A star sensor

Abstract

This invention discloses a star sensor, which extracts the effective star region of a single star map using the following steps: First, the star map is preprocessed to obtain the saliency value of each pixel in the star map and generate a star map saliency map; then, the star map saliency map is initially segmented using the saliency value as a feature to obtain pre-segmented blocks, and features are extracted from the pre-segmented blocks; and finally, the features are clustered and combined to realize the identification and labeling of interference regions.

Description

Technical Field

[0001] This invention relates to the field of aerospace technology, and in particular to a method for extracting an effective star region from a single star map.

[0002] This application is a divisional application of the parent application, "A method for extracting an effective star region from a single star map" (application number: 2021108555323, application date: July 28, 2021). Background Technology

[0003] The surge in satellite constellations has dramatically increased the number of artificial satellites, significantly impacting space exploration missions and ground-based observation experiments. Star sensors, with their advantages of high precision, high autonomy, and low power consumption, play a crucial role in space missions. As the complexity of space exploration missions and ground-based observation experiments increases, higher demands are being placed on star sensors, especially those with high precision and high dynamic performance.

[0004] Star images are the sole data source for star sensors, but they are often rendered useless by various interferences, directly affecting subsequent star point extraction. Therefore, one of the key performance indicators for star sensors is their anti-interference capability.

[0005] Besides background noise, large-area interference in the star sensor's field of view can also affect its attitude output. High-energy protons released from the Van Allen inner radiation belts, cosmic rays, and solar proton events can cause scratch-like interference on the image plane due to velocity differences; regular-shaped interference can be generated on the image plane by satellite components and their secondary diffuse reflection light entering the star sensor's field of view; and irregularly shaped interference, such as thin clouds, inevitably enters the field of view during ground-based stargazing and calibration experiments. Furthermore, the demand for low-cost, high-performance star sensors in the future makes urban stargazing possible. However, the biggest problem facing urban stargazing is the presence of many large, irregularly shaped interference in the field of view. Therefore, removing interference in the field of view is a crucial step in ensuring the normal operation of the star sensor.

[0006] In traditional astronomical image processing research, a significant amount of research has been devoted to addressing interference issues in astronomical images. Interference suppression algorithms based on multi-image overlay can remove these interferences. Methods based on point spread features remove interference by detecting whether the distribution characteristics of cosmic rays and satellite trajectories conform to the point spread function. Multi-feature matching methods remove interference by extracting multiple features from the image and using geometric matching. However, using these methods on star sensors incurs significant memory consumption and requires lengthy comparison times. To remove scratch-like interference from a single astronomical image, methods based on Hough transform and outlier detection are commonly used, but they are only effective for interference conforming to a linear distribution. For other large-area interference, Laplacian edge detection is used to identify scratch-like interference of arbitrary shapes and sizes based on the sharpness of its edges. However, for large-area interference, there are currently no good algorithms or methods. Summary of the Invention

[0007] To address some or all of the problems in existing technologies, and in order to remove large areas of significant interference in star charts and thus extract the effective starry sky region, this invention provides a method for extracting the effective starry sky region from a single star chart, comprising:

[0008] The star map is preprocessed to obtain the saliency value of each pixel in the star map, and a star map saliency map is generated.

[0009] Using the saliency value as a feature, the star map saliency map is initially segmented;

[0010] Perform feature extraction; and

[0011] The features are clustered and combined to identify and label interference regions.

[0012] Furthermore, the preprocessing includes: de-equalizing the star map histogram using the restricted LC algorithm to obtain the saliency value of each pixel.

[0013] Furthermore, the preprocessing also includes: the calculation of the significance value includes the following constraints:

[0014]

[0015] in, This is the pixel value corresponding to the minimum saliency value.

[0016] Furthermore, the initial segmentation includes: using the Simple Linear Iterative Clustering (SLIC) method to cluster pixels with similar saliency values.

[0017] Furthermore, the metric used in the simple linear iterative clustering method is calculated based on saliency distance and spatial distance.

[0018] Furthermore, the extraction method also includes normalizing the saliency value before performing the initial segmentation.

[0019] Furthermore, the features include the saliency mean of the superpixels and the superpixel variance in the star map.

[0020] Furthermore, the clustering combination is implemented using a density-based clustering method with noise (DBSCAN).

[0021] Furthermore, the distance metric in the DBSCAN algorithm uses weighted Minkowski distance.

[0022] This invention provides a method for extracting the effective star region from a single star map. It integrates saliency detection, SLIC superpixel segmentation, and the DBSCAN method to identify and label interference regions in the star map, thereby extracting the effective star region for star point extraction and star map recognition. Specifically, the extraction method first preprocesses the star map using the restricted LC algorithm to increase the contrast between large-area interference and the background; then, based on the idea of ​​SLIC superpixel segmentation, it pre-segments the saliency map; finally, it extracts features from these superpixels and uses DBSCAN clustering to combine superpixels with similar features to obtain the large-area interference region and the effective star region. Test results on real-world star maps show that the extraction method can effectively segment large-area interference and star regions in star maps when dealing with large-area interference in the field of view, and can successfully extract star points even under strong interference, thus improving the usability of star sensors. This extraction method can be extended to image recognition using star cameras and ground-based stargazing experiments. Attached Figure Description

[0023] To further illustrate the above and other advantages and features of the various embodiments of the present invention, a more specific description of the various embodiments of the present invention will be presented with reference to the accompanying drawings. It is to be understood that these drawings depict only typical embodiments of the invention and are therefore not intended to limit its scope. In the drawings, identical or corresponding parts will be indicated by identical or similar reference numerals for clarity.

[0024] Figure 1 a shows a schematic diagram of the ideal star point energy distribution generated by the simulation;

[0025] Figure 1 b shows a schematic diagram of the energy distribution of star points simulating dynamic uniform motion;

[0026] Figure 1 c shows a schematic diagram of the energy distribution of a star point simulating variable angular velocity motion;

[0027] Figure 1 d-1f shows a schematic diagram of the energy distribution of star points in an actual star map in orbit;

[0028] Figure 2 a-2d shows an enlarged schematic diagram of large-area interference objects in various interfered star maps;

[0029] Figure 3 This is a flowchart illustrating a method for extracting an effective starry sky region from a single star map according to an embodiment of the present invention;

[0030] Figure 4 a and 4b show the original star map and its saliency map after saliency calculation using the LC algorithm in one embodiment of the present invention, respectively.

[0031] Figure 5 a-5b respectively show the original star map and the histogram distribution of its saliency map after saliency calculation using the LC algorithm in one embodiment of the present invention;

[0032] Figure 6 a-6b respectively show the correspondence between grayscale and saliency after saliency calculation using the LC algorithm and the restricted LC algorithm in one embodiment of the present invention;

[0033] Figure 7 This diagram illustrates the effective star occupancy rate under different numbers of superpixels in superpixel segmentation according to one embodiment of the present invention.

[0034] Figure 8 This diagram illustrates the boundary recall rates corresponding to different m values ​​in superpixel segmentation according to one embodiment of the present invention.

[0035] Figure 9 a-9d respectively show the superpixel segmentation map and its details representing the salient interference region, the details representing the suspected star point region, and the distribution diagrams of the mean of each superpixel saliency map and the variance of the original image;

[0036] Figures 10a-10e The images show the long stripe-like interference formed by Starlink satellites passing through the sky, the results of extracting star points using the thresholding method, the binary mask generated after clustering, the star point extraction results when the mask is present, and the results of DBSCAN clustering.

[0037] Figure 11a-11f show the starry sky images taken from the ground, the results of star point extraction using the thresholding method, the SLIC superpixel segmentation results, the DBSCAN clustering results, the binary mask generated after clustering, and the star point extraction results when the mask exists.

[0038] Figure 12 a-12f show the star map with linear scratch interference, the results of star point extraction using the thresholding method, the SLIC superpixel segmentation results, the DBSCAN clustering results, the binary mask generated after clustering, and the star point extraction results when the mask exists.

[0039] Figure 13 a-13f respectively show star charts with regular halo interference, and the results of star point extraction using thresholding, SLIC superpixel segmentation, DBSCAN clustering, the binary mask generated after clustering, and star point extraction results when the mask is present; and

[0040] Figure 14 a-14f show star charts affected by interference from the moon and its reflected light, and the results of star point extraction using the thresholding method, SLIC superpixel segmentation, DBSCAN clustering, the binary mask generated after clustering, and star point extraction results when the mask is present. Detailed Implementation

[0041] In the following description, the invention is described with reference to various embodiments. However, those skilled in the art will recognize that the embodiments may be practiced without one or more specific details or with other alternatives and / or additional methods, materials, or components. In other instances, well-known structures, materials, or operations are not shown or described in detail so as not to obscure the inventive points of the invention. Similarly, for illustrative purposes, specific quantities, materials, and configurations are set forth to provide a comprehensive understanding of embodiments of the invention. However, the invention is not limited to these specific details. Furthermore, it should be understood that the embodiments shown in the drawings are illustrative representations and are not necessarily drawn to scale.

[0042] In this specification, references to "an embodiment" or "this embodiment" mean that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of the invention. The phrase "in one embodiment" appearing throughout this specification does not necessarily refer to the same embodiment in all instances.

[0043] It should be noted that the embodiments of the present invention describe the process steps in a specific order; however, this is only for illustrating the specific embodiment and not for limiting the order of the steps. On the contrary, in different embodiments of the present invention, the order of the steps can be adjusted according to the process.

[0044] As more and more satellites are launched into space, interference factors in space are gradually increasing, inevitably causing a great impact on star sensors. Future star sensors will face even more challenges.

[0045] Stellar images are the sole data source for star sensors, and these images are typically subject to various interferences. A typical star image should consist of dozens of bright stars against a dark background. The signal-to-noise ratio (SNR) of a star image is generally between 20 dB and 50 dB, expressed as the point spread function (PSF). In a static state, the energy distribution of the stars on the imaging plane approximates a Gaussian spread function. The static distribution of the stars is represented as:

[0046]

[0047] Where (x,y) and (x0,y0) represent the pixel position on the image plane and the true centroid coordinates of the star point, respectively, σ PSF E is the Gaussian radius, representing the degree of energy concentration. sum It is the energy grayscale coefficient, which is related to the apparent magnitude, quantum efficiency, integration time, lens aperture, and optical transmittance of the corresponding star.

[0048] In dynamic scenarios, the imaging of stars on the image plane can be represented by a model of the centroid motion of the star points, with the following star point energy distribution:

[0049]

[0050] Where (x0(t), y0(t)) represents the centroid coordinates of the star at time t, where t = t0 + Δt (Δt << T), and T is the exposure time of the star sensor, which is in the millisecond range.

[0051] Figure 1 a-1f show schematic diagrams of static and dynamic star point energy distribution under simulated and real conditions, respectively, where:

[0052] Figure 1 a shows a schematic diagram of the ideal star point energy distribution generated by the simulation;

[0053] Figure 1 b shows a schematic diagram of the energy distribution of star points simulating dynamic uniform motion;

[0054] Figure 1 c shows a schematic diagram of the energy distribution of a star point simulating variable angular velocity motion;

[0055] Figure 1 d-1f shows a schematic diagram of the energy distribution of star points in an actual star map in orbit, where, Figure 1 The star point energy distribution shown in f exhibits a slight trailing phenomenon in dynamic imaging.

[0056] Common large-area disturbances in star charts include linear scratch-like disturbances, regular and irregular shapes, etc. Figure 2 a-2d shows a magnified schematic diagram of large-area interference objects in various disturbed star maps:

[0057] Figure 2 Figure a shows linear scratch-like interference. This type of interference often occurs due to high-energy particles, especially protons, scratching the image sensor, or satellite constellations entering the star-sensor's field of view and forming scratches on the image plane. This type of interference is usually not fixed in location and can appear anywhere in the star chart. Furthermore, due to differences in relative velocity, the length of the scratch is also not fixed. Within a limited exposure time, the length of the interference differs significantly from the typical energy distribution of a star point. With the dramatic increase in the number of space satellites and their orbital altitudes in the future, space interference is becoming increasingly complex, and this situation will become more and more common.

[0058] Figure 2 b shows the interference caused by a design flaw in the sunshade, and Figure 2 c illustrates interference caused by satellite components or their reflected light entering the star sensor's field of view. Both types of interference are regularly shaped and cause an increase in grayscale values ​​in localized areas of the image. This type of interference is relatively fixed in location and typically appears around the perimeter of the image; and

[0059] Figure 2 d shows the interference caused by reflected light from celestial objects entering the star sensor's field of view. This type of interference is similar to... Figure 2 The interference shown in b and 2c is similar, but the difference is that this type of interference often appears in the center of the image and has an irregular shape, which has the greatest impact on the star sensor.

[0060] As shown in the figure, compared with typical star points, large-area interference differs significantly in shape and grayscale value. Furthermore, large-area interference consists of continuous pixel blocks with grayscale values ​​close to or higher than those of star points. For common star maps, due to the significant grayscale contrast between star points and the dark background, global thresholding methods, such as the Otsu algorithm, can effectively segment the star points and background. However, when large-area interference exists, the contrast between the significant interference, star points, and background in the star map is relatively small. If the Otsu algorithm is still used, different thresholds need to be set according to different star sensors, which has significant limitations and is not very effective. Directly extracting star points from such interference star maps will extract a large number of false star points near the interference, causing the star sensor to malfunction.

[0061] Therefore, to solve the problem of star sensors malfunctioning due to various interferences, it is first necessary to improve the contrast between significant interference objects, stars, and the background in the star image, and then perform image segmentation. This invention provides a method for extracting the effective starry sky region from a single star image. The technical solution of this invention will be further described below with reference to the accompanying drawings of the embodiments.

[0062] Figure 3 This diagram illustrates a flowchart of a method for extracting an effective star region from a single star map according to an embodiment of the present invention. Figure 3 As shown, a method for extracting an effective starry sky region from a single star map includes:

[0063] First, in step 301, the saliency value is calculated. The star map is preprocessed to obtain the saliency value of each pixel in the star map, and a star map saliency map is generated, thereby increasing the contrast between gray values. The increase in pixel contrast is beneficial for better segmentation.

[0064] In an embodiment of the present invention, the saliency calculation is performed using the LC algorithm. According to the LC algorithm, the saliency value of the star map grayscale value is defined as its contrast with other pixel values ​​in the image.

[0065]

[0066] Where n = 255, representing the total number of gray levels in the image, g s For pixel I k grayscale value, f s For g s The probability of appearing in image I, ‖ ‖ represents the gray-level distance metric, and thus, the saliency value corresponding to each gray level in [0,255] can be calculated. Figure 4 a and 4b respectively show the original star map and its saliency map after saliency calculation using the LC algorithm in one embodiment of the present invention; and Figure 5 Figures a-5b show the histogram distribution of the original star map and the saliency map obtained after saliency calculation using the LC algorithm of this invention. As shown, pixels with larger gray values ​​in the star map receive larger saliency values, while pixels with smaller gray values ​​also have larger saliency values. However, this also leads to darker pixels in the star map having an increased proportion in the saliency map due to saliency calculation, thus affecting the segmentation result. To avoid the influence of dark pixels on the segmentation result, in one embodiment of this invention, constraints are added to the saliency value calculation, which can be called the restricted LC algorithm. The constraints are as follows:

[0067]

[0068] in, This is the pixel value corresponding to the minimum saliency value.

[0069] Figure 6 a-6b respectively show the grayscale and saliency correspondence after saliency calculation using the LC algorithm and the restricted LC algorithm in one embodiment of the present invention. It can be seen that since the background pixels in the star map account for most of the image and the grayscale values ​​are distributed between [0, 130], and the distribution is most concentrated in the histogram, after the restricted saliency value calculation, pixels with larger grayscale values, such as star points and significant interference, obtain large saliency values, while pixels with smaller grayscale values, such as the star map background, obtain small saliency values. The constraint condition can increase the contrast between significant interference objects and star points and the background in the star map.

[0070] Next, in step 302, initial image segmentation is performed. Using the saliency value as a feature, the star map saliency map is initially segmented to obtain pre-segmented blocks. The difference between star points and salient interference lies in the fact that star points occupy fewer pixels, generally between 3x3 and 7x7, and are approximately circular in shape, while salient interference occupies more pixels and consists of continuous pixel regions without a fixed shape. Therefore, in one embodiment of the present invention, the initial segmentation includes: using a Simple Linear Iterative Clustering (SLIC) method to cluster pixels with similar saliency values, obtaining pre-segmented blocks with superpixels as cluster centers. The generated segmented blocks have good boundary fitting characteristics, and each superpixel contains either star points or salient interference.

[0071] The SLIC algorithm, also known as the superpixel segmentation algorithm, has the following specific process:

[0072] Initialize seed point K;

[0073] To avoid the initial seed point being located on a boundary with a large gradient, the seed point is moved to the point with the minimum gradient within a 3x3 neighborhood; and

[0074] Calculate the distance D between seed points, and select the seed point with the smallest distance value within the (2S*2S) grid as the cluster center of that pixel. The method for calculating the distance D is as follows:

[0075]

[0076] Where, d c and d s The color distance and spatial distance are calculated separately as follows:

[0077]

[0078]

[0079] Where i is the cluster center label, j is the one-dimensional index value of the pixel coordinates of the 2Sx2S neighborhood corresponding to cluster center i, and (x, y) are the pixel coordinates;

[0080] Iterative optimization continues until the residual is less than the set threshold, at which point the segmentation is complete.

[0081] In an embodiment of the present invention, the SLIC is classified based on saliency value, and its seed point is the geometric center of the superpixel. Therefore, it cannot continue to use color distance. Based on this, in one embodiment of the present invention, the distance calculation method is improved, which uses the pixel saliency value Sal(I) as the seed point. k The coordinates (x, y) of the vector and its corresponding coordinates in the image constitute a three-dimensional feature vector G = [sal, x, y]. T The distance between the three-dimensional feature vectors is used as the seed point distance, and the distance incorporates the saliency distance metric d. sal and spatial distance d s :

[0082]

[0083] in:

[0084] Let K be the initial segmentation superpixel side length, where K is the seed number, i.e., the number of superpixels. This seed number varies depending on the classification task and the image size. In the star map, the star occupancy A... sky (K), that is, the proportion of the starry sky in the image, can be defined as:

[0085]

[0086] Among them, A i Let C represent the i-th superpixel. sky The superpixels represent the clustering results of the starry sky. This shows that when using superpixel segmentation in a star map, if the value of K is too large, the generated superpixels may only contain star points, which will be removed as interference in subsequent processing; if it is too small, the generated superpixel results will be inaccurate and have a large error. Figure 7 The diagram illustrates the effective star occupancy rate under different numbers of superpixels in superpixel segmentation according to one embodiment of the present invention. It can be seen that as the K value increases, A... sky The K value will become increasingly dispersed, indicating that the superpixel segmentation is oversegmented. In one embodiment of the present invention, the K value is preferably a value within the range of 1 to 2 times the side length of the star map.

[0087] as well as

[0088]

[0089] In yet another embodiment of the invention, to prevent d sal If the distance weight is too large, the pixel saliency value Sal(I) should be adjusted before calculating the saliency distance metric. k The normalization was performed to [0, 255]; and

[0090] 'm' is a positive constant used to control the influence of the saliency distance metric and spatial distance on the distance metric. A larger value indicates greater spatial similarity and a more compact superpixel result; a smaller value results in superpixels closer to the image boundary, but with more irregular shapes. To make the superpixel segmentation result closer to the boundary and to preserve a more complete starry sky region, the inventors used different values ​​of 'm' for superpixel segmentation and statistically analyzed the results. Boundary recall was used as an evaluation metric to assess the degree of fit between the superpixel boundary and the manually labeled boundary. Boundary recall is expressed as the percentage of pixels whose manually labeled image boundary falls within two pixels of the superpixel boundary obtained from the superpixel segmentation, out of the total number of pixels on the manually labeled image boundary. A larger value indicates better boundary detection performance of the superpixel. Figure 8 This diagram illustrates the boundary recall rates corresponding to different m values ​​in superpixel segmentation according to an embodiment of the present invention. Figure 7 As shown, when m is 21, the segmentation result fits the boundary best;

[0091] Next, in step 303, feature extraction. Features are extracted from the pre-segmented blocks. Since the human visual system is most sensitive to information such as scene boundaries, spatial pixel variance distribution, spatial grayscale differences, and boundary fitting, and based on this and the characteristics of common interference in star maps, in one embodiment of the present invention, two features are extracted from the superpixels for superpixel combination:

[0092] The saliency mean M in each superpixel i ;as well as

[0093] Superpixel variance V in the original star map i ;as well as

[0094] Finally, in step 304, clustering and combination are performed. The features are clustered and combined to identify and label interference regions. Since the salience mean and original variance distribution of the starry sky region are relatively small, and the differences between superpixels are small, while the salience mean and original variance of the large-area interference region vary significantly due to fluctuations in pixel values, and the differences between different superpixels are also large. Therefore, in order to combine superpixels of different categories to segment the starry sky region and the large-area interference region, in one embodiment of the present invention, a density-based clustering of applications with noise (DBSCAN) method is used for clustering and combination.

[0095] The DBSCAN algorithm is a density clustering algorithm that assumes that the sample class can be determined by the density of the sample distribution. Figure 9 Figures a-9d show the superpixel segmentation map and its details representing salient interference regions, details representing suspected star point regions, and the distribution diagrams of the mean of each superpixel saliency map and the variance of the original image. As shown in the figures, for star map superpixels, the features of star regions are relatively similar and the feature distribution is concentrated. Therefore, they can be clustered into clusters in DBSCAN clustering. Large-area interference regions have significantly different features and their feature distribution is not concentrated; therefore, they can be treated as noise in DBSCAN. The core of the DBSCAN algorithm is to calculate the distance between samples to find all core objects. The distance can use the Minkowski distance.

[0096]

[0097] Where, S p S represents the feature vector extracted from the superpixel. i =[M i V i ], where q represents the dimension of the feature vector.

[0098] Because the saliency mean and the variance of the original star map have different weights in the distance metric, the variance of saliency interference superpixels fluctuates greatly and therefore carries a larger weight in distance calculation. In contrast, the variance of star region superpixels does not change much, but its value is still much larger than the mean. Therefore, in one embodiment of this invention, ln(V) is used when calculating the distance. i ) replace V i To reduce the weight of variance in the distance metric, the Minkowski distance can be modified as follows:

[0099]

[0100] Finally, we can obtain the cluster set C = {C0, C1, ..., C}.k The set of star region characteristic clusters is the effective star region in the star map after removing large-area interference.

[0101] To verify the effectiveness of the extraction method in the embodiments of the present invention, the inventors used the extraction method to extract the star region from the star map taken by the satellite in orbit and the star map taken from the ground.

[0102] Figures 10a-10e The images show the long stripe-like interference formed by Starlink satellites gliding across the sky, the results of extracting star points using a thresholding method, the binary mask generated after clustering, the star point extraction results with the mask present, and the results of DBSCAN clustering. Figure 10a This is a star map taken at the Farah Observatory in Italy. The Starlink satellite constellation passes through the map, creating linear streaks or interference. Figure 10b The results of using the threshold method to extract star points show that it extracts many false star points at interference locations. However, the extraction method in this embodiment of the invention identifies the significant interference areas, masks them, and then extracts the star points. Here, the parameter radius epd in the DBSCAN algorithm is set to 0.4. After clustering using the DBSCAN algorithm, the following results are obtained: Figure 10e Based on the aforementioned results, superpixels in non-starry regions can be combined to form a binary mask, such as... Figure 10c As shown, the following can be obtained: Figure 10d The star point extraction results shown, compared with the original star map, have largely eliminated the interference of linear scratches in the star map and greatly reduced the existence of false star points.

[0103] Figure 11 a-11f show ground-based images of the starry sky, along with the results of star extraction using thresholding, SLIC superpixel segmentation, DBSCAN clustering, the binary mask generated after clustering, and star extraction results when the mask is present. Strictly speaking, as... Figure 11 The situation described in figure a would not occur on a satellite, but given the future demand for low-cost, high-performance star sensors, ground-based stargazing experiments in cities will become possible. Therefore, the inventors applied the extraction method described in this invention to this situation to test the applicability of the algorithm. As shown in the figure, when using the threshold method to extract star points, many false star points exist at interference locations. However, after clustering using the extraction method described in this invention, a binary mask is formed, and the resulting star point extraction result, compared with the original star map, essentially eliminates interference from buildings in the star map, greatly reducing the existence of false star points.

[0104] Figure 12a-12f show the star map with linear scratch interference, the results of star point extraction using the thresholding method, the SLIC superpixel segmentation results, the DBSCAN clustering results, the binary mask generated after clustering, and the star point extraction results when the mask exists.

[0105] Figure 13 a-13f respectively show star charts with regular halo interference, and the results of star point extraction using thresholding, SLIC superpixel segmentation, DBSCAN clustering, the binary mask generated after clustering, and star point extraction results when the mask is present; and

[0106] Figure 14 a-14f show star charts affected by interference from the moon and its reflected light, and the results of star point extraction using the thresholding method, SLIC superpixel segmentation, DBSCAN clustering, the binary mask generated after clustering, and star point extraction results when the mask is present.

[0107] As can be seen from the figure, when using the threshold method to extract star points, there are many false star points at the interference locations. However, after clustering using the extraction method in this embodiment of the invention to form a binary mask, the star point extraction result obtained, compared with the original star map, basically eliminates large-area interference in the star map and greatly reduces the existence of false star points.

[0108] In summary, it can be seen that the extraction method in this embodiment of the invention can effectively segment large-area interference objects and star regions in the star map when dealing with large-area interference in the field of view, and can successfully extract star points under strong interference conditions, thereby improving the usability of the star sensor.

[0109] Although various embodiments of the invention have been described above, it should be understood that they are presented by way of example only and not as limitations. It will be apparent to those skilled in the art that various combinations, modifications, and alterations can be made without departing from the spirit and scope of the invention. Therefore, the breadth and scope of the invention disclosed herein should not be limited by the exemplary embodiments disclosed above, but should be defined solely by the appended claims and their equivalents.

Claims

1. A star sensor, characterized in that, The effective sky region of a single star chart is extracted using the following steps: The star map histogram is de-equalized using the LC algorithm to obtain the saliency value of each pixel in the star map, and a star map saliency map is generated. The calculation of the saliency value includes the following constraints: in, The pixel value corresponding to the minimum saliency value; Using the saliency value as a feature, a simple linear iterative clustering method is employed to cluster pixels with similar saliency values, performing initial segmentation on the star map saliency map to obtain pre-segmented blocks. The metric D′ used in the simple linear iterative clustering method is based on the saliency distance metric d. sal and spatial distance d s The calculation yielded: in: Let K be the initial segmentation superpixel side length, where K is the number of superpixels; and m is a positive constant, which is used to control the influence of the saliency distance metric and spatial distance on the metric D′; Feature extraction is performed on the pre-segmented blocks; and A noisy density-based clustering method is used to cluster and combine the features to identify and label interfering regions. The distance metric used in this noisy density-based clustering method is weighted Minkowski distance. in, M i The saliency mean in superpixel i; V i Let be the pixel variance of superpixel i in the original star map; i is the cluster center label, and j is the one-dimensional index value of the pixel coordinates of the 2Sx2S neighborhood corresponding to cluster center i.

2. The star sensor as described in claim 1, characterized in that, The saliency distance metric d sal The calculation is as follows: Where i is the cluster center label, and j is the one-dimensional index value of the pixel coordinates of the 2Sx2S neighborhood corresponding to cluster center i.

3. The star sensor as described in claim 1, characterized in that, It also includes normalizing the saliency values ​​before performing the initial segmentation.

4. The star sensor as described in claim 1, characterized in that, The features include the saliency mean of each superpixel and the superpixel variance in the star map.

Images