A system for searching for dwarf galaxies based on dg-adm

Abstract

The application discloses a dwarf galaxy searching system based on a DG-ADM, and has the following technical scheme: firstly, acquiring and preprocessing the dwarf galaxy data based on the cross DESI-cut data set to obtain a training set, and classifying and labeling the training set; secondly, constructing a dwarf galaxy searching model DG-ADM (Dwarf Galaxy Automatic Detection Model); next, constructing and preprocessing the searching data set of a large-scale sky area, and using the DG-ADM to search the dwarf galaxy of the searching data set of the large-scale sky area to obtain a candidate set; finally, screening and analyzing the candidate set, and after obtaining the candidate set, crossing the candidate set with a simbad database, screening by SExtractor, and cross analyzing the un-identified candidate with an SDSS database. The application has the following advantages: the problem that the background and the marginal field information related to the dwarf galaxy are lacked when directly using the cut image to train the model is solved; the situation that the features of the dwarf galaxies are similar and difficult to effectively distinguish, and the features of part of the dwarf galaxy samples are weak and easy to be confused with the background can be handled.

Description

Technical Field

[0001] This invention relates to a dwarf galaxy search system based on DG-ADM, which belongs to the interdisciplinary field of astronomical big data, astronomical spectroscopy, astronomy and parallel computing. Background Technology

[0002] Dwarf galaxies are a class of relatively small, low-luminosity galaxies, typically containing relatively few stars and interstellar matter. The search for dwarf galaxies is of great significance in astronomy because they provide crucial clues for studying cosmic evolution, the distribution of dark matter, and galaxy formation. By studying the stellar composition and chemical elements within dwarf galaxies, we can gain a deeper understanding of the history of cosmic formation and galaxy evolution; the motion and structure of dwarf galaxies provide important clues for the study of dark matter, and the distribution of matter within dwarf galaxies may indicate the distribution and properties of dark matter; some dwarf galaxies are considered satellite galaxies of the Milky Way, and studying these satellite galaxies can help us understand galactic interactions in the environment surrounding the Milky Way and the formation history of the Milky Way itself.

[0003] Astronomers utilize various techniques to search for and study dwarf galaxies, including optical observations, radio observations, and redshift measurements. These studies help reveal hidden structures in the universe and provide a more comprehensive understanding of the formation and evolution of galaxies and the cosmos. However, the ever-increasing volume of data from astronomical projects necessitates automated identification methods. Furthermore, the problem of lost satellite galaxies needs to be addressed: theoretically, there are many such galaxies, but current observations show a very small number. Existing identification methods rely primarily on manual physical judgment, with few solutions incorporating advanced technologies such as deep learning. Summary of the Invention

[0004] In order to overcome the shortcomings of the prior art, the technical problem to be solved by the present invention is to provide a dwarf galaxy search system based on DG-ADM.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a dwarf galaxy search system based on DG-ADM, comprising the following search steps,

[0006] S1: Acquisition, preprocessing, and annotation of dwarf galaxy data based on the cross-referenced DESI-cut dataset: The DSDGC (DESISubset Dwarf Galaxy Catalog) star catalog was processed and cross-referenced with the DESI-cut dataset, and data cleaning was performed based on prior knowledge; data that meets the requirements was downloaded and cropped to generate pseudo-color images in JPEG format; data augmentation was performed to obtain the training dataset; data was classified and manually labeled;

[0007] S2: DG-ADM (Dwarf Galaxy Automatic Detection Model) Model Construction: It consists of four parts: a three-channel dwarf galaxy image input terminal, a dwarf galaxy image heterogeneous nonlinear transformation driver, a dwarf galaxy multi-granularity feature fusion integrator, and a dwarf galaxy mission decision customizer;

[0008] S3: Construction and preprocessing of a large-scale sky search dataset based on the cross-reference of MATLAS and DESI: downloading, format conversion and segmentation of independent bricks after cross-reference of MATLAS and DESI bricks;

[0009] S4: Dwarf galaxy target search in large-scale sky regions based on DG-ADM: The DG-ADM model is used to search for candidate dwarf galaxy targets in the constructed large-scale sky region search dataset;

[0010] S5: Screening and analysis of dwarf galaxy target candidate set: After obtaining the dwarf galaxy target candidate set, it is cross-referenced with the Simbad database; the SExtractor technology is used for screening; and unverified candidates are cross-analyzed with the SDSS database.

[0011] As a preferred technical solution of the present invention, the specific steps of obtaining, preprocessing and labeling dwarf galaxy data based on the cross-DESI-cut dataset in step S1 are as follows:

[0012] S1.1: DSDGC catalog processing: Unconfirmed candidate data were removed from the DSDGC catalog, and only confirmed dwarf galaxy targets were retained; dwarf galaxies within 5 arcseconds were considered as the same dwarf galaxy and deduplication was performed; the data was cross-referenced with the DESI-cut dataset to ensure that the targets had been photographed in DESI; 1744 dwarf galaxy targets were obtained.

[0013] S1.2: Data cleaning based on prior information: Dwarf galaxy targets with sizes and brightness exceeding the standard are excluded according to quantitative standards; finally, 1500 dwarf galaxy targets that meet the requirements are obtained;

[0014] S1.3: Acquisition of Dwarf Galaxy Data: The .fits file was downloaded and extracted from DESIDR9 based on the star catalog. The extracted image was 640*640 pixels centered on the target, with a pixel scale of 0.262. A pseudo-color image in JPEG format was synthesized using the tools provided by DESI.

[0015] S1.4: Data Augmentation to Obtain the Training Dataset: Based on the obtained pseudo-color images of 1500 targets, custom data augmentation is performed, including but not limited to rotation, inversion, translation, etc.; a total of 7500 data points are obtained as the training dataset.

[0016] S1.5: Classification and Data Labeling: Label 7500 images and add information on the size of dwarf galaxies; based on prior knowledge, classify the data into two types: smaller scale and lower brightness, and larger scale and higher brightness.

[0017] As a preferred technical solution of the present invention, the multi-level nonlinear transformation driver for the dwarf galaxy image of the DG-ADM model in step S2 includes a multi-level receptive field fusion layer and a cross-gradient pool aggregation layer, and the layers are combined in a stacked manner.

[0018] As a preferred technical solution of the present invention, the multi-level receptive field fusion layer includes a 640-dimensional input, which is used to fuse the features through two overlapping normalizations and to increase the representativeness of the features through the SiLU activation function. The obtained features are then subjected to quadruple feature decomposition and recombination, and the above process is repeated once to finally obtain 160-dimensional features.

[0019] As a preferred technical solution of the present invention, the cross-gradient pooling aggregation layer includes a 160-dimensional input, which performs cross-gradient fusion on the input features and then performs pooling operation, repeats the process three times, and finally adds cross-gradient fusion again, while outputting features of 3 dimensions.

[0020] As a preferred technical solution of the present invention, the dwarf galaxy multi-granularity feature fusion integrator of the DG-ADM model in step S2 includes hierarchical feature fusion, semantic-guided downsampling, horizontal information interaction and adaptive feature convergence module.

[0021] As a preferred technical solution of the present invention, the hierarchical feature fusion includes feature extraction and fusion in a progressively upgraded manner; semantic-guided downsampling includes downsampling the feature map under the guidance of high-level semantic information; horizontal information interaction includes communication between hierarchical feature fusion and semantic-guided downsampling; and the adaptive feature aggregation module includes adaptive encoding and fusion of spatial and positional information into the features.

[0022] As a preferred technical solution of the present invention, in step S2, the dwarf galaxy mission decision customizer of the DG-ADM model performs scale transformation on the obtained features to obtain the geometric size information and class confidence information of the search samples.

[0023] As a preferred technical solution of the present invention, the specific steps for constructing and preprocessing the large-scale sky area search dataset in step S3 are as follows:

[0024] S3.1: Construct a basic large-scale sky area search dataset: Cross-reference the MATLAS and DESI bricks to obtain 1270 independent bricks, and download them;

[0025] S3.2: Data Processing: Convert the downloaded individual brick data into JPEG format files; cut the JPEG images, maintaining a certain overlap (approximately 5 pixels) between the cut images to prevent the target source from being cut at the boundary.

[0026] As a preferred technical solution of the present invention, the specific steps of screening and analyzing the candidate set in step S5 are as follows:

[0027] S5.1: Statistical results are obtained by cross-referencing the candidate dataset with the Simbad database;

[0028] S5.2: Source detection: Perform two screenings on unverified targets; use different parameters for the two screenings to obtain parameters for targets at different scales;

[0029] S5.3: Target Screening: Further screen the results of S5.2 to exclude erroneous data and stars;

[0030] S5.4: Cross-reference the two sets of data obtained in S5.3 with the SDSSS database to obtain spectral information and perform result analysis.

[0031] The advantages of this invention over the prior art are as follows:

[0032] 1. This system optimizes the problem of lacking background and neighborhood information related to dwarf galaxies when directly using cut maps for model training. It can handle situations where dwarf galaxies have similar features and are difficult to distinguish effectively, as well as situations where some dwarf galaxy samples have weak features and are easily confused with the background, which helps to improve the accuracy of data analysis.

[0033] 2. This system can be used to discover previously undetected dwarf galaxy targets, and their characteristics and distribution can help verify or adjust existing theoretical models.

[0034] 3. Dwarf galaxy targets detected by the Dwarf Galaxy Search System can help enrich astronomical knowledge, provide new insights into the evolution of the universe and galaxy formation, and contribute to a deeper understanding of the nature and distribution of dark matter. Attached Figure Description

[0035] The invention will now be further described with reference to the accompanying drawings.

[0036] Figure 1 This is a flowchart of the dwarf galaxy search system of the present invention.

[0037] Figure 2 This diagram illustrates the pseudo-color image synthesis method during the training set construction process of this invention.

[0038] Figure 3 This is a diagram of the dwarf galaxy search model of the present invention. Detailed Implementation

[0039] This invention provides a dwarf galaxy search system based on DG-ADM, comprising the following steps:

[0040] S1: Acquisition, preprocessing, and annotation of dwarf galaxy data based on the cross-referenced DESI-cut dataset: The DSDGC (DESISubset Dwarf Galaxy Catalog) star catalog was processed and cross-referenced with the DESI-cut dataset, and data cleaning was performed based on prior knowledge; data that meets the requirements was downloaded and cropped to generate pseudo-color images in JPEG format; data augmentation was performed to obtain the training dataset; data was classified and manually labeled;

[0041] S2: DG-ADM (Dwarf Galaxy Automatic Detection Model) Model Construction: It consists of four parts: a three-channel dwarf galaxy image input terminal, a dwarf galaxy image heterogeneous nonlinear transformation driver, a dwarf galaxy multi-granularity feature fusion integrator, and a dwarf galaxy mission decision customizer;

[0042] S3: Construction and preprocessing of a large-scale sky search dataset based on the cross-reference of MATLAS and DESI: downloading, format conversion and segmentation of independent bricks after cross-reference of MATLAS and DESI bricks;

[0043] S4: Dwarf galaxy target search in large-scale sky regions based on DG-ADM: The DG-ADM model is used to search for candidate dwarf galaxy targets in the constructed large-scale sky region search dataset;

[0044] S5: Screening and analysis of dwarf galaxy target candidate set: After obtaining the dwarf galaxy target candidate set, it is cross-referenced with the Simbad database; the SExtractor technology is used for screening; and unverified candidates are cross-analyzed with the SDSS database.

[0045] The specific steps of step S1, which involves the acquisition, preprocessing, and annotation of dwarf galaxy data based on the cross-DESI-cut dataset, are as follows:

[0046] S1.1: DSDGC catalog processing: Unconfirmed candidate data were removed from the DSDGC catalog, and only confirmed dwarf galaxy targets were retained; dwarf galaxies within 5 arcseconds were considered as the same dwarf galaxy and deduplication was performed; the data was cross-referenced with the DESI-cut dataset to ensure that the targets had been photographed in DESI; 1744 dwarf galaxy targets were obtained.

[0047] S1.2: Data cleaning based on prior information: Dwarf galaxy targets with sizes and brightness exceeding the standard are excluded according to quantitative standards; finally, 1500 dwarf galaxy targets that meet the requirements are obtained.

[0048] S1.3: Acquisition of Dwarf Galaxy Data: The .fits file was downloaded and extracted from DESIDR9 based on the star catalog. The extracted image was 640*640 pixels centered on the target, with a pixel scale of 0.262. A pseudo-color image in JPEG format was synthesized using the tools provided by DESI.

[0049] S1.4: Data Augmentation to Obtain the Training Dataset: Based on the obtained pseudo-color images of 1500 targets, custom data augmentation is performed, including but not limited to rotation, inversion, translation, etc.; a total of 7500 data points are obtained as the training dataset.

[0050] S1.5: Classification and Data Labeling: Label 7500 images and add information on the size of dwarf galaxies; based on prior knowledge, classify the data into two types: smaller scale and lower brightness, and larger scale and higher brightness.

[0051] The multi-level nonlinear transformation driver for the dwarf galaxy image in step S2 includes a multi-level receptive field fusion layer and a cross-gradient pooling aggregation layer, with the layers being combined in a stacked manner.

[0052] The multi-level receptive field fusion layer includes a 640-dimensional input. The features are fused by overlapping and normalizing twice and the representativeness is increased by the SiLU activation function. The obtained features are then decomposed into four features and recombined. The above process is repeated once to finally obtain 160-dimensional features.

[0053] The cross-gradient pooling aggregation layer takes a 160-dimensional input, performs cross-gradient fusion on the input features, then performs pooling operations, repeats this process three times, and finally adds cross-gradient fusion again, outputting features in 3 dimensions.

[0054] The dwarf galaxy multi-granularity feature fusion integrator of the DG-ADM model in step S2 includes hierarchical feature fusion, semantically guided downsampling, horizontal information interaction, and adaptive feature convergence module.

[0055] Hierarchical feature fusion includes feature extraction and fusion in a progressively upgraded manner; semantic-guided downsampling includes downsampling of feature maps guided by high-level semantic information; horizontal information interaction includes communication between hierarchical feature fusion and semantic-guided downsampling; and the adaptive feature convergence module includes adaptive encoding and fusion of spatial and positional information into features.

[0056] In step S2, the Dwarf Galaxy Mission Decision Customizer of the DG-ADM model performs a scale transformation on the obtained features to obtain the geometric size information and class confidence information of the search samples.

[0057] The specific steps for constructing and preprocessing the large-scale sky region search dataset in step S3 are as follows:

[0058] S3.1: Construct a basic large-scale sky area search dataset: Cross-reference the MATLAS and DESI bricks to obtain 1270 independent bricks, and download them;

[0059] S3.2: Data Processing: Convert the downloaded individual brick data into JPEG format files; cut the JPEG images, maintaining a certain overlap (approximately 5 pixels) between the cut images to prevent the target source from being cut at the boundary.

[0060] The specific steps for screening and analyzing the candidate dataset in step S5 are as follows:

[0061] S5.1: Statistical results are obtained by cross-referencing the candidate dataset with the Simbad database;

[0062] S5.2: Source detection: Perform two screenings on unverified targets; use different parameters for the two screenings to obtain parameters for targets at different scales;

[0063] S5.3: Target Screening: Further screen the results of S5.2 to exclude erroneous data and stars;

[0064] S5.4: Cross-reference the two sets of data obtained in S5.3 with the SDSSS database to obtain spectral information and perform result analysis.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dwarf galaxy search system based on DG-ADM, characterized in that: The search steps are as follows: S1: Acquisition, preprocessing and labeling of dwarf galaxy data based on the cross-DESI-cut dataset: The DSDGC star catalog is processed and then cross-referenced with the DESI-cut dataset, and the data is cleaned based on prior knowledge; Download and extract data that meets the requirements to generate pseudo-color images in JPEG format; perform data augmentation to obtain the training dataset; classify the data and manually label it; S2: DG-ADM Model Construction: It consists of four parts: a three-channel dwarf galaxy image input terminal, a dwarf galaxy image heterogeneous nonlinear transformation driver, a dwarf galaxy multi-granularity feature fusion integrator, and a dwarf galaxy mission decision customizer; The multi-level nonlinear transformation driver for dwarf galaxy images includes a multi-level receptive field fusion layer and a cross-gradient pooling aggregation layer, with the layers being combined in a stacked manner. The dwarf galaxy multi-granularity feature fusion integrator includes a hierarchical feature fusion module, a semantically guided downsampling module, a horizontal information interaction module, and an adaptive feature aggregation module. The hierarchical feature fusion module extracts and fuses features in a progressively upgraded manner; the semantically guided downsampling module downsamples the feature map under the guidance of high-level semantic information; the horizontal information interaction module communicates between the hierarchical feature fusion module and the semantically guided downsampling module; and the adaptive feature aggregation module adaptively encodes and fuses spatial and positional information into the features. The dwarf galaxy mission decision customizer is used to scale the obtained features to obtain the geometric size information and class confidence information of the search samples; S3: Construction and preprocessing of a large-scale sky search dataset based on the cross-reference of MATLAS and DESI: downloading, format conversion and segmentation of independent bricks after cross-reference of MATLAS and DESI bricks; S4: Dwarf galaxy target search in large-scale sky regions based on DG-ADM: The DG-ADM model is used to search for a set of candidate dwarf galaxy targets in the constructed large-scale sky region search dataset; S5: Screening and analysis of dwarf galaxy target candidate set: After obtaining the dwarf galaxy target candidate set, it is cross-referenced with the Simbad database; the SExtractor technology is used for screening; and unverified candidates are cross-analyzed with the SDSS database.

2. The dwarf galaxy search system based on DG-ADM according to claim 1, characterized in that: The specific steps of step S1, which involves the acquisition, preprocessing, and annotation of dwarf galaxy data based on the cross-DESI-cut dataset, are as follows: S1.1: DSDGC catalog processing: Unconfirmed candidate data were removed from the DSDGC catalog, and only confirmed dwarf galaxy targets were retained; dwarf galaxies within 5 arcseconds were considered as the same dwarf galaxy and deduplication was performed; the data was cross-referenced with the DESI-cut dataset to ensure that the targets had been photographed in DESI; 1744 dwarf galaxy targets were obtained. S1.2: Data cleaning based on prior information: Dwarf galaxy targets with sizes and brightness exceeding the standard are excluded according to quantitative standards; finally, 1500 dwarf galaxy targets that meet the requirements are obtained; S1.3: Acquisition of Dwarf Galaxy Data: The .fits file was downloaded and extracted from the star catalog of DESI DR9. The extracted image was centered on the target and was 640*640 pixels in size with a pixel scale of 0.

262. The pseudo-color image in JPEG format was synthesized using the tools provided by DESI. S1.4: Data Augmentation to Obtain the Training Dataset: Based on the obtained pseudo-color images of 1500 targets, perform custom data augmentation, including but not limited to rotation, inversion, translation, etc. A total of 7,500 images were obtained as the training dataset; S1.5: Classification and Data Labeling: Label 7500 images and add information on the size of dwarf galaxies; based on prior knowledge, classify the data into two types: smaller scale and lower brightness, and larger scale and higher brightness.

3. The dwarf galaxy search system based on DG-ADM according to claim 2, characterized in that: The multi-level receptive field fusion layer includes a 640-dimensional input. The features are fused by overlapping and normalizing twice and the representativeness is increased by the SiLU activation function. The obtained features are then decomposed into four features and recombined. The above process is repeated once to finally obtain 160-dimensional features.

4. A dwarf galaxy search system based on DG-ADM according to claim 3, characterized in that: The cross-gradient pooling aggregation layer takes a 160-dimensional input, performs cross-gradient fusion on the input features, then performs pooling operations, repeats this process three times, and finally adds cross-gradient fusion again, outputting features in 3 dimensions.

5. A dwarf galaxy search system based on DG-ADM according to claim 1, characterized in that: The specific steps for constructing and preprocessing the large-scale sky region search dataset in step S3 are as follows: S3.1: Construct a basic large-scale sky area search dataset: Cross-reference the MATLAS and DESI bricks to obtain 1270 independent bricks, and download them; S3.2: Data Processing: Convert the downloaded standalone brick data into JPEG format files; cut the JPEG images, maintaining a certain overlap (approximately 5 pixels) between the cut images to prevent the target source from being cut at the boundary.

6. A dwarf galaxy search system based on DG-ADM according to claim 1, characterized in that: The specific steps for screening and analyzing the candidate dataset in step S5 are as follows: S5.1: Statistical results are obtained by cross-referencing the candidate dataset with the Simbad database; S5.2: Source detection: Perform two screenings on unverified targets; use different parameters for the two screenings to obtain parameters for targets at different scales; S5.3: Target Screening: Further screen the results of S5.2 to exclude erroneous data and stars; S5.4: Cross-reference the two sets of data obtained in S5.3 with the SDSSS database to obtain spectral information and perform result analysis.

Images