Ionized layer traveling wave disturbance intelligent identification and propagation parameter calculation method and system

Abstract

The invention discloses an ionosphere traveling wave disturbance intelligent identification and propagation parameter calculation method and system, and belongs to the technical field of ionosphere disturbance detection.The method comprises the steps that an adjusted DTEC image is input into a trained YOLO target detection model for target detection, and a Bbox frame label containing a wavy disturbance structure is identified and obtained; preprocessing before YOLO instance segmentation is carried out to obtain segmented images in continuous time, and then the segmented images are input into a YOLO instance segmentation model to carry out accurate segmentation of color blocks to obtain images containing color block masks; performing five-dimensional feature calculation on the color block mask by using an OpenCV library; matching the segmented images in continuous time, calculating a matching score based on the weight of each feature, and obtaining an optimal matching result of a color block mask by adopting a Hungary algorithm; and performing circumscribed ellipse fitting on the successfully matched color block masks in the optimal matching result, and calculating the horizontal speed, azimuth angle and coverage area of the MSTIDs. According to the method, all-parameter calculation of the MSTIDs including the propagation speed can be realized, and the calculation time is short.

Description

TECHNICAL FIELD

[0001] The present application relates to an ionospheric traveling disturbance intelligent identification and propagation parameter calculation method and system, belonging to the technical field of ionospheric disturbance detection. BACKGROUND

[0002] MSTIDs are sudden and periodic wave-like structures in the ionosphere, with a typical period from tens of minutes to nearly 1 hour, a propagation speed of 100-400 m / s, and a wavelength usually within a few hundred kilometers. During the day in autumn and winter, MSTIDs mainly propagate towards the equator, while at night in summer they mainly propagate towards the southwest. The generation of MSTIDs is mainly related to space weather events (such as solar flares, coronal mass ejections, etc.) and natural disasters (such as earthquakes, tsunamis, volcanic eruptions, etc.). Among them, space weather events are excited by the cross-ordinate coupling process between the atmosphere, ionosphere and magnetosphere. The ionospheric electron content changes caused by MSTIDs are usually within a few tenths of TECu, while during the solar maximum period, this change can reach several TECu. This directly leads to the decrease in precision of high-precision satellite positioning technologies such as PPP-RTK and NRTK, accompanied by an increase in the occurrence rate of GNSS cycle slips. With the solar activity entering a new cycle of active peak phase, ionospheric traveling disturbances occur frequently worldwide, seriously affecting the service performance of satellite navigation systems. Therefore, how to accurately identify and accurately express the propagation characteristics of ionospheric traveling disturbances has become a common focus in the fields of satellite navigation and space physics research.

[0003] The detection of MSTIDs is a highly challenging task, with the main difficulties being as follows: First, how to accurately determine whether a disturbance has occurred based on one-dimensional time series signals or two-dimensional image observations, and effectively separate ionospheric traveling disturbances (TIDs) from equatorial plasma bubbles (EPBs), equatorial ionization anomalies (EIA), sporadic E (Es) and other ionospheric anomalies; Second, due to the large differences in the propagation characteristics and scale of TIDs, it is difficult to use a simple and unified criterion to effectively define them, which leads to the reliance on expert subjective experience judgment or the development of specific judgment criteria in previous studies.

[0004] Most of the current methods focus on the detection of TIDs induced by natural disasters such as tsunamis and earthquakes. The excitation source of such MSTIDs is clear, and they have specific propagation characteristics, which can be verified by various observation methods. Therefore, the detection difficulty is not high, and one-dimensional differential ionospheric electron content (DTEC) time series signals can often be used for detection. VARION, a representative algorithm based on real-time ionospheric observations, uses single-time-difference geometry-free linear combinations (GFLC) of carrier phase measurements for TIDs detection. Some methods have also been used to detect MSTIDs with unclear excitation sources in two-dimensional DTEC maps, such as the atomic decomposition MSTIDs detection method (ADDTID). This method can automatically identify and track MSTIDs in the DTEC maps generated by the dense GEONET network in Japan by constructing a reasonable quantized MSTIDs waveform dictionary and using sparse reconstruction technology. However, due to the diversity of MSTIDs excitation sources and the complexity of the propagation process, traditional mathematical methods cannot effectively detect and quantitatively describe the complex spatiotemporal variation characteristics of MSTIDs. Moreover, GNSS and other ionospheric detection technologies have accumulated vast amounts of high-resolution observation data over the past three decades, making it difficult to analyze the propagation characteristics of MSTIDs through manual statistical analysis.

[0005] With the continuous development of deep learning technology, especially the continuous progress of object detection and image segmentation algorithms, it is possible to intelligently identify and express the propagation characteristics of MSTIDs through data-driven methods. Existing methods such as CNN and Mask R-CNN have been applied to the near-real-time detection of MSTIDs, and the propagation characteristics of MSTIDs in the Japanese region over the past 20 years can be statistically analyzed. However, these methods can only analyze the occurrence rate, propagation period, and wavefront direction of the MSTIDs color block structure in a single image, and cannot extract key propagation speed information. In addition, two-stage algorithms such as Mask R-CNN, which first generate candidate regions and then classify and segment them, have a long calculation time, making it difficult to meet the future demand for real-time detection of MSTIDs. SUMMARY

[0006] The technical problem to be solved by the present application is to overcome the defects of the prior art and provide an ionospheric traveling wave disturbance intelligent identification and propagation parameter calculation method and system. By integrating object detection, color block segmentation, and continuous image matching, the present application can calculate all parameters of MSTIDs, including propagation speed, with a shorter calculation time.

[0007] To solve the above technical problems, the technical scheme adopted by the present application is as follows:

[0008] In a first aspect, the present application provides an ionospheric traveling wave disturbance intelligent identification and propagation parameter calculation method, comprising the following steps:

[0009] The DTEC image data of the ionosphere in the dense observation network is acquired, and after the DTEC image data is adjusted in resolution, an adjusted DTEC image is obtained;

[0010] The adjusted DTEC image is input into a trained YOLO target detection model for target detection, and a Bbox frame label containing a wave-like disturbance structure is recognized and obtained;

[0011] The DTEC image in the Bbox frame label is preprocessed before YOLO instance segmentation to obtain a segmented image of continuous time, and the segmented image of continuous time is input into a YOLO instance segmentation model for accurate segmentation of color blocks to obtain an image containing color block masks;

[0012] Five-dimensional feature calculation is performed on the color block mask using the OpenCV library, including color block color, color block area, mask centroid coordinates, Hu matrix and topological relationship between color blocks, to obtain five features;

[0013] The segmented image of continuous time is matched using the five features, a matching score is calculated based on the weight of each feature, and the best matching result of the color block mask is obtained based on the matching score using the Hungarian algorithm;

[0014] The successfully matched color block mask in the best matching result is fitted with an inscribed ellipse, and the pixel coordinates of the mask centroid are converted into geographic coordinates. The size, moving direction, distance and time interval of the standardized inscribed ellipse of the mask are calculated to obtain the horizontal velocity, azimuth angle and coverage area of the MSTIDs. The wavelength and period parameters of the MSTIDs are calculated using the feature relationship between different color blocks in the image containing the color block mask.

[0015] The preprocessing of the DTEC image in the Bbox frame label before YOLO instance segmentation to obtain the segmented image of continuous time includes:

[0016] A group of time-continuous DTEC images is screened out;

[0017] The overlapping area index IoM of the Bbox frames in two consecutive images in the DTEC image group is calculated;

[0018] A pair of consecutive images whose IoM meets the set range is selected, and the images in the pair of consecutive images are cropped according to the corresponding Bbox frame coordinates to obtain images containing only the red and blue color blocks of the MSTIDs, and finally the segmented image of continuous time is obtained;

[0019] The overlapping area index IoM is the ratio of the overlapping area of the two Bboxes to the smaller area Bbox, and the calculation formula is as follows:

[0020] (1),

[0021] In the formula, A and B represent the areas of two Bboxes, respectively, and the threshold of IoM is set to 0.5, and only the continuous image pairs satisfying IoM≥0.5 are selected.

[0022] The color block area is calculated based on the irregular polygon contour of the mask using the shoelace formula, and the specific expression is as follows:

[0023] (2),

[0024] In the formula, C is the area of the polygon, k is the number of polygon vertex coordinates, and is the horizontal and vertical coordinates of the s+1th vertex, and is the horizontal and vertical coordinates of the s+2th vertex.

[0025] The centroid coordinates of the mask are obtained by the geometric moments of the image, and the definition of the geometric moments is as follows:

[0026] (3),

[0027] In the formula, is the geometric moment, p and q are the orders of the moment, x and y are the horizontal and vertical coordinates of the two-dimensional image, is the value of the pixel point in the binary image, and the centroid of the color block is represented as:

[0028] (4),

[0029] and are the horizontal and vertical coordinates of the centroid of the color block, respectively.

[0030] The Hu matrix removes the influence of translation by calculating the central moment, and the calculation formula is as follows:

[0031] (5),

[0032] In the formula, is the centralized geometric moment, and the central distance is normalized to eliminate the scale effect:

[0033] (6),

[0034] In the formula, represents the normalized central moment, and the Hu matrix is represented as:

[0035] (7),

[0036] (8),

[0037] wherein, , , , , , and respectively represent the invariant features as matrices.

[0038] The topological relationship between the extracted color blocks is obtained by comparing the coordinate difference of the two color block centers in the x direction and the y direction: if the absolute value of the x direction difference is greater than the y direction, then according to the positive or negative of the x difference, it is judged as left or right; otherwise, according to the positive or negative of the y difference, it is judged as up or down.

[0039] The matching score based on the weight of each feature includes:

[0040] In the matching process, first set the color block set of the MSTIDs segmented from the continuous image pair respectively as and , then the matching score between the color blocks and is represented as:

[0041] (9),

[0042] wherein, is the feature difference between the color blocks and , which can be specifically represented as:

[0043] (10),

[0044] wherein, and are the barycentric coordinates of the color blocks and respectively; and are the areas of the color blocks and respectively; and are the Hu matrices of the color blocks and respectively; is the topological relationship penalty term, if the adjacent color blocks of the two color blocks are inconsistent in color or relative direction, then otherwise ; , , and respectively represent the weights of each feature difference.​

[0045] The optimal matching result of the color block mask based on the matching score adopts the Hungarian algorithm, including:

[0046] Based on the MSTIDs color block set And Construct Similarity matrix;

[0047] Adopt the Hungarian algorithm to select a matching set from The similarity matrix corresponds to all possible matches So that the sum of the scores of all selected matches is minimized, where the Hungarian algorithm criterion is expressed as:

[0048] (11),

[0049] After matching, validity matching is performed, and the screening condition is:

[0050] (12),

[0051] In the formula, T is the screening score threshold.

[0052] The pixel coordinates of the mask center are converted into geographic coordinates, including:

[0053] Select control points to determine the pixel points in the image and the corresponding latitude and longitude coordinates;

[0054] Convert the known latitude and longitude coordinates into coordinate points in the projection coordinate system, and use a nonlinear polynomial fitting method to establish a nonlinear correspondence between the pixel coordinates in the image and the projection coordinates to form a coordinate conversion model;

[0055] After fitting is completed, the projection coordinates are reversely converted into latitude and longitude coordinates based on the coordinate conversion model.

[0056] In a second aspect, the present application provides an ionospheric traveling wave disturbance intelligent identification and propagation parameter calculation system, comprising:

[0057] An image acquisition and adjustment module is used to acquire DTEC image data of the ionosphere in the dense observation network, and then after resolution adjustment of the DTEC image data, an adjusted DTEC image is obtained;

[0058] An MSTIDs detection module is used to input the adjusted DTEC image into a trained YOLO target detection model for target detection, identify and obtain a Bbox frame label containing a wave-like disturbance structure;

[0059] The MSTID color block instance segmentation module is used for pre-processing of the DTEC image in the Bbox frame label before YOLO instance segmentation, obtaining a continuous-time segmentation image, and then inputting the continuous-time segmentation image into a YOLO instance segmentation model for accurate segmentation of the color block to obtain an image containing a color block mask;

[0060] The MSTID feature extraction module is used for five-dimensional feature calculation of the color block mask by using an OpenCV library, including color block color, color block area, centroid coordinates of the mask, a Hu matrix and a topological relationship between color blocks, to obtain five features.

[0061] The color block mask matching module is used for matching the continuous-time segmentation image by using the five features, calculating a matching score based on the weight of each feature, and then obtaining the best matching result of the color block mask by using a Hungarian algorithm based on the matching score.

[0062] The MSTID propagation parameter calculation module is used for fitting an inscribed ellipse of the successfully matched color block mask in the best matching result, converting the pixel coordinates of the mask centroid into geographic coordinates, calculating the horizontal velocity, azimuth angle and coverage area of the MSTIDs by the size, moving direction and distance of the standardized inscribed ellipse of the mask and the time interval, and then calculating the wavelength and period parameters of the MSTIDs by using the feature relationship between different color blocks in the image containing the color block mask.

[0063] The method and system for intelligently identifying and calculating the propagation parameters of the ionospheric traveling wave disturbance provided by the application have the following advantages:

[0064] (1) Overall, by integrating target detection, color block segmentation and continuous image matching, the full parameter calculation of the MSTIDs including the propagation velocity can be realized, and the calculation time is relatively short.

[0065] (2) Target detection and instance segmentation are coordinated to significantly improve the detection accuracy and robustness. Unlike the color block segmentation of the whole image by only using the instance segmentation method, the application first uses the YOLO model to accurately detect and crop the image of the target area of the MSTIDs, and eliminates the interference of isolated waves, noise and the like, effectively reduces the dependence of the model on the performance of the instance segmentation, and greatly improves the accuracy and robustness of the MSTIDs identification.

[0066] (3) The continuous time domain information is fused to support automatic calculation of the whole process propagation parameters. Based on the continuous time sequence DTEC image, the time sequence tracking and cross-time matching of the MSTIDs color block are realized by analyzing the IoM (overlapping area index) of the detected target area, the continuity of the MSTIDs propagation can be fully utilized, the key physical parameters such as the propagation velocity, direction, wavelength and period can be automatically calculated from the image sequence, and the problem that the dynamic characteristics such as the propagation velocity cannot be obtained by the traditional method is solved.

[0067] (4) Full-process end-to-end intelligent processing, expanding the quantitative analysis capability of MSTIDs parameter. The application not only realizes the pixel-level fine segmentation of the MSTIDs target, but also automatically completes the quantitative calculation of multiple propagation characteristics such as MSTIDs coverage area, wavelength, and period by combining fitting ellipse and geographic coordinate conversion, improving the automation and intelligent analysis level of the spatio-temporal evolution of MSTIDs, and expanding the scientific research means of ionospheric disturbance events.

[0068] (5) Adapt to the needs of automatic processing of massive data, suitable for large-scale application and promotion. Fully combining deep learning and image processing technology, it can cope with the massive DTEC observation data accumulated by dense network such as GNSS, meet the needs of automation, large quantity, and efficient analysis, and provide a solid technical foundation for related ionospheric space weather monitoring, early warning, and scientific research. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure 1 The algorithm flow framework diagram of the application based on YOLO and feature matching for MSTIDs intelligent recognition and propagation parameter calculation;

[0070] Figure 2 The comparison chart of the MSTIDs target detection results in the continuous image pair on April 9, 2024 in the application;

[0071] Figure 3 The loss and performance index change curve graph in the training and verification process of the MSTIDs detection model based on YOLOv8m in the application;

[0072] Figure 4 The loss and performance index change curve graph in the training and verification process of the MSTIDs instance segmentation model based on YOLOv8-seg in the application;

[0073] Figure 5 The result graph of the YOLOv8-seg model for MSTIDs color block segmentation in the application;

[0074] Figure 6 The MSTIDs color block feature matching and propagation parameter calculation result graph in the period from 13:40 to 13:50 on January 22, 2024 in the application;

[0075] Figure 7 The MSTIDs propagation parameter calculation result graph of the three-station isospectral method (traditional mathematical method) in Japan on January 22, 2024 in the application. DETAILED DESCRIPTION

[0076] The application will be further described below with reference to the drawings, and the following examples are only used to more clearly illustrate the technical solutions of the application, and cannot be used to limit the protection scope of the application.

[0077] Example 1

[0078] As Figure 1 shown, the application provides a mesoscale ionospheric traveling wave disturbance intelligent identification and propagation parameter calculation method, comprising the following steps:

[0079] Step one, obtain the ionospheric DTEC image data of the dense observation network and adjust the resolution to 640x640 pixels to obtain the adjusted DTEC image.

[0080] Step two, input the adjusted DTEC image into the trained YOLO target detection model for target detection, identify and obtain the Bbox frame label containing the wave-like disturbance structure.

[0081] Step three, pre-process the DTEC image with the identified Bbox frame before YOLO instance segmentation to obtain a segmented image of continuous time, and then input the segmented image into the YOLO instance segmentation model for accurate segmentation of color blocks to obtain an image containing color block masks (Masks).

[0082] Specifically, the data pre-processing of the DTEC image with the Bbox frame specifically includes:

[0083] 1. Select a set of time-continuous DTEC images; 2. Calculate the overlapping area index (IoM) of the Bbox frame in the two continuous images; 3. Select the continuous image pair with IoM greater than 0.5, and crop the image according to the corresponding Bbox frame coordinates to obtain an image containing only the MSTIDs red and blue color blocks. Wherein, IoM is defined as the ratio of the overlapping area of two Bboxes to the smaller area Bbox, and the calculation formula is as follows:

[0084] (1)

[0085] In the formula, A and B represent the areas of two Bboxes, respectively. In the application, in order to ensure that most of the MSTIDs are not missed, the threshold value of IoM is set to 0.5, that is, only the continuous image pair with IoM greater than 0.5 is selected for subsequent image matching and further analysis.

[0086] Step four, use OpenCV library to calculate five-dimensional features of the segmented color block Mask, including color block color, color block area, centroid coordinates of the mask, Hu matrix and topological relationship between color blocks, which are used to abstractly represent the MSTIDs structure.

[0087] Among them, the color block color is divided into two categories: red and blue.

[0088] Where the area of the color block is calculated based on the irregular polygon contour of the mask using the Shoelace formula in Green's theorem, and the specific expression is as follows:

[0089] (2),

[0090] Where C is the area of the polygon, k is the number of polygon vertex coordinates, and is the horizontal and vertical coordinates of the s+1th vertex, and is the horizontal and vertical coordinates of the s+2th vertex, when s=k-1, that is, the horizontal and vertical coordinates of the 1st vertex.

[0091] The centroid coordinates of the mask are obtained by the geometric moments of the image, and the definition of the geometric moments is:

[0092] (3)

[0093] Where p and q are the orders of the moments, is the value of the pixel point in the binary image (usually 0 or 1, indicating foreground or background), then the centroid of the color block can be represented as:

[0094] (4)

[0095] The Hu matrix is mainly used to describe the 7 invariant features of the image Mask shape, and has the advantages of translation invariance, scale invariance and rotation invariance, which can be used to describe and match the overall shape of the object. The Hu matrix removes the effect of translation by calculating the central moment, and the calculation formula is as follows:

[0096] (2)

[0097] Where is the centralized geometric moment, and the definitions of the remaining variables are the same as formula (3). Then normalize the central distance to eliminate the scale effect:

[0098] (6)

[0099] Where represents the normalized central moment, then the Hu matrix can be represented as:

[0100] (7)

[0101] Where

[0102] (8)

[0103] The core idea of extracting the topological relationship between color blocks is: for each color block, find the nearest adjacent color block, record the color and relative direction of the adjacent color block. In the direction determination process, a simplified strategy is adopted, which is to compare the coordinate difference of the two color block centers in x direction and y direction: if the absolute value of the x direction difference is greater than y direction, then judge "left" or "right" according to the positive or negative of x difference; otherwise, judge "up" or "down" according to the positive or negative of y difference.

[0104] Step five, use the five features in step four to match the continuous time segmented image after preprocessing in step three, calculate the matching score based on the weight of each feature, and obtain the best matching result of color block Mask through Hungarian algorithm.

[0105] The specific idea of calculating the matching score based on the weight of each feature is:

[0106] In the matching process, first set the color block set of MSTIDs segmented from the continuous image pair respectively as and , then the matching score between color blocks and can be expressed as:

[0107] (9)

[0108] In the formula, is the feature difference between color blocks and , which can be specifically expressed as:

[0109] (10)

[0110] In the formula, and are the barycentric coordinates of color blocks and respectively; and are the areas of color blocks and respectively; and are the Hu matrices of color blocks and respectively; is the topological relationship penalty term, if the adjacent color block color or relative direction of the two color blocks is not consistent, then otherwise . The weights represent the differences in features. The purpose of using weights is to unify the scores of different features to the same order of magnitude; in this invention, W is set as the weight. pos = 0.5, W area = 200, W Hu =1000, W topo = 50. Additionally, when two color blocks do not match (color mismatch), the score is set to an outlier of 999.

[0111] The Hungarian algorithm is a classic method for solving bipartite graph matching or linear assignment problems. For the MSTIDs color patch set... and Constructed The similarity matrix algorithm selects a set of matches from all possible matches. The goal is to minimize the total score of all selected matches. The algorithm's criterion can be expressed as:

[0112] (11)

[0113] After matching is complete, a validity matching process is required, with the following filtering criteria:

[0114] (12)

[0115] In the formula, the screening score threshold T is set to 200 to remove color block pairs with excessively high scores (i.e., dissimilar).

[0116] Step six involves fitting the circumscribed ellipse to the successfully matched color patch mask and converting the pixel coordinates of the mask's centroid into geographic coordinates (latitude and longitude). The main steps in converting pixel coordinates to geographic coordinates are: first, control points are selected to determine the pixels in the image and their corresponding latitude and longitude coordinates. Next, the known latitude and longitude coordinates are converted into coordinates in the projected coordinate system, and a nonlinear polynomial fitting method is used to establish a nonlinear correspondence between the pixel coordinates in the image and the projected coordinates, forming a coordinate transformation model. After fitting, the projected coordinates can be converted back to latitude and longitude coordinates for high-precision calculation of MSTIDs propagation parameters.

[0117] The size, moving direction, distance, and time interval of the standardized circumscribed ellipse of the Mask can be used to calculate the horizontal velocity, azimuth angle, and coverage area of the MSTIDs. The wavelength and period parameters of the MSTIDs can be calculated by using the feature relationship between different color blocks in the same image in step three. Specifically, by using the longitude and latitude coordinates of the matched color block centers in the continuous images, the horizontal propagation distance and propagation azimuth angle of the MSTIDs in 10-minute intervals can be calculated by using the inverse solution formula of geodesy (WGS84 ellipsoid model), and the horizontal propagation velocity can be further obtained.

[0118] In addition, the present application needs to make a MSTIDs dataset and train a model before using the YOLO series model for MSTIDs target detection and instance segmentation, including the following steps:

[0119] (1) DTEC image data acquisition. According to the influence of different solar activity levels on the structure of MSTIDs, DTEC original image data above Japan in high solar activity years (2024), medium solar activity years (2022), and low solar activity years (2018) are selected, and the first to 100th day of each year is selected, a total of 43,200 DTEC images are obtained.

[0120] (2) MSTIDs target detection dataset making. The collected DTEC images are manually screened, and images with obvious red and blue alternating wavy structures and possible MSTIDs features are selected, a total of 3,426 effective samples are obtained. The image resolution needs to be adjusted to 640*640 before labeling, and the roboflow website is used for online labeling, and the training set, the verification set, and the test set are divided in a ratio of 7:2:1 to obtain the MSTIDs target detection dataset.

[0121] (3) MSTIDs target detection model training. Using the data set prepared, the YOLO series model is trained. In the model training process, the main parameter settings are as follows: the training round is 150 rounds, the batch size is 32, the input image resolution is 640*640 pixels, the optimizer adopts SGD, and the pre-trained weight is loaded to accelerate convergence. The initial learning rate lr0 is set to 0.01, the maximum learning rate factor lrf is 0.01, the momentum is 0.937, the weight decay is 0.0005, and the learning rate and momentum preheating strategy is used for the first 3 rounds of training. The loss function weight parameter settings are as follows: the target box loss (box) is 7.5, the category loss (cls) is 0.5, and the distributed regression loss (dfl) is 1.5. In addition, data enhancement strategies are used during training, including translation (translate = 0.1), scaling (scale = 0.5), random left-right flipping (fliplr = 0.5), color disturbance (hsv_h = 0.015, hsv_s = 0.7, hsv_v = 0.4), etc., to further improve the generalization ability of the model. The training process adopts the early stopping strategy (patience = 100), and the model performance is monitored on the validation set to ensure that the final model achieves the optimal detection effect.

[0122] (4) MSTIDs instance segmentation data set preparation. The DTEC image with a resolution of 640*640 pixels is input into the trained YOLO target detection model for target detection, and the Bbox frame label containing the wave-like disturbance structure is identified and obtained. Then, the DTEC image group containing the Bbox frame is selected, and the continuous image pairs with IoM greater than 0.5 are calculated and selected. According to the corresponding Bbox frame coordinates, the image is cropped to obtain an image containing only the MSTIDs red and blue blocks. After the above steps, only the MSTIDs wave-like color block structure is retained, and the present application selects 236 cropped pictures of DOY = 71-75 in 2018, 2022 and 2024 for instance segmentation annotation, which are used for YOLO model training. The MSTIDs segmentation data set also needs to adjust the image resolution to 640*640 before annotation, and is annotated and prepared online through the roboflow website. The training set and validation set of the MSTIDs instance segmentation data set are divided in the ratio of 8:2.

[0123] (5) MSTIDs color block instance segmentation model training. Using the annotated instance segmentation dataset, the YOLO series instance segmentation model is trained. The main parameter settings in the model training process and the target detection model parameter settings in step (3) are consistent. The loss function weight parameter settings are as follows: the target box loss (box) is 7.5, the class loss (cls) is 0.5, the distributed regression loss (dfl) is 1.5, and the mask segmentation related parameters (overlap_mask, mask_ratio, etc.) are set according to the model requirements. The data enhancement and early stopping strategy used in the training process are consistent with the target detection training process strategy.

[0124] In order to comprehensively evaluate the performance of the MSTIDs detection and segmentation model, this paper adopts the mean average precision (mAP) as the core evaluation index, focusing on mAP 50 and mAP 50-95 two sub-items. In the target detection or instance segmentation task, four situations will be formed between each prediction result and the true annotation: TP: the IoU of the MSTIDs prediction box and the true box is ≥ threshold and the class prediction is correct; FP: there is a prediction box, but it is not matched with any true box, or the class prediction is wrong; FN: the true box exists, but it is not detected, which is judged as a false negative example; TN: not applicable to the detection task, usually ignored. According to TP, FP and FN, the precision P and recall R can be calculated, and the calculation formula is:

[0125] (13)

[0126] The P-R curve is drawn with different confidence thresholds, and then the area under the curve is calculated as the average precision AP:

[0127] (14)

[0128] mAP 50 When the IoU threshold is set to 0.5, the average of AP of all classes is defined as:

[0129] (15)

[0130] In this invention, the MSTIDs target detection only contains one class, so N=1. In the instance segmentation task, the MSTIDs color block is subdivided into two classes of red block and blue block, so N=2. AP i is the AP value of the ith class.

[0131] mAP 50-95 Then a more stringent evaluation standard is adopted, the IoU threshold is from 0.5 to 0.95, the step is 0.05, a total of 10 thresholds, the AP of all classes at each threshold is calculated and averaged, the formula is:

[0132] (16)

[0133] Figure 2 The results of the MSTIDs structure target detection in a set of consecutive image pairs are shown. It can be seen that the Bbox area difference in 10-minute intervals is large, so the present application adopts the IoM index more suitable for the present study to screen the consecutive image pairs.

[0134] Figure 3 The change curves of various indexes of the MSTIDs detection task based on the YOLOv8m model in the training and verification processes are shown. The loss of the regression frame, the class classification and the distribution focus (DFL) in the training stage and the verification stage all show a continuous downward trend, among which the class loss converges rapidly, which may be related to the small number of classes in the task. At the same time, the evaluation indexes such as P, R, mAP 50 and mAP 50-95 steadily improve, indicating that the model detection performance is continuously optimized and has good generalization ability. The final mAP 50 is about 0.77, and the mAP 50-95 is about 0.51, and there is a certain gap between them, indicating that the model performs better when the IoU threshold is 0.5. Therefore, the present application sets the IoU to 0.5 in the detection process.

[0135] Figure 4 The change curves of various indexes of the MSTIDs instance segmentation task based on the YOLOv8m-seg model in the training and verification processes are shown. Overall, the segmentation loss (seg_loss), the class classification loss (cls_loss) and the distribution focus loss (dfl_loss) in the training and verification stages all show a downward trend, reflecting that the model gradually converges and the learning effect continuously improves in the training process. However, in the early training stage (about the first 20 epochs) and some stages in the middle and late training stages, there are obvious oscillations and local rises in the loss curves, especially in the val / seg_loss and val / cls_loss indexes. This may be caused by the fact that individual batches in the training contain samples that are difficult to learn or the local oscillation of the momentum update of the optimizer in the parameter space. In terms of performance evaluation indexes, P, R, mAP 50 and mAP 50-90 all continuously improve with the increase of the training rounds, indicating that the model detection performance is continuously optimized and has good generalization ability. Finally, on the validation set, the model achieves an mAP 50 of more than 0.90 and an mAP 50-90, which indicates that the detection effect is better at IoU=0.5.

[0136] Figure 5 Five typical MSTIDs structures are shown in the color block segmentation effect comparison of the YOLOv8m instance segmentation model. The first row is the original TEC disturbance image, which is used as the segmentation reference. As can be seen from the figure, the YOLOv8m model can accurately identify the red and blue wave peak regions, and the segmented color blocks have clear outlines and strong coherence. The model also has good recognition ability for weak disturbance regions, and the overall segmentation quality and physical consistency are better, which is beneficial to subsequent color block feature extraction and matching.

[0137] Figure 6 and Figure 7 The propagation parameter inversion results of the MSTIDs intelligent detection and matching method and the cross-spectrum method proposed in the application during 13:40-13:50 on January 22, 2024 are shown in a and b, respectively. Figure 6 The MSTIDs propagation parameters calculated by fitting an ellipse are: propagation speed 253.68 m / s, propagation azimuth 247.76°, period 28.21 minutes, wavelength 429.43 km, and total coverage area 36,009.62 km². Figure 7 a shows the fluctuation structure of MSTIDs in the DTEC image, and there is a clear phase difference between the stations with an interval of about 100 kilometers. Figure 7 b to Figure 7 f respectively show the changes of the propagation parameters with UT time, wherein at about 13:45, the propagation speed of the MSTIDs event is about 214.43 m / s, the propagation azimuth is 252.87°, the period is 40.6 minutes, and the wavelength is 534.34 km. It can be seen that the MSTIDs propagation parameters calculated by different methods have good consistency in general, and the errors of other parameters are less than 20% except the period. The difference in parameters is mainly due to the fact that the method of the application inverts the overall propagation characteristics of MSTIDs in a large range, while the cross-spectrum method only reflects the ionospheric disturbance characteristics in a local area of three stations.

[0138] Embodiment 2

[0139] The application discloses an ionospheric traveling wave disturbance intelligent identification and propagation parameter calculation system, comprising:

[0140] An image acquisition and adjustment module is configured to acquire DTEC image data of the ionosphere in a dense observation network, and to obtain adjusted DTEC images after adjusting the resolution of the DTEC image data.

[0141] The MSTIDs detection module is configured to input the adjusted DTEC image into a trained YOLO target detection model for target detection, and identify and obtain a Bbox frame label containing a wave-like disturbance structure;

[0142] The MSTIDs color block instance segmentation module is configured to perform pre-processing on the DTEC image in the Bbox frame label before YOLO instance segmentation, to obtain a segmented image in continuous time, and then input the segmented image in continuous time into a YOLO instance segmentation model for accurate segmentation of color blocks to obtain an image containing a color block mask;

[0143] The MSTIDs feature extraction module is configured to use an OpenCV library to calculate five-dimensional features of the color block mask, including color block color, color block area, centroid coordinates of the mask, a Hu matrix and a topological relationship between color blocks, to obtain five features.

[0144] The color block mask matching module is configured to match the segmented image in continuous time using the five features, calculate a matching score based on the weight of each feature, and then obtain a best matching result of the color block mask based on the matching score using a Hungarian algorithm.

[0145] The MSTIDs propagation parameter calculation module is configured to perform an ellipse fitting on the successfully matched color block mask in the best matching result, convert pixel coordinates of the centroid of the mask into geographic coordinates, calculate a horizontal velocity, an azimuth angle and a coverage area of the MSTIDs based on the size, moving direction and distance of the standardized ellipse circumscribing the mask and a time interval, and calculate a wavelength and a period parameter of the MSTIDs based on the feature relationship between different color blocks in the image containing the color block mask.

[0146] Embodiment 3

[0147] The embodiment discloses a computer readable storage medium, which stores a computer program / instruction, and the computer program / instruction is executed by a processor to implement the ionospheric traveling wave disturbance intelligent identification and propagation parameter calculation method in the embodiment 1 or 2.

[0148] The present application is described with reference to flowcharts and / or block diagrams according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device for implementing the functions described in the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocks Figure 1means for performing the function specified by the block or blocks.

[0149] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 flow or flows and / or blocks Figure 1 means for performing the function specified by the block or blocks.

[0150] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 flow or flows and / or blocks Figure 1 means for performing the function specified by the block or blocks.

[0151] The above description is merely that of the preferred embodiments of the application and modifications and alterations are possible to those having ordinary skill in the art without departing from the spirit or scope of the application.

Claims

1. An ionospheric traveling wave disturbance intelligent identification and propagation parameter calculation method, characterized in that: The method comprises the following steps: DTEC image data of an ionosphere in a dense observation network is acquired, and after resolution adjustment of the DTEC image data, adjusted DTEC images are obtained; the adjusted DTEC images are input into a trained YOLO target detection model for target detection, and a Bbox frame label containing a wave-like disturbance structure is recognized and obtained; preprocessing of the DTEC images in the Bbox frame label before YOLO instance segmentation is performed to obtain segmented images of continuous time, and the segmented images of continuous time are input into a YOLO instance segmentation model for accurate segmentation of color blocks to obtain images containing color block masks; five-dimensional feature calculation of the color block masks is performed using an OpenCV library, including color block color, color block area, mask centroid coordinates, Hu matrix and topological relationship between color blocks, to obtain five features; the segmented images of continuous time are matched using the five features, a matching score is calculated based on the weights of the features, and the best matching result of the color block masks is obtained based on the matching score using a Hungarian algorithm; the successfully matched color block masks in the best matching result are subjected to an external ellipse fitting, and the pixel coordinates of the mask centroids are converted into geographic coordinates, the horizontal velocity, azimuth angle and coverage area of the MSTIDs are calculated through the size, moving direction, distance and time interval of the standardized external ellipse of the mask, and the wavelength and period parameters of the MSTIDs are calculated using the feature relationship between different color blocks in the images containing the color block masks.

2. The ionospheric traveling wave perturbation intelligent identification and propagation parameter calculation method according to claim 1, characterized in that: The preprocessing of the DTEC images in the Bbox frame label before YOLO instance segmentation to obtain segmented images of continuous time comprises: a group of time-continuous DTEC images is screened out; an overlapping area index IoM of the Bbox frames in two continuous images in the DTEC image group is calculated; a pair of continuous images satisfying a set range of the IoM is selected, and the images in the pair of continuous images are cropped according to the corresponding Bbox frame coordinates to obtain images containing only the red and blue color blocks of the MSTIDs, and finally segmented images of continuous time are obtained; wherein the overlapping area index IoM is the ratio of the overlapping area of the two Bbox frames to the smaller area Bbox, and the calculation formula is as follows: （1）， in the formula, A and B represent the areas of the two Bbox frames respectively, the threshold value of the IoM is set to 0.5, and only a pair of continuous images satisfying IoM≥0.5 will be selected.

3. The ionospheric traveling wave perturbation intelligent identification and propagation parameter calculation method according to claim 1, characterized in that: The color block area is calculated based on the irregular polygon contour of the mask using a shoelace formula, and the specific expression is as follows: （2）， In the formula, C is the area of the polygon, k is the number of coordinates of the vertices of the polygon, and is the horizontal and vertical coordinate of the s+1th vertex, and is the horizontal and vertical coordinate of the s+2th vertex.

4. The ionospheric traveling wave perturbation intelligent identification and propagation parameter calculation method according to claim 1, characterized in that: The centroid coordinates of the mask are obtained through the geometric moments of the image, and the geometric moment is defined as: （3）， wherein is the geometric moment, p and q are the order of the moment, and x and y are the horizontal and vertical coordinates of the two-dimensional image, respectively, is the pixel point in the binary image, the centroid of the color block is represented as: （4）， and are the horizontal and vertical coordinates of the centroid of the color patch, respectively.

5. The ionospheric traveling wave perturbation intelligent identification and propagation parameter calculation method according to claim 4, characterized in that: The Hu matrix is obtained by calculating the central moment to remove the influence of translation, and the calculation formula is as follows: （5）， wherein is the centralized geometric moment, and the central distance is normalized to eliminate the scale effect: （6）， wherein denotes the normalized central moments, then the Hu matrix is expressed as: （7）， （8）， wherein , , , , , and are the invariants of the matrix respectively.

6. The ionospheric traveling wave perturbation intelligent identification and propagation parameter calculation method according to claim 1, characterized in that: The topological relationship between the color blocks is obtained by comparing the coordinate difference of the two color block centroids in the x direction and the y direction: if the absolute value of the x direction difference is greater than the y direction, it is judged as left or right according to the positive or negative of the x difference; otherwise, it is judged as up or down according to the positive or negative of the y difference.

7. The ionospheric traveling wave perturbation intelligent identification and propagation parameter calculation method according to claim 1, characterized in that: The matching score is calculated based on the weights of the features, and the best matching result of the color block masks is obtained based on the matching score using a Hungarian algorithm. In the matching process, first set the segmentation of the continuous image pairs of the MSTIDs color block set respectively and , the matching score between the color block and is expressed as: ​ （9）， wherein, , and are the barycentric coordinates, the area and the eigen-difference of the Hu matrix between the color blocks and , respectively, and can be expressed as: （10）， wherein, and are the barycentric coordinates of the color patches and respectively; and are the areas of the color patches and respectively; and are the Hu matrices of the color patches and respectively; is the topological relationship penalty term, if the adjacent color patches of two color patches are not consistent in color or relative direction, then otherwise ; , , and represent the weights of each feature difference respectively.

8. The ionospheric traveling wave perturbation intelligent identification and propagation parameter calculation method according to claim 7, characterized in that: ​ Mstid tile set and constructing a similarity matrix; The Hungarian algorithm is used to find the best match from all possible matches The similarity matrix selects a match set from all possible matches The sum of scores of all selected matches is minimized, where the Hungarian algorithm criterion is expressed as: （11）， After the matching is completed, validity matching is performed, and the screening condition is: （12）， In the formula, T is a screening score threshold.

9. The ionospheric traveling wave perturbation intelligent identification and propagation parameter calculation method according to claim 1, characterized in that: The conversion of the pixel coordinates of the mask center into geographical coordinates comprises: Selecting control points to determine the pixel points in the image and the corresponding latitude and longitude coordinates; Converting the known latitude and longitude coordinates into coordinate points in a projection coordinate system, and using a nonlinear polynomial fitting method to establish a nonlinear correspondence between the pixel coordinates in the image and the projection coordinates to form a coordinate conversion model; After the fitting is completed, the projection coordinates are reversely converted into latitude and longitude coordinates based on the coordinate conversion model.

10. An ionospheric traveling wave perturbation intelligent identification and propagation parameter calculation system, characterized in that: Comprise: An image acquisition and adjustment module for acquiring DTEC image data of an ionosphere in a dense observation network, and then obtaining an adjusted DTEC image after adjusting the resolution of the DTEC image data; An MSTIDs detection module for inputting the adjusted DTEC image into a trained YOLO target detection model for target detection to identify and obtain a Bbox frame label containing a wave-like disturbance structure; An MSTIDs color block instance segmentation module for pre-processing the DTEC image in the Bbox frame label before YOLO instance segmentation to obtain a segmented image of continuous time, and then inputting the segmented image of continuous time into a YOLO instance segmentation model for accurate segmentation of color blocks to obtain an image containing color block masks; An MSTIDs feature extraction module for using an OpenCV library to calculate five-dimensional features of the color block masks, including color block color, color block area, mask center coordinates, Hu matrix, and topological relationship between color blocks, to obtain five features; A color block mask matching module for matching the segmented image of continuous time using the five features, calculating a matching score based on the weight of each feature, and then obtaining the best matching result of the color block masks based on the matching score using the Hungarian algorithm; An MSTIDs propagation parameter calculation module for fitting a circumscribed ellipse for the successfully matched color block masks in the best matching result, converting the pixel coordinates of the mask center into geographical coordinates, calculating the horizontal velocity, azimuth angle, and coverage area of the MSTIDs by the size, moving direction, distance, and time interval of the standardized circumscribed ellipse of the mask, and then calculating the wavelength and period parameters of the MSTIDs using the feature relationship between different color blocks in the image containing the color block masks.

Images