Tropical cyclone scale monitoring method and device based on deep learning and space-time coding

Abstract

The invention relates to a tropical cyclone scale monitoring method and device based on deep learning and space-time coding. The method comprises the steps of obtaining multi-source satellite image data and an international optimal path data set of the tropical cyclone, and performing preprocessing; performing space-time independent field coding based on the international optimal path data set to generate three-channel time scale field data reflecting annual change characteristics, seasonal change rules and evolution time recording characteristics of the tropical cyclones and two-channel space scale field data representing space position information of the tropical cyclones; and splicing with the three-channel satellite image data obtained by preprocessing to obtain a tropical cyclone multi-scale comprehensive feature data set, inputting the tropical cyclone multi-scale comprehensive feature data set into the trained TC-Resnet model to carry out tropical cyclone scale monitoring, and predicting to obtain a tropical cyclone multi-scale monitoring result at a target moment. By adopting the method, the spatial-temporal information of the tropical cyclone can be fully utilized to realize the scale high-precision monitoring of the tropical cyclone.

Description

Technical Field

[0001] The present application relates to the field of deep learning technology, and in particular to a tropical cyclone scale monitoring method and device based on deep learning and spatiotemporal coding. Background Art

[0002] The scale of tropical cyclones (TCs) is an important indicator for researchers to measure the impact range of tropical cyclone systems, and is of decisive significance for revealing the potential harm that tropical cyclones will bring. However, due to the interaction of various physical processes in the atmosphere and ocean, as well as the interaction between tropical cyclones and climate phenomena at various scales, these complex nonlinear interaction effects make it challenging to monitor the scale of tropical cyclones.

[0003] With the development of remote sensing technology, objective quantitative monitoring of the scale of tropical cyclones has gradually become possible. Some researchers have used geostationary satellite infrared data to estimate the scale of tropical cyclones. They use the maximum wind speed radius (RMAX) and the tangential wind speed at 182 kilometers (V182) as the main parameters, construct a symmetric tangential wind field through the improved Rankine vortex model, and add the storm moving vector to obtain a complete two-dimensional wind field. Some researchers have also built a TC surface wind field analysis system (MTCSWA) based on multi-platform satellite data. The system integrates multi-source observation data such as scatterometers, microwave sounders, and cloud guide winds, and can provide estimates of 34 knots, 50 knots, and 64 knots of wind speed radii. Some researchers have also proposed a relatively simple method to estimate the TC wind field structure. This method is based on conventionally available information, including storm data (location, motion, and intensity) and TC scale parameters. The modified Rankine vortex is used to fit the wind field to estimate the wind speed radii of 34 knots, 50 knots, and 64 knots. The verification results show that the average absolute errors are 54 n mi (nautical miles), 32 n mi, and 20 n mi, respectively. In addition, some researchers have proposed an objective technology based on geostationary satellite infrared images to characterize the structure and intensity changes of tropical cyclones. This method calculates the deviation angle variance (DAV) of the brightness temperature gradient to measure the degree of symmetry of the cloud system.

[0004] These remote sensing and objective estimation techniques provide important technical support for TC scale monitoring in sea areas lacking conventional observation data. However, due to the limitations of the spatial and temporal resolution of observation methods and the complexity of physical relationships, these methods still have certain uncertainties in TC scale estimation. Summary of the invention

[0005] Based on this, it is necessary to provide a tropical cyclone scale monitoring method and device based on deep learning and spatiotemporal coding to address the above-mentioned technical problems, which can make full use of the spatiotemporal information of tropical cyclones to achieve high-precision monitoring of tropical cyclone scales, and support the exploration of the tropical cyclone scale change mechanism based on model parameter visualization.

[0006] A tropical cyclone scale monitoring method based on deep learning and spatiotemporal coding, the method comprising: Obtain multi-source satellite image data and international best path datasets of tropical cyclones, and pre-process the multi-source satellite image data; Based on the international best path data set, the spatiotemporal independent field coding is performed to generate three-channel time scale field data reflecting the interannual variation characteristics, seasonal variation laws and evolution time record characteristics of tropical cyclones and two-channel space scale field data representing the spatial position information of tropical cyclones. The data is then spliced ​​with the three-channel satellite image data obtained by preprocessing to obtain a multi-scale comprehensive characteristic data set of tropical cyclones. The tropical cyclone multi-scale comprehensive feature data set is divided into a training set, a validation set and a test set according to a preset ratio, and the training set and the validation set are input into a pre-built TC-Resnet model for optimization training to obtain a trained TC-Resnet model; wherein, the model consists of a feature extraction layer and a fully connected layer; the feature extraction layer extracts the tropical cyclone scale features of the input image through a sequentially connected convolution module, a maximum pooling module and four groups of pyramid convolution residual modules; the fully connected layer obtains the feature map output by the feature extraction layer, and performs average pooling, flattening and linear layer classification processing in sequence to output multi-type scale parameter estimation of the tropical cyclone; wherein, the size parameters include the maximum wind speed radius, the 34-knot wind speed radius, the 50-knot wind speed radius and the 64-knot wind speed radius; The test set is input into the trained TC-Resnet model for tropical cyclone scale monitoring, and the multi-scale monitoring results of tropical cyclones at the target time are predicted.

[0007] In one embodiment, obtaining multi-source satellite image data and an international best path data set of a tropical cyclone, and preprocessing the multi-source satellite image data include: Acquire multi-source satellite image data and international best path datasets of tropical cyclones; among them, the multi-source satellite image data include the complementary GridSat-B1 satellite image data, CMORPH satellite inverted precipitation image data and Fengyun series meteorological satellite image data, and the GridSat-B1 satellite image data include infrared channel image data and water vapor channel image data; in the time dimension, align all satellite image data to be consistent with the international best path dataset, and in the spatial dimension, interpolate all satellite image data to a spatial resolution consistent with the CMORPH satellite inverted precipitation image data; after time alignment and spatial interpolation, complete the multi-source satellite image data, and splice the completed multi-source satellite image data to construct a satellite image dataset; further perform data cleaning and spatial position screening on the satellite image dataset to obtain the final satellite image dataset.

[0008] In one of the embodiments, a satellite image dataset is cleaned and spatially screened to obtain a final satellite image dataset, including: based on the physical characteristics of the tropical cyclone scale parameter, screening and eliminating satellite image data that does not meet the preset scale parameter constraint conditions in the satellite image dataset; the scale parameter constraint conditions include R34>R50>R64, R34≥RMW, R50≥RMW and R64≥RMW, wherein RMW, R34, R50 and R64 are the maximum wind speed radius, 34-knot wind speed radius, 50-knot wind speed radius and 64-knot wind speed radius of the tropical cyclone, respectively; based on the longitude and latitude positions of the tropical cyclone path center recorded in the international best path dataset, extracting satellite image data of a set area around the longitude and latitude positions of the tropical cyclone path center from the satellite image dataset after data cleaning to obtain the final satellite image dataset.

[0009] In one embodiment, spatiotemporal independent field coding is performed based on the international best path data set to generate three-channel time scale field data reflecting the interannual variation characteristics, seasonal variation laws, and evolution time record characteristics of tropical cyclones and two-channel space scale field data representing the spatial position information of tropical cyclones, and spliced ​​with the three-channel satellite image data obtained by preprocessing to obtain a multi-scale comprehensive feature data set of tropical cyclones, including: Based on the international best path data set, spatiotemporal independent field coding is performed. In the time dimension, three-channel time scale field data are generated, including a long-term time scale field that reflects the interannual variation characteristics of tropical cyclones, a medium-term time scale field that reflects the seasonal variation law of tropical cyclones, and a short-term time scale field that captures the evolution time record characteristics of tropical cyclone weather scales; in the spatial dimension, two-channel spatial scale field data are generated, including latitude and longitude fields that characterize the spatial position information of tropical cyclones; among them, all data obtained by spatiotemporal independent field coding maintain the same spatial resolution as the preprocessed satellite image data, and the value of each grid point obtained by coding represents the spatiotemporal information corresponding to the grid point position; The three-channel time scale field data, the two-channel spatial scale field data, and the infrared channel image data, water vapor channel image data, and CMORPH satellite inverted precipitation image data in the final satellite image data set are spliced ​​to obtain a tropical cyclone multi-scale comprehensive feature data set containing 8 channels.

[0010] In one of the embodiments, the four groups of pyramid convolution residual modules in the TC-Resnet model respectively include 3-layer, 8-layer, 36-layer and 3-layer pyramid convolution residual blocks; wherein each pyramid convolution residual block is composed of a first convolution layer, a second pyramid convolution layer, a third convolution layer and a fourth SE attention mechanism module, and the data input into the pyramid convolution residual block is divided into a main path and a shortcut path. After the data of the main path is processed by three layers of convolution and input into the SE attention mechanism module for attention weighting, it is spliced ​​with the data of the shortcut path and input into the next layer for processing.

[0011] In one of the embodiments, after obtaining the input feature map, the pyramid convolution layer first divides the input feature map into different branches, applies feature extraction capability branches with different convolution kernel sizes to the data of each branch in parallel for convolution, and concatenates the feature map obtained by the convolution of each feature extraction capability branch in the data dimension to obtain the final output feature map; wherein each feature extraction capability branch is allocated a different number of channels, and for a feature extraction capability branch with a larger convolution kernel, the number of channels allocated is smaller and the corresponding number of groups is greater; After obtaining the input feature map, the SE attention mechanism module first performs feature compression conversion, then recalibrates the features of each channel in the compressed feature map to obtain channel attention information, and finally multiplies the channel attention information and the original input feature map by the weight coefficient channel by channel, and finally outputs a feature map with channel attention.

[0012] In one embodiment, the data in the TC-Resnet model is processed by a batch normalization layer and a Mish activation function after each convolution.

[0013] In one embodiment, an additional SE attention mechanism module or attention module is added to the first layer of the TC-Resnet model; wherein the additional SE attention mechanism module is used to assign weights to each type of data input to the model and perform data weight visualization analysis, analyzing the importance of each type of data to the model's multi-scale results of monitoring tropical cyclones; the additional attention module is used to perform model weight visualization analysis after normalizing various types of satellite image data input to the model, analyzing and obtaining the model's key focus areas for various types of satellite image data.

[0014] In one embodiment, the TC-Resnet model is optimized and trained, including: calculating a loss function based on multi-type scale parameter estimates of tropical cyclones predicted and output by the TC-Resnet model and the true values ​​of the scale parameters of the corresponding types, and optimizing the TC-Resnet model based on the loss function and the Adam optimizer; wherein the loss function is expressed as ; in, is the number of input feature maps, is the feature map number and ; Estimates of scale parameters of tropical cyclones for model forecast output; is the true value of the scale parameter.

[0015] A tropical cyclone scale monitoring device based on deep learning and spatiotemporal coding, the device comprising: A preprocessing module is used to obtain multi-source satellite image data and international best path data sets of tropical cyclones, and to preprocess the multi-source satellite image data; The spatiotemporal coding module is used to perform spatiotemporal independent field coding based on the international best path dataset, generate three-channel time-scale field data reflecting the interannual variation characteristics, seasonal variation laws, and evolution time record characteristics of tropical cyclones, and two-channel space-scale field data representing the spatial position information of tropical cyclones, and obtain a multi-scale comprehensive feature dataset of tropical cyclones by splicing it with the three-channel satellite image data obtained by preprocessing; A model training module is used to divide the tropical cyclone multi-scale comprehensive feature data set into a training set, a validation set and a test set according to a preset ratio, and input the training set and the validation set into a pre-built TC-Resnet model for optimization training to obtain a trained TC-Resnet model; wherein the model consists of a feature extraction layer and a fully connected layer; the feature extraction layer extracts the tropical cyclone scale features of the input image through a sequentially connected convolution module, a maximum pooling module and four groups of pyramid convolution residual modules; the fully connected layer obtains the feature map output by the feature extraction layer, performs average pooling, flattening and linear layer classification processing in sequence, and outputs multi-type scale parameter estimation of the tropical cyclone; wherein the size parameters include the maximum wind speed radius, the 34-knot wind speed radius, the 50-knot wind speed radius and the 64-knot wind speed radius; The scale monitoring module is used to input the test set into the trained TC-Resnet model for tropical cyclone scale monitoring and predict the multi-scale monitoring results of tropical cyclones at the target time.

[0016] The above tropical cyclone scale monitoring method and device based on deep learning and spatiotemporal coding have the following beneficial effects: 1. Based on satellite image data, we further spliced ​​three-channel time-scale field data reflecting the interannual variation characteristics, seasonal variation laws and evolution time record characteristics of tropical cyclones generated based on the spatiotemporal coding of the International Best Path Dataset and two-channel space-scale field data representing spatial position information to construct an 8-channel multi-scale comprehensive characteristic dataset of tropical cyclones, which made up for the limitations of a single data source, effectively integrated the historical spatiotemporal evolution information of tropical cyclones with real-time satellite observation characteristics, and significantly improved the multi-scale monitoring performance of the model.

[0017] 2. In view of the nonlinear data situation that tropical cyclone wind fields often present an asymmetric structure and different data samples have large differences, the constructed TC-Resnet model is based on the pyramid convolution residual module with feature extraction capability branches including different convolution receptive field sizes, and the SE attention mechanism module that can recalibrate the features of each channel of the input data. In this way, it can fully pay attention to the multi-scale comprehensive characteristic data of tropical cyclones input to the model and realize multi-scale intelligent collaborative monitoring of tropical cyclones.

[0018] 3. By adding the SE attention mechanism module or attention module to the first layer of the model, it supports the visualization analysis of input data weights and model weights, which can reveal the data types and areas that the model focuses on, thereby further improving the monitoring performance of the model and optimizing model training. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 11 is a flow chart of a tropical cyclone scale monitoring method based on deep learning and spatiotemporal coding in one embodiment; Figure 2 A schematic diagram of the composition of a tropical cyclone multi-scale comprehensive characteristic data set in one embodiment; Figure 3 Schematic diagram of the architecture of a TC-Resnet model in one embodiment; Figure 4 FIG. 1 is a schematic diagram showing a comparison between the TC-Resnet model fitting result and the existing other model fitting results in one embodiment; wherein, Figure 4 (a) is a schematic diagram showing the comparison of fitting results for RMW scale parameters. Figure 4 (b) is a schematic diagram showing the comparison of the fitting results for the R64 scale parameter. Figure 4 (c) is a schematic diagram showing the comparison of the fitting results for the R50 scale parameter. Figure 4 (d) is a schematic diagram showing the comparison of the fitting results for the R34 scale parameter; Figure 5 : is a schematic diagram of the distribution of the mean error (ME) and the mean absolute error (ME-weight) weighted based on the true value of the TC-Resnet model in one embodiment; wherein, Figure 5 (a) is a schematic diagram of the error distribution for the RMW scale parameter. Figure 5 (b) is a schematic diagram of the error distribution for the R64 scale parameter. Figure 5 (c) is a schematic diagram of the error distribution for the R50 scale parameter. Figure 5 (d) Schematic diagram of the error distribution for the R34 scale parameter. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0021] In one embodiment, Figure 1 As shown, a tropical cyclone scale monitoring method based on deep learning and spatiotemporal coding is provided, comprising the following steps: Step S1, obtaining multi-source satellite image data and international best path data set of tropical cyclones, and preprocessing the multi-source satellite image data.

[0022] Specifically, this application selects GridSat-B1 satellite image data, CMORPH satellite inversion precipitation image data and Fengyun series meteorological satellite image data from 90°E to 160°E and 5°S to 55°N in the northwest Pacific Ocean. The three types of satellite image data complement each other, among which GridSat-B1 satellite image data include infrared channel image data (IR, close to 11μm) and water vapor channel image data (WV, close to 6.7μm).

[0023] The International Best Track Archive for ClimateStewardship (IBTrACS) is the most authoritative and complete tropical cyclone historical data record database in the world, compiled in cooperation with the World Meteorological Organization (WMO) and other organizations and individuals from all over the world. These complete records with extremely high temporal resolution can provide important records for researchers to analyze tropical cyclones.

[0024] Preprocessing of multi-source satellite image data includes the following steps: First, in order to properly put the data into the model training, all satellite image data were aligned to the 3-hour record consistent with the international best path dataset in the temporal dimension; further, in order to enable the data to be put into the model training and the multi-channel data influence to be integrated with each other, in the spatial dimension, all satellite image data were interpolated to the spatial resolution consistent with the CMORPH satellite inversion precipitation image data, with a grid longitude and latitude of 0.25°×0.25°.

[0025] After time alignment and spatial interpolation, the multi-source satellite image data are completed, and the completed multi-source satellite image data are dimensionally spliced ​​to construct a satellite image data set. The completion includes: for some satellite image data with missing values, reasonable interpolation and completion are performed using methods such as cubic splines, and due to task adjustments such as satellite observation, for some satellite image data with completely missing values, Fengyun satellite data are used to complete the data, so as to ensure the integrity and continuity of the data.

[0026] The satellite image dataset is further cleaned and spatially screened to obtain the final satellite image dataset.

[0027] Among them, data cleaning is to screen and eliminate satellite image data that do not meet the preset scale parameter constraints in the satellite image data set according to the physical characteristics of the tropical cyclone scale parameters; the scale parameter constraints include R34>R50>R64, R34≥RMW, R50≥RMW and R64≥RMW, where RMW, R34, R50 and R64 are the maximum wind speed radius, 34-knot wind speed radius, 50-knot wind speed radius and 64-knot wind speed radius of the tropical cyclone respectively. Spatial position screening refers to extracting the satellite image data in the set area around the longitude and latitude position of the center of the tropical cyclone path from the satellite image data set after data cleaning based on the longitude and latitude position of the center of the tropical cyclone path recorded in the international best path data set, and obtaining the final satellite image data set. The size of the set area around the longitude and latitude position of the center of the tropical cyclone path set in this application is 10°×10°, that is, the satellite image data in the box area of ​​5 longitudes and latitudes above and below and to the left and right of the center point of the tropical cyclone path is extracted.

[0028] Step S2, based on the international best path data set, performs spatiotemporal independent field encoding to generate three-channel time-scale field data reflecting the interannual variation characteristics, seasonal variation laws, and evolution time record characteristics of tropical cyclones and two-channel space-scale field data representing the spatial position information of tropical cyclones, and then splices them with the three-channel satellite image data obtained by preprocessing to obtain a multi-scale comprehensive characteristic data set of tropical cyclones.

[0029] Specifically, the process of constructing a multi-scale comprehensive characteristic dataset of tropical cyclones includes: Based on the international best path dataset, spatiotemporal independent field encoding is performed. In the time dimension, three-channel time scale field data are generated, including the long-term time scale field (year field) reflecting the interannual variation characteristics of tropical cyclones, the medium-term time scale field (month field) reflecting the seasonal variation law of tropical cyclones, and the short-term time scale field (short-term time series field) capturing the evolution time record characteristics of tropical cyclone weather scales; in the spatial dimension, two-channel spatial scale field data are generated, including the latitude field and longitude field representing the spatial position information of tropical cyclones; among them, all data obtained by spatiotemporal independent field encoding maintain the same spatial resolution (0.25°×0.25°) as the preprocessed satellite image data, and the value of each grid point obtained by encoding represents the spatiotemporal information corresponding to the grid point position.

[0030] The three-channel time scale field data, the two-channel space scale field data, and the infrared channel image data, water vapor channel image data, and CMORPH satellite inversion precipitation image data in the final satellite image data set are spliced ​​to obtain a tropical cyclone multi-scale comprehensive feature data set containing 8 channels, such as Figure 2The 8-channel tropical cyclone multi-scale comprehensive characteristic dataset effectively integrates the historical spatiotemporal evolution information of tropical cyclones recorded in the international best track dataset and the real-time satellite observation characteristics. This multi-dimensional data fusion method enables subsequent models to fully capture the multi-scale spatiotemporal evolution characteristics of tropical cyclones, providing a solid data foundation for improving model performance. Figure 2 The four circles in the middle correspond from inside to outside: RMW (red), R64 (green), R50 (blue), and R34 (purple).

[0031] Step S3, dividing the tropical cyclone multi-scale comprehensive feature data set into a training set, a validation set and a test set according to a preset ratio, inputting the training set and the validation set into a pre-built TC-Resnet model for optimization training, and obtaining a trained TC-Resnet model; wherein the model is composed of a feature extraction layer and a fully connected layer; the feature extraction layer extracts the tropical cyclone scale features of the input image through a sequentially connected convolution module, a maximum pooling module and four groups of pyramid convolution residual modules; the fully connected layer obtains the feature map output by the feature extraction layer, performs average pooling, flattening and linear layer classification processing in sequence, and outputs multi-type scale parameter estimates of the tropical cyclone.

[0032] The size parameters of tropical cyclones include the maximum wind speed radius, the 34-knot wind speed radius, the 50-knot wind speed radius, and the 64-knot wind speed radius. The specific structure of the TC-Resnet model constructed in this application is as follows: Figure 3 As shown in the figure, the four groups of pyramid convolution residual modules in the model contain 3 layers, 8 layers, 36 layers and 3 layers of pyramid convolution residual blocks respectively. Among them, each pyramid convolution residual block consists of the first convolution layer (Conv1), the second pyramid convolution layer (Pyramidal Convolution, PyConv), the third convolution layer (Conv2) and the fourth SE attention mechanism module. The data input to the pyramid convolution residual block is divided into the main path X and the shortcut path Xshortcut. After the data of the main path is processed by three layers of convolution and input into the SE attention mechanism module for attention weighting, it is spliced ​​with the data of the shortcut path and input into the next layer for processing. The residual module designed in this way can enhance the model's ability to extract features for field variable data.

[0033] Compared with the ordinary convolution process, the input feature map enters the standard convolution kernel for convolution processing, and then obtains the output feature map. This convolution design cannot take into account the defects of features of different scales. The present application adds a pyramid convolution layer to the residual module. After obtaining the input feature map, the pyramid convolution layer first divides the input feature map into different branches. For the data of each branch, feature extraction branches with different convolution kernel sizes are applied in parallel for convolution, and different numbers of channels are assigned to these convolution kernels. The feature maps obtained by convolution of each feature extraction branch are spliced ​​in the data dimension to obtain the final output feature map. Such a design is equivalent to considering the extraction branch routes for features of different scales separately. Those larger convolution kernels have larger receptive fields, but the amount of calculated parameters is also larger, and the machine performance requirements are higher. Therefore, for feature extraction branches with larger convolution kernels, fewer channels are allocated, which can well balance the model performance. At the same time, a hierarchical grouping strategy is introduced in the pyramid convolution layer, that is, for branches with larger feature extraction capabilities of the convolution kernel, the corresponding number of groups is larger, so that the hierarchical expression of features can be constructed at different scales and the demand for computing resources can be further reduced. The multi-scale convolution branch route design such as the pyramid convolution layer can show greater advantages in the multi-scale intelligent monitoring task of tropical cyclones that requires attention to multi-scale image features.

[0034] The last layer of each pyramid convolution residual block is added with an SE (Squeeze-and-Excitation) attention mechanism module. After obtaining the input feature map, the SE attention mechanism module first performs feature compression conversion, and then recalibrates the features of each channel in the compressed feature map to obtain channel attention information. Finally, the channel attention information and the original input feature map are multiplied by the weight coefficient channel by channel, and finally the feature map with channel attention is output. This module can give different weights to each channel of the input feature data, so that the model pays more attention to the data of important channels. In addition, there is a reduction ratio design in the middle of this module, which not only significantly reduces the number of parameters, but also plays the role of an information bottleneck, forcing the network model to learn more discriminative channel attention. Such a design can reduce the influence of feature channels that are not related to the model fitting results, and give greater weight to feature channels with higher utility.

[0035] Furthermore, in the TC-Resnet model, in order to make the model pay more attention to the key areas on the feature map, the data is processed by the batch normalization layer (BN) and the Mish activation function after each convolution, that is, a batch normalization layer and a Mish activation function are added after each convolution layer in the model. Compared with the traditional ReLU activation function and its variants shown in formula (1), the Mish activation function selected in this application exhibits unique mathematical properties, which is a smooth, non-monotonic, self-normalized activation function. The function form of Mish is shown in formula (2): (1); (2).

[0036] The unboundedness of the Mish activation function can alleviate the gradient vanishing problem. x When it is less than 0, y Taking a smaller negative value can alleviate the phenomenon of neuron death to a certain extent, and retain some neurons with negative weights. At the same time, the Mish activation function is smoother than the ReLU activation function, so a more stable gradient can be obtained, which has a better effect on the application of the optimizer. Moreover, the Mish activation function has a quasi-linear nature. On the one hand, it can maintain its nonlinear expression ability, while maintaining the characteristics of some linear activation functions. For deep networks such as Resnet-152, Mish can have particularly good gradient flow properties. The TC-Resnet model constructed in this application is particularly deep, so all activation functions are replaced with Mish activation functions to obtain better performance than Relu.

[0037] When training the TC-Resnet model, in order to facilitate the evaluation of model performance, the tropical cyclone multi-scale comprehensive feature dataset is divided into training set, validation set and test set in a ratio of 80:5:15. During the training process, the training set is used to iteratively adjust the parameters of the model; the validation set is used to guide the tuning of hyperparameters such as the learning rate, to facilitate the selection of an appropriate learning rate and to discover problems in model training; the test set is used to evaluate the performance of the model after each iterative training. This three-stage training-validation-testing process can ensure the reliability of the trained model and facilitate understanding of the actual effect of the model. The test set data that is not involved in the training can also be used to evaluate the multi-scale monitoring capabilities of the model in actual scenarios.

[0038] When optimizing the TC-Resnet model, the loss function is calculated based on the multi-type scale parameter estimates of tropical cyclones predicted and output by the TC-Resnet model and the true values ​​of the scale parameters of the corresponding types, and the TC-Resnet model is optimized and trained based on the loss function and the Adam optimizer. The Adam optimizer combines the advantages of the momentum method and the adaptive learning rate method and can adaptively adjust the learning rate of the parameters. The loss function is expressed as ; in, is the number of input feature maps, is the feature map number and ; Estimates of scale parameters of tropical cyclones for model forecast output; is the true value of the scale parameter. Each time the model is trained, the loss function can be calculated based on the estimated results of the four types of scale parameters monitored by the model and the label true value data, thereby optimizing the next training iteration of the model and obtaining a model with continuously optimized performance. This enables the model to continuously learn from the input data, optimize the internal parameter weights, and gradually improve the model's effect on multi-scale monitoring of tropical cyclones. With this end-to-end processing flow, the trained model can achieve direct monitoring of multi-scale parameters of tropical cyclones from raw satellite image data.

[0039] Step S4, input the test set into the trained TC-Resnet model to perform tropical cyclone scale monitoring, and predict the multi-scale monitoring results of the tropical cyclone at the target time.

[0040] Furthermore, the method further includes: adding an additional SE attention mechanism module or attention module to the first layer of the TC-Resnet model; wherein the additional SE attention mechanism module is used to assign weights to each type of data input to the model and perform data weight visualization analysis, analyzing and obtaining the importance of each type of data to the multi-scale results of the model monitoring tropical cyclones; the additional attention module is used to perform model weight visualization analysis after normalizing various types of satellite image data input to the model, analyzing and obtaining the model's key focus areas for various types of satellite image data. By adding the SE attention mechanism module or attention module, input data weight visualization analysis and model weight visualization analysis are supported, which can reveal the type of data and area that the model focuses on, thereby further improving the monitoring performance of the model and optimizing model training.

[0041] Table 1 Comparison of TC-Resnet model with other existing models

[0042] In order to further verify the beneficial effects of the tropical cyclone scale monitoring method based on deep learning and spatiotemporal coding provided by this application, the TC-Resnet model constructed in this application was compared with other existing models. The results of the comparison test are shown in Tables 1 and Figure 4 As shown in Table 1 and Figure 4 It can be seen that the TC-Resnet model constructed in this application can monitor four tropical cyclone scale parameters simultaneously, and its accuracy has reached the current optimal level.

[0043] In order to further verify the monitoring accuracy of the TC-Resnet model constructed in this application in different intervals of the four types of scale parameters, for the four types of scales, considering the physical characteristics of the tropical cyclone scale and the quantiles of the four types of scale parameters, for RMW, due to the small overall change of RMW, the values ​​of less than 12, 12-15, 15-20, and greater than 20 were selected, and the four value ranges accounted for 25.79%, 26.47%, 28.12%, and 19.62% of the total samples respectively (considering the integer effect and only retaining 2 decimal places). For R34, due to the large range of changes in the wind speed of 34 knots, four ranges of less than or equal to 100, 100-130, 130-160, and greater than or equal to 160 were selected, and each range accounted for 17.97%, 34.57%, 26.06%, and 21.40% of the total samples respectively. Among them, 100-130 is the area with the most samples in the R34 division interval this time. For R50, considering the numerical range, it is divided into less than or equal to 50 (24.42%), 50-70 (35.94%), 70-90 (21.67%), and greater than or equal to 90 (17.97%). For R64, it is divided into less than or equal to 30 (35.67%), 30-40 (25.24%), 40-50 (18.79%), and greater than or equal to 50 (20.30%). In the evaluation process of the TC-Resnet model, in addition to the mean absolute error (MAE), the mean absolute error weighted based on the true value (MAE-weight) is also added as an evaluation indicator. This indicator can give greater weight to larger tropical cyclone scales based on real data. The results are as follows: Figure 5 shown.

[0044] from Figure 5 It can be seen that for the maximum wind speed radius RMW, the model monitoring error is low due to the overall small value range. In the more typical scale range of 12-15 nautical miles, the model effect reaches the best state. In the scale range of 15-20 nautical miles, the MAE-weight of the model monitoring result is only 3.09 nautical miles. These two RMW scale division ranges account for 54.59% of the total samples, indicating that the model is more accurate in monitoring the typical RMW scale.

[0045] For the outer tropical cyclone scale R34, the prediction error increases significantly with increasing distance, especially for scales such as R34. Starting from the first range, the error gradually increases and reaches a maximum value above 160 nautical miles. Compared with the other scales, the R34 value range is larger and the error is also larger, which reflects the inherent challenges of monitoring the outer wind field scale. R50 and R64 also follow this rule, and the error from the first division range to the fourth division range shows a gradual increase. RMW is different. For extremely large RMW values ​​(greater than or equal to 20 nautical miles), the model monitoring error is large (MAE-weight is 9.48). For smaller values ​​(less than or equal to 12 nautical miles), the model error is larger than the typical cases of the second and third ranges (MAE-weight is 4.57).

[0046] The monitoring errors of the four types of scale parameters all show the characteristic of increasing negative deviations as the values ​​increase, that is, the scale of tropical cyclones monitored by the model is smaller than the actual label data. This is because the sample size of the extreme scale values ​​is small and unbalanced. In the first divided range, ME and ME-weight are positive values. For example, for R34, within the range of less than or equal to 100 nautical miles, the ME-weight error of the model monitoring value is 8.16, indicating that the model monitors values ​​greater than the actual values ​​in relatively small values ​​of this type of scale, which may also be due to the above reasons. In the typical value range of the four types of scales, the accuracy of model monitoring is good.

[0047] There is a high consistency between the mean absolute error and the weighted mean absolute error, the mean error weighted by the true value, and the mean error, indicating that the model remains stable in the process of monitoring the scale of tropical cyclones after considering the weights assigned by the true values ​​of the samples.

[0048] Table 2 Experiments with different input data

[0049] Furthermore, during the verification, the data set was divided into multiple satellite single-channel data, which were input into the model for training and parameter adjustment. The model's multi-scale monitoring effect of tropical cyclones was greatly reduced. However, by integrating the three satellite data channels into the model training, the monitoring performance of the model was improved to a certain extent. After that, the three satellite data channels were integrated, and the multi-scale spatiotemporal independent field variable information of the five channels generated based on IBTrACS was introduced to form an 8-channel data set, which was put into the model for training, and the best effect of this experiment was achieved, as shown in Table 2. This shows that multi-source information fusion plays a great role in improving the accuracy of multi-scale monitoring of tropical cyclones, and the model performance has also been greatly improved.

[0050] In one embodiment, a tropical cyclone scale monitoring device based on deep learning and spatiotemporal coding is provided, comprising: A preprocessing module is used to obtain multi-source satellite image data and international best path data sets of tropical cyclones, and to preprocess the multi-source satellite image data; The spatiotemporal coding module is used to perform spatiotemporal independent field coding based on the international best path dataset, generate three-channel time-scale field data reflecting the interannual variation characteristics, seasonal variation laws, and evolution time record characteristics of tropical cyclones, and two-channel space-scale field data representing the spatial position information of tropical cyclones, and obtain a multi-scale comprehensive feature dataset of tropical cyclones by splicing it with the three-channel satellite image data obtained by preprocessing; A model training module is used to divide the tropical cyclone multi-scale comprehensive feature data set into a training set, a validation set and a test set according to a preset ratio, and input the training set and the validation set into a pre-built TC-Resnet model for optimization training to obtain a trained TC-Resnet model; wherein the model consists of a feature extraction layer and a fully connected layer; the feature extraction layer extracts the tropical cyclone scale features of the input image through a sequentially connected convolution module, a maximum pooling module and four groups of pyramid convolution residual modules; the fully connected layer obtains the feature map output by the feature extraction layer, performs average pooling, flattening and linear layer classification processing in sequence, and outputs multi-type scale parameter estimation of the tropical cyclone; wherein the size parameters include the maximum wind speed radius, the 34-knot wind speed radius, the 50-knot wind speed radius and the 64-knot wind speed radius; The scale monitoring module is used to input the test set into the trained TC-Resnet model for tropical cyclone scale monitoring and predict the multi-scale monitoring results of tropical cyclones at the target time.

[0051] For the specific limitations of the tropical cyclone scale monitoring device based on deep learning and spatiotemporal coding, please refer to the limitations of the tropical cyclone scale monitoring method based on deep learning and spatiotemporal coding above, which will not be repeated here. Each module in the above-mentioned tropical cyclone scale monitoring device based on deep learning and spatiotemporal coding can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0052] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0053] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several modifications and improvements may be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A tropical cyclone scale monitoring method based on deep learning and spatiotemporal coding, characterized in that: The method comprises: Acquire multi-source satellite image data and international best path data set of tropical cyclones, and pre-process the multi-source satellite image data; Based on the international best path data set, time-space independent field coding is performed to generate three-channel time-scale field data reflecting the interannual variation characteristics, seasonal variation laws, and evolution time record characteristics of tropical cyclones and two-channel space-scale field data representing the spatial position information of tropical cyclones, and the multi-scale comprehensive characteristic data set of tropical cyclones is obtained by splicing the data with the three-channel satellite image data obtained by preprocessing; The tropical cyclone multi-scale comprehensive feature data set is divided into a training set, a validation set and a test set according to a preset ratio, and the training set and the validation set are input into a pre-constructed TC-Resnet model for optimization training to obtain a trained TC-Resnet model; wherein the model consists of a feature extraction layer and a fully connected layer; the feature extraction layer extracts the tropical cyclone scale features of the input image through a convolution module, a maximum pooling module and four groups of pyramid convolution residual modules connected in sequence; the fully connected layer obtains the feature map output by the feature extraction layer, performs average pooling, flattening and linear layer classification processing in sequence, and outputs multi-type scale parameter estimation of the tropical cyclone; wherein the size parameters include the maximum wind speed radius, the 34-knot wind speed radius, the 50-knot wind speed radius and the 64-knot wind speed radius; The test set is input into the trained TC-Resnet model for tropical cyclone scale monitoring, and the multi-scale monitoring results of the tropical cyclone at the target time are predicted.

2. The method according to claim 1, characterized in that Acquire multi-source satellite image data and international best path data set of tropical cyclones, and pre-process the multi-source satellite image data, including: Acquire multi-source satellite image data and international best path data sets of tropical cyclones; wherein the multi-source satellite image data include complementary GridSat-B1 satellite image data, CMORPH satellite inversion precipitation image data and Fengyun series meteorological satellite image data, and the GridSat-B1 satellite image data include infrared channel image data and water vapor channel image data; In the temporal dimension, all satellite image data are aligned to be consistent with the international best path dataset, and in the spatial dimension, all satellite image data are interpolated to a spatial resolution consistent with the CMORPH satellite inverted precipitation image data; After time alignment and spatial interpolation, the multi-source satellite image data are completed, and the completed multi-source satellite image data are dimensionally spliced ​​to construct a satellite image dataset; The satellite image dataset is further cleaned and spatially screened to obtain a final satellite image dataset.

3. The method according to claim 2, characterized in that The satellite image dataset is cleaned and spatially screened to obtain a final satellite image dataset, including: According to the physical characteristics of the tropical cyclone scale parameter, the satellite image data that does not meet the preset scale parameter constraint conditions in the satellite image data set are screened and eliminated; the scale parameter constraint conditions include R34>R50>R64, R34≥RMW, R50≥RMW and R64≥RMW, wherein RMW, R34, R50 and R64 are the maximum wind speed radius, 34-knot wind speed radius, 50-knot wind speed radius and 64-knot wind speed radius of the tropical cyclone respectively; According to the longitude and latitude positions of the tropical cyclone path center recorded in the international best path dataset, satellite image data of a set area around the longitude and latitude positions of the tropical cyclone path center are extracted from the satellite image dataset after data cleaning to obtain a final satellite image dataset.

4. The method according to claim 3, characterized in that Based on the international best path dataset, time-space independent field coding is performed to generate three-channel time scale field data reflecting the interannual variation characteristics, seasonal variation laws and evolution time record characteristics of tropical cyclones and two-channel space scale field data representing the spatial position information of tropical cyclones. The data is then spliced ​​with the three-channel satellite image data obtained by preprocessing to obtain a multi-scale comprehensive feature dataset of tropical cyclones, including: Based on the international best path data set, spatiotemporal independent field coding is performed to generate three-channel time scale field data in the time dimension, including a long-term time scale field reflecting the interannual variation characteristics of tropical cyclones, a medium-term time scale field reflecting the seasonal variation law of tropical cyclones, and a short-term time scale field capturing the evolution time record characteristics of tropical cyclone weather scales; in the spatial dimension, two-channel spatial scale field data are generated, including a latitude field and a longitude field representing the spatial position information of tropical cyclones; wherein all data obtained by spatiotemporal independent field coding maintain the same spatial resolution as the preprocessed satellite image data, and the numerical value of each grid point obtained by coding represents the spatiotemporal information corresponding to the grid point position; The three-channel time scale field data, the two-channel space scale field data, and the infrared channel image data, water vapor channel image data, and CMORPH satellite inverted precipitation image data in the final satellite image data set are spliced ​​to obtain a tropical cyclone multi-scale comprehensive feature data set containing 8 channels.

5. The method according to claim 1, characterized in that The four groups of pyramid convolution residual modules in the TC-Resnet model include 3-layer, 8-layer, 36-layer and 3-layer pyramid convolution residual blocks respectively; Among them, each pyramid convolution residual block is composed of the first convolution layer, the second pyramid convolution layer, the third convolution layer and the fourth SE attention mechanism module. The data input into the pyramid convolution residual block is divided into a main path and a shortcut path. After the data of the main path is processed by three layers of convolution and input into the SE attention mechanism module for attention weighting, it is spliced ​​with the data of the shortcut path and input into the next layer for processing.

6. The method according to claim 5, characterized in that After obtaining the input feature map, the pyramid convolution layer first divides the input feature map into different branches, applies feature extraction capability branches with different convolution kernel sizes to the data of each branch in parallel for convolution, and concatenates the feature maps obtained by convolution of each feature extraction capability branch in the data dimension to obtain the final output feature map; wherein each feature extraction capability branch is allocated a different number of channels, and for a feature extraction capability branch with a larger convolution kernel, the number of allocated channels is less and the corresponding number of groups is more; After obtaining the input feature map, the SE attention mechanism module first performs feature compression conversion, then recalibrates the features of each channel in the compressed feature map to obtain channel attention information, and finally multiplies the channel attention information and the original input feature map by the weight coefficient channel by channel, and finally outputs a feature map with channel attention.

7. The method according to claim 1 or 5, characterized in that: The data in the TC-Resnet model is processed by a batch normalization layer and a Mish activation function after each convolution.

8. The method according to claim 7, characterized in that The first layer of the TC-Resnet model is also added with an additional SE attention mechanism module or attention module; wherein the additional SE attention mechanism module is used to assign weights to each type of data input to the model and perform data weight visualization analysis, analyzing the importance of each type of data to the model's multi-scale results of monitoring tropical cyclones; the additional attention module is used to perform model weight visualization analysis after normalizing various types of satellite image data input to the model, analyzing and obtaining the model's key focus areas for various types of satellite image data.

9. The method according to claim 1, characterized in that: Optimize the training of the TC-Resnet model, including: The loss function is calculated based on the multi-type scale parameter estimates of tropical cyclones predicted and output by the TC-Resnet model and the true values ​​of the scale parameters of the corresponding types, and the TC-Resnet model is optimized and trained based on the loss function and the Adam optimizer; wherein the loss function is expressed as ； in, is the number of input feature maps, is the feature map number and ; Estimates of scale parameters of tropical cyclones for model forecast output; is the true value of the scale parameter.

10. A tropical cyclone scale monitoring device based on deep learning and spatiotemporal coding, characterized in that: The device comprises: A preprocessing module, used to obtain multi-source satellite image data and an international best path data set of tropical cyclones, and preprocess the multi-source satellite image data; A spatiotemporal coding module is used to perform spatiotemporal independent field coding based on the international best path dataset, generate three-channel time scale field data reflecting the interannual variation characteristics, seasonal variation laws and evolution time record characteristics of tropical cyclones and two-channel space scale field data representing the spatial position information of tropical cyclones, and obtain a multi-scale comprehensive feature dataset of tropical cyclones by splicing the data with the three-channel satellite image data obtained by preprocessing; A model training module, for dividing the tropical cyclone multi-scale comprehensive feature data set into a training set, a validation set and a test set according to a preset ratio, and inputting the training set and the validation set into a pre-built TC-Resnet model for optimization training to obtain a trained TC-Resnet model; wherein the model consists of a feature extraction layer and a fully connected layer; the feature extraction layer extracts the tropical cyclone scale features of the input image through a convolution module, a maximum pooling module and four groups of pyramid convolution residual modules connected in sequence; the fully connected layer obtains the feature map output by the feature extraction layer, performs average pooling, flattening and linear layer classification processing in sequence, and outputs multi-type scale parameter estimation of the tropical cyclone; wherein the size parameters include the maximum wind speed radius, the 34-knot wind speed radius, the 50-knot wind speed radius and the 64-knot wind speed radius; The scale monitoring module is used to input the test set into the trained TC-Resnet model to perform tropical cyclone scale monitoring and predict the multi-scale monitoring results of the tropical cyclone at the target time.

Images