Celestial body data retrieval method and device, storage medium and electronic equipment

Abstract

The invention discloses a celestial body data retrieval method and device, a storage medium and electronic equipment, and the method comprises the steps: matching the coordinate information of each celestial body image in a celestial body image header file with the coordinate information contained in each celestial body star catalogue record in a star catalogue database of each different data source according to the coordinate similarity; and determining each celestial body data pair formed by the celestial body image and the celestial body star catalogue record which have a matching relationship, and for each celestial body data pair, carrying out feature coding on the celestial body image and the celestial body star catalogue record in the celestial body data pair to obtain the multi-modal features of the celestial body data pair. A multi-modal feature is constructed by associating a celestial body image in a celestial body image header file with celestial body star catalog records in a star catalog database. And when celestial body data retrieval needs to be carried out, obtaining a retrieval result according to the multi-modal features. According to the method, the similarity of data features in the celestial body image and the celestial body star catalogue record is considered at the same time, and more accurate retrieval precision can be achieved.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a celestial data retrieval method, device, storage medium, and electronic device. Background Art

[0002] In astronomy, various parameters of celestial objects, such as their positions, motions, and spectra, are stored in star catalog databases. Currently, when searching for similar celestial objects, the spherical cosine theorem or the semi-rectangular error formula is often used to calculate the difference in coordinates (right ascension and declination) between the object being searched and the objects in the catalog databases. This is an example of the object search method used by tools such as TOPCAT and Aladin.

[0003] Due to differences in observation conditions, resolution and format of data sources in different star catalog databases, there are differences in the data recorded in different star catalog databases. The coordinates of the same celestial body recorded in different star catalog databases are often different, resulting in poor accuracy of the existing method of searching based only on celestial body coordinates.

[0004] Therefore, this specification provides a celestial body data retrieval method. Summary of the Invention

[0005] This specification provides a celestial body data retrieval method, device, storage medium and electronic device to at least partially solve the above-mentioned problems existing in the prior art.

[0006] This manual adopts the following technical solutions:

[0007] This specification provides a celestial body data retrieval method, including:

[0008] Based on coordinate similarity, the coordinate information of each celestial image in the celestial image header file is matched with the coordinate information contained in each celestial catalog record in the star catalog database of each different data source, and each celestial data pair consisting of a celestial image and a celestial catalog record having a matching relationship is determined;

[0009] For each celestial data pair, feature encoding is performed on the celestial image and celestial catalog record in the celestial data pair to obtain multimodal features of the celestial data pair;

[0010] In response to a retrieval task for celestial body data to be retrieved, performing similarity matching on features to be retrieved obtained by encoding features of the celestial body data to be retrieved and multimodal features of each celestial body data pair, and determining a target multimodal feature that matches the features to be retrieved, wherein the celestial body data to be retrieved is a celestial body image or at least a partial celestial body catalog record;

[0011] A search result is determined based on the celestial body data pairs corresponding to the target multimodal features.

[0012] Optionally, according to coordinate similarity, coordinate information of each celestial image in the celestial image header file is matched with coordinate information contained in each celestial catalog record in the catalog database of each different data source, to determine each celestial data pair composed of a celestial image and a celestial catalog record that have a matching relationship, and the method specifically comprises:

[0013] In the catalog database of each different data source, coordinate information contained in each celestial catalog record is extracted as to-be-matched coordinate information;

[0014] For each celestial image in the celestial image header file, to-be-matched coordinate information that has a difference from coordinate information of the celestial image within a specified range is screened out in the catalog database of each different data source;

[0015] A celestial catalog record indicated by the screened out to-be-matched coordinate information and the celestial image are taken as a celestial data pair.

[0016] Optionally, for each celestial data pair, a celestial image and a celestial catalog record in the celestial data pair are feature-encoded to obtain a multi-modal feature of the celestial data pair, and the method specifically comprises:

[0017] For each celestial data pair, an image feature of the celestial image in the celestial data pair is obtained through a preset image feature embedding model;

[0018] For the celestial catalog record in the celestial data pair, according to a table numerical type contained in the celestial catalog record, a table numerical feature of the celestial catalog record is obtained by using a set encoding mode corresponding to the numerical type;

[0019] The image feature and the table numerical feature are spliced to obtain the multi-modal feature of the celestial data pair.

[0020] Optionally, the celestial catalog record in the celestial data pair has an abnormal value;

[0021] For each celestial data pair, a celestial image and a celestial catalog record in the celestial data pair are feature-encoded to obtain a multi-modal feature of the celestial data pair, and the method specifically comprises:

[0022] In a catalog database to which the celestial catalog record in the celestial data pair belongs, data values of other celestial catalog records of a data column corresponding to the abnormal value are determined;

[0023] According to the data values of the other celestial catalog records, a mathematical probability model to which each data value of the data column corresponding to the abnormal value is subjected is fitted;

[0024] According to the mathematical probability model, an estimated data value of the abnormal value is sampled;

[0025] Filling the celestial object catalog record of the celestial object data pair based on the estimated data value;

[0026] Feature encoding is performed on the celestial body image and the filled celestial body catalog record in the celestial body data pair to obtain the multimodal features of the celestial body data pair.

[0027] Optionally, performing similarity matching on the features to be retrieved obtained by encoding the features of the celestial body data to be retrieved and the multimodal features of each celestial body data pair specifically includes:

[0028] Using a plurality of specified indexing algorithms, respectively calculating the vector distances between the features to be retrieved obtained by encoding the features of the celestial body data to be retrieved and the multimodal features of each celestial body data pair;

[0029] For each indexing algorithm, according to a set recall threshold, candidate celestial body data pairs are returned from the celestial body data pairs, and the vector distance ranking between the multimodal features of the candidate celestial body data and the features to be retrieved is within the recall threshold range;

[0030] Integrating the candidate celestial body data pairs returned by each indexing algorithm to obtain candidate result pairs of the celestial body data to be retrieved;

[0031] For each candidate result pair, the vector distances of the candidate result pair calculated based on the indexing algorithms are weighted to determine a similarity score between the candidate result pair and the celestial body data to be retrieved.

[0032] Optionally, before weighting the vector distances calculated based on the indexing algorithms for each candidate result pair and determining a similarity score between the candidate result pair and the celestial body data to be retrieved, the method further includes:

[0033] Within each class of celestial bodies in a known astronomical data set, celestial body data are extracted in pairs to construct sample pairs with matching relationships; within two different classes of celestial bodies in the known astronomical data set, celestial body data are extracted respectively to construct sample pairs without matching relationships;

[0034] For each sample pair, a plurality of pre-specified indexing algorithms are used to calculate the vector distance between the tabular numerical features of the celestial catalog records of the two celestial bodies in the sample pair obtained by each indexing algorithm;

[0035] Input the obtained vector distances into the regression model to be trained to obtain the similarity score of the sample pair;

[0036] Training the regression model based on the similarity score of the sample pair and the difference between the class label of the sample pair, wherein the class label of the sample pair with a matching relationship is 1, and the class label of the sample pair with no matching relationship is 0;

[0037] For each candidate result pair, weighting the vector distances calculated based on each indexing algorithm for the candidate result pair to determine a similarity score between the candidate result pair and the celestial body data to be retrieved specifically includes:

[0038] For each candidate result pair, the vector distance calculated based on each indexing algorithm for the candidate result pair is input into the trained regression model to obtain a similarity score between the candidate result pair and the celestial body data to be retrieved.

[0039] Optionally, there are missing values ​​in the vector distances of the candidate result pairs calculated based on the respective indexing algorithms;

[0040] Weighting the vector distances calculated based on each indexing algorithm for the candidate result pair to determine a similarity score between the candidate result pair and the celestial body data to be retrieved, specifically comprising:

[0041] Obtaining a maximum value of a distance vector calculated in each candidate result pair based on a target indexing algorithm, wherein the target indexing algorithm is an indexing algorithm for calculating the missing value;

[0042] Fill the missing values ​​with the maximum value;

[0043] The vector distances calculated based on the candidate result pair and each indexing algorithm are weighted to determine a similarity score between the candidate result pair and the celestial body data to be retrieved.

[0044] This specification provides a celestial body data retrieval device, the device comprising:

[0045] A matching module matches the coordinate information of each celestial image in the celestial image header file with the coordinate information contained in each celestial catalog record in the star catalog database of each different data source based on coordinate similarity, and determines each celestial data pair consisting of a celestial image and a celestial catalog record having a matching relationship;

[0046] An encoding module, for each celestial data pair, performs feature encoding on the celestial image and celestial catalog record in the celestial data pair to obtain multimodal features of the celestial data pair;

[0047] a retrieval module, in response to a retrieval task for celestial body data to be retrieved, performing similarity matching between features to be retrieved obtained by encoding features of the celestial body data to be retrieved and multimodal features of each celestial body data pair, and determining a target multimodal feature that matches the features to be retrieved, wherein the celestial body data to be retrieved is a celestial body image or at least a partial celestial body catalog record;

[0048] The result returning module determines the search result according to the celestial body data pair corresponding to the target multimodal feature.

[0049] This specification provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned celestial body data retrieval method is implemented.

[0050] This specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned celestial body data retrieval method is implemented.

[0051] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:

[0052] In the celestial data retrieval method provided in this specification, the coordinate information of each celestial image in the celestial image header file is matched with the coordinate information contained in each celestial star catalog record in the star catalog database of each different data source based on coordinate similarity, and each celestial data pair consisting of a celestial image and a celestial star catalog record with a matching relationship is determined. For each celestial data pair, feature encoding is performed on the celestial image and the celestial star catalog record in the celestial data pair to obtain the multimodal features of the celestial data pair.

[0053] By associating the celestial image in the celestial image header file with the celestial catalog records in the star catalog database, a multimodal feature representation of the celestial image and the celestial catalog records is constructed. When celestial data is retrieved, the multimodal feature representation is used to obtain search results. This method also considers the similarity of various data features in the celestial image and the celestial catalog records, achieving more accurate search precision. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The drawings described herein are used to provide a further understanding of this specification and constitute a part of this specification. The exemplary embodiments and descriptions of this specification are used to explain this specification and do not constitute an improper limitation of this specification. In the drawings:

[0055] Figure 1 This is a flowchart of a celestial body data retrieval method in this manual;

[0056] Figure 2 A schematic diagram of the structure of a retrieval system in this specification;

[0057] Figure 3 A schematic diagram of a celestial data retrieval device provided in this specification;

[0058] Figure 4 The corresponding Figure 1 Schematic diagram of electronic equipment. DETAILED DESCRIPTION

[0059] To make the purpose, technical solutions, and advantages of this specification more clear, the technical solutions of this specification will be clearly and completely described below in conjunction with the specific embodiments of this specification and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0060] The technical solutions provided by the embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0061] Figure 1 The following is a flowchart of a celestial body data retrieval method in this specification, which specifically includes the following steps:

[0062] S100: According to coordinate similarity, the coordinate information of each celestial body image in the celestial body image header file is matched with the coordinate information contained in each celestial body star catalog record in the star catalog database of each different data source, and each celestial body data pair consisting of a celestial body image and a celestial body star catalog record having a matching relationship is determined.

[0063] All steps in the celestial data retrieval method provided in this specification can be implemented by any electronic device with computing capabilities, such as a terminal, server, or other device. For ease of description, the following description of the celestial data retrieval method provided in this specification uses a server as the execution entity.

[0064] In astronomy, star catalog databases are used to store structured data such as celestial coordinate information (right ascension, declination), magnitude, proper motion, and spectral type. Astronomical image header files are used to store celestial image data, such as film scans and raw telescope observation data. Astronomical image header files also contain the coordinates of the observation center of the celestial image and are typically stored in dedicated storage systems as FITS files.

[0065] Therefore, under the existing storage framework, the image data and star catalog data of celestial bodies are stored separately in different forms, and a connection cannot be established between the two types of information, making it difficult to apply the image information of celestial bodies in celestial body similarity retrieval.

[0066] Based on this limitation, at present, the data processing of celestial catalog data and the data processing of celestial image are often carried out respectively. For example, the tools such as TOPCAT and Aladin mainly match the catalog data, and rely on the calculation of celestial coordinates (right ascension and declination) (usually using the spherical cosine theorem or the formula of the secant of half an arc) to complete the matching of similar celestial bodies. The Galaxy Zoo project uses a deep learning model to classify celestial images, thereby extracting morphological features of celestial bodies.

[0067] In the present specification, in order to associate the celestial image and the celestial catalog record representing the same celestial body in the catalog database, the coordinate information of each celestial image in the celestial image header file is matched with the coordinate information contained in each celestial catalog record in the catalog database of each different data source according to the coordinate similarity, and each celestial data pair composed of the celestial image and the celestial catalog record with a matching relationship is determined.

[0068] In the public data, there are catalog databases established by multiple different data sources, and these catalog databases are also difficult to be commonly applied to the same retrieval task due to differences in observation conditions, data processing standards, etc. However, in the present method,

[0069] Specifically, in the catalog database of each different data source, the coordinate information contained in each celestial catalog record is extracted as the to-be-matched coordinate information.

[0070] The coordinate information of a celestial body is identified by right ascension and declination, and the to-be-matched coordinate information of the present step is the right ascension and declination of all celestial bodies contained in each celestial database. The celestial catalog records of multiple catalog databases are collected, and the matching range is expanded to catalog databases of different sources, so as to achieve a more accurate retrieval effect.

[0071] Secondly, for each celestial image in the celestial image header file, the to-be-matched coordinate information with a difference within a specified range from the coordinate information of the celestial image is screened in the catalog database of each different data source.

[0072] The specified range is a pre-specified tolerance range of coordinate difference, and the difference within the specified range can be considered as the same celestial body described by the celestial image and the celestial catalog record. The present specification does not limit the calculation method of the coordinate difference, and methods such as the formula of the secant of half an arc and the spherical cosine theorem can be used.

[0073] It is generally believed in the field of astronomy that two celestial bodies in different catalogs with an angular distance of less than 3 arcseconds are the same celestial body. Therefore, in an example, 3 arcseconds can be set as the specified range.

[0074] After the coordinate information to be matched is determined, the celestial catalog record containing the selected coordinate information and the celestial image are combined into a celestial data pair in the celestial catalog database. For a celestial image, multiple matching celestial catalog records may be matched in this step. In this case, the celestial image and each matching celestial catalog record will form a celestial data pair.

[0075] The image and catalog records of an object in an object data pair are considered to represent the same object. This allows the image data and catalog data of the same object to be associated.

[0076] S102: For each celestial body data pair, feature encoding is performed on the celestial body image and the celestial body catalog record in the celestial body data pair to obtain multimodal features of the celestial body data pair.

[0077] Celestial body catalog records are structured tabular data, and celestial body images are image data. The two belong to different data types and need to be feature encoded separately to construct a multimodal vector database for retrieval.

[0078] For the celestial image in the celestial data pair, a preset image embedding model can be used to extract the image features of the celestial image. This specification does not restrict the model type of this image embedding model, and it can be selected as needed.

[0079] A celestial body catalog record contains data values ​​of multiple celestial body attributes, such as an celestial body's identification code, right ascension, declination, magnitude, proper motion, and spectral type. The numerical types of different celestial body attribute data values ​​are different. When performing feature coding, celestial body attributes of different numerical types need to be processed separately.

[0080] Therefore, for the celestial body catalog record in the celestial body data pair, according to the table value type contained in the celestial body catalog record, a set encoding method corresponding to the value type is adopted to obtain the table value feature of the celestial body catalog record.

[0081] Specifically, if the numerical type of the celestial attribute is a floating-point continuous variable, the corresponding encoding method is set to Z-score normalization, which can eliminate the influence of different dimensions. If the numerical type of the celestial attribute is a multidimensional array, the corresponding encoding method is set to Principal Component Analysis (Flatten-PCA), and combined with dimensionality reduction to achieve embedding from tensor space to Euclidean space. If the numerical type of the celestial attribute is a type label, when the number of type label categories is less than the preset cardinality, the corresponding encoding method is set to generate orthogonal vectors using one-hot encoding. When the number of type label categories is not less than the preset cardinality, the corresponding encoding method is set to binary encoding to compress the dimension through bit operations. If the numerical type of the celestial attribute is a Boolean variable, the corresponding encoding method is set to binary value {0, 1} mapping.

[0082] The data values ​​of each celestial attribute in the celestial catalog record are encoded using a set encoding method to obtain the attribute characteristics of each celestial attribute. Then, the attribute characteristics are spliced ​​together to obtain a one-dimensional dense feature vector, which serves as the tabular numerical feature of the celestial catalog record.

[0083] Finally, the image features of the celestial image and the numerical features of the celestial table records are combined to obtain the multimodal features of the celestial data pair. This multimodal feature contains the information of the attributes of each celestial object in the celestial catalog record and the image information of the celestial object.

[0084] In this manual, a multimodal retrieval database is constructed based on the multimodal features of each celestial body pair, in the hope that in subsequent celestial body retrieval tasks, retrieval will not only be based on the similarity of the celestial body coordinate information, which is a single celestial body attribute, but will also fully consider the similarity of the image information and the similarity of all celestial body attributes to perform retrieval.

[0085] S104: In response to a retrieval task for celestial data to be retrieved, similarity matching is performed on the features to be retrieved obtained by encoding the features of the celestial data to be retrieved and the multimodal features of each celestial data pair, and a target multimodal feature that matches the features to be retrieved is determined, where the celestial data to be retrieved is a celestial image or at least a partial celestial catalog record.

[0086] Because multimodal features contain information about the properties of each celestial body in the celestial catalog records, as well as image information of the celestial body, the retrieval method in this manual can be applied to the retrieval of astronomical images and the retrieval of celestial body properties.

[0087] When a celestial body retrieval task is required, the celestial body data to be retrieved is obtained. The celestial body data to be retrieved is a celestial body image or at least a partial celestial body catalog record, where the partial celestial body catalog record is a data value of one or more celestial body attributes.

[0088] The server performs feature coding on the data to be retrieved in the same manner as in step S102 to obtain the features of the data to be retrieved.

[0089] Through the feature to be retrieved, similarity matching is performed among the multimodal features of each celestial body data pair obtained in advance, and the feature to be retrieved obtained by encoding the feature of the celestial body data to be retrieved is used to determine the target multimodal feature that matches the feature to be retrieved.

[0090] S106: Determine a search result based on the celestial body data pairs corresponding to the target multimodal features.

[0091] In the multimodal retrieval database, each celestial object pair contains an associated celestial object image, celestial object catalog record, and multimodal features. After determining a target multimodal feature that matches the target feature of the data to be retrieved, the corresponding celestial object image and celestial object catalog record can be associated and determined in the multimodal retrieval database.

[0092] In practical applications, at least one of the celestial object image and the celestial object catalog record associated with the target multimodal feature can be fed back as a search result. Generally speaking, when the data to be searched is an astronomical image, the search requirement is to obtain the celestial object catalog record describing the same celestial object stored in the catalog database. Therefore, in this scenario, only the celestial object catalog record associated with the target multimodal feature can be used as a search result.

[0093] based on Figure 1 The celestial data retrieval method matches the coordinate information of each celestial image in the celestial image header file with the coordinate information contained in each celestial catalog record in a star catalog database from different data sources based on coordinate similarity. This method identifies celestial data pairs consisting of matching celestial images and celestial catalog records. For each celestial data pair, the celestial image and celestial catalog record in the pair are feature-encoded to obtain a multimodal feature of the celestial data pair. By associating the celestial image in the celestial image header file with the celestial catalog record in the star catalog database, a multimodal feature representing the celestial image and celestial catalog record is constructed. When celestial data retrieval is required, retrieval results are obtained based on the multimodal feature. This method simultaneously considers the similarity of the data features in the celestial image and the celestial catalog record, achieving more accurate retrieval precision.

[0094] In the above step S100 , when calculating the coordinate difference, in order to speed up the calculation, a rough screening may be performed first, and then a precise coordinate difference calculation may be performed. The specific steps are as follows.

[0095] First, the coordinate information recorded in the celestial catalog database is used as candidate coordinate information. For each celestial image in the celestial image header file, the absolute difference in right ascension and declination between the celestial image and each candidate coordinate information is calculated. The calculation process can be expressed as follows:

[0096] For a celestial body image and a candidate coordinate information, the longitude of the celestial body image is denoted as α1, and the latitude is denoted as δ1; the longitude of the candidate coordinate information is denoted as α2, and the latitude is denoted as δ2. Then the absolute difference in right ascension between the celestial body image and the candidate coordinate information is: ΔRA = |α1-α2|, and the absolute difference in declination is: ΔDec = |δ1-δ2|.

[0097] Because 3 arc seconds corresponds to a decimal value of 0.000833°, a value slightly greater than 0.000833° can be used as the coarse screening threshold, for example, 0.001°. When either the absolute right ascension difference or the absolute declination difference exceeds this coarse screening threshold, it can be determined that the coordinate difference between the celestial object image and the candidate coordinate information is definitely greater than 3 arc seconds. This allows the elimination of celestial catalog records that clearly do not match the celestial object image.

[0098] Then, candidate coordinate information whose coordinate information difference from the celestial image is within the coarse screening threshold range is determined, and the celestial star catalog records indicated by these candidate coordinate information and the celestial image are taken as celestial data pairs that have passed the preliminary screening.

[0099] Finally, for the celestial data pairs that have passed the preliminary screening, the coordinate differences are accurately calculated based on the half-vector formula, the spherical cosine theorem and other methods to verify whether the coordinate difference between the coordinate information of the celestial object image and the candidate coordinate information is actually within the specified range of 3 arc seconds.

[0100] Taking the haversine formula as an example, the haversine formula is hav(θ)=hav(δ2-δ1)+cos(δ1)cos(δ2)hav(α2-α1). According to the haversine function Then there is

[0101]

[0102] Where hav is the calculation using the haversine formula, θ is the angular distance in radians, and α1, α2, δ1, and δ2 are also in radians.

[0103] After calculating θ according to the above formula, the unit of θ needs to be converted into arc seconds. The conversion formula is as follows:

[0104]

[0105] Among them, d arcsecIndicates the coordinate difference in arc seconds. Compare d arcsec The size between the specified range of 3 arc seconds, if d arcsec If the accuracy is within 3 arc seconds, the celestial data pair that has passed the preliminary screening is determined to be a matching celestial data pair, and it is determined that the celestial image and the celestial catalog record in the celestial data pair that has passed the preliminary screening describe the same celestial object.

[0106] In step S102, because different celestial body catalog data may contain different celestial body attributes, the data type is unified to enhance the consistency of the encoding results across different catalog databases. The celestial body attributes that require feature encoding can be pre-set to ensure that the same celestial body attributes are encoded for celestial body catalog records from celestial body databases of different sources, thereby obtaining tabular numerical features for the celestial body catalog records.

[0107] Specifically, we can use file parsing tools such as configparser to read the celestial attributes that need to be coded and the numerical types of each celestial attribute in each celestial database. Then, when encoding the characteristics of the celestial catalog records in the database, we encode the celestial attributes that need to be coded according to the corresponding data type.

[0108] In some celestial body catalog records in the star catalog database, there may be missing data values ​​of celestial body attributes, or abnormal values ​​such as noise values. The existence of abnormal values ​​will affect the accuracy of the feature coding in step S102. Therefore, this specification provides a method for processing abnormal values.

[0109] In the celestial catalog database to which the celestial catalog record of the celestial data pair belongs, the data values ​​of other celestial catalog records in the data column corresponding to the outlier are determined. Based on the data values ​​of the other celestial catalog records, a mathematical probability model is fitted to the data values ​​in the data column corresponding to the outlier. Then, based on the mathematical probability model, an estimated data value for the outlier is sampled.

[0110] Based on the estimated data value, the celestial object catalog records of the celestial object data pair are filled in. Feature encoding is performed on the celestial object image and the filled celestial object catalog records in the celestial object data pair to obtain multimodal features of the celestial object data pair.

[0111] The data values ​​of celestial body attributes in the star catalog database include discrete values ​​and continuous values.

[0112] For discrete values, a multinomial distribution is used for fitting.

[0113] Specifically, suppose the discrete variable X follows a multinomial distribution: X~Multinomial(θ), where θ=(θ1,…,θ K ) represents the probability of each discrete category.

[0114] Assume that θ follows the Dirichlet prior distribution: θ ~ Dirichlet(α), where the hyperparameter α = (α1,…,α K ) controls the sparsity of the discrete category distribution.

[0115] For missing values ​​x i , from its posterior distribution P(x i |x i-1 ) to obtain the estimated data value. i-1 Represents other observation values, that is, the data values ​​recorded in other celestial star catalogs in the data column corresponding to the abnormal value. i ) is the assumed prior distribution (Dirichlet multinomial distribution), P(x i-1 |x i ) represents the joint probability of other variables given the current sample value.

[0116] The specific steps are as follows:

[0117] The first step is initialization. For each category, count the frequency of the category in the data values ​​of other astronomical star catalog records in the data column corresponding to the outlier, and initialize the missing value x i Random category.

[0118] The second step is iterative sampling. From the Dirichlet posterior distribution θ to Dirichlet(α+n k ) Sample θ,n k is the frequency of each category. i , sample its category according to the current θ: P(x i =k|θ)=θ k Then, the new sampled x i , update frequency n k , recalculate θ.

[0119] Monitor the changes of θ during iterations, and stop the iteration when θ converges.

[0120] For continuous values, Gaussian Mixture Module (GMM) fitting is used.

[0121] Specifically, assume that the discrete variable X follows a multinomial distribution: Among them, π k Represents the weight of the k-th Gaussian component, which obeys the Dirichlet prior distribution: θ~Dirichlet(α), where the hyperparameter α=(α1,…,α K ) controls the sparsity of the discrete category distribution. μ k represents the mean of the kth Gaussian component, represents the variance of the kth Gaussian component.

[0122] For missing values ​​x i , from its posterior distribution P(x i |x i-1 ) to obtain the estimated data value. i-1 Represents other observation values, that is, the data values ​​recorded in other celestial star catalogs in the data column corresponding to the abnormal values.

[0123] The specific steps are as follows:

[0124] The first step is to fit the Gaussian mixture model based on the data values ​​recorded in other celestial catalogs of the data column corresponding to the outlier to obtain μ k , and the category distribution π k .

[0125] The second step is to calculate the missing value x. i , calculate the posterior probability of the given known part of the data, that is, each component k to x i The contribution of the posterior probability P(x i |x i-1 ) represents the kth Gaussian component for the missing value x i importance.

[0126] The third step is to select a category k based on the posterior probability, and then select the corresponding Gaussian distribution Sample fill value.

[0127] In one embodiment of the present specification, the image embedding model used for image feature encoding in step S102 may be trained using a residual network structure.

[0128] Specifically, weakly supervised learning is performed on labeled astronomical image data to train a deep model based on morphological classification. After training, the output of the feature layer before the fully connected layer of the residual network is intercepted as the output of the image feature.

[0129] The astronomical image to be encoded is input into the trained deep model, and through forward propagation, the high-dimensional feature vector of the feature layer before the fully connected layer of the residual network is extracted as the image feature of the astronomical image.

[0130] Based on this embodiment, the classification result of the deep model for the astronomical image can be further obtained as the category label of the astronomical image.

[0131] When constructing a multimodal feature database, for each celestial object pair, the category labels and image features of the astronomical images in the celestial object pair, as well as the tabular numerical features of the celestial object catalog records in the celestial object pair, are stored together.

[0132] In the above step S104, a multi-index algorithm may be used to calculate the similarity to improve the accuracy of the index result.

[0133] First, using multiple specified indexing algorithms, the vector distances between the features to be retrieved, obtained from the feature encoding of the celestial data to be retrieved, and the multimodal features of each celestial data pair are calculated. This specification does not limit the specific algorithm type or number of specified indexing algorithms; for example, both the IndexIVFPQ indexing algorithm and the IndexIVFFlat indexing algorithm may be used simultaneously.

[0134] For the celestial body data Q to be retrieved, the vector distances of the i-th multimodal feature stored in the multimodal database and calculated based on n indexing algorithms are d i (1) d i (2) ,…,d i (n) .

[0135] In order to make the calculation results of different indexing algorithms comparable, the calculated vector distances can be normalized.

[0136] In one embodiment, Min-Max normalization may be used, and the specific calculation method is as follows:

[0137]

[0138] Among them, min(d (k) ) is the minimum value of all vector distances returned by the k-th indexing method, max(d (k) ) is the maximum of all vector distances returned by the k-th indexing method.

[0139] In this manual, Min-Max normalization is calculated for each indexing method separately to ensure that their ranges are consistent and that the fusion calculation is not affected by the difference in distance scales.

[0140] Secondly, for each indexing algorithm, according to the set recall threshold, candidate celestial data pairs among the celestial data pairs are returned, and the vector distance ranking between the multimodal features of the candidate celestial data and the features to be retrieved is within the recall threshold range.

[0141] Then, the candidate celestial body data pairs returned by each indexing algorithm are integrated to obtain the candidate result pairs of the celestial body data to be retrieved. Specifically, the union or intersection of the candidate celestial body data pairs returned by each indexing algorithm can be taken as the candidate result pairs.

[0142] Finally, for each candidate result pair, the vector distances calculated based on the indexing algorithms for the candidate result pair are weighted to determine a similarity score between the candidate result pair and the celestial body data to be retrieved.

[0143] In this step, the preset weights of each index algorithm can be used for weighting, or the regression model can be used for weighting.

[0144] When regression model weighting is adopted, the regression model can be trained as follows.

[0145] First, it is necessary to construct the sample pairs required for training based on known astronomical datasets.

[0146] Within each class of celestial objects in the known astronomical dataset, data for each of two objects is extracted to construct matching pairs. Furthermore, data for two different classes of celestial objects in the known astronomical dataset is extracted to construct matching pairs. The matching pairs are assigned a class label of 1, while the matching pairs are assigned a class label of 0.

[0147] Next, for each sample pair, a pre-specified set of indexing algorithms is used to calculate the vector distances between the tabular numerical features of the celestial catalog records for the two celestial bodies in that sample pair, as obtained by each indexing algorithm. Each of these vector distances is input into the regression model to be trained to obtain a similarity score for the sample pair. The regression model is then trained based on the difference between the similarity score and the class label of the sample pair.

[0148] After the regression model is trained, for each candidate result pair, the vector distance calculated based on each indexing algorithm is input into the trained regression model to obtain a similarity score between the candidate result pair and the celestial body data to be retrieved.

[0149] Based on this embodiment, since each indexing algorithm returns an indexing result separately, for each candidate result pair, the vector distances calculated by each indexing algorithm for the candidate result pair are not necessarily all included.

[0150] For example, when using indexing algorithm 1-2, the recall threshold is set to 100. If the vector distance calculated by indexing algorithm 1 for a candidate pair ranks in the top 100 of all vector distances calculated by indexing algorithm 1, the returned index results will include the vector distance calculated by indexing algorithm 1 for the candidate pair. However, if the vector distance calculated by indexing algorithm 2 for the candidate pair does not rank in the top 100 of all vector distances calculated by indexing algorithm 1, the returned index results will not include the vector distance calculated by indexing algorithm 2 for the candidate pair.

[0151] Subsequent weighting is performed based on the vector distances obtained by each indexing algorithm. If the calculation results of a certain indexing algorithm are missing, the data format during weighting is not standardized. This will exclude candidate celestial body pairs with missing vector distances from the indexing algorithm from the returned results. This practice may lead to incorrect exclusion of manuscript similarity results. Therefore, this manual provides a method to compensate for missing vector distances.

[0152] First, obtaining the maximum value of the distance vector calculated in each candidate result pair based on the target index algorithm, wherein the target index algorithm is an index algorithm for calculating the missing value;

[0153] Then, the missing values ​​are filled by the maximum value. The vector distances calculated based on each indexing algorithm for the candidate result pair are weighted to determine a similarity score between the candidate result pair and the celestial body data to be retrieved.

[0154] Based on the above embodiments, a method such as Figure 2 The end-to-end astronomical data retrieval system shown in Figure 2 As shown, the system includes a data matching unit, a feature encoding unit and a retrieval unit.

[0155] In the matching unit, the right ascension and declination coordinate information is extracted based on the celestial image header file and the star catalog database. A preliminary screening is performed using the set initial screening threshold, and then the absolute difference between the right ascension and declination is calculated to accurately determine the celestial data pairs with a matching relationship.

[0156] In the feature coding unit, the abnormal values ​​in the celestial body deformation records are filled by mathematical model fitting, and then the celestial body images and complete celestial body catalog records in the celestial body data pair are feature coded respectively, and the coding results are stored to construct a multimodal retrieval database.

[0157] In the retrieval unit, the celestial body data to be retrieved is obtained, and the target multimodal features are retrieved in the multimodal retrieval database according to multiple indexing algorithms. The missing values ​​in the vector distance returned by the indexing algorithm are compensated, and then the similarity score is calculated and the retrieval results are returned.

[0158] The above is the celestial data retrieval method provided in this specification. Based on the same idea, this specification also provides a corresponding celestial data retrieval device, such as Figure 3 shown.

[0159] Figure 3 A schematic diagram of a celestial data retrieval device provided in this specification, specifically comprising:

[0160] A matching module 200 is configured to match the coordinate information of each celestial image in the celestial image header file with the coordinate information contained in each celestial catalog record in the star catalog database of each different data source based on coordinate similarity, and determine celestial data pairs consisting of celestial images and celestial catalog records that have a matching relationship;

[0161] An encoding module 202 is configured to perform feature encoding on the celestial body image and celestial body catalog record in each celestial body data pair to obtain multimodal features of the celestial body data pair;

[0162] A retrieval module 204 is configured to, in response to a retrieval task for celestial body data to be retrieved, perform similarity matching between features to be retrieved obtained by encoding features of the celestial body data to be retrieved and multimodal features of each celestial body data pair, and determine a target multimodal feature that matches the features to be retrieved, wherein the celestial body data to be retrieved is a celestial body image or at least a partial celestial body catalog record;

[0163] The result returning module 206 is configured to determine a search result based on the celestial body data pairs corresponding to the target multimodal features.

[0164] Optionally, the matching module 200 is specifically used to extract the coordinate information contained in each celestial body star catalog record in the star catalog database of each different data source as the coordinate information to be matched, and for each celestial body image in the celestial body image header file, screen the coordinate information to be matched whose difference from the coordinate information of the celestial body image is within a specified range in the star catalog database of each different data source, and use the celestial body star catalog record indicated by the screened coordinate information to be matched and the celestial body image as a celestial body data pair.

[0165] Optionally, the encoding module 202 is specifically used to obtain, for each celestial data pair, image features of the celestial image in the celestial data pair through a preset image feature embedding model, and for the celestial star catalog record in the celestial data pair, according to the table value type contained in the celestial star catalog record, adopt a set encoding method corresponding to the value type to obtain the table value features of the celestial star catalog record, splice the image features and the table value features to obtain the multimodal features of the celestial data pair.

[0166] Optionally, there are outliers in the celestial star catalog records in the celestial body data pair, and the encoding module 202 is specifically used to determine the data values ​​of other celestial star catalog records in the data column corresponding to the outlier in the celestial star catalog database to which the celestial star catalog records of the celestial body data pair belong, fit a mathematical probability model to which the data values ​​of the data column corresponding to the outlier obey according to the data values ​​of the other celestial star catalog records, sample estimated data values ​​of the outlier according to the mathematical probability model, fill in the celestial star catalog records of the celestial body data pair based on the estimated data values, perform feature encoding on the celestial body images and the filled celestial star catalog records in the celestial body data pair, and obtain multimodal features of the celestial body data pair.

[0167] Optionally, the retrieval module 204 is specifically configured to adopt a specified plurality of indexing algorithms to respectively calculate the vector distances between the features to be retrieved obtained by encoding the features of the celestial data to be retrieved and the multimodal features of each celestial data pair; for each indexing algorithm, based on a set recall threshold, return candidate celestial data pairs in each celestial data pair; the vector distances between the multimodal features of the candidate celestial data and the features to be retrieved are sorted within the recall threshold; the candidate celestial data pairs returned by each indexing algorithm are integrated to obtain candidate result pairs of the celestial data to be retrieved; for each candidate result pair, the candidate result pair is weighted based on the vector distances calculated by each indexing algorithm to determine a similarity score between the candidate result pair and the celestial data to be retrieved.

[0168] Optionally, the retrieval module 204 is specifically configured to extract celestial data from each class of celestial objects in the known astronomical dataset, pairwise, to construct sample pairs with matching relationships; and to extract celestial data from two different classes of celestial objects in the known astronomical dataset, respectively, to construct sample pairs without matching relationships. For each sample pair, a plurality of pre-specified indexing algorithms are used to calculate the vector distances between the tabular numerical features of the celestial catalog records of the two celestial objects in the sample pair, obtained by each indexing algorithm. The obtained vector distances are input into the regression model to be trained to obtain a similarity score for the sample pair. The regression model is trained based on the difference between the similarity score of the sample pair and the class label of the sample pair, wherein the class label of the sample pair with matching relationships is 1, and the class label of the sample pair with matching relationships is 0. For each candidate result pair, the vector distances calculated for the candidate result pair based on the respective indexing algorithms are input into the trained regression model to obtain a similarity score between the candidate result pair and the celestial data to be retrieved.

[0169] Optionally, there are missing values ​​in the vector distances of the candidate result pairs calculated based on the indexing algorithms. The retrieval module 204 is specifically used to obtain the maximum value of the distance vectors calculated in each candidate result pair based on the target indexing algorithm, and the target indexing algorithm is an indexing algorithm for calculating the missing values. The missing values ​​are filled in by the maximum value, and the vector distances of the candidate result pairs calculated based on the indexing algorithms are weighted to determine the similarity score between the candidate result pairs and the celestial body data to be retrieved.

[0170] This specification also provides a computer-readable storage medium, which stores a computer program that can be used to execute the above Figure 1 Provided astronomical data retrieval methods.

[0171] This manual also provides Figure 4 The schematic structure diagram of the electronic device shown in FIG. Figure 4 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. In other words, the execution body of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0172] Improvements to a technology can be clearly distinguished as either hardware improvements (for example, improvements to circuit structures such as diodes, transistors, and switches) or software improvements (improvements to process flows). However, with technological advancements, many process flow improvements today can now be considered direct improvements to hardware circuit structures. Designers almost always program the improved process flow into the hardware circuit to obtain the corresponding hardware circuit structure. Therefore, it cannot be said that a process flow improvement cannot be implemented using a hardware module. For example, a programmable logic device (PLD), such as a field programmable gate array (FPGA), is an integrated circuit whose logical function is determined by user programming. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to hire a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly done using "logic compiler" software. This is similar to the software compiler used when developing programs. Before compilation, the original code must also be written in a specific programming language, called a hardware description language (HDL). There is not just one HDL, but many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that by simply programming the method flow in one of these hardware description languages ​​and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.

[0173] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that in addition to implementing the controller in a purely computer-readable program code format, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules that implement the method and structures within the hardware component.

[0174] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0175] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0176] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0177] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0178] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0179] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0180] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0181] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0182] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0183] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not only include those elements, but also other elements not explicitly listed or inherent to such process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0184] Those skilled in the art will appreciate that embodiments of the present specification can be provided as methods, systems or computer program products. Therefore, the present specification can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0185] The present specification can be described in the general context of computer-executable instructions, such as program modules, executed by computers. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The present specification can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in both local and remote computer storage media, including storage devices.

[0186] The various embodiments described in this specification are described using a numbering of embodiments approach: these are each individually integrated contributions pertaining to different aspects of the description. For each embodiment, the description focuses on the differences from the other embodiments. Each embodiment is to be read in isolation, with the understanding that the same or similar features from other embodiments can be combined with the features of the respective embodiment. In particular, the description of the system embodiments is kept relatively short, as the system embodiments are largely analogous to the method embodiments.

[0187] The above description is embodied in the form of only a description of embodiments of the present specification, and is not intended to limit the present specification. Various changes and modifications can be made by those skilled in the art based on the present specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present application.

Claims

1. A celestial body data retrieval method, characterized in that: include: Based on coordinate similarity, the coordinate information of each celestial image in the celestial image header file is matched with the coordinate information contained in each celestial catalog record in the star catalog database of each different data source, and each celestial data pair consisting of a celestial image and a celestial catalog record having a matching relationship is determined; For each celestial data pair, feature encoding is performed on the celestial image and celestial catalog record in the celestial data pair to obtain multimodal features of the celestial data pair; In response to a retrieval task for celestial body data to be retrieved, performing similarity matching on features to be retrieved obtained by encoding features of the celestial body data to be retrieved and multimodal features of each celestial body data pair, and determining a target multimodal feature that matches the features to be retrieved, wherein the celestial body data to be retrieved is a celestial body image or at least a partial celestial body catalog record; A search result is determined based on the celestial body data pairs corresponding to the target multimodal features.

2. The method according to claim 1, wherein Based on coordinate similarity, the coordinate information of each celestial image in the celestial image header file is matched with the coordinate information contained in each celestial catalog record in the star catalog database of each different data source, and each celestial data pair consisting of a celestial image and a celestial catalog record with a matching relationship is determined, specifically including: Extracting coordinate information contained in the star catalog records of each celestial body from the star catalog databases of different data sources as coordinate information to be matched; For each celestial image in the celestial image header file, select coordinate information to be matched from star catalog databases of different data sources, the coordinate information of which is within a specified range of difference from the coordinate information of the celestial image; The celestial body catalog record indicated by the filtered coordinate information to be matched and the celestial body image are used as a celestial body data pair.

3. The method according to claim 1, wherein For each celestial data pair, feature encoding is performed on the celestial image and celestial catalog records in the celestial data pair to obtain the multimodal features of the celestial data pair, specifically including: For each celestial body data pair, the image features of the celestial body image in the celestial body data pair are obtained by using a preset image feature embedding model; For the celestial body catalog record in the celestial body data pair, according to the table value type contained in the celestial body catalog record, a set encoding method corresponding to the value type is used to obtain the table value feature of the celestial body catalog record; The image features and the table numerical features are spliced ​​together to obtain the multimodal features of the celestial data pair.

4. The method according to claim 1, wherein There are abnormal values ​​in the celestial catalog records of the celestial body data pair; The celestial image and celestial catalog records in the celestial data pair are feature coded to obtain the multimodal features of the celestial data pair, specifically including: Determining data values ​​of other celestial body catalog records in the data column corresponding to the abnormal value in the celestial body catalog database to which the celestial body catalog record of the celestial body data pair belongs; Fitting a mathematical probability model obeyed by each data value of the data column corresponding to the abnormal value according to the data values ​​recorded in the other celestial body catalog; sampling estimated data values ​​of the outliers according to the mathematical probability model; Filling the celestial object catalog record of the celestial object data pair based on the estimated data value; Feature encoding is performed on the celestial body image and the filled celestial body catalog record in the celestial body data pair to obtain the multimodal features of the celestial body data pair.

5. The method according to claim 1, wherein Performing similarity matching on the features to be retrieved obtained by encoding the features of the celestial body data to be retrieved and the multimodal features of each celestial body data pair, specifically including: Using a plurality of specified indexing algorithms, respectively calculating the vector distances between the features to be retrieved obtained by encoding the features of the celestial body data to be retrieved and the multimodal features of each celestial body data pair; For each indexing algorithm, according to a set recall threshold, candidate celestial body data pairs are returned from the celestial body data pairs, and the vector distance ranking between the multimodal features of the candidate celestial body data and the features to be retrieved is within the recall threshold range; Integrating the candidate celestial body data pairs returned by each indexing algorithm to obtain candidate result pairs of the celestial body data to be retrieved; For each candidate result pair, the vector distances of the candidate result pair calculated based on the indexing algorithms are weighted to determine a similarity score between the candidate result pair and the celestial body data to be retrieved.

6. The method according to claim 5, wherein Before weighting the vector distances of each candidate result pair calculated based on the indexing algorithms to determine a similarity score between the candidate result pair and the celestial body data to be retrieved, the method further includes: Within each class of celestial bodies in a known astronomical data set, celestial body data are extracted in pairs to construct sample pairs with matching relationships; within two different classes of celestial bodies in the known astronomical data set, celestial body data are extracted respectively to construct sample pairs without matching relationships; For each sample pair, a plurality of pre-specified indexing algorithms are used to calculate the vector distance between the tabular numerical features of the celestial catalog records of the two celestial bodies in the sample pair obtained by each indexing algorithm; Input the obtained vector distances into the regression model to be trained to obtain the similarity score of the sample pair; Training the regression model based on the similarity score of the sample pair and the difference between the class label of the sample pair, wherein the class label of the sample pair with a matching relationship is 1, and the class label of the sample pair with no matching relationship is 0; For each candidate result pair, weighting the vector distances calculated based on each indexing algorithm for the candidate result pair to determine a similarity score between the candidate result pair and the celestial body data to be retrieved specifically includes: For each candidate result pair, the vector distance calculated based on each indexing algorithm for the candidate result pair is input into the trained regression model to obtain a similarity score between the candidate result pair and the celestial body data to be retrieved.

7. The method according to claim 5, wherein There are missing values ​​in the vector distances of the candidate result pairs calculated based on the respective index algorithms; Weighting the vector distances calculated based on each indexing algorithm for the candidate result pair to determine a similarity score between the candidate result pair and the celestial body data to be retrieved, specifically comprising: Obtaining a maximum value of a distance vector calculated in each candidate result pair based on a target indexing algorithm, wherein the target indexing algorithm is an indexing algorithm for calculating the missing value; Fill the missing values ​​with the maximum value; The vector distances calculated based on the candidate result pair and each indexing algorithm are weighted to determine a similarity score between the candidate result pair and the celestial body data to be retrieved.

8. A celestial body data retrieval device, characterized in that: include: A matching module matches the coordinate information of each celestial image in the celestial image header file with the coordinate information contained in each celestial catalog record in the star catalog database of each different data source based on coordinate similarity, and determines each celestial data pair consisting of a celestial image and a celestial catalog record having a matching relationship; An encoding module, for each celestial data pair, performs feature encoding on the celestial image and celestial catalog record in the celestial data pair to obtain multimodal features of the celestial data pair; a retrieval module, in response to a retrieval task for celestial body data to be retrieved, performing similarity matching between features to be retrieved obtained by encoding features of the celestial body data to be retrieved and multimodal features of each celestial body data pair, and determining a target multimodal feature that matches the features to be retrieved, wherein the celestial body data to be retrieved is a celestial body image or at least a partial celestial body catalog record; The result returning module determines the search result according to the celestial body data pair corresponding to the target multimodal feature.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

Images