Tropical cyclone strength grade determination method and device, storage medium and electronic device

Abstract

The invention discloses a tropical cyclone strength grade determination method, and the method comprises the steps: obtaining a plurality of collected historical satellite cloud pictures; classifying the plurality of historical satellite cloud pictures, and determining historical satellite cloud pictures respectively corresponding to a plurality of intensity grades; carrying out data equalization processing on the historical satellite cloud pictures corresponding to the plurality of intensity levels to obtain a target satellite cloud picture; inputting the target satellite cloud picture into a to-be-trained neural network model, performing grade prediction by the neural network model based on the target satellite cloud picture, and training the neural network model according to a grade prediction result to obtain a grade prediction model; wherein the grade prediction model is used for predicting the tropical cyclone strength grade. The problem that the tropical cyclone strength grade determination accuracy is low in the related technology is solved.

Description

Technical Field

[0001] The present application relates to the technical field of offshore wind farms, and in particular to a method and device for determining the intensity level of a tropical cyclone, a storage medium, and an electronic device. Background Art

[0002] A tropical cyclone (TC) refers to an air mass rotating over a tropical ocean. It uses the tropical ocean as its heat source and after gaining energy, it rotates and rises over the tropical ocean, forming a giant "typhoon" over the tropical ocean, forming a high-pressure area on the ground. The outer airflow of the high-pressure area rotates around it, forming the eye of the typhoon.

[0003] The intensity level of tropical cyclones directly determines the extent of their impact on offshore wind farms. Weak tropical cyclones can bring continuous winds and increase power generation, while high-intensity tropical cyclones may cause damage to wind farms, requiring protective measures such as shutdown in advance. Therefore, accurate and timely determination of the intensity level of tropical cyclones is crucial to the operation and safety of offshore wind farms.

[0004] At present, the intensity level of tropical cyclones is mainly estimated through expert analysis and a series of complex observation indicators. However, this subjective judgment method is easily affected by the operator's experience and has poor repeatability at different times and locations. In addition, traditional methods require manual extraction of tropical cyclone characteristics, such as cloud structure and central wind speed. This process is cumbersome and inefficient, and it is difficult to meet the needs of real-time prediction.

[0005] Therefore, how to accurately predict the intensity level of cyclones has become an urgent problem to be solved. Summary of the invention

[0006] The embodiment of the present application provides a method for determining the intensity level of a tropical cyclone, so as to at least solve the problem of low accuracy in determining the intensity level of a tropical cyclone in the related art.

[0007] According to an embodiment of the embodiments of the present application, a method for determining the intensity level of a tropical cyclone is provided, comprising: acquiring a plurality of collected historical satellite cloud images; classifying the plurality of historical satellite cloud images to determine historical satellite cloud images corresponding to a plurality of intensity levels; performing data balancing processing on the historical satellite cloud images corresponding to the plurality of intensity levels to obtain a target satellite cloud image; inputting the target satellite cloud image into a neural network model to be trained, having the neural network model perform a level prediction based on the target satellite cloud image, and training the neural network model according to the level prediction results to obtain a level prediction model; wherein the level prediction model is used to predict the intensity level of a tropical cyclone.

[0008] In an exemplary embodiment, the classifying of the plurality of historical satellite cloud images to determine the historical satellite cloud images corresponding to the plurality of intensity levels includes: determining the central wind speeds corresponding to the plurality of historical satellite cloud images respectively; and determining the historical satellite cloud images corresponding to the plurality of intensity levels respectively based on the central wind speeds.

[0009] In an exemplary embodiment, performing data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels to obtain a target satellite cloud image includes: determining the number of cloud images of the historical satellite cloud images corresponding to the multiple intensity levels; determining the number of target cloud images from the multiple numbers of cloud images; and performing data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels based on the target number of cloud images to obtain a target satellite cloud image.

[0010] In an exemplary embodiment, based on the target cloud image number, data balancing processing is performed on the historical satellite cloud images corresponding to the multiple intensity levels to obtain the target satellite cloud image, including: comparing the target cloud image number with the number of other cloud images except the target cloud image number; data enhancement is performed on the historical satellite cloud images of the intensity level associated with the number of cloud images less than the target cloud image number, and random under-sampling is performed on the historical satellite cloud images of the intensity level associated with the number of cloud images greater than the target cloud image number to obtain the target satellite cloud image.

[0011] In an exemplary embodiment, the target satellite cloud image is input into a neural network model to be trained, the neural network model performs grade prediction based on the target satellite cloud image, and the neural network model is trained according to the grade prediction result to obtain a grade prediction model, including: inputting the target satellite cloud image into the neural network model to be trained, and processing the target satellite cloud image by multiple channels of the neural network model respectively to obtain the target satellite cloud image after channel processing; performing feature extraction on the target satellite cloud image after channel processing based on the feature extraction layer of the neural network model, and performing grade prediction according to the feature extraction result to obtain predicted grade strength; comparing the predicted grade strength with the true label to determine the model loss, training the neural network model according to the model loss to obtain a grade prediction model.

[0012] In an exemplary embodiment, the target satellite cloud image is processed by multiple channels of the neural network model respectively to obtain the target satellite cloud image after channel processing, including: setting weights corresponding to the multiple channels respectively based on a channel attention mechanism; and performing weighted processing on the target satellite cloud image according to the channels after setting the weights to obtain the target satellite cloud image after channel processing.

[0013] In an exemplary embodiment, the method further includes: acquiring real-time satellite cloud images collected in real time; inputting the real-time satellite cloud images into the level prediction model, and the level prediction model predicts the tropical cyclone intensity level based on the real-time satellite cloud images.

[0014] According to another embodiment of the embodiment of the present application, a device for determining the intensity level of a tropical cyclone is also provided, including: an acquisition module for acquiring multiple historical satellite cloud images; a classification module for classifying the multiple historical satellite cloud images to determine the historical satellite cloud images corresponding to multiple intensity levels; a processing module for performing data equalization processing on the historical satellite cloud images corresponding to the multiple intensity levels to obtain a target satellite cloud image; a training module for inputting the target satellite cloud image into a neural network model to be trained, and the neural network model performs a level prediction based on the target satellite cloud image, and trains the neural network model according to the level prediction results to obtain a level prediction model; wherein the level prediction model is used to predict the intensity level of a tropical cyclone.

[0015] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above method when running.

[0016] According to another aspect of the embodiments of the present application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the method through the computer program.

[0017] According to another aspect of the embodiments of the present application, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0018] In an embodiment of the present application, by acquiring multiple historical satellite cloud images, classifying the historical satellite cloud images and performing data balancing processing for different intensity levels, the problem of data imbalance is solved, ensuring that the model can fully learn the characteristics of tropical cyclones of different intensities during training. The target satellite cloud image is input into the neural network model to be trained, and the neural network model performs a grade prediction based on the target satellite cloud image, and then the neural network model is predicted according to the grade prediction result to obtain a grade prediction model. The grade prediction model The trained grade prediction model can quickly process and predict new satellite cloud image data, which makes it possible to monitor and predict tropical cyclone intensity levels in real time, greatly improves the level of automation of the prediction process, and reduces reliance on manual intervention and subjective judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 It is a hardware structure block diagram of a computer terminal of a method for determining a tropical cyclone intensity level according to an embodiment of the present application;

[0021] Figure 2 is a flow chart of a method for determining a tropical cyclone intensity level according to an embodiment of the present application;

[0022] Figure 3 is a flow chart of a method for determining a tropical cyclone intensity level according to an embodiment of the present application;

[0023] Figure 4 is a structural block diagram of a channel attention mechanism according to an embodiment of the present application;

[0024] Figure 5 It is a structural block diagram of a device for determining the tropical cyclone intensity level according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices; "plurality" means two or more.

[0027] The method embodiments provided in the embodiments of the present application can be executed in a computer terminal or similar computing device or cloud platform or independent physical server or software platform, wherein the above software platform is run through one or more servers. Taking running on a computer terminal as an example, Figure 1 is a hardware structure block diagram of a method for determining the intensity level of a tropical cyclone according to an embodiment of the present application. Figure 1 As shown, the computer terminal may include one or more ( Figure 1 Only one is shown in the figure) processor 102 and memory 104 for storing data. In an exemplary embodiment, the computer terminal may also include a transmission device 106 for communication functions and an input and output device 108, wherein the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA. A person skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. Figure 1 More or fewer components as shown, or with Figure 1 Equivalent functions or comparisons shown Figure 1 A different configuration with more features is shown.

[0028] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as computer programs corresponding to the methods in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, that is, the above-mentioned methods are implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0029] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0030] In this embodiment, a method for determining the intensity level of a tropical cyclone is provided, which is applied to the above-mentioned computer device. Figure 2 : is a flow chart of a method for determining a tropical cyclone intensity level according to an embodiment of the present application, the process comprising the following steps:

[0031] Step S202, obtaining a plurality of collected historical satellite cloud images;

[0032] It should be noted that historical satellite cloud images can be obtained from the meteorological satellite database. Historical satellite cloud images can include observation images of multiple infrared channels, each channel corresponding to a different infrared wavelength, which can reveal the characteristics of the cloud layer such as temperature, thickness and water vapor content, which is crucial for the identification and intensity assessment of tropical cyclones. The multiple historical satellite cloud images can specifically be thousands of historical satellite cloud images.

[0033] In some embodiments, the original satellite cloud images obtained may have different resolutions, image sizes and formats. Therefore, the preprocessing step includes image standardization, such as unifying the image size to 300*300 pixels, which helps maintain data consistency and facilitates subsequent model training. In addition, image enhancement, noise removal, contrast adjustment and other operations may be required to improve image quality so that the model can recognize features more accurately.

[0034] Step S204, classifying the plurality of historical satellite cloud images to determine historical satellite cloud images corresponding to a plurality of intensity levels;

[0035] Step S206, performing data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels to obtain a target satellite cloud image;

[0036] It is understandable that the target satellite cloud image is used to train the neural network model.

[0037] Step S208, inputting the target satellite cloud image into the neural network model to be trained, having the neural network model perform grade prediction based on the target satellite cloud image, and training the neural network model according to the grade prediction result to obtain a grade prediction model; wherein the grade prediction model is used to predict the tropical cyclone intensity grade.

[0038] The above steps S202-S208, by acquiring multiple historical satellite cloud images, classifying the historical satellite cloud images and performing data balancing processing for different intensity levels, solve the problem of data imbalance and ensure that the model can fully learn the characteristics of tropical cyclones of different intensities during training. The target satellite cloud image is input into the neural network model to be trained, and the neural network model performs grade prediction based on the target satellite cloud image, and then predicts the neural network model according to the grade prediction result to obtain a grade prediction model. Grade prediction model The trained grade prediction model can quickly process and predict new satellite cloud image data, which makes it possible to monitor and predict tropical cyclone intensity levels in real time, greatly improves the level of automation of the prediction process, and reduces dependence on manual intervention and subjective judgment.

[0039] In an exemplary embodiment, the classification of the multiple historical satellite cloud images to determine the historical satellite cloud images corresponding to multiple intensity levels includes: determining the central wind speeds corresponding to the multiple historical satellite cloud images; and determining the historical satellite cloud images corresponding to multiple intensity levels based on the central wind speeds.

[0040] It should be noted that each historical satellite cloud image is an observation record of a tropical cyclone at a specific point in time, and the intensity level of a tropical cyclone is directly related to the central wind speed. Therefore, the central wind speed data corresponding to each historical satellite cloud image can be obtained from meteorological data. These data usually come from reports or databases of professional meteorological agencies, recording the maximum wind speed near the bottom center of the tropical cyclone at a specific moment. The determination of the central wind speed is the basis for subsequent classification, ensuring that each cloud image is correctly classified into the corresponding intensity level category.

[0041] Specifically, after determining the central wind speed of each historical satellite cloud image, the historical satellite cloud images are classified into different intensity levels according to the definition of tropical cyclone intensity levels. According to the classification standard, the central wind speed is divided into six different ranges, corresponding to different intensity levels: tropical depression (TD, 6-7): the central wind speed is relatively small, and the impact on the environment is relatively limited; tropical storm (TS, 8-9): the central wind speed increases further and begins to cause potential damage to buildings and facilities; strong tropical storm (STS, 10-11): the central wind speed reaches a relatively strong level, and the impact on coastal areas is significant; typhoon (TY, 12-13): the central wind speed reaches the typhoon level, causing serious impacts on coastal and inland areas; strong typhoon (STY, 14-15): the central wind speed is stronger and the destructive power is extremely great; super typhoon (SuperTY, 16 or above): the central wind speed reaches the highest level, the destructive power is extremely strong, and it poses a great threat to human society and the natural environment.

[0042] By comparing the actual central wind speed corresponding to each historical satellite cloud image with the above standards, the historical satellite cloud images can be accurately classified into the corresponding intensity levels. This process ensures the accuracy of the data set and the rationality of the classification, providing structured data support for subsequent deep learning model training.

[0043] In the above embodiment, the collected historical satellite cloud images can be scientifically and reasonably classified, which provides a basis for subsequent data balancing and model training, and helps to improve the efficiency and accuracy of tropical cyclone intensity level determination. This process not only reduces the subjectivity and errors of manual operation, but also provides structured, high-quality training data for deep learning models, which is the key guarantee for the reliability and performance of the entire prediction system.

[0044] In an exemplary embodiment, performing data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels to obtain the target satellite cloud image includes: determining the number of cloud images of the historical satellite cloud images corresponding to the multiple intensity levels; determining the number of target cloud images from the multiple numbers of cloud images; and performing data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels based on the target number of cloud images to obtain the target satellite cloud image.

[0045] Among them, after the classification is completed, the data volume of historical satellite cloud images at each intensity level can be counted. Since the intensity distribution of tropical cyclones is often uneven, for example, cyclones at the tropical depression and tropical storm levels usually occur more frequently, while cyclones at the strong typhoon and super typhoon levels occur less frequently, which will directly lead to huge differences in the number of cloud images at different intensity levels. Counting the number of cloud images can be done by checking the number of images in each classification folder or database to ensure that the number of historical satellite cloud images at each intensity level is accurately recorded. This data will be used for the next step of data balancing processing decision.

[0046] In order to achieve data balance, it is necessary to determine a target number of cloud maps, that is, the ideal number of cloud map data for each intensity level during model training. The selection of this number is based on the analysis of the original data distribution, and usually a value is selected that can ensure the model training effect without significantly increasing the data processing burden.

[0047] For example, the intensity level with the second largest amount of data is used as the benchmark. This is because the number of cloud images of extreme intensity levels (such as super typhoons) is often small, while the number of high-frequency intensity levels (such as tropical depressions) is too large, which may cause the model to tend to recognize these common features and ignore the features in extreme cases. By selecting the intermediate intensity level as the reference, the learning ability of the model can be guaranteed while avoiding the problems caused by data imbalance.

[0048] The purpose of data balancing is to make the amount of cloud data for each intensity level as close as possible to the target number of cloud images. For intensity levels with a small number of cloud images, a generative adversarial network (GAN) can be used to generate additional similar cloud images, which will increase the amount of data in these categories. Generative Adversarial Networks (GAN): An automated unsupervised deep learning algorithm consisting of a generator and a discriminator. Through continuous adversarial training, the performance of both the generator and the discriminator is continuously improved, and finally the generator can generate very realistic sample data.

[0049] GAN can learn the statistical characteristics of real cloud maps and generate new cloud maps with similar features. Although these new cloud maps are not actually observed, they can help the model learn more features from different angles and increase the diversity of data. For intensity levels with too many cloud maps, a random undersampling method can be used to randomly select cloud maps from these categories to reduce their data volume to the target number of cloud maps. This method avoids overfitting of the model to a certain category and ensures that cloud map features of all intensity levels can be represented equally.

[0050] After data balancing, the amount of historical satellite cloud image data of different intensity levels will be closer, forming a balanced target satellite cloud image set. This set will be used for subsequent neural network model training to ensure that the model can take into account all possible characteristics and intensity distribution when predicting the intensity level of tropical cyclones, thereby improving the accuracy and comprehensiveness of the prediction.

[0051] By balancing the data, a common problem in deep learning model training, namely the problem of class imbalance, is solved. This is particularly important for improving the model's ability to identify rare or extreme events, thereby achieving higher performance and reliability in determining the intensity level of tropical cyclones. This method is not only applicable to tropical cyclone prediction, but also has broad reference significance for other deep learning applications involving multi-class imbalanced data sets.

[0052] In an exemplary embodiment, the target satellite cloud image is input into a neural network model to be trained, the neural network model performs grade prediction based on the target satellite cloud image, and the neural network model is trained according to the grade prediction result to obtain a grade prediction model, including: inputting the target satellite cloud image into the neural network model to be trained, and processing the target satellite cloud image by multiple channels of the neural network model respectively to obtain the target satellite cloud image after channel processing; performing feature extraction on the target satellite cloud image after channel processing based on the feature extraction layer of the neural network model, and performing grade prediction according to the feature extraction result to obtain predicted grade strength; comparing the predicted grade strength with the true label to determine the model loss, training the neural network model according to the model loss to obtain a grade prediction model.

[0053] Specifically, the target satellite cloud image can be standardized and formatted into an input form readable by a neural network. The neural network model used in this application is a convolutional neural network (CNN), which can efficiently process image data. The target satellite cloud image usually contains three different infrared band information, each band corresponds to an input channel, which means that the CNN model will have three input channels to process information from near-mid-infrared, mid-infrared and far-infrared satellite cloud images respectively.

[0054] Each input channel of the CNN model will be processed by a series of convolutional layers. The convolutional layers can automatically learn and extract spatial features in the image, such as the shape, structure, and distribution pattern of the cloud layer. Through the channel attention mechanism (SE), the model can assign different weights to tropical cyclone images of different channels, which means that the model can adaptively adjust the degree of attention to information according to the characteristics of the infrared band, so as to more efficiently extract features related to the intensity level from the cloud image.

[0055] After channel processing, the deep convolutional layers and pooling layers of the neural network model will further extract and compress the features in the cloud image to achieve the prediction of the intensity level of tropical cyclones. The addition of pooling layers helps to reduce the dimension of the data while keeping the most important features unchanged, which is especially important when processing high-resolution satellite cloud images. The deep convolutional layers can capture the complex patterns in the image, which are crucial for judging the intensity of tropical cyclones.

[0056] After feature extraction is completed, the output layer of the neural network model will classify the extracted features to predict the intensity level of the tropical cyclone. The output layer usually uses a Softmax function, which can convert the output of multiple nodes into a probability distribution, indicating the probability that the cloud image belongs to different intensity levels. In the present invention, the output layer will have six nodes, corresponding to six intensity levels (TD, TS, STS, TY, STY, SuperTY), and by calculating the output of these nodes, the model will give a predicted intensity level.

[0057] The predicted intensity level needs to be compared with the actual intensity level (true label) corresponding to each cloud image to evaluate the accuracy of the model prediction. The model loss is determined by calculating the difference between the predicted intensity level and the true intensity level, and cross-entropy loss is usually used as the evaluation metric. Cross-entropy loss can quantify the difference between two probability distributions and is suitable for multi-classification problems.

[0058] The purpose of model training is to minimize model loss and improve prediction accuracy by adjusting the weights and biases of the neural network. This process is achieved through the back propagation algorithm, which propagates the model loss from the output layer to the input layer, calculates the gradient of each layer, and then uses the gradient descent method (such as stochastic gradient descent SGD or Adam optimization algorithm) to update the model parameters. The training process is iterative until the model loss drops to a preset threshold or the maximum number of training rounds is reached.

[0059] After enough training rounds, the neural network model can achieve high prediction accuracy and stability. At this time, the structure and parameters of the model can be saved to form the final grade prediction model. This model can be used to predict the intensity grade of real-time or future tropical cyclone cloud images, providing key information for meteorological warning and decision support.

[0060] In the above embodiment, deep learning technology is used to realize automatic prediction of tropical cyclone intensity levels, which not only greatly reduces the dependence on manual feature extraction, but also improves the accuracy and automation level of prediction, providing strong technical support for offshore wind farm operations, storm warning and scientific research.

[0061] In an exemplary embodiment, the target satellite cloud image is processed by multiple channels of the neural network model respectively to obtain the target satellite cloud image after channel processing, including: setting weights corresponding to the multiple channels respectively based on a channel attention mechanism; and performing weighted processing on the target satellite cloud image according to the channels after setting the weights to obtain the target satellite cloud image after channel processing.

[0062] It should be noted that the core idea of ​​the channel attention mechanism (Squeeze and Excitation, SE) is to enable the network to focus on specific channels, which are often crucial to the recognition task. In this application, three channels are set: near-mid-infrared, mid-infrared and far-infrared, each of which carries different information about tropical cyclones. The SE mechanism constructs a channel-level attention model, evaluates the importance of each channel, and then adjusts the channel weight according to the evaluation results. Specifically, the SE module first compresses the feature map of each channel into a vector of fixed length through a global average pooling operation (Squeeze), and then generates channel weights (Excitation) through two fully connected layers (which may include a ReLU activation function and a Sigmoid activation function). This process is actually learning the relationship between channels to determine which channel information is more critical for predicting the intensity level of tropical cyclones.

[0063] When the channel weights are determined, the target satellite cloud images are weighted. The feature map of each channel will be multiplied by the corresponding weight. This step is dynamically performed during the forward propagation of the neural network, ensuring that the model can take into account the latest state of the channel weights when processing each cloud image. The weighted channel feature maps will be combined to form a feature representation that integrates all channel information. This feature representation not only contains the details of the original image, but also takes into account the importance of different infrared bands for tropical cyclone identification, so that the model can use the information more accurately in the subsequent feature extraction and classification process.

[0064] After weighted processing, channel feature maps are obtained, which are regarded as target satellite cloud images after channel processing. They will be sent to the subsequent layers of the neural network (such as convolutional layers and pooling layers) for deeper feature extraction and processing. The channel attention mechanism ensures that the model can focus on the most informative features for determining the intensity level of tropical cyclones from multi-channel satellite cloud images by adaptively adjusting the weights of different channels.

[0065] By introducing the channel attention mechanism, not only the accuracy and robustness of the deep learning model in determining the intensity level of tropical cyclones is improved, but also the interpretability of the model is enhanced, making the final prediction results more reliable. This is because the model can clearly show which infrared bands play a key role in the prediction process, which is very important for understanding the model decision-making process and model tuning.

[0066] In an exemplary embodiment, the method further includes: acquiring real-time satellite cloud images collected in real time; inputting the real-time satellite cloud images into the level prediction model, and the level prediction model predicts the tropical cyclone intensity level based on the real-time satellite cloud images.

[0067] It can be understood that in order to better understand the process of the above-mentioned method for determining the intensity level of a tropical cyclone, the above-mentioned method for determining the intensity level of a tropical cyclone is described below in combination with an optional embodiment, but it is not used to limit the technical solution of the embodiment of the present application.

[0068] refer to Figure 3 As shown, it is a flow chart of the method for determining the intensity level of a tropical cyclone proposed in this application, which mainly consists of 6 steps: 100-dataset import; 200-dataset classification; 300-dataset balancing; 400-CNN network training; 500-CNN network preservation, saving the CNN network that meets the performance requirements for tropical cyclone intensity level determination; 600-CNN network application, obtaining multi-infrared channel tropical cyclone observation data, converting the tropical cyclone observation data of each infrared channel into a standard satellite cloud image of 300*300 pixels through a bilinear interpolation algorithm, and using the saved CNN network to determine the intensity level of the tropical cyclone.

[0069] In the data set import part, the historical satellite multi-infrared channel tropical cyclone observation data of region A, namely the historical satellite cloud images, are imported, and each historical satellite cloud image is converted into a standard satellite cloud image of 300*300 pixels through a bilinear interpolation algorithm.

[0070] In the data set classification part, the data set is divided into six categories: SupperTY, STY, TY, STS, TS, and TD based on the central wind speed of each tropical cyclone infrared cloud image, and the number of tropical cyclone infrared cloud images of different categories is counted.

[0071] In the data set balancing part, tropical cyclone data sets of different categories are substituted into the corresponding GAN for network training. After the training is completed, the samples of the second highest category are used as the base, and GAN is used for data enhancement for sample categories with less than the number of samples in this category, and random under-sampling is performed for sample categories with more than the number of samples in this category, so as to finally balance the number of tropical cyclone cloud images of each category.

[0072] In the CNN network training part, the tropical cyclone infrared cloud image data sets of different categories after data equalization are divided into training set, validation set and test set. The training set is used to train the CNN network, the validation set is used to test the generalization ability of the model and perform parameter tuning at different training stages, and the test set is the final evaluation of the model performance after the model training is completed. If the model accuracy meets the requirements, the CNN network is saved and used for tropical cyclone intensity level determination. If the model accuracy does not meet the requirements, the CNN network will continue to be trained until the model accuracy requirements are met.

[0073] Adversarial neural networks mainly consist of two networks, the generator and the discriminator, and a set of training sets. The generator network contains a Gaussian noise input layer, two fully connected points, multiple convolutional layers, multiple upsampling layers, and an image generation layer. The discriminator network contains an image input layer, multiple convolutional layers, multiple pooling layers, two fully connected layers, and an image authenticity judgment output layer. Adversarial neural networks GAN are trained using the following method: (1) During training, the discriminator network is trained first. The training set data is labeled with true labels and the fake images generated by the generator are labeled with fake labels to form a batch and sent to the discriminator for training. When calculating the loss, the discriminator's judgment on the true data input is made to be close to true, and the judgment on the fake images generated by the generator is made to be close to false. In this process, only the parameters of the discriminator are updated, and the parameters of the generator are not updated. (2) Then the generator network is trained again. Gaussian noise is sent to the generator, and then the fake images generated by the generator are labeled with true labels and sent to the discriminator. When calculating the loss, the discriminator is made to judge the fake images generated by the generator as close to the real ones. In this process, only the parameters of the generator are updated, and the parameters of the discriminator are not updated.

[0074] The convolutional neural network is mainly composed of a meteorological satellite multi-infrared channel input layer, a SE mechanism processing module, multiple convolutional layers, multiple pooling layers, two fully connected layers, and a tropical cyclone intensity level classification output layer. The input layer is a three-channel meteorological satellite infrared cloud image, each channel corresponds to the near-mid-infrared, mid-infrared and far-infrared cloud images, and then the channel attention mechanism SE is used to assign weights to the tropical cyclone images of different channels. Finally, multi-layer convolution and pooling are used to extract the local and global features of tropical cyclones, thereby realizing the classification of tropical cyclone intensity levels. The output layer includes 6 nodes, corresponding to the six categories of SupperTY, STY, TY, STS, TS, and TD. The output layer uses Softmax for feature extraction. The Softmax function is defined as follows:

[0075]

[0076] Among them, z iis the output value of the ith node, and C is the number of output nodes, that is, the number of categories. The Softmax function can be used to convert the output values ​​of multiple categories into a probability distribution ranging from [0, 1] and 1.

[0077] Channel attention mechanism SE module see Figure 4 . SE mainly includes two major operations: Squeeze and Excitation. The input image set X∈RH×W×C, H is the input image height, W is the input image width, and C is the number of channels in the input image set. Squeeze performs global average pooling on the feature X in the spatial dimension H×W, compressing each channel into 1 node. Then use the fully connected layer to compress the C channel nodes into C / r channel nodes, and use the ReLU function to activate the features. The ReLU activation function is defined as follows:

[0078] ReLU(x)=max(0,x)

[0079] Then, a fully connected layer is used to restore C / r channel nodes to C channel nodes, and the sigmoid function is used to activate the features to obtain the channel scale factor sc. The sigmoid function is defined as follows:

[0080]

[0081] Finally, the input tropical cyclone infrared cloud atlas X is multiplied by the channel scale factor sc to obtain the tropical cyclone infrared cloud atlas processed by the SE module.

[0082] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0083] The embodiment of the present application also provides a structural block diagram of a tropical cyclone intensity level determination device, Figure 5 is a structural block diagram of a tropical cyclone intensity level determination device according to an embodiment of the present application; Figure 5 As shown, including:

[0084] An acquisition module 52 is used to acquire a plurality of historical satellite cloud images;

[0085] A classification module 54 is used to classify the plurality of historical satellite cloud images and determine the historical satellite cloud images corresponding to the plurality of intensity levels;

[0086] A processing module 56 is used to perform data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels to obtain a target satellite cloud image;

[0087] The training module 58 is used to input the target satellite cloud image into the neural network model to be trained, and the neural network model performs grade prediction based on the target satellite cloud image, and trains the neural network model according to the grade prediction result to obtain a grade prediction model; wherein the grade prediction model is used to predict the intensity grade of tropical cyclones.

[0088] Through the above device, by obtaining multiple historical satellite cloud images, classifying the historical satellite cloud images and performing data balancing processing for different intensity levels, the problem of data imbalance is solved, ensuring that the model can fully learn the characteristics of tropical cyclones of different intensities during training. The target satellite cloud image is input into the neural network model to be trained, and the neural network model predicts the level based on the target satellite cloud image, and then predicts the neural network model according to the level prediction result to obtain the level prediction model. The trained level prediction model can quickly process and predict new satellite cloud image data, which makes it possible to monitor and predict the intensity level of tropical cyclones in real time, greatly improves the automation level of the prediction process, and reduces the dependence on manual intervention and subjective judgment.

[0089] In an exemplary embodiment, the classification module 54 is further used to determine the central wind speeds corresponding to the multiple historical satellite cloud images respectively; based on the central wind speed, determine the historical satellite cloud images corresponding to multiple intensity levels respectively.

[0090] In an exemplary embodiment, the processing module 56 is also used to determine the number of cloud images of the historical satellite cloud images corresponding to the multiple intensity levels respectively; determine the number of target cloud images from the multiple numbers of cloud images; and based on the target number of cloud images, perform data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels respectively to obtain the target satellite cloud images.

[0091] In an exemplary embodiment, the processing module 56 is further used to compare the target cloud image number with the number of other cloud images other than the target cloud image number; perform data enhancement on historical satellite cloud images of the intensity level associated with the number of cloud images less than the target cloud image number, and perform random under-sampling on historical satellite cloud images of the intensity level associated with the number of cloud images greater than the target cloud image number, to obtain the target satellite cloud image.

[0092] In an exemplary embodiment, the training module 58 is also used to input the target satellite cloud image into the neural network model to be trained, and the target satellite cloud image is processed by multiple channels of the neural network model respectively to obtain the target satellite cloud image after channel processing; based on the feature extraction layer of the neural network model, feature extraction is performed on the target satellite cloud image after channel processing, and grade prediction is performed according to the feature extraction result to obtain predicted grade strength; the predicted grade strength is compared with the true label to determine the model loss, and the neural network model is trained according to the model loss to obtain a grade prediction model.

[0093] In an exemplary embodiment, the training module 58 is also used to set the weights corresponding to the multiple channels respectively based on the channel attention mechanism; according to the channels after setting the weights, the target satellite cloud image is weighted processed to obtain the target satellite cloud image after channel processing.

[0094] In an exemplary embodiment, the device also includes an application module; the application module is used to obtain real-time satellite cloud images collected in real time; the real-time satellite cloud images are input into the level prediction model, and the level prediction model predicts the tropical cyclone intensity level based on the real-time satellite cloud images.

[0095] An embodiment of the present application further provides a storage medium, which includes a stored program, wherein the program executes any of the above methods when it is run.

[0096] Optionally, in this embodiment, the storage medium may be configured to store program codes for executing the following steps:

[0097] S1, obtain multiple historical satellite cloud images;

[0098] S2, classifying the plurality of historical satellite cloud images to determine the historical satellite cloud images corresponding to the plurality of intensity levels;

[0099] S3, performing data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels to obtain a target satellite cloud image;

[0100] S4, inputting the target satellite cloud image into the neural network model to be trained, having the neural network model perform grade prediction based on the target satellite cloud image, and training the neural network model according to the grade prediction result to obtain a grade prediction model; wherein the grade prediction model is used to predict the intensity grade of tropical cyclones.

[0101] An embodiment of the present application further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0102] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0103] Optionally, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0104] S1, obtain multiple historical satellite cloud images;

[0105] S2, classifying the plurality of historical satellite cloud images to determine the historical satellite cloud images corresponding to the plurality of intensity levels;

[0106] S3, performing data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels to obtain a target satellite cloud image;

[0107] S4, inputting the target satellite cloud image into the neural network model to be trained, having the neural network model perform grade prediction based on the target satellite cloud image, and training the neural network model according to the grade prediction result to obtain a grade prediction model; wherein the grade prediction model is used to predict the intensity grade of tropical cyclones.

[0108] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store program codes.

[0109] An embodiment of the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps in any one of the above method embodiments are implemented.

[0110] An embodiment of the present application further provides another computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0111] An embodiment of the present application also provides a computer program, which includes computer instructions stored in a computer-readable storage medium; a processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps in any one of the above method embodiments.

[0112] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.

[0113] Obviously, those skilled in the art should understand that the above modules or steps of the present application can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that herein, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0114] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for determining the intensity level of a tropical cyclone, characterized in that: include: Obtain multiple historical satellite cloud images; Classifying the plurality of historical satellite cloud images to determine historical satellite cloud images corresponding to a plurality of intensity levels; Performing data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels to obtain a target satellite cloud image; The target satellite cloud image is input into the neural network model to be trained, the neural network model performs grade prediction based on the target satellite cloud image, and the neural network model is trained according to the grade prediction result to obtain a grade prediction model; wherein the grade prediction model is used to predict the intensity grade of tropical cyclones.

2. The method according to claim 1, characterized in that The classifying of the plurality of historical satellite cloud images to determine the historical satellite cloud images corresponding to the plurality of intensity levels comprises: Determine the central wind speeds corresponding to the plurality of historical satellite cloud images respectively; Based on the central wind speed, historical satellite cloud images corresponding to a plurality of intensity levels are determined.

3. The method according to claim 1, characterized in that The performing data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels to obtain the target satellite cloud image includes: Determine the number of cloud images of the historical satellite cloud images corresponding to the multiple intensity levels respectively; Determine a target cloud image quantity from a plurality of cloud image quantities; Based on the number of target cloud images, data balancing processing is performed on the historical satellite cloud images corresponding to the multiple intensity levels to obtain the target satellite cloud images.

4. The method according to claim 3, characterized in that The step of performing data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels based on the number of target cloud images to obtain the target satellite cloud images includes: Compare the target cloud image quantity with the cloud image quantities other than the target cloud image quantity; Data enhancement is performed on historical satellite cloud images of intensity levels associated with a number of cloud images less than the number of target cloud images, and random undersampling is performed on historical satellite cloud images of intensity levels associated with a number of cloud images greater than the number of target cloud images, to obtain target satellite cloud images.

5. The method according to claim 1, characterized in that The step of inputting the target satellite cloud image into a neural network model to be trained, having the neural network model perform grade prediction based on the target satellite cloud image, and training the neural network model according to the grade prediction result to obtain a grade prediction model comprises: The target satellite cloud image is input into the neural network model to be trained, and the target satellite cloud image is processed by multiple channels of the neural network model respectively to obtain the target satellite cloud image after channel processing; Based on the feature extraction layer of the neural network model, feature extraction is performed on the target satellite cloud image after the channel processing, and grade prediction is performed according to the feature extraction result to obtain the predicted grade strength; The predicted grade strength is compared with the true label, the model loss is determined, and the neural network model is trained according to the model loss to obtain a grade prediction model.

6. The method according to claim 5, characterized in that The target satellite cloud image is processed by the multiple channels of the neural network model respectively to obtain the target satellite cloud image after channel processing, including: Based on the channel attention mechanism, setting weights corresponding to the multiple channels respectively; According to the channels with set weights, weighted processing is performed on the target satellite cloud image to obtain the target satellite cloud image after channel processing.

7. The method according to claim 1, characterized in that The method further comprises: Get real-time satellite cloud images collected in real time; The real-time satellite cloud image is input into the grade prediction model, and the grade prediction model predicts the tropical cyclone intensity grade based on the real-time satellite cloud image.

8. A device for determining the intensity level of a tropical cyclone, characterized in that: include: The acquisition module is used to obtain multiple historical satellite cloud images; A classification module, used to classify the plurality of historical satellite cloud images, and determine the historical satellite cloud images corresponding to the plurality of intensity levels; A processing module, used for performing data balancing processing on the historical satellite cloud images corresponding to the multiple intensity levels to obtain a target satellite cloud image; A training module is used to input the target satellite cloud image into a neural network model to be trained, and the neural network model performs grade prediction based on the target satellite cloud image, and trains the neural network model according to the grade prediction result to obtain a grade prediction model; wherein the grade prediction model is used to predict the intensity grade of tropical cyclones.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the method described in any one of claims 1 to 7 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 7 through the computer program.

Images