A star map recognition method based on radial and ring angular distance

Abstract

The application provides a star map recognition method based on radial and ring angular distance, and the main steps include selection of a star table and construction of a navigation database, construction of an observation star radial and ring angular distance mode feature, initial matching of a radial mode, further matching of a ring angular distance mode, and link angular distance verification. The method realizes one-to-one identification of star points in a star map with large density and multiple pseudo stars, still maintains a high recognition rate when large noise interference and edge field loss are encountered, and makes the correct rate of star map recognition reach 98.7%, and has strong robustness to pseudo stars and background noise interference.

Description

TECHNICAL FIELD

[0001] The application provides a star map recognition method based on radial and ring angular distance, relates to a novel star pattern recognition algorithm, and belongs to the technical field of spaceflight. BACKGROUND

[0002] At present, star map recognition is an indispensable key technology for accurate attitude determination of satellites and astronomical positioning of space debris in the field of view. In order to achieve accurate determination of satellite attitude and high-precision positioning of space debris, a sufficient number of star targets need to be recognized as data support, and the recognition effect significantly affects the accuracy of the two.

[0003] Star map recognition algorithms mainly include construction of a navigation database and matching of star map features, aiming to determine a unique feature quantity to correspond star points in an unknown observation star map with stars in a navigation star database, and then realize acquisition of a probe visual axis direction. At present, common star map recognition algorithms mainly include three categories: subgraph isomorphism algorithms, pattern recognition algorithms and artificial intelligence algorithms.

[0004] Subgraph isomorphism algorithms often construct features of star points based on distances and positions between star points in a field of view. Commonly used algorithms include a triangle algorithm, a frequency counting algorithm and a pyramid algorithm. The triangle algorithm is simple and easy to implement, but the navigation database to be constructed is extremely large, the matching time is relatively long, the feature dimension is low, and the algorithm is easily affected by noise, resulting in redundancy and errors; the frequency counting algorithm needs to count the frequency of all subgraphs, and for a large star map, the number of subgraphs to be traversed is extremely large, and the algorithm efficiency is not high.

[0005] Pattern recognition algorithms are a popular development direction at present, which often take the geometric spatial distribution of nearby stars as the feature pattern of the star to be recognized, and common algorithms include a grid algorithm, a radial algorithm, a radial and ring algorithm and the like. Such algorithms have fast matching speed, but have low accuracy in selecting calibration stars, resulting in matching errors. Subsequent optimized algorithms solve this problem, but when there is a large amount of pseudo-star interference or the geometric spatial distribution of nearby stars at the edge of the star map is greatly different from the ideal star map, a large number of redundant situations occur, and the recognition effect is poor.

[0006] Artificial intelligence algorithms are a technology developed in recent years, which mainly use machine learning, deep learning and self-adaptive models for recognition. The advantage of this kind of algorithm is that it can learn and optimize based on a large amount of data, but the long training time and low recognition efficiency limit the application of this kind of algorithm.

[0007] In summary, in order to realize accurate satellite attitude determination and high-precision astronomical positioning of space debris, more information of calibration stars needs to be used, which involves the identification of dimmer stars. Although there are identification methods based on radial and dynamic ring patterns in existing star map identification technologies, when the star map density is large, there are false star interferences or edge field missing, the existing methods are still difficult to guarantee high accuracy. SUMMARY

[0008] (I) Invention purposes

[0009] In order to provide data support for accurate satellite attitude determination and high-precision astronomical positioning of space debris, realize one-to-one identification of all star points in the star map, and solve the problem of decreasing identification accuracy when a large number of false stars interfere or the edge field is missing in the pattern recognition star map identification algorithm, the present application proposes a star map identification method based on radial and ring angular distance, which is an optimized algorithm based on the radial and dynamic ring star map identification method. A new angular distance feature is added to the ring pattern, the link angular distance verification and maximum similarity matching algorithm are introduced, which significantly improves the accuracy of star map identification, especially in the case of false star interference and field missing, and has good robustness.

[0010] (II) Technical solutions

[0011] The present application proposes a star map identification method based on radial and ring angular distance, which comprises the following steps:

[0012] Step 1: Create a basic star table suitable for large-density star map identification

[0013] Integrate Tycho2 star table and UCAC4 star table, and select star objects with a star magnitude lower than 14mv, remove duplicate targets, obtain a star table with 17.58 million stars, as star identification reference targets of the star map. And select the star number, star magnitude, right ascension, declination, right ascension proper motion and declination proper motion to form a new star table, to reduce the storage space of the star table.

[0014] Step 2: Construct a navigation database

[0015] (1) According to the initial attitude provided by the attitude sensor and the field of view range R of the camera (if the field of view is rectangular, take R as half of the diagonal line of the matrix), the range is circled with the initial attitude center as the center and 1.5R as the radius, and the stars within the limit magnitude that can be detected by the detector are selected as the navigation stars.

[0016] (2) Select different radial pattern radii Rr (5-15 nearest neighbor stars in the range of most star pattern radii are appropriate) according to different star map densities with the selected navigation stars as the center, that is, get Nr nearest neighbor stars for generating the radial pattern features of the navigation stars.

[0017] (3) Take the navigation star as the center to divide the ring zone, calculate the angular distance between the navigation star and each neighbor star, and distribute it in the corresponding ring zone to construct the radial mode vector. The vector length is the number of ring zones. Set 1 for the ring zone with neighbor star, and 0 for the ring zone without neighbor star, so that it becomes a vector composed of 0 and 1.

[0018] (4) Take the selected navigation star as the center, select different ring mode radius Rc according to different star map density (5-15 neighbor stars in the mode radius range of most stars are appropriate), that is, obtain Nc neighbor stars to generate the ring angular distance mode feature of the navigation star.

[0019] (5) Take any star as the starting point, calculate the included angle θ between each neighbor star and the previous neighbor star in the counterclockwise direction ij , recorded as V θ ={θ 12 ,θ 23 ,...,θ k(k+1) ,...,θ (n-1)n ,θ n1}, and calculate the length of the terminal side of the included angle, recorded as V r ={d r2 ,d r3 ,...,d r(k+1) ,...,d rn ,d r1}, which together constitute the ring angular distance mode vector, which is calculated by the following formula:

[0020]

[0021] In the formula, (α i ,δ i ) and (α i ,δ j ) are the right ascension and declination coordinates of stars i and j in the celestial coordinate system.

[0022] All radial and ring angular distance mode features of the navigation stars are combined to form the navigation database.

[0023] Step 3: Establish the radial and ring angular distance feature mode corresponding to the observed stars

[0024] Sort the star points in the star map by brightness, and establish the corresponding radial and ring angular distance mode features of the observed stars in the star map according to the same mode parameters in step two. Among them, the angular distance between two stars is the distance between star points in the star map,

[0025]

[0026] In the formula, (x i ,y i ) and (x jy j are the corresponding coordinates of the star i, j centroid position in the image coordinate system, respectively, and f is the focal length of the optical system.

[0027] Step 4: Radial mode initial matching

[0028] The radial mode vector of the observed star is compared with the radial mode vectors of all navigation stars stored in the navigation database, and the number of position elements with the same vector is recorded. If the number is greater than the radial matching threshold, the navigation star is determined as a candidate star. If the number of candidate stars is unique, the matched navigation star is taken as the initial identification result; if not, further matching is performed on the azimuthal angular distance mode.

[0029] Step 5: Azimuthal angular distance mode further matching

[0030] The azimuthal angular distance mode similarity between the observed star and the candidate selected in the previous step is compared using the method of circular shift.

[0031] (1) Let the azimuthal mode features of the navigation star be V θ and V r , the azimuthal mode features of the observed star be v θ and v r , ε θ be the angle matching threshold, and ε r be the angular distance matching threshold. The initial matching items between the two stars are found in a traversal manner, and if the following formula is satisfied

[0032]

[0033] it indicates that the i-th included angle and the terminal edge angular distance of the navigation star can be successfully matched with the j-th included angle and the terminal edge angular distance of the observed star. The two positions are taken as the starting items of the vectors after circular shift.

[0034] (2) The angle vectors after circular shift are accumulated, and the k-th item of the accumulated vector is the sum of the first k items of the vectors before accumulation, i.e.

[0035]

[0036] The angular distance vectors of the two are unchanged.

[0037] (3) The similarity of the azimuthal feature vectors of the navigation star and the observed star is matched, and if the following formula is satisfied

[0038]

[0039] it indicates that the i-th azimuthal feature of the navigation star is matched with the j-th azimuthal feature of the observed star. At the same time, the number of matched groups is the similarity value, and the candidate star corresponding to the maximum similarity is taken as the candidate matching star of the observed star.

[0040] (4) If the similarity value of the candidate matching star is greater than the ring matching threshold, the candidate star is output as the initial identification result.

[0041] Step 6: Link angle distance verification

[0042] (1) The initial identification results are sorted according to the matching degree, and the principle is that the radial mode matching candidate star is unique in front; the radial matching degree of the candidate star is unique in front; and the ring matching degree of the candidate star is not unique in front.

[0043] (2) After sorting, the link connection is performed in sequence according to the order, and the connection condition is that the angle distance between the previous star and the next star in the image and the angle distance between the identification results in the sky is less than the angle distance verification threshold, that is, it meets

[0044]

[0045] In the formula, d ij is the angle distance of the i,jth star point in the star map, is the angle distance of the initial identification result of the two star points in the sky, and ε is the angle distance verification threshold.

[0046] (3) According to the method, the link connection is performed on all initial identification results and is expanded, so as to search for the longest link in the identification result.

[0047] (4) The angle distance error of each star on the longest link and all other stars is calculated, and the combinations exceeding the angle distance verification threshold are counted. The star point with the highest frequency is removed from the initial matching result, and the longest link is reconnected in (2), until all angle distance errors in this step are less than the angle distance verification threshold. The final identification result is the final identification result.

[0048] (Three) Advantages

[0049] The star map identification method based on radial and ring angle distances provided by the application has the following advantages:

[0050] ① The method provided by the application more accurately describes the geometric distribution characteristics of the navigation stars and the observation stars around the near neighbor stars from the radial and ring dimensions, and can accurately screen the candidate stars in the case of a large number of star points.

[0051] ② The method provided by the application can still accurately identify the star points by using the ring angle distance characteristics when encountering false star interference or edge field loss, and reduces the occurrence of redundancy.

[0052] ③ The method provided by the application is more sufficient and reasonable in the angle distance verification step, greatly reducing the cases of false matching result output and correct matching result loss.

[0053] Therefore, the method has great advantages in realizing the star map recognition task of large density and multiple pseudo stars, and can achieve high recognition accuracy and robustness to noise and pseudo stars. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 is a flowchart of the star map recognition method of the present application.

[0055] Figure 2 is a schematic diagram of the radial mode feature construction of the present application.

[0056] Figure 3 is a schematic diagram of the ring angle distance mode feature construction of the present application.

[0057] Figure 4 is the identification result of the ground-based real star map by different identification methods.

[0058] Figure 5 is a curve comparison diagram of the number of star points identified by different identification methods on multiple ground-based real star maps.

[0059] Figure 6 is a curve comparison diagram of the star point identification result error of different identification methods on multiple ground-based real star maps. DETAILED DESCRIPTION

[0060] In order to make the purpose, advantages and characteristics of the present application clearer, the star map recognition method based on radial and ring angle distance mode proposed by the present application is further described in detail below in combination with the drawings, specific embodiments and test verification.

[0061] EMBODIMENT

[0062] As shown in Figure 1 , one embodiment of the present application discloses a star map recognition method based on radial and ring angle distance, and the method steps include:

[0063] Step 1: Create a basic star catalog suitable for large density star map recognition

[0064] Integrate Tycho2 star catalog and UCAC4 star catalog, and select star objects with a magnitude lower than 14mv, remove duplicate targets, obtain a star catalog with 17.58 million stars as star identification reference targets of the star map. And select the star number, magnitude, right ascension, declination, right ascension proper motion, and declination proper motion to form a new star catalog, to reduce the storage space of the star catalog.

[0065] Step 2: Construct a navigation database

[0066] The initial three-axis attitude of the attitude sensor is (152°, -5.5°, 18°), and the field of view is 4.57°×3.04°, i.e., R is 2.74°. The range is delineated with the initial attitude center as the center and 1.5R as the radius. A star with a magnitude less than 12mv is selected as the navigation star.

[0067] like Figure 2 As shown, the specific method for establishing the radial mode corresponding to the navigation star is as follows:

[0068] Using the selected navigation star as the center, a radial mode radius Rr = 0.5° is set to delineate the surrounding neighboring stars. A total of 100 rings are created around the navigation star. The angular distance between the navigation star and each neighboring star is calculated and assigned to the corresponding ring. A radial mode vector is constructed, with a length equal to the number of rings. Rings with neighboring stars are set to 1, and others to 0, making it a vector composed of only 0 and 1 elements.

[0069] like Figure 3 As shown, the specific method for establishing the circumferential angular distance mode corresponding to the navigation star is as follows:

[0070] Select the navigation star as the center and set the circumferential angular distance mode radius Rc = 0.4° to circle the surrounding nearest neighbor stars. Starting from star s1, calculate the angle θ between each nearest neighbor star and the previous nearest neighbor star in a counterclockwise direction. ij , denoted as V θ ={θ 12 ,θ 23 ,...,θ k(k+1) ,...,θ (n-1)n ,θ n1} and calculate the length of the terminal side of the included angle, denoted as V. r ={d r2 ,d r3 ,...,d r(k+1) ,...,d rn ,d r1 Together, they constitute the circumferential angular distance pattern vector.

[0071] The radial and circumferential angular distance pattern features of all navigation satellites are combined to form the navigation database.

[0072] Step 3: Establish radial and circumferential angular distance characteristic patterns corresponding to the observed stars.

[0073] To capture an actual star chart, the optical system has a focal length of f = 300mm. The stars in the star chart are then extracted as the observed stars, such as... Figure 4 As shown, a total of 306 observed stars were extracted. Based on the positional distribution characteristics of the observed stars and their surrounding neighboring stars, the radial and circumferential angular distance characteristics corresponding to the observed stars were established in the same way as in step 2, which will not be elaborated here.

[0074] Step 4: Radial mode initial matching

[0075] Compare the radial mode vector of the observed star with the radial mode vectors of all the navigation stars stored in the navigation database, record the number of elements in the vector corresponding to the same position, and if it is greater than the radial matching threshold, the navigation star is determined as a candidate star. If the number of candidate stars is unique, the matched navigation star is taken as the initial identification result; if it is not unique, further matching of the ring angular distance mode is performed.

[0076] Step 5: Ring angular distance mode further matching

[0077] The method of circular shift is used to compare the ring angular distance mode similarity between the observed star and the candidate selected in the previous step.

[0078] Let the navigation star ring mode feature be V θ and V r , the ring mode feature of the observed star be v θ and v r , ε θ be the angle matching threshold (value 1), and ε r be the angular distance matching threshold (value 0.02). The initial matching item between the two stars is found in a traversal manner, such as

[0079]

[0080] The angle vector is circularly shifted, the i-th item of the angle vector and the j-th item of the angular distance vector are taken as the starting items of the vector, and the circularly shifted angle vector is accumulated.

[0081] The similarity of the accumulated navigation star and the observed star ring feature vectors is matched, and if the formula

[0082]

[0083] is satisfied, the i-th ring feature of the navigation star and the j-th ring feature of the observed star are matched. At the same time, the number of matched groups is the similarity value, and the candidate star corresponding to the maximum similarity is taken as the candidate matching star of the observed star. If the similarity value of the candidate matching star is greater than the ring matching threshold, the candidate star is taken as the initial identification result output.

[0084] Step 6: Link angular distance verification

[0085] The initial identification result is sorted according to the matching degree, and the link connection is performed in turn according to the order. The connection condition is that the angular distance between the previous star and the next star in the image and the angular distance between the identification results on the celestial sphere are less than the angular distance verification threshold (value 0.02).

[0086] According to the method, all initial identification results are linked and expanded, and the longest link in the identification results is searched. Each star on the longest link is calculated with angle error with all other stars, and combinations of errors exceeding the angle verification threshold are counted. The star point with the highest frequency is removed from the initial matching result, the longest link is reconnected, and the process is repeated until all angle errors in the step are less than the angle verification threshold.

[0087] Finally, the identification result satisfying the condition is the final star map identification result of the method.

[0088] Test verification

[0089] To evaluate the performance of the method, a self-created star table is used for star map identification verification. The parameters of the actual star map and the key parameters of the algorithm used are shown in the following table.

[0090]

[0091] A large number of star maps are taken using this set of parameters and the star map identification algorithm is verified. The shooting mode is to take a star map every 3s, and the exposure time of each star map is 300ms. As shown in Figure 4 The identification result of the first star map in this series is shown. Among them, the blue square represents a non-constant star false target, the green circle represents a successfully identified target based on the radial and dynamic ring star map identification algorithm, and the red diamond represents a newly identified star point using the algorithm of the application. According to statistics, there are 306 star points in the figure, and there are 13 false stars. The algorithm before optimization identified 274 star points, and the algorithm of the application successfully identified 292 star points. The star map identification achieved good results.

[0092] The star point extraction and star map identification of the next 20 consecutive star maps are carried out. The number of extracted stars and the number of stars identified by the algorithm before and after optimization are shown in Figure 5 , and the mean square error of the star point position is calculated for the identified results to verify the correctness of the identification results, and the error results are shown in Figure 6 .

[0093] After statistics, 5894 star points are extracted in the 20 star maps, the number of real star points is 5640 after removing false stars, the proportion of false stars is 4.31% (254 / 5894), 5140 stars are recognized by the algorithm before optimization, the recognition rate is 91.13% (5140 / 5640), 5567 stars are recognized by the algorithm after optimization, the correct recognition rate is 98.71% (5567 / 5640), the optimization rate of the algorithm is about 7.83%, and the failure rate is reduced by 85.4%. And the root mean square error of the recognition result of the algorithm is within 8 pixels (about 21 arc seconds), which proves that all the recognition results are correct, but the algorithm before optimization has a large root mean square error in some star maps, which shows that the recognition result of the star map is wrong, and this kind of mismatching is unacceptable in the process of high-precision attitude calculation and astronomical positioning.

[0094] In summary, the algorithm proposed in the application has a high recognition rate, which is better than other algorithms, and can ensure that all the recognition results are correct to a great extent, and provides basic data support for high-precision attitude determination and high-precision astronomical positioning.

[0095] Obviously, the above embodiments of the application are only examples for clearly illustrating the application, and are not a limitation on the embodiments of the application, and any obvious changes or variations derived from the technical solutions of the application are still within the protection scope of the application.

Claims

1. A star map recognition method based on radial and circumferential angular distances, characterized in that, Includes the following steps: Building a navigation database involves the following steps: Create a basic star catalog and filter navigation stars that meet the requirements. Select nearby stars within a certain range and construct the radial and circumferential mode vectors of the navigation stars. The radial mode vector indicates whether there are nearby stars within the radial circumference of the navigation star, denoted as P. r ={r1,r2,...,r m }, where the elements are 0 or 1, and the circumferential mode vector represents the angle between two nearest neighbor stars and the navigation star and the terminal angular distance, denoted as V. θ ={θ 12 ,θ 23 ,...,θ k(k+1) ,...,θ (n-1)n ,θ n1 } and V r ={d r2 ,d r3 ,...,d r(k+1) ,...,d rn ,d r1 }; Performing radial and circumferential angular distance pattern matching for observed stars includes the following steps: The radial and circumferential angular distance feature vectors of the observed star are established using the method and parameters for establishing the feature pattern of the navigation star. First, radial pattern matching is performed. If the candidate star is unique, it is directly used as the initial identification result of the observed star. Otherwise, circumferential angular distance pattern matching is performed. After circumferential angular distance pattern matching, the candidate star with the largest circumferential matching degree is output as the initial identification result of the observed star. The link angular distance verification of the initial identification results includes the following steps: After sorting the initial identification results according to reliability, the longest link is searched and the full link angular distance error is verified. If the maximum angular distance error is greater than the angular distance verification threshold, the star identification result with the most errors is deleted and the link search and verification are repeated until the final identification result meets the angular distance verification conditions.

2. The method according to claim 1, characterized in that, The selection of a suitable navigation satellite and the construction of its mode features include the following steps: Based on the initial attitude provided by the attitude sensor and the camera's field of view R, a radius of 1.5R is defined with the initial attitude center as the center. Within this radius, stars with magnitudes less than the limiting magnitude that the detector can detect are selected as navigation stars. Using the selected navigation star as the center, a radial mode radius R is defined. r Choose N as the radius. r For each nearest neighbor star, calculate the angular distance between the navigation star and each nearest neighbor star, and assign them to the corresponding rings. Construct a radial mode vector with a length equal to the number of rings. Set the rings containing nearest neighbors to 1, and set them to 0 otherwise, making it a vector composed of only 0 and 1 elements. Using the selected navigation star as the center, extend the radial mode vector with a radius R. c Choose N as the radius. c Starting with any one of the nearest neighboring stars, calculate the angle θ between each nearest neighboring star and the previous nearest neighboring star in a counter-clockwise direction. ij , denoted as V θ ={θ 12 ,θ 23 ,...,θ k(k+1) ,...,θ (n-1)n ,θ n1 } and calculate the length of the terminal side of the included angle, denoted as V. r ={d r2 ,d r3 ,...,d r(k+1) ,...,d rn ,d r1 Together, they constitute the circumferential angular distance pattern vector.

3. The method according to claim 1, characterized in that, The matching of radial and circumferential angular distances and the verification of link angular distances include the following steps: The radial pattern vector of the observed star is compared with the radial pattern vectors of all navigation stars stored in the navigation database. The number of identical elements at corresponding positions in the vectors is recorded. If the number exceeds the radial matching threshold, the navigation star is identified as a candidate star. Angle matching thresholds and angular distance matching thresholds are set. A pair of values ​​between the navigation star and the observed star that satisfy the set threshold conditions is found. The vectors are then cyclically shifted, and the cyclically shifted angle vectors are accumulated. The similarity between the accumulated circumferential feature vectors of the navigation star and the observed star is matched. Each position that meets the threshold requirement is considered a similarity score. The candidate star corresponding to the highest similarity score is selected as the candidate star. For candidate matching stars of the observed star, if the similarity value of the candidate matching star is greater than the circumferential matching threshold, the candidate star is output as the initial recognition result. The condition for link connection is that the angular distance between the previous star and the next star in the image and the angular distance of the recognition result on the celestial sphere are less than the angular distance verification threshold. The angular distance error of each star on the longest link with all other stars is calculated, and the combinations of these errors that exceed the angular distance verification threshold are counted. The star with the highest frequency is removed from the initial matching result. Then the longest link is reconnected until all angular distance errors in this step are less than the angular distance verification threshold.

Images