Intelligent astronomical image quality evaluation method, system and device based on convolutional neural network

Abstract

The invention discloses an astronomical image quality intelligent evaluation method, system and device based on a convolutional neural network. The method comprises the following steps: S1, constructing a data set based on astronomical image quality evaluation; s2, performing feature extraction and data processing based on the images in the data set; s3, establishing a convolutional neural network model, and establishing an AQSA-Net model; step S4, the data set in the step S1 is sent to an AQSA-Net model for training, and an optimal AQSA-Net network weight model is generated; and S5, applying the AQSA-Net model obtained by training to an automatic evaluation and classification task of astronomical image quality. The image feature extraction and data processing method has universality and expandability, and the grid division density and the image cutting size of the image can be flexibly adjusted according to the specific configuration and observation requirements of the telescope; the constructed AQSA-Net model is a highly lightweight model, can significantly reduce memory usage, accelerates model training and reasoning, and is suitable for an environment with low computing resource requirements in deep space exploration.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and astronomical image processing, and in particular to a method, system and device for intelligently evaluating astronomical image quality based on a convolutional neural network. Background Art

[0002] With the rapid development of astronomical observation technology, from ground-based telescopes to space probes, from visible light to infrared, radio and other multi-band observations, this process has not only greatly broadened the human vision of exploring the universe, but also caused the amount of astronomical data to grow at an explosive rate, resulting in a doubling of the complexity of the data. Faced with complex, multi-dimensional and massive astronomical data, traditional manual evaluation methods can no longer meet the dual requirements of timeliness and accuracy. There is an urgent need for an efficient and automated data processing mechanism to quickly screen out high-quality data and provide strong support for scientific research.

[0003] In addition, the observation data of ground-based telescopes are often affected by many factors, such as meteorological conditions, detector noise, atmospheric turbulence, sky background, etc., so it is particularly important to accurately evaluate the data quality. This evaluation process is an important step to ensure the credibility and effective use of data. High-quality data not only provides a solid foundation for effective analysis and scientific research, but also helps to improve the accuracy and scientific value of research results. Low-quality data will not only affect the accuracy of research results, but may also reduce the credibility and value of scientific results. In space telescope observations, data quality assessment is even more important. Since the bandwidth and resources for data transmission are very valuable in the space environment, transmitting low-quality data is a great waste of valuable resources. Therefore, conducting effective data quality assessment and screening out high-quality data for transmission is the key to improving resource utilization efficiency. However, due to the very limited computing resources for deep space exploration, existing methods often cannot meet computing needs, making it difficult to ensure the accuracy of data quality assessment.

[0004] Therefore, it is urgent to provide an intelligent assessment method for astronomical image quality based on convolutional neural networks. Summary of the invention

[0005] In order to solve the above problems, the technical solution of the present invention provides a method, system and device for intelligent assessment of astronomical image quality based on convolutional neural network, which can solve the problem.

[0006] According to a first aspect of the technical solution of the present invention, there is provided an intelligent astronomical image quality assessment method based on a convolutional neural network, comprising:

[0007] Step S1, constructing a data set based on astronomical image quality assessment;

[0008] Step S2, performing feature extraction and data processing based on the images in the data set;

[0009] Step S3: Build a convolutional neural network model and establish an AQSA-Net model;

[0010] Step S4, sending the data set in step S1 to the AQSA-Net model for training to generate an optimal AQSA-Net network weight model;

[0011] Step S5: Apply the trained AQSA-Net model to the automatic assessment and classification task of astronomical image quality.

[0012] In the above scheme, step S1 includes:

[0013] S1.1. Acquire astronomical images;

[0014] S1.2, performing quality assessment on the astronomical images, classifying the images into categories according to the quality assessment results, storing the images according to the categories and generating data labels;

[0015] S1.3. According to the different categories of the images, at least some categories of images are divided into different categories, the number of samples is increased by image enhancement, and data labels are attached to obtain labeled images to form a data set, and the processed data set is divided into a training set and a test set.

[0016] In the above scheme, step S2 includes:

[0017] S2.1, obtaining local information of the image and constructing a first input channel of the model input image;

[0018] S2.2, obtaining global information of the image and constructing a second input channel of the model input image;

[0019] S2.3, stitching the images by channel to form a dual-channel image;

[0020] S2.4, performing data processing on the dual-channel image, eliminating the scale difference between the two channels, and normalizing the input data, thereby improving the training effect and data consistency of the model.

[0021] In the above scheme, step S3 includes:

[0022] S3.1. Configure the convolutional neural network operating environment;

[0023] S3.2. Based on the optimization of the ShuffleNetV2 model, the AQSA-Net model is established.

[0024] In the above scheme, step S4 includes:

[0025] S4.1. Initializing the training parameters of the AQSA-Net model;

[0026] S4.2, training is performed on the training set in the data set obtained in step S1.3, and after each round of training, verification is performed on the test set, and the weight of the model during each training is saved;

[0027] S4.3. After the training is completed, the accuracy, precision, recall and F1 score of each round on the test set are compared to obtain the optimal AQSA-Net model weights for evaluating the quality of astronomical images with the highest generalization performance.

[0028] In the above scheme, step S5 includes:

[0029] S5.1, performing feature extraction and data processing on the astronomical image to be evaluated in step S2;

[0030] S5.2, send the processed image to the model and load the optimal AQSA-Net model weights;

[0031] S5.3. Predict the category of the image. Use the Softmax function in the output layer of the model to convert the output into a probability distribution, obtain the confidence of each category, traverse all category probabilities output by Softmax, and select the category with the highest probability as the prediction result.

[0032] In the above solution, obtaining the local information of the image in step S2.1 includes:

[0033] Read the image, focus directly on the core area of ​​the image, and center-crop it to a 256×256 pixel area as the first channel of the model input image.

[0034] In the above solution, obtaining the global information of the image in step S2.2 includes:

[0035] S2.21, performing source detection on the entire image to obtain the centroid coordinates of each source;

[0036] S2.22, divide the entire image into multiple grids, traverse each grid, check whether it contains a source, and for grids that do not contain a source, randomly crop an image block;

[0037] S2.23, randomly select a source in the grid containing the source, take the centroid of the selected source as the cropping center, and determine a cropping area;

[0038] S2.24. All the cropped image blocks are stitched together according to their original grid positions to obtain an image with the same size as the center cropped image. The stitched image is used as the second channel of the model input.

[0039] According to a second aspect of the technical solution of the present invention, there is provided an intelligent astronomical image quality assessment system based on a convolutional neural network, the system being used to implement the astronomical image quality intelligent assessment method of the convolutional neural network described in any one of the above schemes, the system comprising:

[0040] Dataset construction module, which constructs a dataset based on astronomical image quality assessment;

[0041] A feature extraction and data processing module, which performs feature extraction and data processing based on the images in the data set and combines local information and global information of the images;

[0042] Model building module, build convolutional neural network model, and establish AQSA-Net model;

[0043] The training module sends the data set to the AQSA-Net model for training to generate the optimal AQSA-Net network weight model;

[0044] The evaluation module applies the trained AQSA-Net model to the automatic assessment and classification task of astronomical image quality.

[0045] According to a third aspect of the technical solution of the present invention, there is provided an electronic device, the electronic device comprising:

[0046] A memory storing executable instructions;

[0047] A processor, wherein the processor runs the executable instructions in the memory to implement the method described in any one of the above schemes.

[0048] Beneficial effects of the present invention:

[0049] The present invention discloses an intelligent astronomical image quality assessment method, system and device based on convolutional neural network, which have universality and scalability for image feature extraction and data processing methods, and can flexibly adjust the image grid division density and image cropping size according to the specific configuration of the telescope and observation requirements; the local and global information of the image are combined to effectively reduce the dimension of the high-resolution image and realize efficient data input; the constructed AQSA-Net model is a highly lightweight model, which can significantly reduce memory usage, accelerate model training and reasoning, and greatly accelerate the real-time processing and rapid response capabilities of the model, with high classification accuracy, and is suitable for environments with low computing resource requirements in deep space exploration. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0051] Figure 1 A flow chart of the intelligent astronomical image quality assessment method based on convolutional neural network provided by the present invention.

[0052] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0053] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0054] The terms "first", "second", etc. in the specification and claims of the present disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein, for example.

[0055] In addition, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or elements is not necessarily limited to those steps or elements explicitly listed, but may include other steps or elements not explicitly listed or inherent to such process, method, product, or apparatus.

[0056] Multiple includes two or more.

[0057] It should be understood that the term "and / or" used in this disclosure is only a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone.

[0058] like Figure 1As shown, an embodiment of the technical solution of the present invention provides an intelligent astronomical image quality assessment method based on a convolutional neural network, comprising:

[0059] Step S1, constructing a data set based on astronomical image quality assessment;

[0060] Step S2: performing feature extraction and data processing based on the images in the data set;

[0061] Step S3: Build a convolutional neural network model and establish an AQSA-Net model;

[0062] Step S4, sending the data set in step S1 to the AQSA-Net model for training to generate an optimal AQSA-Net network weight model;

[0063] Step S5: Apply the trained AQSA-Net model to the automatic assessment and classification task of astronomical image quality.

[0064] Step S1 specifically includes the following steps:

[0065] S1.1. Acquire astronomical images;

[0066] It includes information such as observation target, exposure time, pixel scale, effective field of view, etc. The image data format is .fits or .fit.

[0067] S1.2. Perform quality assessment on astronomical images, classify the images according to the quality assessment results, store them according to the categories and generate data labels;

[0068] Among them, the quality assessment of astronomical images includes comprehensive consideration of the quality of the entire image and star imaging. Combined with statistical laws and visual inspection, the images are subdivided into eight categories in the target tracking mode, namely: image categories with point images of stars, image categories with elongated images of stars, image categories with unstable telescope tracking, image categories with image rotation problems, image categories with source defocus or poor seeing, image categories with detector problems, image categories with no sources or very few sources in the image, and image categories with abnormal shutter closure. When storing data, a separate folder is created for each category to store image data, and the data format is .fits or .fit.

[0069] S1.3. According to the categories of images, for the image categories with fewer numbers, the number of samples is increased by image enhancement, and data labels are attached to obtain labeled images to form a data set, and the processed data set is divided into a training set and a test set.

[0070] Specifically, it is necessary to process the classified image data and expand the data samples. For example, for the three categories of images, namely, elongated images of stars, unstable telescope tracking, and image rotation problems, image enhancement is performed by flipping, rotating, adding Gaussian noise and Poisson noise, and data labels are attached to obtain labeled images; for the four categories of images, namely, source defocus or poor seeing, detector problems, no source in the image or very few sources, and abnormal shutter closure, image expansion is performed only by rotation and flipping, and the respective data labels are maintained to obtain labeled images. After that, the expanded images are merged and sorted with the original images by category to obtain the overall data set, and the data set is divided into a training set and a test set for training the convolutional neural network model and verifying the model accuracy. The data set is randomly divided, but when dividing, the training set and the test set must have the same distribution to ensure that the number of images in each category is distributed in the same proportion. Preferably, the division ratio of the training set and the test set is 8:2.

[0071] Step S2 specifically includes the following steps:

[0072] S2.1, obtaining local information of the image and constructing the first input channel of the model input image;

[0073] Specifically, the image is read, the core area of ​​the image is directly focused on, and the center is cropped into a 256×256 pixel area as the first channel of the model input image.

[0074] S2.2, obtaining global information of the image and constructing a second input channel of the model input image;

[0075] Step 2.2 specifically includes the following steps:

[0076] S2.21, performing source detection on the entire image to obtain the centroid coordinates of each source;

[0077] S2.22, divide the entire image into grids, traverse each grid, check whether it contains a source, and for grids that do not contain a source, randomly crop an image block;

[0078] In this embodiment, the image is preferably divided into 8×8 grid areas to facilitate subsequent processing and analysis. For a grid without a source, a 32×32 image block is randomly cropped.

[0079] S2.23, randomly select a source in the grid containing the source, take the centroid of the selected source as the cropping center, and determine a cropping area;

[0080] The determined cropping area is a 32×32 area.

[0081] Furthermore, it is necessary to determine whether the cropping boundary exceeds the image boundary. If not, crop directly; if so, mark the source as a "boundary source" and remove it from the source list of the current grid. Continue to check whether there are still sources in the grid. If yes, repeat step S2.23. If no, randomly select a source from the marked "boundary sources" and crop a 32×32 area containing the source. The cropping area must adapt to the boundary conditions, so the cropping center is not the centroid of the source.

[0082] S2.24. All the cropped 32×32 image blocks are concatenated into a 256×256 image according to their original grid positions, and the concatenated image is used as the second channel of the model input.

[0083] S2.3, stitching the images by channel to form a dual-channel image;

[0084] Specifically, the center-cropped 256×256 image and the stitched 256×256 image are stitched by channel to form a dual-channel image.

[0085] S2.4. Perform data processing on the dual-channel image to eliminate the scale difference between the two channels and normalize the input data, thereby improving the training effect and data consistency of the model.

[0086] The data processing of the dual-channel image specifically includes the following steps:

[0087] S2.41, maximum and minimum normalization: adjust the grayscale value range of the two-channel images to the range of 0-1 to ensure that all images are at the same scale before being input into the model;

[0088]

[0089] Among them, f1 is the normalized pixel value, f0 is the original value of each pixel in the image, and f max and f min are the maximum and minimum values ​​of the pixels in the data set, respectively. In order to simplify the processing flow and adapt to the characteristics of 16-bit astronomical images, f is set max is 65535, set f min is 0.

[0090] S2.42, logarithmically stretching the normalized pixel values;

[0091] f2=c·log v+1 (1+v·f1) (2)

[0092] Among them, f2 is the pixel value after logarithmic stretching, the value of the constant is set to 1, and the value of the base adjustment factor is set to 100.

[0093] S2.43. Convert image data into tensor format.

[0094] S2.44. Perform data augmentation on the data in the training set by randomly flipping the image vertically with a probability of 0.5, randomly flipping the image horizontally with a probability of 0.5, and randomly rotating the image between 0° and 90°.

[0095] S2.45. Perform Z-Score normalization on the image by channel, and perform this operation independently on each channel.

[0096] Since the image feature extraction and data processing include local information and global information of the image, and the local information and global information are combined, the dimension of the high-resolution image is effectively reduced, and efficient data input is achieved.

[0097] Step S3 includes:

[0098] S3.1. Configure the convolutional neural network operating environment;

[0099] S3.2. Based on the optimization of the ShuffleNetV2 model, the AQSA-Net model is established.

[0100] Specifically, the optimization of the ShuffleNetV2 model includes: optimizing the number of network layers and channels; introducing the spatial attention unit (SAU) into the Block structure of the AQSA-Net model, so that the AQSA-Net model can pay more attention to important areas in the image and enhance the model's ability to extract key information.

[0101] The establishment of the backbone structure of the AQSA-Net model includes: using an initial convolutional layer with a 7×7 convolution kernel, which outputs a feature map of size 128×128×24; then, using a maximum pooling layer of size 3×3 and stride 2 to reduce the spatial dimension of the feature map while retaining important features. Subsequently, the AQSA-Net model enters the core part consisting of three main stages, namely Stage 2, Stage 3, and Stage 4. Each stage integrates multiple blocks. Specifically, Stage 2 contains two blocks, Block 1 and Block 2. Stage 3 contains a combination of 1 Block 1 and 3 Block 2. Stage 4 is simplified in design and contains 1 Block 1 and 1 Block 2. After that, the model realizes cross-channel information fusion through a 1×1 convolution. Since the 1×1 convolution linearly combines all channel features of the input, it can realize cross-channel information fusion. It can also be used to change the number of channels in the feature map. Here, it is used to perform dimensionality increase to increase the diversity of feature expression. The number of channels in the feature map becomes 512. Subsequently, the global average pooling layer is used to further compress the feature space and extract the most representative global features. Finally, these global features are passed to a fully connected layer containing 8 neurons, which acts as a classifier and outputs the prediction results.

[0102] Furthermore, in order to enhance the information exchange in the spatial dimension, a new SAU is proposed with reference to the spatial attention module in CBAM, and is introduced into Block 2, after the 3×3 depth convolution of the Block 2 structure. In SAU, maximum pooling and average pooling operations are performed along the channel axis. At the same time, 1×1 convolution is used to compress the channel dimension information into a single channel. Then, the feature map after channel pooling and the feature map after 1×1 convolution are spliced ​​by channel to obtain a three-channel feature map. Then, a 5×5 convolution is used to capture a larger range of spatial information, realize information fusion, and generate a new feature map. Finally, the Sigmoid activation function is used to normalize the output to the [0, 1] interval to obtain the spatial attention weight map, and the generated spatial weight map is multiplied element by element with the original input feature map to achieve feature weighting and enhance the model's attention to important areas.

[0103] Step S4 specifically includes the following steps:

[0104] S4.1. Initializing the training parameters of the AQSA-Net model;

[0105] Among them, the parameters are set as follows: the loss function uses the cross entropy loss function, the optimizer uses the stochastic gradient descent algorithm with momentum and weight decay, the momentum is 0.9, the weight decay is 4E-5, the learning rate scheduler uses cosine annealing learning rate decay, the number of training iterations is 150 epochs (rounds), the batch size is 128, and the initial learning rate is 0.1.

[0106] S4.2, training is performed on the training set in the data set obtained in step S1.3. After each epoch (round) of training, verification is performed on the test set, and the weight of the model during each training is saved;

[0107] S4.3. After training, the accuracy, precision, recall and F1 score of each round on the test set are compared to obtain the optimal AQSA-Net model weights for evaluating astronomical image quality with the highest generalization performance.

[0108] Step S5 specifically includes the following steps:

[0109] S5.1, performing feature extraction and data processing on the astronomical image to be evaluated in step S2;

[0110] S5.2, send the processed image to the model and load the optimal AQSA-Net model weights;

[0111] S5.3. Predict the category of the image. Use the Softmax function in the output layer of the model to convert the output into a probability distribution, obtain the confidence of each category, traverse all category probabilities output by Softmax, and select the category with the highest probability as the prediction result.

[0112] According to a second aspect of the technical solution of the present invention, there is provided an intelligent astronomical image quality assessment system based on a convolutional neural network, the system being used for the above-mentioned intelligent astronomical image quality assessment method based on a convolutional neural network, the system comprising:

[0113] Dataset construction module, which constructs a dataset based on astronomical image quality assessment;

[0114] The feature extraction and data processing module performs feature extraction and data processing based on the images in the data set and combines the local information and global information of the images;

[0115] Model building module, build convolutional neural network model, and establish AQSA-Net model;

[0116] The training module sends the data set to the AQSA-Net model for training to generate the optimal AQSA-Net network weight model;

[0117] The evaluation module applies the trained AQSA-Net model to the automatic assessment and classification task of astronomical image quality.

[0118] According to a third aspect of the technical solution of the present invention, there is provided an electronic device, the electronic device comprising:

[0119] A memory storing executable instructions;

[0120] The processor runs the executable instructions in the memory to implement the above method.

[0121] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0122] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0123] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0124] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. An intelligent astronomical image quality assessment method based on convolutional neural network, characterized in that: include: Step S1, constructing a data set based on astronomical image quality assessment; Step S2, performing feature extraction and data processing based on the images in the data set; Step S3: Build a convolutional neural network model and establish an AQSA-Net model; Step S4, sending the data set in step S1 to the AQSA-Net model for training to generate an optimal AQSA-Net network weight model; Step S5: Apply the trained AQSA-Net model to the automatic assessment and classification task of astronomical image quality.

2. The method for intelligent assessment of astronomical image quality based on convolutional neural network according to claim 1, characterized in that: Step S1 includes: S1.

1. Acquire astronomical images; S1.2, performing quality assessment on the astronomical images, classifying the images into categories according to the quality assessment results, storing the images according to the categories and generating data labels; S1.

3. According to the different categories of the images, at least some categories of images are divided into different categories, the number of samples is increased by image enhancement, and data labels are attached to obtain labeled images to form a data set, and the processed data set is divided into a training set and a test set.

3. The method for intelligent assessment of astronomical image quality based on convolutional neural network according to claim 1, characterized in that: Step S2 includes: S2.1, obtaining local information of the image and constructing a first input channel of the model input image; S2.2, obtaining global information of the image and constructing a second input channel of the model input image; S2.3, stitching the images by channel to form a dual-channel image; S2.4, performing data processing on the dual-channel image, eliminating the scale difference between the two channels, and normalizing the input data, thereby improving the training effect and data consistency of the model.

4. The method for intelligently evaluating astronomical image quality based on convolutional neural network according to claim 3, characterized in that: Step S3 includes: S3.

1. Configure the convolutional neural network operating environment; S3.

2. Based on the optimization of the ShuffleNetV2 model, the AQSA-Net model is established.

5. The method for intelligent assessment of astronomical image quality based on convolutional neural network according to claim 4, characterized in that: Step S4 includes: S4.

1. Initializing the training parameters of the AQSA-Net model; S4.2, training is performed on the training set in the data set obtained in step S1.3, and after each round of training, verification is performed on the test set, and the weight of the model during each training is saved; S4.

3. After the training is completed, the accuracy, precision, recall and F1 score of each round on the test set are compared to obtain the optimal AQSA-Net model weights for evaluating the quality of astronomical images with the highest generalization performance.

6. The method for intelligent assessment of astronomical image quality based on convolutional neural network according to claim 5, characterized in that: Step S5 includes: S5.1, performing feature extraction and data processing on the astronomical image to be evaluated in step S2; S5.2, send the processed image to the model and load the optimal AQSA-Net model weights; S5.

3. Predict the category of the image. Use the Softmax function in the output layer of the model to convert the output into a probability distribution, obtain the confidence of each category, traverse all category probabilities output by Softmax, and select the category with the highest probability as the prediction result.

7. The method for intelligent astronomical image quality assessment based on convolutional neural network according to claim 3, characterized in that: Acquiring the local information of the image in step S2.1 includes: Read the image, focus directly on the core area of ​​the image, and center-crop it to a 256×256 pixel area as the first channel of the model input image.

8. The method for intelligently evaluating astronomical image quality based on convolutional neural network according to claim 3, characterized in that: Acquiring the global information of the image in step S2.2 includes: S2.21, performing source detection on the entire image to obtain the centroid coordinates of each source; S2.22, divide the entire image into multiple grids, traverse each grid, check whether it contains a source, and for grids that do not contain a source, randomly crop an image block; S2.23, randomly select a source in the grid containing the source, take the centroid of the selected source as the cropping center, and determine a cropping area; S2.

24. All the cropped image blocks are stitched together according to their original grid positions to obtain an image with the same size as the center cropped image. The stitched image is used as the second channel of the model input.

9. An intelligent astronomical image quality assessment system based on convolutional neural network, characterized in that: The system is used to implement the astronomical image quality intelligent assessment method of the convolutional neural network according to any one of claims 1 to 8, and the system comprises: Dataset construction module, which constructs a dataset based on astronomical image quality assessment; A feature extraction and data processing module, which performs feature extraction and data processing based on the images in the data set and combines local information and global information of the images; Model building module, build convolutional neural network model, and establish AQSA-Net model; The training module sends the data set to the AQSA-Net model for training to generate the optimal AQSA-Net network weight model; The evaluation module applies the trained AQSA-Net model to the automatic assessment and classification task of astronomical image quality.

10. An electronic device, characterized in that: The electronic device comprises: A memory storing executable instructions; A processor, wherein the processor runs the executable instructions in the memory to implement the method according to any one of claims 1 to 8.

Images