Image analysis method based on deep learning and application thereof

Abstract

The invention provides an image analysis method based on deep learning and an application thereof, relates to the field of solar activity image analysis, and aims to improve the monitoring and prediction capability of solar burst activities, more accurately extract the picture feature information of the solar burst activities through the image analysis method based on deep learning, and improve the accuracy of the solar burst activities. And therefore, prediction is carried out. The risk influence of the space environment on the aircraft is effectively reduced, the influence range of the solar burst activity in the space environment is judged by combining the association state analysis module, then the safe flight path of the target aircraft is planned, the aircraft can be prevented from being directly influenced by the solar burst activity, and the safety of the aircraft is improved. The safety navigation operation probability of the aircraft is improved, and the space weather is monitored in real time by using the result of the association state analysis module. When solar burst activities affect the space environment, early warning information is issued in time, and guarantee is provided for spacecraft operation and satellite communication.

Description

Technical Field

[0001] The present invention relates to the field of solar activity image analysis, and in particular to an image analysis method based on deep learning and its application. Background Art

[0002] As human beings continue to explore the space, space activities are becoming more frequent. Solar flares are a common natural phenomenon in space, which release a large amount of energy and matter, and have a significant impact on the space environment and aircraft. In existing technologies, there are certain limitations in monitoring and predicting solar flares and evaluating their impacts.

[0003] A solar burst activity prediction and analysis platform based on machine learning methods is proposed in the Chinese invention patent application number: CN202410995643.8. It includes a supervision prediction module, a correlation state analysis module, and an impact risk determination module. The time, location, and intensity of solar burst activities are predicted in advance through a machine learning model to clarify the time, location, and intensity of solar burst activities. Combined with the correlation state analysis module, the impact range of solar burst activities in the space environment and the impact range in the earth environment are judged, and then the target aircraft affected by the radiation in the affected range is obtained, and the flight path planning of the aircraft is formed. When a sudden solar burst activity is encountered in space, the aircraft in the space environment is avoided to achieve safe navigation.

[0004] However, the existing solar burst activity prediction and analysis process lacks the problem of extracting image feature information from images obtained from solar burst activities for analysis and prediction. There is still room for improvement. Traditional image analysis cannot accurately capture the detailed characteristics of solar burst activities, resulting in inaccurate predictions of the time, intensity and type of occurrence, making it difficult to make adequate preparations in advance.

[0005] Existing technologies do not fully consider the combined effects of multiple factors, such as radiation intensity and magnetic field changes, when analyzing the impact of solar flare activities on the space environment through images, and thus cannot accurately determine the scope and extent of the impact.

[0006] The current image analysis is insufficient in analyzing and predicting the relationship between solar bursts and the space environment. Due to the lack of understanding and prediction of solar bursts, space equipment operators lack scientific basis when making decisions and find it difficult to take effective protective measures to ensure the safe operation of space equipment. Therefore, the present invention makes improvements to this and proposes an image analysis method based on deep learning and its application. Summary of the invention

[0007] The purpose of the present invention is to address the problems raised by the existing background technology. In order to achieve the above invention purpose, the present invention provides the following technical solutions: an image analysis method based on deep learning, comprising the following steps: step 1, data collection and preprocessing: collecting a large amount of image data of solar burst activities, the image data including different solar burst types, intensities and times; Step 2: Build a deep learning model: Use a deep learning convolutional neural network (CNN) model for image analysis. The architecture of the deep learning convolutional neural network (CNN) model includes a convolutional layer, a pooling layer, and a fully connected layer. The hyperparameters of the deep learning convolutional neural network model include the number of iterations and the batch size. The model performance is optimized through experiments and adjustments. Step 3, model training: divide the preprocessed image data into training set, validation set and test set; use the training set to train the model, and continuously adjust the model parameters through the back propagation algorithm to minimize the loss function; during the training process, use the validation set to monitor the performance of the model, adjust the hyperparameters in time, and prevent overfitting; Step 4: Image feature extraction: Use the trained model to extract features from the image of solar burst activity to obtain high-level feature representation of the image; the high-level features of the image include the shape, intensity, and location information of the solar burst; Step 5: Solar burst activity image prediction: Based on the extracted image features, use machine learning models including recurrent neural network RNN ​​or long short-term memory network LSTM to predict solar burst activity; input solar burst activity image feature sequence is in Represents the image feature vector at the Tth time point, where T is the length of the time series. The model will output the prediction results of future solar eruption activities; Step 6, correlation state analysis module: Combine the predicted results of solar burst activities and the space environment to build a correlation state analysis module; the correlation state analysis module analyzes the impact of solar burst activities on the space environment, including radiation intensity and magnetic field changes; and determines the scope of the impact of solar burst activities in the space environment through simulation.

[0008] As a preferred technical solution of the present invention, step 1 preprocesses the image, including image enhancement, denoising, cropping and normalization operations, and the enhanced image is represented as: ;in: is the original pixel value, The pixel value after enhancement; and are the minimum and maximum pixel values ​​in the image, respectively. is the maximum number of gray levels of the output image.

[0009] As a preferred technical solution of the present invention, step 2, in the convolutional neural network, the deep learning convolutional neural network CNN for image analysis is used to extract the features of the image, and the expression of the convolution operation of the deep learning convolutional neural network is: in: is the output value after convolution, located at position ; is the input image at position The pixel value of is the weight of the convolution kernel at the position; m and n are the height and width of the convolution kernel respectively.

[0010] As a preferred technical solution of the present invention, step 3, in model training: the preprocessed image data is divided into a training set, a validation set and a test set; the model is trained using the training set, and the parameters of the model are continuously adjusted through the back propagation algorithm to minimize the average loss function, and the average loss function is expressed as: in: represents the average loss function; is the real image data, is the predicted image data, N is the number of image data samples, is the pixel coordinate of the image data.

[0011] As a preferred technical solution of the present invention, it also includes safe flight path planning: designing a safe flight path planning algorithm for the affected target aircraft; the safe flight path planning algorithm takes into account the predicted results of solar flare activities, the impact range of the space environment, and the performance and constraints of the target aircraft; and through the optimization algorithm, generating a safe flight path that can avoid the impact of solar flare activities.

[0012] As a preferred technical solution of the present invention, the specific steps of step 5, solar burst activity image prediction, are as follows: Step 51, data collection and preprocessing: collect historical solar flare activity image data; perform feature extraction on the image to extract the number of sunspots, flare area, and magnetic field intensity features; perform data cleaning and preprocessing on the extracted features to handle missing values ​​and outliers; Step 52: Build a machine learning model: Use a recurrent neural network (RNN) as a prediction model and determine the structure of the model, including the number of hidden layer neurons and layer parameters; Step 53, model training: Use the training set data to train the model; during the training process, adjust the model parameters according to the value of the loss function, and use the test set data to evaluate the trained model.

[0013] As a preferred technical solution of the present invention, step 6, the association state analysis module: through the image data pixels Combined with the analysis and prediction of solar burst activities, the correlation state analysis and prediction function expression is constructed as follows: ;in: Association state analysis prediction function; Explosive intensity level: expressed as a value from 1 to 5. Magnetic field change; the intensity of solar eruptive activity; Radiation intensity, is the pixel coordinate of the image data.

[0014] As a preferred technical solution of the present invention, it also includes space weather monitoring and early warning: using the results of the correlation state analysis module to monitor space weather in real time. When solar flare activities affect the space environment, early warning information is issued in a timely manner to provide protection for spacecraft operation and satellite communications.

[0015] As a preferred technical solution of the present invention, the image analysis method based on deep learning is applied in the analysis and prediction of the relationship between solar burst activities and space environment.

[0016] As a preferred technical solution of the present invention, the image analysis method based on deep learning is applied in space weather monitoring and early warning.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: In the scheme of the present invention: the present invention improves the monitoring and prediction capabilities of solar flare activities, and through the deep learning image analysis method, it is able to more accurately extract the image feature information of solar flare activities, thereby predicting them.

[0018] The present invention effectively reduces the risk impact of the space environment on aircraft. By combining the associated state analysis module to determine the impact range of solar burst activities in the space environment, and then planning the safe flight path of the target aircraft, the aircraft can be prevented from being directly affected by solar burst activities, thereby improving the probability of safe navigation operation of the aircraft.

[0019] The present invention provides auxiliary decision-making for the safe operation of space equipment. By analyzing and predicting solar burst activities and evaluating the impact on aircraft, it can provide a scientific basis for space equipment operators and managers, helping them make reasonable decisions and ensure the safe operation of space equipment.

[0020] The present invention enhances the safety of space exploration and utilization. By studying solar flare activities and implementing countermeasures, it can reduce uncertainties in the space environment, provide more reliable protection for human activities in space, and promote the sustainable development of space exploration and utilization.

[0021] The present invention enhances the understanding and cognition of the space environment. By studying the relationship between solar flare activities and the space environment, the present invention can deepen the understanding of the space environment and lay the foundation for further exploring the mysteries of the universe and developing and utilizing space resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The present invention provides a flow chart of the method. DETAILED DESCRIPTION

[0023] To make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be described clearly and completely in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them.

[0024] Therefore, the following detailed description of the embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely represents some embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. It should be noted that the embodiments of the present invention and the features and technical solutions in the embodiments can be combined with each other without conflict. It should be noted that similar numbers and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0025] Embodiment 1: An image analysis method based on deep learning, comprising the following steps: Step 1, data collection and preprocessing: collecting a large amount of image data of solar burst activities, wherein the image data includes different solar burst types, intensities and times; Step 2: Build a deep learning model: Use a deep learning convolutional neural network (CNN) model for image analysis. The architecture of the deep learning convolutional neural network (CNN) model includes a convolutional layer, a pooling layer, and a fully connected layer. The hyperparameters of the deep learning convolutional neural network model include the number of iterations and the batch size. The model performance is optimized through experiments and adjustments. Step 3, model training: divide the preprocessed image data into training set, validation set and test set; use the training set to train the model, and continuously adjust the model parameters through the back propagation algorithm to minimize the loss function; during the training process, use the validation set to monitor the performance of the model, adjust the hyperparameters in time, and prevent overfitting; Step 4: Image feature extraction: Use the trained model to extract features from the image of solar burst activity to obtain high-level feature representation of the image; the high-level features of the image include the shape, intensity, and location information of the solar burst; Step 5: Solar burst activity image prediction: Based on the extracted image features, use machine learning models including recurrent neural network RNN ​​or long short-term memory network LSTM to predict solar burst activity; input solar burst activity image feature sequence is in Represents the image feature vector at the Tth time point, where T is the length of the time series. The model will output the prediction results of future solar eruption activities; Step 6, correlation state analysis module: Combine the predicted results of solar burst activities and the space environment to build a correlation state analysis module; the correlation state analysis module analyzes the impact of solar burst activities on the space environment, including radiation intensity and magnetic field changes; and determines the scope of the impact of solar burst activities in the space environment through simulation.

[0026] Step 1 preprocesses the image, including image enhancement, denoising, cropping and normalization operations, where the enhanced image is represented as: ;in: is the original pixel value, The pixel value after enhancement; and are the minimum and maximum pixel values ​​in the image, respectively. is the maximum number of gray levels of the output image.

[0027] Step 2. In the convolutional neural network, the deep learning convolutional neural network CNN for image analysis is used to extract the features of the image. The expression of the convolution operation of the deep learning convolutional neural network is: in: is the output value after convolution, located at position ; is the input image at position The pixel value of is the weight of the convolution kernel at the position; m and n are the height and width of the convolution kernel respectively.

[0028] Step 3: Model training: divide the preprocessed image data into a training set, a validation set, and a test set; use the training set to train the model, and continuously adjust the parameters of the model through the back propagation algorithm to minimize the average loss function, which is expressed as: in: represents the average loss function; is the real image data, is the predicted image data, N is the number of image data samples, is the pixel coordinate of the image data.

[0029] It also includes safe flight path planning: designing a safe flight path planning algorithm for the affected target aircraft; the safe flight path planning algorithm takes into account the predicted results of solar burst activities, the impact range of the space environment, and the performance and constraints of the target aircraft; and through the optimization algorithm, generating a safe flight path that can avoid the impact of solar burst activities.

[0030] Step 5: The specific steps of solar burst activity image prediction are as follows: Step 51, data collection and preprocessing: collect historical solar flare activity image data; perform feature extraction on the image to extract the number of sunspots, flare area, and magnetic field intensity features; perform data cleaning and preprocessing on the extracted features to handle missing values ​​and outliers; Step 52: Build a machine learning model: Use a recurrent neural network (RNN) as a prediction model and determine the structure of the model, including the number of hidden layer neurons and layer parameters; Step 53, model training: Use the training set data to train the model; during the training process, adjust the model parameters according to the value of the loss function, and use the test set data to evaluate the trained model.

[0031] Step 6: Association state analysis module: through image data pixels Combined with the analysis and prediction of solar burst activities, the correlation state analysis and prediction function expression is constructed as follows: ;in: Association state analysis prediction function; Explosive intensity level: expressed as a value from 1 to 5. Magnetic field change; the intensity of solar eruptive activity; Radiation intensity, is the pixel coordinate of the image data.

[0032] Space weather monitoring and early warning: Use the results of the correlation state analysis module to monitor space weather in real time. When solar eruptions affect the space environment, early warning information is issued in a timely manner to provide protection for spacecraft operations and satellite communications.

[0033] An application of deep learning-based image analysis methods, which is applied in the analysis and prediction of the relationship between solar flare activity and space environment. The deep learning-based image analysis method is applied in space weather monitoring and early warning.

[0034] The working principle of the deep learning-based image analysis method is as follows: First, data collection and preprocessing are performed. A large amount of image data of solar burst activities is collected extensively, covering different solar burst types, intensities, and times. The collected data is preprocessed to prepare for subsequent analysis, and the image is enhanced by enhancing the image. The enhanced image is represented as: ;in: is the original pixel value, The pixel value after enhancement; and are the minimum and maximum pixel values ​​in the image, respectively. is the maximum number of gray levels of the output image.

[0035] Next, we build a deep learning model. We use a deep learning convolutional neural network (CNN) model for image analysis. The architecture of this model includes convolutional layers, pooling layers, and fully connected layers. We optimize the performance of the model by experimenting and adjusting hyperparameters, such as the number of iterations and batch size.

[0036] In the convolutional neural network, the deep learning convolutional neural network CNN for image analysis is used to extract the features of the image. The expression of the convolution operation of the deep learning convolutional neural network is: in: is the output value after convolution, located at position ; is the input image at position The pixel value of is the weight of the convolution kernel at the position; m and n are the height and width of the convolution kernel respectively.

[0037] Then, the preprocessed image data is divided into a training set, a validation set, and a test set. The model is trained using the training set, and the parameters of the model are continuously adjusted using the back propagation algorithm to minimize the loss function. During the training process, the performance of the model is monitored using the validation set, and the hyperparameters are adjusted in a timely manner to avoid overfitting. The preprocessed image data is divided into a training set, a validation set, and a test set; the model is trained using the training set, and the parameters of the model are continuously adjusted using the back propagation algorithm to minimize the average loss function, which is expressed as: in: represents the average loss function; is the real image data, is the predicted image data, N is the number of image data samples, is the pixel coordinate of the image data.

[0038] After the training is completed, the trained model is used to extract features from images of solar burst activities to obtain high-level feature representations of the images. These features include the morphology, intensity, and location information of the solar burst.

[0039] Based on the extracted image features, machine learning models such as recurrent neural networks (RNNs) or long short-term memory networks (LSTMs) are used to predict solar flare activity. The model takes the solar flare activity image feature sequence as input and outputs the prediction results for future solar flare activity.

[0040] Finally, the correlation state analysis module is constructed. Combining the prediction results of solar burst activities and the space environment, this module analyzes the impact of solar burst activities on the space environment, such as radiation intensity and magnetic field changes, and determines the impact range of solar burst activities in the space environment through simulation. Combined with the analysis and prediction of solar burst activities, the correlation state analysis and prediction function expression is constructed as follows: in: Association state analysis prediction function; Explosive intensity level: expressed as a value from 1 to 5. Magnetic field change; the intensity of solar eruptive activity; Radiation intensity, is the pixel coordinate of the image data.

[0041] A safe flight path planning algorithm; the safe flight path planning algorithm takes into account the predicted results of solar flare activities, the impact range of the space environment, and the performance and constraints of the target aircraft; through an optimization algorithm, a safe flight path that can avoid the impact of solar flare activities is generated; The steps of the safe flight path planning algorithm are as follows: 1. Import the required libraries: import numpy as np and from scipy.optimize import minimize.

[0042] 2. Define the prediction model of solar flare activity solar_flare_prediction_model: Modeling is done based on the prediction data of solar flare activity. The intensity of solar flare activity varies randomly within a certain time range, and the function returns a random value.

[0043] 3. Define the space environment influence range model space_environment_influence_model: Modeling is done based on actual space environment data. The space environment influence range is a sphere with a certain radius centered on the earth. Assume that the center of the earth is np.array([0, 0, 0]) and the influence range radius is 10000. By calculating the distance between the position and the center of the earth, determine whether it is within the influence range. If the distance is less than or equal to the radius, 1 is returned, indicating that it is within the influence range; otherwise, 0 is returned, indicating that it is not within the influence range.

[0044] 4. Define the performance and constraint model of the target aircraft: aircraft_performance_and_constraints_model: Model the target aircraft based on its actual performance and constraints, taking into account factors such as the aircraft's speed, fuel consumption, and maximum acceleration. Set the speed limit to 0 <= velocity <= 1000, and the fuel consumption is proportional to the path length. Assuming a fixed speed of 500, the fuel consumption is calculated as the sum of the distances of adjacent points on the path. If the speed on the path exceeds the limit, or the fuel consumption exceeds the limit (assuming it is 10000), or the acceleration exceeds the limit (assuming it is 10), then return infinity; otherwise, return the fuel consumption value as the objective function value.

[0045] 5. Define the safe flight path planning algorithm safe_flight_path_planning: Initialize the path: Generate num_points points at equal intervals between the starting position and the ending position to form the initial path path.

[0046] Define the objective function objective_function: This function comprehensively considers the impact of solar flare activity, the impact of the space environment, and the performance and constraints of the spacecraft. By calling the three model functions defined above, the intensity of solar flare activity, the impact of the space environment, and the values ​​of the spacecraft's performance and constraints are calculated, and they are multiplied as the value of the objective function.

[0047] Use the optimization algorithm to solve the optimal path: use the minimize function, objective_function as the objective function, path as the initial value, and use the SLSQP method to solve the optimal path.

[0048] Return the optimal path: Return the result result.x obtained by solving the problem as the optimal path.

[0049] Example 2: A deep learning-based image analysis method, data collection and preprocessing: 10,000 images of solar burst activities were collected, covering a variety of solar burst types (such as flares, coronal mass ejections), different intensities (weak, medium, strong) and times (different years and seasons). The images were preprocessed by denoising, cropping, and normalization.

[0050] Constructing a deep learning model: Convolutional neural network (CNN) model architecture: 3 convolutional layers, each with a kernel size of 3x3, a step size of 1, and an activation function of ReLU; 2 pooling layers, using maximum pooling, a pooling window size of 2x2, and a step size of 2; 1 fully connected layer, with 128 neurons and an activation function of Sigmoid. Hyperparameters: 50 iterations and 32 batch sizes.

[0051] Model training: The preprocessed image data is divided into training set, validation set and test set in a ratio of 7:2:1. Stochastic gradient descent SGD is used as the optimizer, the learning rate is 0.01, and the loss function is the cross entropy loss function. During the training process, the validation set is used to evaluate the model performance every 5 iterations, and the hyperparameters are adjusted according to the evaluation results.

[0052] Image feature extraction: Use the trained CNN model to extract features from images of solar burst activities to obtain high-level feature representation of the image, including the shape, intensity, and location information of the solar burst.

[0053] Solar burst activity image prediction: Long short-term memory network LSTM is used as a machine learning model to predict solar burst activity. The length T of the input solar burst activity image feature sequence is 10, which means that the image features of the past 10 time points are considered. The number of hidden layer neurons in the LSTM model is 64, and the number of layers is 2.

[0054] Correlation state analysis module: Combine the prediction results of solar burst activities with space environment data to build a correlation state analysis module. Consider the radiation intensity (unit: millisievert / hour) and magnetic field change (unit: gauss) factors. Through numerical simulation methods, determine the impact range of solar burst activities in the space environment (with the Earth orbit as a reference, unit: kilometers).

[0055] A large amount of solar eruption activity image data was collected and preprocessed. Then, the present invention constructed a deep learning convolutional neural network (CNN) model and trained it according to the set architecture and hyperparameters. During the training process, the present invention used the training set to adjust the model parameters, continuously optimized the model performance through the back propagation algorithm, and used the validation set to monitor the performance of the model, and adjusted the hyperparameters in time to prevent overfitting.

[0056] After training, the present invention obtains a model that can effectively extract the features of solar burst activity images. Next, the present invention inputs the pre-processed new solar burst activity image data into the trained model to extract image features. These features include the shape, intensity and location information of the solar burst.

[0057] Then, the present invention inputs the extracted image feature sequence into the long short-term memory network LSTM model to predict solar flare activities. Assuming that the length of the input image feature sequence is 10, the LSTM model outputs the prediction results of future solar flare activities based on these historical feature information, such as predicting the type and intensity of solar flare at a certain time point in the future.

[0058] Finally, the present invention combines the prediction results of solar burst activities with space environment data to construct an associated state analysis module. By analyzing the impact of solar burst activities on the space environment, including radiation intensity and magnetic field changes, the present invention uses a numerical simulation method to determine the impact range of solar burst activities in the space environment. It is predicted that there will be a strong solar burst activity in the next week. Through the simulation of the associated state analysis module, the present invention determines that this solar burst activity will cause the radiation intensity near the earth's orbit to increase to 5 millisieverts / hour, the magnetic field change reaches 10 gauss, and will have a certain impact on space equipment within 5,000 kilometers from the earth.

[0059] Experimental example Experiment Name: Image Analysis and Prediction of Solar Eruption Activities Based on Deep Learning Purpose: Use deep learning methods to analyze and predict images of solar eruption activities and study their impact on the space environment.

[0060] Experimental Materials: 1. A large amount of solar burst activity image data, covering different solar burst types, intensities and times.

[0061] 2. Computer equipment with sufficient computing power to run deep learning models and perform data analysis.

[0062] Experimental steps: 1. Data collection and preprocessing: A large amount of solar burst activity image data is collected from relevant astronomical observation agencies and databases.

[0063] The image data is cleaned, cropped, and normalized to improve data quality and model training results.

[0064] 2. Build a deep learning model: A deep learning convolutional neural network (CNN) model is used for image analysis.

[0065] Design the architecture of convolutional layers, pooling layers, and fully connected layers, and determine the appropriate number of layers and neurons.

[0066] Through experiments and adjustments, the hyperparameters of the model are optimized, such as the number of iterations is set to 50 and the batch size is set to 32.

[0067] 3. Model training: The preprocessed image data is divided into training set, validation set and test set in the ratio of 7:2:1.

[0068] The model is trained using the training set, and the back-propagation algorithm is used to continuously adjust the model parameters to minimize the loss function.

[0069] During the training process, the validation set is used to evaluate the performance of the model every certain number of iterations, and the hyperparameters are adjusted in time according to the evaluation results to prevent overfitting.

[0070] 4. Image feature extraction: Use the trained model to extract features from images of solar burst activity.

[0071] Obtain high-level feature representations of the image, such as the shape, intensity, and location information of solar bursts.

[0072] 5. Solar eruption activity image prediction: Based on the extracted image features, recurrent neural network RNN ​​or long short-term memory network LSTM is used as a machine learning model to predict solar burst activities.

[0073] Input the solar burst activity image feature sequence, where represents the image feature vector at the Tth time point, and T is the length of the time series.

[0074] The model will output predictions of future solar eruption activity.

[0075] 6. Association status analysis module: Experimental data: Table 1 Solar burst activity image data Image number Solar Burst Types strength time 1 Flare high 2024-1-05 10:00 2 Coronal Mass Ejection middle 2024-2-06 12:00 3 Flare Low 2024-3-07 08:00 4 Coronal Mass Ejection high 2024-4-07 15:00 5 Flare middle 2024-5-9 10:00 Table 2 Deep learning model training data Dataset Number of images Training set 3500 Validation set 1000 Test Set 500 Table 3. Deep learning model hyperparameters Hyperparameters value Iterations 50 Batch size 32 Learning Rate 0.001 Table 4. Results of feature extraction of solar burst activity images Image number Morphological characteristics Strength characteristics Location Features 1 Round light spot with bright edges high Central region of the sun's surface 2 Large-scale mass ejection, fan-shaped middle The edge of the sun 3 Smaller light spot, brighter Low The southern position of the sun 4 Strong material ejection, irregular shape high Sun Northwest 5 Medium-sized light spot, moderate brightness middle Eastern part of the sun Table 5. Solar burst activity image prediction results Time Predicted solar eruption types Forecast strength 2024-7-09 10:00 Flare middle 2024-8-11 12:00 Coronal Mass Ejection high 2024-10-20 08:00 Flare Low Combine the prediction results of solar flare activities with space environment data to build a correlation state analysis module to analyze the impact of solar flare activities on the space environment, including radiation intensity and magnetic field changes.

[0076] Through simulation calculation, the impact range of solar burst activity in the space environment is determined. The present invention successfully collects and preprocesses a large amount of solar burst activity image data and constructs a deep learning convolutional neural network (CNN) model. By adjusting the hyperparameters, the model performance is optimized, so that it achieves better results in solar burst activity image classification and feature extraction.

[0077] The trained model is used to extract features from solar burst activity images, and high-level feature representations of the images are obtained, including morphology, intensity, and location information. These features provide an important basis for the prediction of solar burst activity.

[0078] Based on the extracted image features, the recurrent neural network RNN ​​or long short-term memory network LSTM is used to predict solar burst activities, and certain prediction results are achieved. Although there are certain errors in the prediction results, it provides a direction for further improving the prediction accuracy; The constructed correlation state analysis module can analyze the impact of solar burst activities on the space environment, including radiation intensity and magnetic field changes, and determine the scope of the impact of solar burst activities in the space environment. This has important application value for space science research and space missions.

[0079] The above embodiments are only used to illustrate the present invention and are not intended to limit the technical solutions described in the present invention. Although the present invention has been described in detail with reference to the above embodiments, the present invention is not limited to the above specific implementation methods. Therefore, any modification or replacement of the present invention; and all technical solutions and improvements thereof that do not depart from the spirit and scope of the invention are included in the scope of the claims of the present invention.

Claims

1. An image analysis method based on deep learning, characterized in that: The following steps are included: Step 1, data collection and preprocessing: collect a large amount of image data of solar burst activities, the image data includes different solar burst types, intensities and times; Step 2: Build a deep learning model: Use a deep learning convolutional neural network (CNN) model for image analysis. The architecture of the deep learning convolutional neural network (CNN) model includes a convolutional layer, a pooling layer, and a fully connected layer. The hyperparameters of the deep learning convolutional neural network model include the number of iterations and the batch size. The model performance is optimized through experiments and adjustments. Step 3, model training: divide the preprocessed image data into training set, validation set and test set; use the training set to train the model, and continuously adjust the model parameters through the back propagation algorithm to minimize the loss function; during the training process, use the validation set to monitor the performance of the model, adjust the hyperparameters in time, and prevent overfitting; Step 4: Image feature extraction: Use the trained model to extract features from the image of solar burst activity to obtain high-level feature representation of the image; the high-level features of the image include the shape, intensity, and location information of the solar burst; Step 5: Solar burst activity image prediction: Based on the extracted image features, use machine learning models including recurrent neural network RNN ​​or long short-term memory network LSTM to predict solar burst activity; input solar burst activity image feature sequence is ,in Represents the image feature vector at the Tth time point, where T is the length of the time series. The model will output the prediction results of future solar eruption activities; Step 6, correlation state analysis module: Combine the prediction results of solar burst activities and the space environment to build a correlation state analysis module; The correlation state analysis module analyzes the impact of solar flare activities on the space environment, including radiation intensity and magnetic field changes; and determines the scope of the impact of solar flare activities on the space environment through simulation.

2. The image analysis method based on deep learning according to claim 1, characterized in that: Step 1 preprocesses the image, including image enhancement, denoising, cropping and normalization operations, where the enhanced image is represented as: ;in: is the original pixel value, The enhanced pixel value; and are the minimum and maximum pixel values ​​in the image, respectively. is the maximum number of gray levels of the output image.

3. The image analysis method based on deep learning according to claim 2, characterized in that: Step 2. In the convolutional neural network, the deep learning convolutional neural network CNN for image analysis is used to extract the features of the image. The expression of the convolution operation of the deep learning convolutional neural network is: in: is the output value after convolution, located at position ; is the input image at position The pixel value of is the weight of the convolution kernel at the position; m and n are the height and width of the convolution kernel respectively.

4. The image analysis method based on deep learning according to claim 3, characterized in that: Step 3: Model training: divide the preprocessed image data into a training set, a validation set, and a test set; use the training set to train the model, and continuously adjust the parameters of the model through the back propagation algorithm to minimize the average loss function, which is expressed as: in: represents the average loss function; is the real image data, is the predicted image data, N is the number of image data samples, is the pixel coordinate of the image data.

5. The image analysis method based on deep learning according to claim 4, characterized in that: It also includes safe flight path planning: designing a safe flight path planning algorithm for the affected target aircraft; the safe flight path planning algorithm takes into account the predicted results of solar burst activities, the impact range of the space environment, and the performance and constraints of the target aircraft; and through the optimization algorithm, generating a safe flight path that can avoid the impact of solar burst activities.

6. The image analysis method based on deep learning according to claim 5, characterized in that: Step 5: The specific steps of solar burst activity image prediction are as follows: Step 51, data collection and preprocessing: collect historical solar flare activity image data; perform feature extraction on the image to extract the number of sunspots, flare area, and magnetic field intensity features; perform data cleaning and preprocessing on the extracted features to handle missing values ​​and outliers; Step 52: Build a machine learning model: Use a recurrent neural network (RNN) as a prediction model and determine the structure of the model, including the number of hidden layer neurons and layer parameters; Step 53, model training: Use the training set data to train the model; during the training process, adjust the model parameters according to the value of the loss function, and use the test set data to evaluate the trained model.

7. The image analysis method based on deep learning according to claim 6, characterized in that: Step 6: Association state analysis module: through image data pixels Combined with the analysis and prediction of solar burst activities, the correlation state analysis and prediction function expression is constructed as follows: ;in: Association state analysis prediction function; Explosive intensity level: expressed as a value from 1 to 5. Magnetic field change; the intensity of solar eruptive activity; Radiation intensity, is the pixel coordinate of the image data.

8. The image analysis method based on deep learning according to claim 7, characterized in that: It also includes space weather monitoring and early warning: using the results of the correlation state analysis module to monitor space weather in real time; when solar flare activities have an impact on the space environment, early warning information is issued in a timely manner to provide protection for spacecraft operations and satellite communications.

9. Application of the deep learning-based image analysis method according to any one of claims 1 to 8 in the analysis and prediction of the relationship between solar flare activities and space environment.

10. Application of the deep learning-based image analysis method according to any one of claims 1 to 8 in space weather monitoring and early warning.

Images