Typhoon identification tracking and prediction method based on ground wave radar wind field data

Abstract

The invention relates to the technical field of marine environment monitoring, in particular to a typhoon identification tracking and prediction method based on ground wave radar wind field data, which can acquire the moving track and speed of a typhoon in real time and provide timely and accurate information support for disaster prevention and reduction work. According to the typhoon identification tracking and prediction method based on the ground wave radar, radar data are processed and analyzed through an image processing technology and a mode identification algorithm, and the position change of the typhoon wind eye center can be efficiently tracked by using the wind eye center with the minimum distance in continuous time as a tracking object and combining a wind eye center identification model.

Description

Technical Field

[0001] The present invention relates to the technical field of marine environment monitoring, and in particular to a typhoon identification, tracking and prediction method based on ground wave radar wind field data. Background Art

[0002] Typhoons are a common extreme weather phenomenon, and the disasters they bring, such as strong winds, heavy rains, and storm surges, pose a serious threat to human activities, economic development, and the ecological environment in coastal areas. Therefore, accurately and timely identifying and tracking typhoons, as well as predicting their movement paths and intensity changes, are crucial for disaster prevention and mitigation. In recent years, with the continuous development of radar technology, ground wave radar, as a new type of sea surface detection tool, has shown great potential in the field of typhoon monitoring.

[0003] Ground wave radar is a radar system that uses high-frequency electromagnetic waves propagating along the sea surface for long-range detection. Compared with traditional microwave radar, ground wave radar has the advantages of long detection distance, wide coverage, and strong anti-interference ability, and is particularly suitable for marine environmental monitoring. The ocean echo signal reflected by the waves is processed, and environmental information such as sea surface wind, waves, and flow fields can be extracted from the first-order and second-order Doppler spectra generated.

[0004] Traditional typhoon identification and tracking mainly rely on satellite remote sensing and manual identification methods. However, these methods have certain limitations, such as low data update frequency, great influence by weather conditions, and strong subjectivity. Summary of the invention

[0005] In view of this, the present invention provides a typhoon identification, tracking and prediction method based on ground wave radar wind field data, which can obtain the movement trajectory and speed of the typhoon in real time, and provide timely and accurate information support for disaster prevention and mitigation work.

[0006] To achieve the above purpose, the technical solution of the present invention is as follows:

[0007] A typhoon identification, tracking and prediction method based on ground wave radar wind field data, obtaining sea surface wind field information within a detection area through a ground wave radar system; and drawing a wind field map based on the wind field information;

[0008] Constructing a wind circle recognition model; marking wind circles in the wind field map to obtain a wind circle detection data set, and using the wind circle detection data set to train the wind circle recognition model; using the trained wind circle recognition model to perform wind circle detection to obtain a wind circle detection result;

[0009] Construct an eye center recognition model; use the wind circle detection result to mark the eye center position in the wind field map to obtain an eye center detection data set; use the eye center detection data set to train the eye center recognition model; use the trained eye center recognition model to track the eye center, where the eye center with the smallest distance in continuous time is the tracking object;

[0010] Constructing a typhoon eye trajectory prediction model; using the tracked longitude and latitude trajectory of the eye center as a typhoon trajectory prediction data set, and using the typhoon trajectory prediction data set to train the typhoon eye trajectory prediction model; using the trained typhoon eye trajectory prediction model to predict the typhoon eye trajectory and analyze the results;

[0011] Among them, the identified typhoon eye position and wind field data are combined to obtain the typhoon center wind speed, rotation direction, wind circle radius, moving direction and speed.

[0012] Among them, the Yolov5 neural network model framework is used to build a wind circle identification model.

[0013] Among them, the wind circle recognition model includes input, Backbone, Neck, Head and output; Among them, Backbone is responsible for extracting the features of the input image, including Conv module, C3 module and SPPF module, Conv module includes convolution layer, BN layer and activation function, C3 module adaptively aggregates the input feature map, SPPF module obtains spatial information through weighted fusion of global features and local features; Neck performs multi-scale feature fusion on the features extracted by Backbone and passes these features to the prediction layer; Head is responsible for the final regression prediction; Among them, a layer of SE attention mechanism is added before SPPF in Backbone.

[0014] Among them, the wind circle recognition model includes a level-7 wind circle recognition model and a level-10 wind circle recognition model; the level-7 wind circles and level-10 wind circles in the wind field map are marked to obtain a level-7 wind circle detection data set and a level-10 wind circle detection data set; the level-7 wind circle recognition model is trained using the level-7 wind circle detection data set, and the level-10 wind circle recognition model is trained using the level-10 wind circle detection data set; level-7 wind circle detection is performed using the trained level-7 wind circle recognition model, and level-10 wind circle detection is performed using the trained level-10 wind circle recognition model; the wind circle detection results include level-7 wind circle detection results and level-10 wind circle detection results.

[0015] Among them, the Hrnet neural network model framework is used to build a storm eye center identification model.

[0016] Among them, Hrnet predicts the key point of the eye center based on the heatmap method; Hrnet first passes through two convolutional layers with a kernel size of 3×3 and a stride of 2, and then passes through the Layer1 module, which is composed of repeatedly stacked Bottleneck, and then passes through a series of transition structures and stage structures. The transition adds a branch of a scale, and the stage fuses information of different scales; after the last stage fusion branch, it enters a convolutional layer with a kernel size of 3×3, a stride of 1, and a channel number of 1, and finally outputs a feature map of 1 channel, that is, the center position of the eye.

[0017] Among them, the LSTM neural network model framework is used to construct a typhoon eye trajectory prediction model. By introducing input gates, forget gates, output gates and cell states, LSTM processes long-term dependencies in the sequence; the eye trajectory prediction model performs single-step and multi-step predictions on the eye trajectory; among them, wind field maps with typhoon eye in continuous time periods are automatically grouped as input to the eye trajectory prediction model to perform single-step and multi-step predictions on the eye trajectory.

[0018] Among them, when the typhoon lands in the detection range of the ground wave radar, the eye center recognition model detects the center position of the eye, and judges and combines them according to the minimum eye distance between consecutive times to form a typhoon grouping; when the eye center position of 5 time periods is detected, these 5 time series are input into the LSTM to predict the eye center position of the next 3 time periods. When the eye detection and eye center detection neural networks detect the new eye center position at the next moment, the eye center position of the next 3 time periods is predicted again according to the previous steps.

[0019] The sea surface wind field information includes the wind speed, wind direction and confidence level of each grid in the radar detection area.

[0020] The tracked position of the eye center is mapped to the wind field map, and the moving direction and speed of the typhoon within a set time are calculated based on the frequency of the wind field data.

[0021] Beneficial effects:

[0022] 1. The present invention is a typhoon identification, tracking and prediction method based on ground wave radar. The ground wave radar can obtain sea surface wind field information in real time, providing timely data support for typhoon identification and tracking; the method of the present invention processes and analyzes radar data through image processing technology and pattern recognition algorithm, and by using the eye center with the smallest distance in continuous time as the tracking object, combined with the eye center recognition model, it can efficiently track the position change of the typhoon eye center. This method not only reduces the amount of calculation, but also improves the real-time tracking, can accurately extract characteristic information such as the typhoon eye, and improve the accuracy of identification and tracking. The present invention establishes an identification model and a prediction model through machine learning or deep learning algorithms, which can realize the automatic identification and prediction of typhoons, improve the prediction efficiency and accuracy, and provide timely and accurate information support for disaster prevention and mitigation. Deep learning eye detection based on computer artificial intelligence can realize eye detection anytime and anywhere, while being compatible with calculation efficiency and calculation accuracy. By comprehensively utilizing the sea surface wind field information obtained by the ground wave radar system, combined with advanced machine learning and deep learning models, the accuracy and efficiency of typhoon identification, tracking and prediction are significantly improved.

[0023] 2. The present invention uses the Yolov5 neural network model framework to build a wind circle recognition model, which can identify and mark wind circles with high accuracy. In particular, in terms of wind circle detection, the model can distinguish wind circles of different levels, such as level 7 wind circles and level 10 wind circles, thereby improving the accuracy of wind circle detection.

[0024] 3. The present invention uses the Hrnet neural network model framework to construct an eye center identification model, which can accurately locate the eye center position. The Hrnet model predicts the key points of the eye center based on a heatmap, thereby improving the robustness and accuracy of the eye center detection. When the typhoon lands within the detection range of the ground wave radar, the eye center identification model can detect the eye center position and form a typhoon group by judging and combining the minimum distance between consecutive times. This method improves the accuracy of typhoon grouping and helps to more accurately predict the future path of the typhoon.

[0025] 4. The present invention uses the LSTM neural network model framework to construct a typhoon eye trajectory prediction model, which can accurately predict the future position of the typhoon eye. The LSTM model can handle long-term dependencies in the sequence and improve the accuracy of typhoon trajectory prediction through single-step and multi-step predictions.

[0026] 5. In the method of the present invention, by combining the identified typhoon eye position with the wind field data, key information such as the typhoon center wind speed, rotation direction, wind circle radius, moving direction and speed can be obtained. This information is of great significance for the comprehensive analysis and early warning of typhoons.

[0027] 6. The method of the present invention can timely update the status information of the typhoon by processing the wind speed, wind direction and confidence of each grid in the radar detection area in real time, thereby improving the real-time performance and reliability. By automatically processing the wind circle and eye center in the wind field map, the need for manual intervention is reduced, and the automation level and work efficiency of the system are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of data set labeling in an embodiment of the present invention.

[0030] Figure 3 Schematic diagram of wind eye center, wind circle identification and trajectory tracking according to an embodiment of the present invention.

[0031] Figure 4 This is a schematic diagram of the prediction of the eye trajectory according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0033] Ground wave radar technology can realize real-time monitoring of sea surface wind fields, providing a new means for typhoon identification and tracking. In the typhoon identification and tracking method based on ground wave radar, the sea surface wind field information in the detection area is first obtained through the ground wave radar system, including parameters such as wind speed and wind direction. Then, the radar data is processed and analyzed using image processing technology and pattern recognition algorithms to extract characteristic information such as the typhoon eye. Then, the typhoon is tracked by the principle of minimum eye distance between continuous times. This method can obtain the movement trajectory and speed of the typhoon in real time, providing timely and accurate information support for disaster prevention and mitigation work.

[0034] This embodiment provides a typhoon identification, tracking and prediction method based on ground wave radar wind field data. The process is as follows: Figure 1 As shown, the following steps are included:

[0035] Step 1: Obtain sea surface wind field information within the detection area through the ground wave radar system, including wind speed, wind direction, confidence level and other information of each grid in the radar detection area.

[0036] The ground wave radar is a radar system that uses high-frequency electromagnetic waves propagating along the sea surface for long-range detection. Compared with traditional microwave radar, ground wave radar has the advantages of long detection distance, wide coverage, strong anti-interference ability, etc., and is particularly suitable for marine environmental monitoring. The ocean echo signal reflected by the waves is processed, and environmental information such as sea surface wind, waves, and flow fields can be extracted from the first-order and second-order Doppler spectra generated.

[0037] Step 2, preprocessing the data information obtained in step 1, that is, the wind field drawing software draws a wind field map according to the wind field information, the size and color of the arrow symbol in the wind field represent the wind speed, and the direction of the arrow symbol represents the wind direction.

[0038] Step 3, mark the level 7 wind circle and level 10 wind circle in the wind field map obtained in step 2, produce a level 7 wind circle and level 10 wind circle detection data set, and divide the data set into a training set, a test set, and a validation set. Figure 2 This is a schematic diagram of data set labeling in an embodiment of the present invention.

[0039] Among them, the data set division is to divide 50% of the data into a training set, 30% of the data into a test set, and 20% of the data into a validation set. In this embodiment, a total of 5000 wind field maps were obtained in step 2, of which 3500 contained wind field maps with eye of the wind. Use labelimg software to select the eye of the wind and produce label data. Randomly select 2500 wind field maps as a training set, 1000 wind field maps as a test set, and 1500 wind field maps as a validation set.

[0040] Step 4: Build an air outlet recognition model based on the Yolov5 neural network framework, add the SE attention mechanism on the basis of the Yolov5 neural network framework to increase the accuracy of the network. Input the data set obtained in step 3 into the neural network for model training, and perform 100 rounds of training.

[0041] Step 5: Use the neural network model trained in step 4 to detect wind circles and analyze the results. After 100 rounds of training, the box_loss, obj_loss, and cls_loss of the training and test sets all reach a stable state, with mAP_0.5 reaching above 0.95 and mAP_0.5:0.95 reaching above 0.9. The validation set is input into the trained model. There are 800 wind field maps containing wind circles in the validation set, and 785 of them have wind circles accurately detected.

[0042] Step 6: Because the wind circle may be an asymmetric shape in the wind field map, after selecting the wind circle in step 5, it is necessary to further detect the center position of the eye. Use the wind circle detection result in step 5 to mark the center position of the eye in the wind field map, select the wind field map containing the eye, and cut out the eye image. There are 3500 eye images in total, and use labelme to mark the center position of the eye. Randomly select 1500 data as the training set, 800 data as the test set, and 1200 data as the validation set to obtain the eye center detection data set.

[0043] Step 7: construct a storm eye center recognition model based on the Hrnet neural network framework, and use the storm eye center detection data set obtained in step 6 to train the storm eye center recognition model. In the process of training the network, a strategy of gradually reducing the learning rate is adopted. The learning rate suitable for the network can make the loss converge quickly during the training process. In order to avoid overfitting and underfitting, multiple parameter adjustments will be made and the initial values ​​of the weights will be initialized with a standard normal distribution. After 200 epochs of training, the mean square error loss function values ​​of the training set and the test set reach a balance.

[0044] Step 8: Input the validation set into the trained model to obtain the center position of each image in the validation set and map it back to the longitude and latitude coordinates in the non-wind field map. Since the distance resolution of the ground wave radar wind field data is 5km, the Euclidean distance between this coordinate and the marked coordinate is less than 2.5km.

[0045] Step 9, map the level 7 wind circle and level 10 wind circle detection frames obtained by Yolov5 back to the latitude and longitude coordinates in the wind field map, and obtain the radius of the level 7 wind circle and level 10 wind circle according to the position of the wind circle detection frame and the center of the eye of the wind.

[0046] Step 10, using the trained eye center recognition model to track the eye center, wherein the eye center with the smallest distance in the continuous time is the tracking object;

[0047] The position of the typhoon eye center at each continuous time is tracked, the position of the eye center is mapped to the original image, and the moving direction and speed of the typhoon in a short time are calculated; wherein, multiple typhoon moving directions and speeds can be calculated according to each continuous time. In order to ensure the accuracy and stability of the calculation of the typhoon moving direction and speed, in this embodiment, based on the frequency of 10 minutes / time of wind field data, 6 times are accumulated each time, and the typhoon moving direction and speed are calculated as the final result.

[0048] Step 11, obtain the longitude and latitude trajectory data of the tracked eye center, including the trajectory data of Typhoon Surakarta and Typhoon Tali. Save each moment of each typhoon eye and the corresponding longitude and latitude coordinates in the form of a csv file to form a typhoon trajectory prediction data set. Divide 50% of the data set into a training set, 30% of the data into a test set, and 20% of the data into a validation set.

[0049] Step 12: Build a typhoon eye trajectory prediction model based on LSTM and use the typhoon trajectory prediction dataset for training. Take longitude and latitude as input, and predict the positions of the next three time points for every five time points.

[0050] In step 13, the verification set is input into the neural network model trained in step 12 to predict the typhoon eye trajectory. The overall predicted trajectory has a small deviation from the actual trajectory, which proves the reliability and accuracy of the method of the present invention. Figure 3 This is a schematic diagram of wind eye center, wind circle identification and trajectory tracking according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the prediction of the eye trajectory according to an embodiment of the present invention.

[0051] Furthermore, the present invention can use three neural networks, namely, a wind circle recognition model based on Yolov5, a wind eye center recognition model based on Hrnet, and a wind eye trajectory prediction model based on LSTM.

[0052] Among them, the Yolov5 neural network model is a single-stage target detection network. The attention mechanism is added to the network, which can detect the location of wind circles in the wind field map with high efficiency and accuracy. The wind circle recognition model, i.e., the circle detection neural network construction principle, is as follows:

[0053] The model is mainly composed of input, Backbone, Neck, Head, and output. The backbone is responsible for extracting the features of the input image, and its structure mainly includes Conv module, C3 module, and SPPF module. The Conv module is mainly composed of convolution layer, BN layer, and activation function. The C3 module adaptively aggregates the previous feature map. The SPPF module obtains more comprehensive spatial information through weighted fusion of global features and local features. The Neck part is responsible for multi-scale feature fusion of the features extracted by the Backbone and passes these features to the prediction layer. The Head is mainly responsible for the final regression prediction. In order to improve the accuracy of the model, the model adds a layer of SE attention mechanism before SPPF in the Backbone.

[0054] The construction principle of the eye center recognition model, i.e. the eye center detection neural network:

[0055] Hrnet predicts key points based on heatmap. Hrnet first passes through two convolutional layers with a kernel size of 3×3 and a step size of 2, and then passes through the Layer1 module. The Layer1 module consists of repeatedly stacked Bottlenecks, and then passes through a series of transition structures and stage structures. The transition will add a branch of a scale, and the stage will fuse information of different scales. After the last stage fusion branch, it will enter a convolutional layer with a kernel size of 3×3, a step size of 1, and a channel number of 1, and finally output a feature map of 1 channel, that is, the center position of the eye of the wind. The Hrnet neural network model used in this embodiment is a high-resolution network. The output of the network will be changed from the original 17 key points to 1 key point, that is, the position of the center of the eye of the wind. The eye center recognition model can detect multiple eyes in a wind field data. In the experiment, 5,000 wind field data were used. The direct use of the model can achieve the detection of wind field data of this order of magnitude in about 5 minutes.

[0056] The eye track prediction model, i.e. the eye track neural network construction principle:

[0057] LSTM is a special RNN network, which is often used in deep learning models to process sequence data. By introducing input gates, forget gates, output gates and cell states, LSTM can better handle long-term dependencies in sequences. First, when the typhoon lands in the detection range of the ground wave radar, the eye detection and eye center detection neural networks will detect the eye center position. After the eye center position of 5 time periods is detected, these 5 time series are input into LSTM to predict the eye center position of the next 3 time periods. When the eye detection and eye center detection neural networks detect the new eye center position at the next moment, the eye center position of the next 3 time periods is predicted again according to the previous steps.

[0058] In summary, in terms of weather forecasting, traditional meteorological forecasting methods mainly rely on numerical weather forecast models (NWP). However, the NWP model has certain uncertainties in typhoon forecasting, such as uncertainty in initial conditions, uncertainty in model parameters, etc. Ground wave radar technology can provide real-time and accurate sea surface wind field information, providing a new data source for typhoon forecasting. In the typhoon forecasting method based on ground wave radar of the present invention, radar data is processed and analyzed using machine learning or deep learning algorithms to establish a typhoon forecasting model. The model can predict the movement path and intensity changes of typhoons in the future period of time based on historical typhoon data and real-time radar data. By continuously updating and optimizing the forecasting model, the accuracy and reliability of the forecast can be improved. The advantage of the method of the present invention is that for traditional typhoon identification and tracking methods, which generally use manual discrimination methods, the method of the present invention can automatically complete the typhoon eye identification, obtain the typhoon center wind speed, rotation direction, wind circle size, and judge and combine the eye of continuous time periods, and obtain the typhoon movement trajectory, movement direction and speed in real time. At the same time, the typhoon eye prediction model can be used to predict single-step and multi-step typhoon trajectories, so as to achieve the purpose of early warning and disaster prevention and mitigation. In addition, considering that the frequency of ground wave radar data is higher than that of meteorological satellite data, it can obtain information such as the size, location, moving direction and speed of the typhoon more quickly; at the same time, compared with the wind field data of meteorological satellite data at an altitude of several kilometers to tens of kilometers above the ground, the wind field data detected by ground wave radar is generally within 10 meters above the sea surface, which is closer to the data needs of fishing boats and other ships, because the typhoon eye information of meteorological satellites and the sea surface will have a certain degree of offset and size. The typhoon eye obtained by the typhoon identification, tracking and prediction method based on ground wave radar may be more conducive to the relevant marine departments to issue relevant warnings and disaster prevention and mitigation information, and is more in line with actual needs.

[0059] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A typhoon identification, tracking and prediction method based on ground wave radar wind field data, characterized in that: Acquire sea surface wind field information within the detection area through a ground wave radar system; draw a wind field map according to the wind field information; construct a wind circle recognition model; mark the wind circles in the wind field map to obtain a wind circle detection data set, and use the wind circle detection data set to train the wind circle recognition model; use the trained wind circle recognition model to perform wind circle detection to obtain a wind circle detection result; Construct an eye center recognition model; use the wind circle detection result to mark the eye center position in the wind field map to obtain an eye center detection data set; use the eye center detection data set to train the eye center recognition model; use the trained eye center recognition model to track the eye center, where the eye center with the smallest distance in continuous time is the tracking object; Constructing a typhoon eye trajectory prediction model; using the tracked longitude and latitude trajectory of the eye center as a typhoon trajectory prediction data set, and using the typhoon trajectory prediction data set to train the typhoon eye trajectory prediction model; using the trained typhoon eye trajectory prediction model to predict the typhoon eye trajectory and analyze the results; Among them, the identified typhoon eye position and wind field data are combined to obtain the typhoon center wind speed, rotation direction, wind circle radius, moving direction and speed.

2. The method according to claim 1, characterized in that The Yolov5 neural network model framework is used to build the wind circle identification model.

3. The method according to claim 2, characterized in that The wind circle recognition model includes input, Backbone, Neck, Head and output; the Backbone is responsible for extracting the features of the input image, including Conv module, C3 module and SPPF module. The Conv module includes convolution layer, BN layer and activation function. The C3 module adaptively aggregates the input feature map. The SPPF module obtains spatial information through weighted fusion of global features and local features. Neck performs multi-scale feature fusion on the features extracted by Backbone and passes these features to the prediction layer. Head is responsible for the final regression prediction. In the Backbone, a layer of SE attention mechanism is added before SPPF.

4. The method according to claim 3, characterized in that The wind circle recognition model includes a level-7 wind circle recognition model and a level-10 wind circle recognition model; the level-7 wind circles and level-10 wind circles in the wind field map are marked to obtain a level-7 wind circle detection data set and a level-10 wind circle detection data set; the level-7 wind circle recognition model is trained using the level-7 wind circle detection data set, and the level-10 wind circle recognition model is trained using the level-10 wind circle detection data set; the level-7 wind circle detection is performed using the trained level-7 wind circle recognition model, and the level-10 wind circle detection is performed using the trained level-10 wind circle recognition model; the wind circle detection results include level-7 wind circle detection results and level-10 wind circle detection results.

5. The method according to claim 4, characterized in that The Hrnet neural network model framework is used to build a storm eye center identification model.

6. The method according to claim 5, characterized in that Hrnet predicts the key point of the eye center based on the heatmap method; Hrnet first passes through two convolutional layers with a kernel size of 3×3 and a stride of 2, and then passes through the Layer1 module, which is composed of repeatedly stacked Bottleneck, and then passes through a series of transition structures and stage structures. The transition adds a branch of a scale, and the stage fuses information of different scales; after the last stage fusion branch, it enters a convolutional layer with a kernel size of 3×3, a stride of 1, and a channel number of 1, and finally outputs a feature map of 1 channel, that is, the center position of the eye.

7. The method according to claim 5, characterized in that The LSTM neural network model framework is used to construct a typhoon eye trajectory prediction model. By introducing input gates, forget gates, output gates and cell states, LSTM processes long-term dependencies in sequences. The eye trajectory prediction model performs single-step and multi-step predictions on the eye trajectory. Among them, wind field maps with typhoon eye in continuous time periods are automatically grouped as input to the eye trajectory prediction model for single-step and multi-step predictions of the eye trajectory.

8. The method according to claim 7, characterized in that When a typhoon lands within the detection range of the ground wave radar, the eye center recognition model detects the eye center position, and judges and combines them according to the minimum eye distance between consecutive times to form a typhoon grouping; when the eye center positions of five time periods are detected, these five time series are input into the LSTM to predict the eye center positions for the next three time periods. When the eye detection and eye center detection neural networks detect the new eye center position at the next moment, the eye center positions for the next three time periods are predicted again according to the previous steps.

9. The method according to any one of claims 1 to 8, characterized in that: The sea surface wind field information includes the wind speed, wind direction and confidence level of each grid in the radar detection area.

10. The method according to claim 9, characterized in that The tracked position of the eye center is mapped to the wind field map, and the moving direction and speed of the typhoon within a set time are calculated based on the frequency of the wind field data.

Images