An astronomical image dataset labeling method

Abstract

The present application belongs to the technical field of astronomical image data processing, and particularly relates to a kind of astronomical image data set marking method, to solve the problem that the single threshold segmentation marking and star map projection marking based on existing cannot meet the demand of high-quality data set for deep model.The present application is based on threshold segmentation marking and star map projection marking, and the first marking result and difference set are used as marking result A for the target that can be detected with high probability in observation image, the second marking result is used as marking result B for the target that can be detected with low probability in observation image, and the final marking result is obtained by integrating marking result A and marking result B.The present application can improve the integrity of target marking, reduce the probability of missing and wrong marking, meet the actual fit degree of marking result and image, realize the unity of integrity and accuracy, and improve the performance of deep model for the data set produced.

Description

Technical Field

[0001] This invention belongs to the field of astronomical image data processing technology, specifically relating to a method, system, electronic device, and computer storage medium for labeling astronomical image datasets. Background Technology

[0002] With the advancement of observation instruments and technologies, the field of astronomical observation has undergone tremendous changes. One notable feature is the significant improvement in the speed of astronomical image acquisition. The amount of data generated daily by various observation instruments can reach the petabyte (PB) level. An urgent problem to be solved is how to quickly label these massive image data to support subsequent deep model processing and analysis of astronomical image data.

[0003] Due to the massive amount of data, manual image annotation is impractical. Threshold-based segmentation and star map projection are feasible methods for automatic annotation of astronomical image datasets. Threshold-based segmentation classifies objects into targets or background by comparing their grayscale range. This method can fully reflect image information and has high execution efficiency and simple implementation, but it suffers from the problem of missed or false detection of targets (especially low magnitude targets). Star map projection first performs star map matching on the image. After registration and location determination, it directly projects all targets in the star catalog within that range onto the image. Since this method directly uses the star catalog to annotate targets, there is no problem of missed or false detection of targets. However, the accuracy of the annotation differs from the actual image (observation equipment has limiting magnitudes, and not all targets in the star catalog are reflected in the image). In addition, the star catalog database only contains information on natural celestial bodies and cannot annotate artificial celestial bodies.

[0004] Therefore, neither of the above two automatic annotation methods can provide a high-quality labeled dataset for deep models. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, namely the current feasible methods for automatically annotating astronomical image datasets based on threshold segmentation and star map projection, the former suffers from target omission and false detection, while the latter has discrepancies between the annotation accuracy and the actual image and cannot annotate man-made celestial objects. Both methods fail to meet the requirements of deep learning models for high-quality datasets, resulting in low accuracy in celestial object detection. Therefore, in the first aspect, this invention proposes an annotation method for astronomical image datasets, which includes the following steps:

[0006] S1, Obtain observation images at a preset resolution from the observation image dataset;

[0007] S2, The star map registration method is used to process the salient celestial targets in the observed image and output the first annotation result; the salient celestial targets include natural celestial objects;

[0008] A threshold segmentation method is used to process all celestial targets in the observed image, and a second annotation result is output.

[0009] S3, Statistical analysis of the first annotation result and the second annotation result to generate the star magnitude segmentation threshold K:

[0010] S31, using the magnitude of the celestial target as the independent variable, statistically analyze the magnitude distribution of the target in the first annotation result and the second annotation result respectively, and generate the first magnitude distribution histogram and the second magnitude distribution histogram;

[0011] S32, determine the limiting magnitude of the observed image based on the second magnitude distribution histogram, and define the target magnitude range [S0, S1], where S0 represents the lower limit magnitude value of the target magnitude range and S1 represents the upper limit magnitude value of the target magnitude range.

[0012] S33, Traverse the candidate star magnitude values ​​s within the target star magnitude range [S0, S1] and calculate the probability value P0 corresponding to each s;

[0013] S34, when P0 ≤ preset probability confidence level P, terminate the traversal and use the current s value as the star magnitude segmentation threshold K;

[0014] S4, within the target magnitude range [S0, S1], compare each target magnitude value with the magnitude segmentation threshold K. For targets whose target magnitude values ​​are less than or equal to K, use the first annotation result and the difference set as the annotation result A.

[0015] For targets with a star magnitude greater than K, the second annotation result is used as annotation result B;

[0016] S5, integrate the annotation result A and the annotation result B to obtain the final annotation result of the observed image.

[0017] In some preferred embodiments, the method of processing salient celestial targets in the observed image using star map registration to output a first annotation result is as follows:

[0018] The observed image is segmented based on a preset threshold to extract significant celestial targets; the threshold is N times the standard deviation of the sky background of the observed image; N includes 4.5;

[0019] Geometric features between salient celestial objects are extracted, and these features are matched with the geometric features between objects in a reference star catalog. Geometric transformation is then used to align the observed image with the reference star catalog.

[0020] The aligned salient celestial objects are transformed to the image coordinate system through coordinate projection.

[0021] In some preferred embodiments, the matching includes matching using a polygon angular distance matching algorithm;

[0022] The geometric transformations include translation and rotation;

[0023] The coordinate projection transformation includes the Mercator projection method.

[0024] In some preferred embodiments, the method of using threshold segmentation to process all celestial targets in the observed image and outputting a second annotation result is as follows: the observed image is converted to grayscale, a grayscale value threshold is determined on the grayscale observed image, celestial targets and background are extracted, and a second annotation result is output; the threshold is M times the standard deviation of the skylight background of the observed image; M includes 1.5.

[0025] In some preferred embodiments, the preset probability confidence level P ranges from 90% to 95%.

[0026] In some preferred embodiments, the probability value P0 is the ratio of the number of significant celestial objects with a magnitude less than or equal to K in the first magnitude distribution histogram to the number of all labeled objects with a magnitude less than or equal to K in the second magnitude distribution histogram.

[0027] In some preferred embodiments, the difference set is the set of differences between the second annotation result and the targets in the first annotation result whose star magnitude values ​​are in the range [S0, K].

[0028] In some preferred embodiments, the initial value of s is set with reference to the magnitude histogram and empirical values.

[0029] The second invention proposes an annotation system for astronomical image datasets. Based on the aforementioned annotation method for astronomical image datasets, the system includes:

[0030] The image acquisition module is configured to acquire observation images at a preset resolution from the observation image dataset;

[0031] The target processing module is configured to: process salient celestial targets in the observed image using star map registration method and output the first annotation result; and process all celestial targets in the observed image using threshold segmentation method and output the second annotation result.

[0032] The threshold generation module is configured to perform statistical analysis on the first and second annotation results to generate the star magnitude segmentation threshold K.

[0033] Using the magnitude of celestial targets as the independent variable, the magnitude distribution of targets in the first and second annotation results is statistically analyzed to generate the first and second magnitude distribution histograms.

[0034] The limiting magnitude of the observed image is determined based on the second magnitude distribution histogram, and the target magnitude range [S0,S1] is defined.

[0035] Traverse the candidate star magnitude values ​​s within the target star magnitude range [S0, S1] and calculate the probability value P0 corresponding to each s;

[0036] When P0 ≤ preset probability confidence level P, terminate the traversal and use the current s value as the star magnitude segmentation threshold K;

[0037] The result selection module is configured as follows: within the target star magnitude range [S0, S1], the target star magnitude value is compared with the star magnitude segmentation threshold K. For targets with a star magnitude value less than or equal to K, the first annotation result and the difference set are used as annotation result A, where the difference set is the set of differences between the second annotation result and the targets with star magnitude values ​​within the range [S0, K] in the first annotation result; for targets with a star magnitude value greater than K, the second annotation result is used as annotation result B.

[0038] The integration module is configured to integrate annotation results A and B to obtain the final annotation results of the observed image.

[0039] The third invention proposes an electronic device, comprising:

[0040] At least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor to implement the aforementioned method for labeling an astronomical image dataset.

[0041] The fourth invention proposes a computer-readable storage medium storing computer instructions for execution by a computer to implement the aforementioned annotation method for an astronomical image dataset.

[0042] The beneficial effects of this invention are:

[0043] Targets that are likely to be detected in the observed image are labeled using the first annotation result and the difference set as annotation result A, while targets that are unlikely to be detected in the observed image are labeled using the second annotation result B. Annotation results A and B are integrated to obtain the final annotation result. This annotation method can improve the completeness of target annotation, reduce the probability of missing or incorrect annotation, and also meet the requirement of the annotation result being consistent with the actual image, achieving a balance between completeness and accuracy. The resulting dataset can improve the performance of the deep model and improve the accuracy of celestial target detection. Attached Figure Description

[0044] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0045] Figure 1 This is a flowchart illustrating the steps of an astronomical image dataset annotation method according to the present invention.

[0046] Figure 2 This is an example diagram of an annotation method for an astronomical image dataset according to the present invention.

[0047] Figure 3 This invention relates to a histogram of an annotation method for an astronomical image dataset. Figure 1 Example image.

[0048] Figure 4 This invention relates to a histogram of an annotation method for an astronomical image dataset. Figure 2 Example image. Detailed Implementation

[0049] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0050] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0051] To more clearly illustrate the annotation method for an astronomical image dataset of the present invention, the following will be combined with... Figures 1 to 4 The steps in the embodiments of the present invention will be described in detail below.

[0052] With the rapid popularization of artificial intelligence, celestial object detection currently relies heavily on deep learning models. A crucial prerequisite for good detection performance is a high-quality labeled dataset. This invention proposes a labeling method for astronomical image datasets. This method integrates the advantages of multiple labeling methods, closely reflecting the actual images while minimizing the probability of missed and false detections. In most scenarios, this method can be used to quickly generate the required datasets; details are as follows:

[0053] S1, Obtain observation images at a preset resolution from the observation image dataset;

[0054] The pixel resolutions of commonly observed images are 4k, 6k, and 8k;

[0055] S2, the star map registration method is used to process salient celestial objects in the observed image and output the first annotation result; the salient celestial objects include natural celestial objects (e.g., stars).

[0056] In this embodiment, the method of processing salient celestial targets in the observed image using star map registration to output the first annotation result is as follows:

[0057] The observed image is segmented based on a preset threshold to extract salient celestial targets (large and bright point targets, usually natural celestial bodies such as stars; artificial celestial bodies are generally small targets. Since the field of view of the observation equipment is usually very large, an astronomical image covers tens of thousands of celestial targets, so there are multiple salient celestial targets in an image); the threshold is N times the standard deviation of the sky background of the observed image; N includes 4.5;

[0058] Geometric features between salient celestial objects are extracted, and these features are matched with the geometric features between objects in a reference star catalog. Geometric transformation is then used to align the observed image with the reference star catalog.

[0059] The matching process includes using a polygonal angular distance matching algorithm. Specifically, the polygonal angular distance matching algorithm calculates the distance and angular similarity indices between the extracted salient celestial targets and the topological structure of the reference star catalog to find the best matching pair. The reference star catalog can be obtained from a star catalog database based on the observation direction and contains information such as the number, type, position, brightness, and spectrum of natural celestial objects. Star catalog databases are generally published by professional astronomical institutions. In addition, the star map matching algorithm can be chosen arbitrarily and is not limited to the above method. Generally speaking, the more salient celestial targets extracted, the more unique the spatial information they present, and the higher the matching accuracy. After determining the matching pair, the observed target is aligned with the target in the reference star catalog through translation and rotation operations to complete the star map registration.

[0060] The aligned salient celestial objects are transformed to the image coordinate system via coordinate projection.

[0061] Specifically, after star map registration is completed, the aligned salient celestial targets are combined with the reference star catalog to project all natural celestial targets within the observation range into the observation image. The observation range is determined by the pixel resolution of the observation equipment, which is typically 4k, 6k, and 8k. Different resolutions map different ranges, with higher resolutions mapping to larger ranges. Projection realizes the conversion of astronomical coordinates to pixel coordinates. Common projection methods include equidistant cylindrical projection, stereographic projection, and Mercator projection. The image coordinates of all celestial targets in the observation image can be accurately obtained by combining the projected pixel coordinates with the range limited by the pixel resolution of the image. This results in the complete annotation of salient celestial targets in the observation image, which serves as the first annotation result.

[0062] A threshold segmentation method is used to process all celestial targets in the observed image and output a second annotation result. The method is as follows: the observed image is converted to grayscale, and based on the grayscale value threshold judgment of the grayscale observed image, celestial targets and background are extracted (grayscale values ​​exceeding the threshold are targets, otherwise they are background), and the second annotation result is output (the operation is performed directly on the image, so the extracted targets already have image coordinate information); the threshold is 1.5 times the standard deviation of the sky light and background of the observed image, and iterative thresholding method or OTSU method can also be used;

[0063] S3, Statistical analysis of the first annotation result and the second annotation result to generate the star magnitude segmentation threshold K:

[0064] S31, see also Figure 3 , Figure 4 Using the magnitude of celestial targets as the independent variable, the magnitude distribution of targets in the first and second annotation results is statistically analyzed to generate a first magnitude distribution histogram (i.e., histogram). Figure 1 ) and the second star distribution histogram (i.e., histogram) Figure 2 Then, from the magnitude distribution histogram, the number of celestial targets corresponding to each magnitude value in the two types of annotation methods and the limiting magnitude value of the observed image (i.e. the observable magnitude range of the observed image) can be directly obtained;

[0065] S32, determine the limiting magnitude of the observed image based on the second magnitude distribution histogram, and define the target magnitude range [S0, S1], where S0 represents the lower limit magnitude value of the target magnitude range and S1 represents the upper limit magnitude value of the target magnitude range.

[0066] S33, Traverse the candidate star magnitude values ​​s within the target star magnitude range [S0, S1]. Calculate the probability value P0 corresponding to each s;

[0067] S34, when P0 ≤ preset probability confidence level P, terminate the traversal and use the current s value as the star magnitude segmentation threshold K;

[0068] Specifically, compare the size of P0 and P. If P0 is greater than P, increase s and repeat the above operation until P0 is less than or equal to P. Conversely, if P0 is less than P, decrease s until it is exactly at the critical value where P0 is less than or equal to P. The value of s at this time is the star magnitude segmentation threshold K.

[0069] In this embodiment, the process is detailed below with the target star magnitude range being [4.414, 13.697] and the probability confidence level P being set to 95%: The target star magnitude range of the observed image is [4.414, 13.697], the probability confidence level P is set to 95%, and the selected... (The initial value of s can be set based on the magnitude histogram, and it is not necessary to start from the starting point of the interval each time, which will increase the traversal time.) Compare the probability value P0 and the probability confidence P obtained under different values ​​of s. When P0≤P, stop traversing. The magnitude value corresponding to the current s is the magnitude segmentation threshold K in this example, which is 10.288.

[0070] S4, within the target magnitude range [S0, S1], compare the target magnitude value with the magnitude segmentation threshold K. For targets less than or equal to K, it is considered that the observed image has a high probability of detecting the target within the magnitude range, and the first annotation result and the difference set are used as the annotation result A.

[0071] The difference set is the set of differences between the second annotation result and the first annotation result for targets with magnitude values ​​in the range [S0, K]; the set of differences is also the set of artificial celestial targets;

[0072] Within the observed target magnitude range [4.414, 13.697], when the magnitude value is less than or equal to the magnitude threshold, the first annotation result (i.e., the result of star map projection) is used as the annotation result; when the magnitude value is greater than the magnitude threshold, the second annotation result (i.e., the result of threshold segmentation) is used as the annotation result. In particular, in order to supplement the missing artificial celestial targets in the star map projection result, the results below the magnitude threshold range are subtracted. In this example, the difference is made for all targets with magnitude values ​​in the range [4.414, 10.288] in both annotation methods, and the difference result is added to the first annotation result.

[0073] The initial value of s is set with reference to the magnitude histogram and in combination with practical experience. The standard for determining the value of s is that the corresponding probability value P0 should be less than or equal to the set probability confidence level P. Therefore, based on the number of targets of each magnitude in the two magnitude histograms and the judgment of practical experience, a suitable initial value can be roughly selected. For example, according to the magnitude histogram, when the magnitude is less than a, the number of histograms for threshold segmentation is significantly less than the number of star map registrations under the same range. It can be determined that the value of s must be greater than a. That is, based on the total number of histograms in the two methods, the approximate range of the value of s can be determined first, and then the specific value of s can be determined by iterating through the histograms.

[0074] For targets with a magnitude greater than K, it is considered that there is a probability problem in the observation of these targets in the image, and the image cannot fully reflect all targets in this magnitude range. Therefore, the second annotation result is used as annotation result B.

[0075] In this embodiment, the preset probability confidence level P ranges from 90% to 95%.

[0076] The probability value P0 is the number of significant celestial objects with a magnitude less than or equal to K in the first magnitude distribution histogram. The number of all labeled targets with a magnitude value less than or equal to K in the second magnitude distribution histogram. The ratio, P0 expression:

[0077] ;

[0078] S5, integrate the annotation result A and the annotation result B to obtain the final annotation result of the observed image; wherein, the integration includes directly taking the union to achieve integration, the targets that are likely to be detected in the observed image are used as the first annotation result and the difference set as the annotation result A, and the targets that are unlikely to be detected in the observed image are used as the second annotation result B. Integrating the annotation result A and the annotation result B, the final annotation result is obtained. This annotation method can not only improve the completeness of the target annotation and reduce the probability of missing or incorrect annotation, but also meet the requirement of the annotation result being consistent with the actual image.

[0079] Although the steps in the above embodiments are described in the above order, those skilled in the art will understand that in order to achieve the effect of this embodiment, different steps do not need to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple variations are all within the protection scope of this invention.

[0080] A second embodiment of the present invention proposes an annotation system for astronomical image datasets. Based on the aforementioned annotation method for astronomical image datasets, the system includes:

[0081] The image acquisition module is configured to acquire observation images at a preset resolution from the observation image dataset;

[0082] The target processing module is configured to: process salient celestial targets in the observed image using star map registration method and output the first annotation result; and process all celestial targets in the observed image using threshold segmentation method and output the second annotation result.

[0083] The threshold generation module is configured to perform statistical analysis on the first and second annotation results to generate the star magnitude segmentation threshold K.

[0084] Using the magnitude of celestial targets as the independent variable, the magnitude distribution of targets in the first and second annotation results is statistically analyzed to generate the first and second magnitude distribution histograms.

[0085] The limiting magnitude of the observed image is determined based on the second magnitude distribution histogram, and the target magnitude range [S0,S1] is defined.

[0086] Traverse the candidate star magnitude values ​​s within the target star magnitude range [S0, S1] and calculate the probability value P0 corresponding to each s;

[0087] When P0 ≤ the preset probability confidence level P, the traversal is terminated and the current s value is used as the star magnitude segmentation threshold K;

[0088] The result selection module is configured as follows: within the target star magnitude range [S0, S1], the target star magnitude value is compared with the star magnitude segmentation threshold K. For targets with a star magnitude value less than or equal to K, the first annotation result and the difference set are used as annotation result A, where the difference set is the set of differences between the second annotation result and the targets with star magnitude values ​​within the range [S0, K] in the first annotation result; for targets with a star magnitude value greater than K, the second annotation result is used as annotation result B.

[0089] The integration module is configured to integrate annotation results A and B to obtain the final annotation results of the observed image.

[0090] It should be noted that the astronomical image dataset annotation system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0091] A third embodiment of the present invention provides an electronic device comprising:

[0092] At least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor to implement the aforementioned method for labeling an astronomical image dataset.

[0093] In a fourth embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, which are executed by the computer to implement the aforementioned annotation method for an astronomical image dataset.

[0094] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the electronic device and computer-readable storage medium described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0095] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the invention.

[0096] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0097] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0098] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for labeling astronomical image datasets, characterized in that, The method includes the following steps: S1, Obtain observation images at a preset resolution from the observation image dataset; S2, The star map registration method is used to process the salient celestial targets in the observed image and output the first annotation result; the salient celestial targets include natural celestial objects; A threshold segmentation method is used to process all celestial targets in the observed image, and a second annotation result is output. S3, Statistical analysis of the first annotation result and the second annotation result to generate the star magnitude segmentation threshold K: S31, using the magnitude of the celestial target as the independent variable, statistically analyze the magnitude distribution of the target in the first annotation result and the second annotation result respectively, and generate the first magnitude distribution histogram and the second magnitude distribution histogram; S32, determine the limiting magnitude of the observed image based on the second magnitude distribution histogram, and define the target magnitude range [S0, S1], where S0 represents the lower limit magnitude value of the target magnitude range and S1 represents the upper limit magnitude value of the target magnitude range. S33, within the target magnitude range [S0, S1], traverse the candidate magnitude values ​​s and calculate the probability value P0 corresponding to each s; the probability value P0 is the ratio of the number of significant celestial targets with magnitude values ​​less than or equal to K in the first magnitude distribution histogram to the number of all labeled targets with magnitude values ​​less than or equal to K in the second magnitude distribution histogram; S34, when P0 ≤ preset probability confidence level P, terminate the traversal and use the current s value as the star magnitude segmentation threshold K; S4, within the target magnitude range [S0, S1], compare each target magnitude value with the magnitude segmentation threshold K. For targets whose target magnitude value is less than or equal to K, use the first annotation result and the difference set as the annotation result A; the difference set is the set of differences between the second annotation result and the targets in the first annotation result whose magnitude values ​​are within the range [S0, K]. For targets with a star magnitude greater than K, the second annotation result is used as annotation result B; S5, integrate the annotation result A and the annotation result B to obtain the final annotation result of the observed image.

2. The annotation method for an astronomical image dataset according to claim 1, characterized in that, The method of processing salient celestial targets in the observed image using star map registration to output the first annotation result is as follows: The observed image is segmented based on a preset threshold to extract significant celestial targets; the preset threshold is N times the standard deviation of the sky background of the observed image; N includes 4.5; Geometric features between salient celestial objects are extracted, and these features are matched with the geometric features between objects in a reference star catalog. Geometric transformation is then used to align the observed image with the reference star catalog. The aligned salient celestial objects are transformed to the image coordinate system through coordinate projection.

3. The annotation method for an astronomical image dataset according to claim 2, characterized in that, The matching includes matching using a polygon angular distance matching algorithm; The geometric transformations include translation and rotation; The coordinate projection transformation includes the Mercator projection method.

4. The annotation method for an astronomical image dataset according to claim 1, characterized in that, The threshold segmentation method is used to process all celestial targets in the observed image and output a second annotation result. The method is as follows: the observed image is converted to grayscale, a grayscale value threshold is determined on the grayscale observed image, celestial targets and background are extracted, and a second annotation result is output; the grayscale value threshold is M times the standard deviation of the sky light background of the observed image; M includes 1.

5.

5. The annotation method for an astronomical image dataset according to claim 1, characterized in that, The preset probability confidence level P ranges from 90% to 95%.

6. The annotation method for an astronomical image dataset according to claim 1, characterized in that, The initial value of s is set with reference to the first magnitude distribution histogram, the second magnitude distribution histogram, and empirical values.

7. A system for labeling an astronomical image dataset, comprising the labeling method for an astronomical image dataset as described in any one of claims 1-6, characterized in that, The system includes: The image acquisition module is configured to acquire observation images at a preset resolution from the observation image dataset; The target processing module is configured to: process salient celestial targets in the observed image using star map registration method and output the first annotation result; and process all celestial targets in the observed image using threshold segmentation method and output the second annotation result. The threshold generation module is configured to perform statistical analysis on the first and second annotation results to generate the star magnitude segmentation threshold K. Using the magnitude of celestial targets as the independent variable, the magnitude distribution of targets in the first and second annotation results is statistically analyzed to generate the first and second magnitude distribution histograms. The limiting magnitude of the observed image is determined based on the second magnitude distribution histogram, and the target magnitude range [S0, S1] is defined. Within the target magnitude range [S0, S1], traverse the candidate magnitude values ​​s and calculate the probability value P0 corresponding to each s; the probability value P0 is the ratio of the number of significant celestial objects with magnitude values ​​less than or equal to K in the first magnitude distribution histogram to the number of all labeled objects with magnitude values ​​less than or equal to K in the second magnitude distribution histogram; When P0 ≤ preset probability confidence level P, terminate the traversal and use the current s value as the star magnitude segmentation threshold K; The result selection module is configured as follows: within the target star magnitude range [S0, S1], the target star magnitude value is compared with the star magnitude segmentation threshold K. For targets with a star magnitude value less than or equal to K, the first annotation result and the difference set are used as annotation result A, where the difference set is the set of differences between the second annotation result and the targets with star magnitude values ​​within the range [S0, K] in the first annotation result; for targets with a star magnitude value greater than K, the second annotation result is used as annotation result B. The integration module is configured to integrate annotation results A and B to obtain the final annotation results of the observed image.

8. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor to implement a method for labeling an astronomical image dataset as described in any one of claims 1-6.

Images