and its corresponding background area ; Step 2: Construct the geometric distance field: for the foreground region Each valid pixel within Calculate its distance to the nearest background region The geometric distances of the boundaries form a geometric distance field characterizing the topological structure. ; Step 3: Extract the physical light intensity field: Extract the true physical light intensity distribution of the corresponding foreground pixels from the original observed image. , as a physical constraint on photometric continuity; Step 4: Dimensionless normalization: for the geometric distance field With physical light intensity field Perform range normalization separately within the foreground region to obtain a dimensionless normalized geometric field. With normalized photometric field ; Step 5: Constructing Hybrid Potential Energy Terrain: Introducing Regularization Weighting Coefficients Normalized geometric field With normalized photometric field By performing weighted fusion and taking its negative mapping, a hybrid potential surface facing the watershed immersion transformation is constructed. ; Step 6: Mixed watershed segmentation: at the mixed potential energy surface Perform watershed submersion transformation to find potential energy minimum points and extract catchment basins and their dividing ridges to achieve robust separation of overlapping star points or complex trailing star points. Step 7: Output Results: The segmentation results are labeled with connected components, numbered, and output as candidate sub-regions to obtain a set of target regions for subsequent detection, matching, or localization processing.

2. The complex star spot segmentation method based on photometric-geometric hybrid terrain for star sensors with large aberrations according to claim 1, characterized in that, In step 1, the observed image is processed, specifically as follows: Local adaptive thresholding, or Equivalent prospect extraction, or Local background estimation subtraction, or Regional growth, or Significance detection.

3. The complex star spot segmentation method based on photometric-geometric hybrid terrain for star sensors with large aberrations according to claim 1, characterized in that, Step 2 targets the foreground region Each valid pixel within Calculate its distance to the nearest background region The geometric distance of the boundary is specifically: Minimum Euclidean distance, or Weighted distance, or Geodetic distance, or Segmented distance constraints.

4. The complex star spot segmentation method based on photometric-geometric hybrid terrain for star sensors with large aberrations according to claim 1, characterized in that, In step 5, the normalized geometric field With normalized photometric field Weighted fusion is performed, specifically as follows: Linear weighted fusion, or Segmented weighted fusion, or Adaptive weight fusion.

5. The complex star spot segmentation method based on photometric-geometric hybrid terrain for star sensors with large aberrations according to claim 3, characterized in that, Step 3 specifically involves: For any pixel belonging to the foreground region The minimum Euclidean distance to the nearest background boundary is defined as: In the formula, Background area arbitrary pixel coordinates in; For geometric distance field.

6. The complex star spot segmentation method based on photometric-geometric hybrid terrain for star sensors with large aberrations according to claim 3, characterized in that, Step 4 specifically involves: For geometric distance field With physical light intensity field Each is uniformly mapped to In the dimensionless probability space, the dimensionless normalized geometric fields are obtained respectively. With normalized photometric field : In the formula, and These are the minimum and maximum values ​​within the connected region, respectively.

7. The complex star spot segmentation method based on photometric-geometric hybrid terrain for star sensors with large aberrations according to claim 6, characterized in that, Step 5 specifically involves: Normalized geometric field With normalized photometric field A linear weighted fusion is performed, and its negative mapping is taken to construct a hybrid potential surface oriented towards the watershed immersion transformation. : In the formula, To adjust the normalized geometric field With normalized photometric field The regularization coefficients of the weights of both.

8. The complex star spot segmentation method based on photometric-geometric hybrid terrain for star sensors with large aberrations according to claim 7, characterized in that, Step 6 specifically involves: Mixed potential energy surface The topological surface is input into the watershed immersion algorithm; during the immersion process, the water flow will converge along the region that has both real light intensity trough and geometric contraction characteristics; the final generated segmentation ridge is anchored at the boundary of real physical energy.