Typhoon detection model training method, typhoon detection method, and typhoon tracking method

Abstract

The present disclosure provides a typhoon detection model training method, a typhoon detection method and a typhoon tracking method. The method comprises: obtaining a plurality of sets of samples, each set of samples comprising a sample flow field bitmap and corresponding label information, the sample flow field bitmap representing a bitmap comprising a cyclone information, the cyclone information being located in the middle of the sample flow field bitmap, the label information representing position information corresponding to the cyclone information, and the plurality of sets of samples comprising training set samples and test set samples; performing multi-round iteration training on the target detection model using the training set samples, recording a first target loss function value and a first target network weight of the target detection model after the end of the target round iteration training; testing the target detection model with the network weight being the first target network weight using the test set samples to obtain an average accuracy; and determining the typhoon detection model obtained by training according to the average accuracy, the first target loss function value and the first target network weight.

Description

Technical Field

[0001] This disclosure relates to the fields of meteorological detection and deep learning technology, and more specifically, to a training method for a typhoon detection model, a typhoon detection method, and a typhoon tracking method. Background Technology

[0002] Typhoon tracking technology is an important tool in typhoon research. This technology can identify typhoons in meteorological data and track important information such as their movement path and landfall location, and has been widely used in typhoon research.

[0003] In typhoon tracking technology, typhoon detection is a crucial issue. Currently, a common method is to use computational programs to detect typhoons in a variable field, grid-by-grid. However, typhoons have relatively small temporal and spatial scales, and some areas may be typhoon-free at certain times. Therefore, when tracking typhoons over large areas, the grid-by-grid detection method results in a relatively slow tracking speed, which is detrimental to typhoon research. Summary of the Invention

[0004] In view of this, this disclosure provides a training method for a typhoon detection model, a typhoon detection method, and a typhoon tracking method.

[0005] One aspect of this disclosure provides a training method for a typhoon detection model, comprising: acquiring multiple sets of samples representing typhoon flow field information, wherein each set of samples includes a sample flow field bitmap and corresponding annotation information, the sample flow field bitmap representing a bitmap including cyclone information located in the middle of the sample flow field bitmap, and the annotation information representing the location information corresponding to the cyclone information; the multiple sets of samples include training set samples and test set samples; performing multiple rounds of iterative training on a target detection model using the training set samples, recording the first target loss function value and the first target network weight corresponding to the target detection model after the completion of the target rounds of iterative training; testing the target detection model with the network weights of the first target network weights using the test set samples to obtain an average accuracy; and determining the trained typhoon detection model based on at least one of the average accuracy, the first target loss function value, and the first target network weights.

[0006] Another aspect of this disclosure provides a typhoon detection method, comprising: acquiring a first flow field bitmap corresponding to a region to be detected; inputting the first flow field bitmap into a typhoon detection model to obtain target bitmap coordinates of a typhoon identification region, wherein the typhoon detection model is trained based on the training method of the typhoon detection model described in this disclosure; determining the initial latitude and longitude coordinates of the typhoon identification region according to the target bitmap coordinates; and determining the typhoon center region according to typhoon characteristics and the initial latitude and longitude coordinates.

[0007] Another aspect of this disclosure provides a typhoon tracking method, comprising: acquiring a second flow field bitmap corresponding to a region to be detected at multiple detection times; determining a target typhoon center region in the second flow field bitmap based on the typhoon detection method described in this disclosure; and determining a typhoon path based on typhoon characteristics and the target typhoon center region.

[0008] Another aspect of this disclosure provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a training method for a typhoon detection model, a typhoon detection method, and a typhoon tracking method according to this disclosure.

[0009] According to embodiments of this disclosure, a method for training a typhoon detection model and detecting typhoons based on the typhoon detection model is proposed. This method at least partially overcomes the technical problem of slow typhoon tracking speed during grid-by-grid detection, significantly improving the speed of typhoon identification and location, and can contribute to typhoon research. The method employs multiple sets of samples to acquire typhoon flow field information. Each set includes a sample flow field bitmap and corresponding annotation information. The sample flow field bitmap represents a bitmap containing cyclone information located in the center of the sample flow field bitmap. The annotation information represents the location information corresponding to the cyclone information. The multiple sets of samples include training set samples and test set samples. The target detection model is iteratively trained multiple times using the training set samples. The first target loss function value and the first target network weight are recorded after each round of iterative training. The test set samples are used to test the target detection model with the first target network weight, obtaining the average accuracy. Based on the average accuracy, the first target loss function value, and the first target network weight, the technical means for determining the trained typhoon detection model are determined. This method proposes a typhoon detection model training and typhoon detection method based on the typhoon detection model. This method at least partially overcomes the technical problem of slow typhoon tracking speed during grid-by-grid detection, significantly improving the speed of typhoon identification and location, and can contribute to typhoon research. Attached Figure Description

[0010] The above and other objects, features, and advantages of this disclosure will become clearer from the following description of embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0011] Figure 1 An exemplary system architecture is illustrated, according to embodiments of the present disclosure, in which at least one of the following methods can be applied: a training method for a typhoon detection model, a typhoon detection method, and a typhoon tracking method.

[0012] Figure 2 A flowchart illustrating a training method for a typhoon detection model according to an embodiment of the present disclosure is shown schematically.

[0013] Figure 3A flowchart illustrating a typhoon detection method according to an embodiment of the present disclosure is shown schematically.

[0014] Figures 4A-4D A schematic diagram illustrating the location of a typhoon identification zone according to an embodiment of the present disclosure is shown.

[0015] Figure 5 A flowchart illustrating a typhoon tracking method according to an embodiment of the present disclosure is shown schematically.

[0016] Figure 6A The illustration shows the typhoon track tracked based on a point-by-point detection method (detection time range: June-September 2020);

[0017] Figure 6B The illustration shows a typhoon path tracked by the typhoon tracking method according to an embodiment of this disclosure;

[0018] Figure 7 A block diagram of a training apparatus for a typhoon detection model according to an embodiment of the present disclosure is shown schematically.

[0019] Figure 8 A block diagram of a typhoon detection device according to an embodiment of the present disclosure is shown schematically.

[0020] Figure 9 A block diagram of a typhoon tracking device according to an embodiment of the present disclosure is schematically shown; and

[0021] Figure 10 A block diagram of a computer system suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Detailed Implementation

[0022] The embodiments of this disclosure will now be described with reference to the accompanying drawings. In this disclosure, "typhoon" can refer generally to a tropical cyclone. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of this disclosure.

[0023] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0024] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0025] When using expressions such as "at least one of A, B, and C," the expression should generally be interpreted in accordance with the meaning commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.). Similarly, when using expressions such as "at least one of A, B, or C," the expression should generally be interpreted in accordance with the meaning commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, or C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.).

[0026] This disclosure provides a method for training a typhoon detection model, a typhoon detection method, and a typhoon tracking method. The method includes acquiring multiple sets of samples representing typhoon flow field information. Each set of samples includes a sample flow field bitmap and corresponding annotation information. The sample flow field bitmap represents a bitmap including cyclone information located in the center of the sample flow field bitmap. The annotation information represents the location information corresponding to the cyclone information. The multiple sets of samples include training set samples and test set samples. The target detection model is iteratively trained multiple times using the training set samples. The first target loss function value and the first target network weights are recorded after each round of iterative training. The target detection model with the first target network weights is tested using the test set samples to obtain an average accuracy. The trained typhoon detection model is determined based on the average accuracy, the first target loss function value, and the first target network weights.

[0027] Figure 1 An exemplary system architecture 100 is illustrated, according to embodiments of the present disclosure, in which at least one of a typhoon detection model training method, a typhoon detection method, and a typhoon tracking method can be applied. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, in order to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments or scenarios.

[0028] like Figure 1As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, and 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the terminal devices 101, 102, and 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0029] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software, etc. (for example only).

[0030] Terminal devices 101, 102, and 103 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0031] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using terminal devices 101, 102, and 103 (for example only). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0032] It should be noted that at least one of the typhoon detection model training method, typhoon detection method, and typhoon tracking method provided in this disclosure embodiment can generally be executed by server 105. Correspondingly, at least one of the typhoon detection model training device, typhoon detection device, and typhoon tracking device provided in this disclosure embodiment can generally be located in server 105. At least one of the typhoon detection model training method, typhoon detection method, and typhoon tracking method provided in this disclosure embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103, and / or server 105. Correspondingly, at least one of the typhoon detection model training device, typhoon detection device, and typhoon tracking device provided in this disclosure embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103, and / or server 105. Alternatively, at least one of the methods in the typhoon detection model training method, typhoon detection method, and typhoon tracking method provided in this embodiment of the disclosure can be executed by terminal devices 101, 102, or 103, or by other terminal devices different from terminal devices 101, 102, or 103. Correspondingly, at least one of the devices in the typhoon detection model training device, typhoon detection device, and typhoon tracking device provided in this embodiment of the disclosure can be disposed in terminal devices 101, 102, or 103, or in other terminal devices different from terminal devices 101, 102, or 103.

[0033] For example, multiple sets of samples characterizing typhoon flow field information can be originally stored in any one of terminal devices 101, 102, or 103 (e.g., terminal device 101, but not limited thereto), or stored on an external storage device and can be imported into terminal device 101. Then, terminal device 101 can locally execute the training method of the typhoon detection model provided in the embodiments of this disclosure, or send multiple sets of samples to other terminal devices, servers, or server clusters, and have the other terminal devices, servers, or server clusters that receive the multiple sets of samples execute the training method of the typhoon detection model provided in the embodiments of this disclosure.

[0034] For example, the first flow field bitmap corresponding to the area to be detected can be originally stored in any one of terminal devices 101, 102, or 103 (e.g., terminal device 101, but not limited thereto), or stored on an external storage device and can be imported into terminal device 101. Then, terminal device 101 can execute the typhoon detection method provided in the embodiments of this disclosure locally, or send the first flow field bitmap to other terminal devices, servers, or server clusters, and have the other terminal devices, servers, or server clusters that receive the first flow field bitmap execute the typhoon detection method provided in the embodiments of this disclosure.

[0035] For example, the second flow field bitmap corresponding to the area to be detected at multiple detection times can be originally stored in any one of terminal devices 101, 102, or 103 (e.g., terminal device 101, but not limited thereto), or stored on an external storage device and can be imported into terminal device 101. Then, terminal device 101 can execute the typhoon tracking method provided in the embodiments of this disclosure locally, or send the second flow field bitmap to other terminal devices, servers, or server clusters, and have the other terminal devices, servers, or server clusters that receive the second flow field bitmap execute the typhoon tracking method provided in the embodiments of this disclosure.

[0036] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0037] Figure 2 A flowchart illustrating a training method for a typhoon detection model according to an embodiment of the present disclosure is shown.

[0038] like Figure 2 As shown, the method includes operations S201 to S204.

[0039] In operation S201, multiple sets of samples representing typhoon flow field information are obtained. Each set of samples includes a sample flow field bitmap and corresponding annotation information. The sample flow field bitmap represents a bitmap containing cyclone information located in the middle of the sample flow field bitmap. The annotation information represents the location information corresponding to the cyclone information. The multiple sets of samples include training set samples and test set samples.

[0040] According to embodiments of this disclosure, the sample flow field bitmap may include at least one of the following: a bitmap determined based on typhoon observation data and observation grid data, a bitmap determined based on typhoon observation data and reanalysis grid data, etc., and may not be limited thereto.

[0041] According to embodiments of this disclosure, typhoon observation data may include data characterizing the location information of the typhoon center, and observation grid data and reanalysis grid data may characterize information such as wind field and temperature in the area where the typhoon is located or within the typhoon identification area.

[0042] According to embodiments of this disclosure, a method for determining a bitmap based on typhoon observation data and observation grid data may include: selecting a rectangular geographical area to be detected as a typhoon, such as a rectangular area of ​​0°-50°N and 90°E-160°E, but not limited to this. Acquiring typhoon observation data (such as IBTrACS typhoon data) generated by the ocean region or major ocean region where this area is located within at least three years (January-December), for example, acquiring IBTrACS typhoon data corresponding to this area within the periods of January-December 2010, January-December 2014, and January-December 2018. The major ocean region may include, for example, the Northwest Pacific region, but the time and area limitations are not limited to this. Statistically determining the typhoon center position at each recording time point in the typhoon observation data. Acquiring observation grid data with a horizontal resolution greater than or equal to 0.5° × 0.5° within a preset area corresponding to the typhoon center position. Based on the observed grid data, a rectangular plotting area is defined, centered on the typhoon center and extending M degrees of longitude east-west and N degrees of latitude north-south. This allows for the creation of an initial flow field map of s pixels × s pixels for each recorded time point. Based on predefined conditions, a first initial flow field map meeting the conditions can be selected from this first initial flow field map and used as a sample flow field map determined from the typhoon observation data and the observed grid data.

[0043] According to embodiments of this disclosure, a method for determining a bitmap based on typhoon observation data and reanalysis grid data may include: selecting a rectangular geographical area for typhoon detection, such as a rectangular area of ​​0°-50°N and 90°E-160°E, but not limited to this; acquiring typhoon observation data generated in the ocean area or major ocean area where this area is located for at least 3 years (January-December), and statistically analyzing the typhoon center position at each recording time point in the typhoon observation data; acquiring reanalysis grid data (such as ERA5 grid data) with a horizontal resolution greater than or equal to 0.5°×0.5°, for example, acquiring ERA5 grid data with a horizontal resolution of 0.25°×0.25°; and, based on the reanalysis grid data, drawing a rectangular drawing area centered on the typhoon center position, extending M degrees of longitude east-west and N degrees of latitude north-south, and drawing a second initial flow field bitmap of s pixels × s pixels for each recording time point. Based on predefined conditions, a second initial flow field map that meets the conditions can be determined from the second initial flow field map, and used as a sample flow field map determined based on typhoon observation data and reanalysis grid data.

[0044] It should be noted that the values ​​of M and N can be the same, for example, both can be 10. The value of s can be 300. The flow field bitmaps obtained above can all be 850-hPa flow field bitmaps. Predefined conditions can include: each map has only one vortex, and this vortex is a cyclone and located in the middle of the map. This method can avoid the same cyclone being labeled repeatedly.

[0045] According to embodiments of this disclosure, cyclones in the first and second initial flow field bitmaps that meet the conditions can be labeled using a labeling tool. The longitude and latitudinal ranges of the labeling rectangle can be within 5 degrees of longitude and latitude, respectively. A labeling file can be generated based on the position coordinates of the four points of the labeling rectangle. The labeling tool can be the image labeling tool LabelImg (labeled image) software, but is not limited to it. The LabelImg software can use the rectangle tool to label cyclones. By setting the label name to vortex and saving it as an XML (Extensible Markup Language) file, a labeling file is obtained. The labeling information corresponding to each sample flow field bitmap can be determined based on the labeling file. Thus, one set of samples can be determined based on one sample flow field bitmap and its corresponding labeling information. The labeled samples can include at least 1500 sets. For example, there can be a total of 1594 labeled samples.

[0046] According to embodiments of this disclosure, after obtaining multiple sets of labeled samples, these sets of samples can be divided into training set samples and test set samples. For example, a larger number of samples can be randomly extracted from all samples as training set samples. Then, the remaining samples are used as test set samples. The number of training set samples can include at least 1000 sets, and the number of test set samples can include at least 100 sets. The ratio of the number of training set samples to the number of test set samples can be 9:1.

[0047] In operation S202, the target detection model is trained through multiple rounds of iterative training using training set samples. The first target loss function value and the first target network weight are recorded after the target detection model has completed the iterative training of the target rounds.

[0048] According to embodiments of this disclosure, the object detection model may include a detection model based on SSD (Single Shot MultiBoxDetector). SSD is an object detection model based on convolutional neural networks, and SSD300 can be used in this embodiment. The target round can be one or more rounds from multiple rounds.

[0049] According to embodiments of this disclosure, when training an SSD is required, parameters such as the learning rate, the number of samples per iteration, the number of iterations, the IOU (Intersection over Union) threshold for prior box matching in loss calculation, and the ratio of positive to negative samples in classification loss calculation can be set first. Then, the SSD can be trained using the training set samples; in this embodiment, GPU acceleration can be used for training. Whenever the number of iterations is an integer multiple of a set value, the network weights and loss function values ​​from one training iteration can be saved to obtain the corresponding first target loss function value and first target network weights.

[0050] It should be noted that the learning rate can be set to, for example, 0.0001. The number of samples in each iteration can be set to, for example, batch_size = 32. The number of iterations can satisfy, for example, the number of iterations > training set sample size ÷ batch_size × 3000, and in a specific embodiment, it can be set to 150000. The IOU threshold can be set to, for example, 0.5. The ratio of positive to negative samples can be set to, for example, 1:3. The setting value can be determined by: setting value > training set sample size ÷ batch_size × 10. In some embodiments, the setting value can be 500. The setting values ​​of various parameters are not limited to these. The number of samples in multiple groups of samples in the above operation S201 can be adjusted appropriately. For example, increasing the number of samples by 1500 groups can improve the detection capability of SSD.

[0051] In operation S203, the target detection model with network weights set as the first target network weights is tested using test set samples to obtain the average accuracy.

[0052] According to embodiments of this disclosure, after obtaining the corresponding first target loss function value and first target network weights, the SSD can be evaluated. When it is necessary to evaluate the trained SSD, the parameters of non-maximum suppression of the detection results, the average accuracy (AP), and the intersection-union (IU) threshold in AP calculation can be set first. The parameters of non-maximum suppression of the detection results may include, for example, the number of selected prediction boxes, a confidence threshold, and an IU threshold. The method for setting the AP calculation can include the VOC 0711 point method. Then, the SSD can be evaluated using test set samples; in this embodiment, GPU acceleration can be used for evaluation. The AP corresponding to each first target network weight can be calculated by sequentially reading all the first target network weights saved during the training steps.

[0053] It should be noted that the crossover ratio (CUP) threshold in the AP calculation can be set to 0.5. The number of predicted boxes can be set to 200. The confidence threshold can be set to 0.001. The CUP threshold in the non-maximum suppression parameters of the detection results can be set to 0.45. However, the settings for various parameters are not limited to these.

[0054] In operation S204, the trained typhoon detection model is determined based on at least one of the average accuracy, the first objective loss function value, and the first objective network weights.

[0055] According to embodiments of this disclosure, a trained typhoon detection model can be determined based on the average accuracy and the weights of the first target network. For example, the SSD corresponding to the first target network weights when the AP is maximum can be determined as the typhoon detection model. A trained typhoon detection model can also be determined based on the first target loss function value and the weights of the first target network. For example, the SSD corresponding to the first target network weights when the first target loss function value is minimum can be determined as the typhoon detection model. Alternatively, a trained typhoon detection model can be determined based on the average accuracy, the first target loss function value, and the weights of the first target network. For example, the SSD corresponding to the first target network weights when the first target loss function value is minimum among multiple larger APs can be determined as the typhoon detection model. Likewise, the SSD corresponding to the first target network weights when the AP is maximum among multiple smaller APs can be determined as the typhoon detection model.

[0056] Through the embodiments of this disclosure, the target detection technology based on convolutional neural networks can be applied to train a typhoon detection model by combining the positional information corresponding to the sample flow field bitmap and the cyclone information located in its center. This model significantly improves the speed of typhoon identification and location, and contributes to typhoon research, thus benefiting disaster prevention and mitigation in coastal areas.

[0057] The following describes specific embodiments. Figure 2 The method shown will be further explained.

[0058] According to embodiments of this disclosure, the target rounds include multiple rounds, the first target loss function value includes multiple loss function values ​​corresponding one-to-one with the multiple rounds, the first target network weight includes multiple network weights corresponding one-to-one with the multiple rounds, and the average accuracy includes multiple average accuracy rates corresponding one-to-one with the multiple network weights; the above operation S204 includes: determining a target average accuracy rate whose value is greater than a first preset value from the multiple average accuracy rates; determining a second target loss function value whose value is less than a second preset value from the multiple loss function values; determining a second target network weight that corresponds to both the target average accuracy rate and the second target loss function value; determining a third target network weight corresponding to the minimum third target loss function value based on the third target loss function value corresponding to the second target network weight; and determining the target detection model with the third target network weight as the typhoon detection model.

[0059] According to embodiments of this disclosure, taking a first preset value of 0.8 and a second preset value of 1 as an example, the method for selecting a first target network weight that meets the conditions may include: if there exists a second target network weight with an AP greater than 0.8 and a loss function value less than 1, then the network weight corresponding to the smallest loss function value is selected as the third target network weight, and the trained typhoon detection model is determined accordingly for typhoon detection, thus ending the debugging process. Otherwise, if there is no third target network weight that meets the aforementioned conditions, the SSD can be re-debugged, including re-dividing the training set samples and test set samples, and re-executing the aforementioned process of training the SSD using the training set samples and evaluating the SSD using the test set samples, until a third target network weight that meets the aforementioned conditions is obtained, thereby determining the trained typhoon detection model.

[0060] Through the above embodiments of this disclosure, a typhoon detection model with higher accuracy can be trained, thereby improving the accuracy of typhoon detection results.

[0061] Figure 3 A flowchart illustrating a typhoon detection method according to an embodiment of the present disclosure is shown.

[0062] like Figure 3 As shown, the method includes operations S301 to S304.

[0063] In operation S301, the first flow field bitmap corresponding to the region to be detected is obtained.

[0064] According to embodiments of this disclosure, the first flow field bitmap includes at least one of the following: a bitmap determined based on observed grid data, a bitmap determined based on reanalysis grid data, etc., and may not be limited thereto.

[0065] According to embodiments of this disclosure, the time range and rectangular geographical area for typhoon detection can be set first. For example, the time range can be set to June-September 2020, and the rectangular geographical area can be set to (0°-50°N, 90°E-160°E), etc., and is not limited to these. Then, the geographical area at each moment in the time period can be divided into several sub-regions. Each sub-region can be a 20°×20° square area, and adjacent sub-regions can have a 10° overlap in the longitude and latitude directions. For example, for a rectangular geographic region (0°-30°N, 100°E-130°E) used to detect a typhoon, if its sub-region includes (0°-20°N, 100°E-120°E), then its three adjacent sub-regions can include (10°N-30°N, 100°E-120°E), (10°N-30°N, 110°E-130°E), and (0°-20°N, 110°E-130°E). Then, observation grid data or reanalysis grid data with a horizontal resolution greater than or equal to 0.5°×0.5° can be obtained for the above time period and region; for example, ERA5 reanalysis grid data with a horizontal resolution of 0.25°×0.25° can be obtained. By drawing a 300-pixel × 300-pixel 850-hPa flow field bitmap for each sub-region at each time point, the first flow field bitmap corresponding to the above-mentioned region to be detected can be obtained.

[0066] It should be noted that the parameter values ​​set in the above embodiments, such as the time range of June-September 2020, the rectangular geographical area of ​​(0°-50°N, 90°E-160°E), the 20°×20° square area, the 10° overlapping area, the 300 pixels×300 pixels, and the 850-hPa flow field bitmap, are not limited to these.

[0067] In operation S302, the first flow field bitmap is input into the typhoon detection model to obtain the target bitmap coordinates of the typhoon identification area. The typhoon detection model is trained based on the aforementioned training method for the typhoon detection model.

[0068] According to embodiments of this disclosure, when it is necessary to locate the typhoon identification area using a typhoon detection model, the parameters for non-maximum suppression of the detection results can be set first, such as selecting 200 prediction boxes, setting the confidence threshold to 0.001, and setting the intersection-over-union (IoU) threshold to 0.45. Then, the weights of the third target network selected based on the aforementioned method can be loaded into SSD, and SSD can be run to locate the typhoon identification area in the first flow field bitmap, obtaining the target bitmap coordinates. In this embodiment, GPU-accelerated detection can also be used.

[0069] It should be noted that the parameter values ​​set in the above embodiments, such as the number of prediction boxes (200), confidence threshold (0.001), and intersection-union ratio (IU) threshold (0.45), are not limited to these.

[0070] In operation S303, the initial latitude and longitude coordinates of the typhoon identification zone are determined based on the target bitmap coordinates.

[0071] According to embodiments of this disclosure, the target bitmap coordinates can characterize the pixel coordinates of the typhoon identification area in the first flow field bitmap. Given a known mapping relationship between pixel coordinates and latitude / longitude coordinates, the initial latitude / longitude coordinates of the typhoon identification area can be determined based on this mapping relationship and the target bitmap coordinates. Alternatively, the target bitmap coordinates can also characterize the latitude / longitude coordinates of the typhoon identification area in the first flow field bitmap; in this case, the target bitmap coordinates can be determined as the initial latitude / longitude coordinates of the typhoon identification area.

[0072] In operation S304, the typhoon center area is determined based on typhoon characteristics and initial latitude and longitude coordinates.

[0073] According to embodiments of this disclosure, when it is necessary to determine typhoon characteristics, a square region with a defined range can be established for each grid point in the typhoon identification zone, for example, two square regions with a range of 3.0° and 6.0°. Each grid point can be referred to as the center grid point of the corresponding square region. By examining the key meteorological variable fields of all grid points within each square region and determining whether they possess the following characteristics, it can be determined whether the corresponding square region has typhoon characteristics.

[0074] It should be noted that the range and number of the square regions established above can be adjusted appropriately. For example, by adding a 2.0° square region, the multi-scale typhoon detection capability can be improved.

[0075] According to embodiments of this disclosure, typhoon characteristics may manifest as at least one of the following:

[0076] a. When the typhoon identification zone is located in the Northern Hemisphere: Within a defined range (in this example, a 3.0° square area corresponds to a 1.5° range in the four cardinal directions, and a 6.0° square area corresponds to a 3.0° range in the four cardinal directions), there are grid points in the four cardinal directions of the isobaric surface below 700 hPa (in this example, an 850 hPa isobaric surface). The meridional wind components of the east-west grid points are positive and negative, respectively, and the zonal wind components of the north-south grid points are positive and negative, respectively. When the typhoon identification zone is located in the Southern Hemisphere: The positive and negative values ​​of the wind components are opposite to those in the Northern Hemisphere.

[0077] b. When the typhoon identification zone is located in the Northern Hemisphere: A rectangular coordinate system is established with the central grid point as the origin. This rectangular coordinate system is extended from the origin along the positive and negative directions of the horizontal and vertical axes to the aforementioned set range. The average values ​​of the zonal and meridional wind components of the wind speed at isobaric surfaces below 700 hPa (850 hPa in this example) within the four quadrants of this set range are as follows: all negative values ​​in the Northwest quadrant, positive and negative values ​​respectively in the Southwest quadrant, all positive values ​​in the Southeast quadrant, and negative and positive values ​​respectively in the Northeast quadrant. When the typhoon identification zone is located in the Southern Hemisphere: the positive and negative values ​​of the wind components in the four quadrants are opposite to those in the Northern Hemisphere.

[0078] c. Within the designated area where the central grid point is located, the maximum wind speed on isobaric surfaces below 700 hPa is not less than the first set value (in this example, the maximum wind speed on isobaric surfaces at 850 hPa is not less than 15 m / s). -1 ).

[0079] d. The average wind speed of isobaric surfaces below 700 hPa (850 hPa in this example) within the set range of the central grid point is greater than the average wind speed of isobaric surfaces above 400 hPa (300 hPa in this example).

[0080] e. The sum of the differences between the air temperature of the central grid point on isobaric surfaces below 700-hPa, isobaric surfaces between 700-hPa and 400-hPa, and isobaric surfaces above 400-hPa (in this example, isobaric surfaces of 700-hPa, 500-hPa, and 300-hPa are used respectively) and the average air temperature of the corresponding isobaric surfaces within the set range area where the central grid point is located is a positive value.

[0081] f. When the designated area where the central grid point is located is in the Northern Hemisphere, the maximum relative vorticity of isobaric surfaces below 700 hPa within that area is not less than the second designated value (in this example, the maximum relative vorticity of the 850 hPa isobaric surface is not less than 5.0 × 10⁻⁶). -5 s -1 When the set range area where the central grid point is located is in the Southern Hemisphere, the maximum relative vorticity of the isobaric surface below 700 hPa within it shall not exceed the second set value.

[0082] g. The sea level pressure at the center grid point is not greater than the third set value (1005-hPa in the example) and is less than the average sea level pressure in the set range area where the center grid point is located.

[0083] It should be noted that the definition of typhoon characteristics is not limited to those mentioned above, but may also include other characteristics known in the art, as long as they conform to the characteristics of a typhoon.

[0084] According to embodiments of this disclosure, after determining the typhoon identification area from the first flow field bitmap based on the typhoon detection model, the typhoon characteristics can be further combined to determine the area with typhoon characteristics from the first flow field bitmap or the typhoon identification area as the typhoon center area.

[0085] Through the above embodiments of this disclosure, the method for typhoon detection based on the typhoon detection model can significantly improve the speed of typhoon identification and location.

[0086] The following describes specific embodiments. Figure 3 The method shown will be further explained.

[0087] According to an embodiment of this disclosure, the above operation S302 may include: inputting a first flow field bitmap into a typhoon detection model to obtain the initial bitmap coordinates of the typhoon identification area and the confidence information corresponding to the initial bitmap coordinates. The initial bitmap coordinates corresponding to the target confidence information whose value is greater than a third preset value are determined as the target bitmap coordinates.

[0088] According to embodiments of this disclosure, after locating the initial bitmap coordinates of the typhoon identification area in the first flow field bitmap based on the typhoon detection model, the initial bitmap coordinates of the located typhoon identification area with a confidence level not less than a third preset value can be saved as the target bitmap coordinates. The third preset value can be, for example, 0.008, and is not limited to this. For example, the third preset value can be appropriately adjusted, such as decreasing it by 0.001, to improve typhoon detection capability.

[0089] Figures 4A-4D A schematic diagram of a typhoon identification zone according to an embodiment of the present disclosure is shown.

[0090] like Figures 4A-4D The figures show examples of typhoon identification zones for rectangular geographic areas (0°-50°N, 90°E-160°E) located based on ERA5 grid data during June-September 2020. The numbers and rectangles represent the confidence information and target location map coordinates of the typhoon identification zone located using the typhoon detection model, respectively. This indicates that the typhoon identification zone located by this typhoon detection method can contain various vortices with typhoon characteristics.

[0091] According to embodiments of this disclosure, the above operation S303 may include: determining the latitude and longitude information corresponding to the area to be detected represented by the first flow field bitmap, wherein the latitude and longitude information includes starting longitude information, starting latitude information, ending longitude information, and ending latitude information; determining the conversion rate between the pixels corresponding to the first flow field bitmap and the latitude and longitude information based on the latitude and longitude information and the pixel information corresponding to the first flow field bitmap; and determining the initial latitude and longitude coordinates based on the target bitmap coordinates, the latitude and longitude information, and the conversion rate.

[0092] According to embodiments of this disclosure, the first flow field bitmap includes, for example, a 300-pixel × 300-pixel flow field bitmap drawn over a square region of 20° × 20°. This yields a conversion rate of 1° = 15 pixels. The target bitmap coordinates may include the pixel coordinates of the typhoon identification region within the detection region corresponding to the first flow field bitmap. For example, it may include the start and end pixel coordinates of the typhoon identification region in the latitude direction, and the start and end pixel coordinates of the typhoon identification region in the longitude direction.

[0093] According to embodiments of this disclosure, when the bitmap coordinates are as follows Figures 4A-4D As shown (the horizontal axis values ​​monotonically increase from left to right, and the vertical axis values ​​monotonically decrease from bottom to top), the latitude and longitude coordinates are as follows: Figures 6A-6B In the scenario shown (where the abscissa representing longitude monotonically increases from left to right, and the ordinate representing latitude monotonically increases from bottom to top), the process of converting the target map coordinates of the located typhoon identification zone into latitude and longitude information can include: First, calculating the starting (ending) latitude information of the typhoon identification zone = the ending latitude information of the area to be detected corresponding to the first flow field map where the typhoon identification zone is located - (the ending (starting) ordinate of the typhoon identification zone ÷ 15), and the starting (ending) longitude information of the typhoon identification zone = the starting longitude information of the area to be detected corresponding to the first flow field map where the typhoon identification zone is located + (the starting (ending) abscissa of the typhoon identification zone ÷ 15), where the ordinate can represent the pixel coordinates of the typhoon identification zone in the latitude direction within the sub-region, and the abscissa can represent the pixel coordinates of the typhoon identification zone in the longitude direction within the sub-region. Then, the calculated latitude and longitude coordinates of the typhoon identification zone can be saved as the initial latitude and longitude coordinates of the determined typhoon identification zone.

[0094] According to embodiments of this disclosure, in the bitmap coordinates, the horizontal coordinate values ​​decrease monotonically from left to right, and the vertical coordinate values ​​increase monotonically from bottom to top; the latitude and longitude coordinates are as follows: Figures 6A-6B In the scenario shown (where the abscissa representing longitude monotonically increases from left to right, and the ordinate representing latitude monotonically increases from bottom to top), the process of converting the target map coordinates of the located typhoon identification zone into latitude and longitude information can include: First, calculating the starting (ending) latitude information of the typhoon identification zone = the starting latitude information of the area to be detected corresponding to the first flow field map where the typhoon identification zone is located + (the starting (ending) ordinate of the typhoon identification zone ÷ 15), and the starting (ending) longitude information of the typhoon identification zone = the ending longitude information of the area to be detected corresponding to the first flow field map where the typhoon identification zone is located - (the ending (starting) abscissa of the typhoon identification zone ÷ 15). Then, the calculated latitude and longitude coordinates of the typhoon identification zone can be saved as the initial latitude and longitude coordinates of the determined typhoon identification zone.

[0095] It should be noted that the calculation methods for the starting (ending) latitude information and the starting (ending) longitude information of the typhoon identification zone can be adjusted according to the changes and corresponding relationships of the bitmap coordinates and latitude and longitude coordinates, and are not limited to those described above.

[0096] According to an embodiment of this disclosure, the above operation S304 may include: in response to determining that at least one of the initial longitude interval and the initial latitude interval corresponding to the initial latitude and longitude coordinates is less than a fourth preset value, determining the target latitude and longitude coordinates corresponding to the typhoon identification area corresponding to the initial latitude and longitude coordinates based on the fourth preset value, the initial latitude and longitude coordinates, the initial longitude interval, and the initial latitude interval, wherein the target longitude interval corresponding to the target latitude and longitude coordinates is closer to the fourth preset value than the initial longitude interval, and the target latitude interval corresponding to the target latitude and longitude coordinates is closer to the fourth preset value than the initial latitude interval. If a grid point with typhoon characteristics is detected within the area corresponding to the target latitude and longitude coordinates, the area corresponding to the target latitude and longitude coordinates is determined as the typhoon center area.

[0097] According to embodiments of this disclosure, a method for determining target latitude and longitude coordinates may include: reading the initial latitude and longitude coordinates of the typhoon identification zone saved in the aforementioned process; and determining the initial longitude interval (initial and final longitude Lon) of the typhoon identification zone. e -Initial starting longitude Lon s )R lon and the initial latitude interval (initial and final latitude Lat) e -Initial starting latitude Lat s )R lat If R Lon If <n°, then modify the initial longitude coordinates of the typhoon identification zone: Lon' s =Lon s -int((nR Lon )÷2),Lon' e =Lon e +int((nR Lon )÷2). If R Lat If <n°, then modify the initial latitude coordinates of the typhoon identification zone: Lat' s =Lat s -int((nR Lat )÷2),Lat' e =Lat e +int((nR Lat )÷2). Lon' e Lon' s Lat' e Lat' sThese can represent the target's ending longitude, starting longitude, ending latitude, and starting latitude, respectively. `n` can represent a fourth preset value, whose range can include values ​​in the interval [1, 6], for example, `n` can be 5. `int()` can retrieve the integer part of a real number.

[0098] It should be noted that the method of updating the initial latitude and longitude coordinates to the target latitude and longitude coordinates is not limited to the above method. Any method that can make the target longitude interval closer to the fourth preset value than the initial longitude interval, and the target latitude interval closer to the fourth preset value than the initial latitude interval, is applicable.

[0099] According to the embodiments of this disclosure, if it is determined that there is a central grid point with the above-mentioned typhoon characteristics in the target typhoon identification area corresponding to the target latitude and longitude coordinates, then the target typhoon identification area can be determined to be the area where the typhoon center is located, that is, a typhoon is detected.

[0100] Figure 5 A flowchart illustrating a typhoon tracking method according to an embodiment of the present disclosure is shown.

[0101] like Figure 5 As shown, the method includes operations S501 to S503.

[0102] In operation S501, the second flow field bitmap corresponding to the region to be detected at multiple detection times is obtained.

[0103] In operation S502, based on the typhoon detection method, the target typhoon center region is determined in the second flow field bitmap.

[0104] When operating S503, the typhoon path is determined based on the typhoon characteristics and the target typhoon's central area.

[0105] According to embodiments of this disclosure, the second flow field bitmap may include at least one of the following: a bitmap determined based on observed grid data, or a bitmap determined based on reanalysis grid data. The specific determination method can be referred to the method for determining the first flow field bitmap described in the foregoing embodiments, and will not be repeated here. After determining the target typhoon center region in the second flow field bitmap, the coordinates of the center grid points with the aforementioned typhoon characteristics within the typhoon identification area corresponding to the second flow field bitmap can be recorded for typhoon tracking.

[0106] For example, in a tracking experiment targeting typhoons (with a lifespan of no less than 30 hours) that formed in the summer of 2020 (June-August) within a rectangular geographic area (0°-50°N, 90°E-160°E) based on ERA5 grid data (6 hours, 0.25°×0.25° grid), the typhoon detection method described above was first used to locate the typhoon identification area from June to September 2020. The location results can be found in [reference needed]. Figures 4A-4D As shown in the diagram. Secondly, this typhoon detection method can be used to identify typhoons within the typhoon identification area. It should be noted that the typhoon detection method can identify multiple candidate center grid points with typhoon characteristics. Based on this, the center grid point representing the typhoon center can be identified from these multiple candidate center grid points using the typhoon center identification method. Finally, according to the typhoon track compilation method, by compiling the coordinates of the center grid point representing the typhoon center at multiple time points, the typhoon track generated from June to August 2020 can be obtained. Thus, typhoon tracking based on the above typhoon detection method can be completed.

[0107] Figure 6A The illustration shows the typhoon track tracked using a point-by-point detection method (detection time range: June-September 2020).

[0108] Figure 6B The illustration shows a typhoon path tracked by the typhoon tracking method according to an embodiment of the present disclosure.

[0109] exist Figures 6A-6B In Chinese, a typhoon symbol indicates the location where a typhoon formed, and a number indicates the time of its formation, in the format mmddhh. Here, mm represents the month, dd the day, and hh the hour.

[0110] based on Figures 6A-6B It can be determined that the typhoon paths tracked by the typhoon detection method described in this embodiment and the point-by-point detection method are consistent. Experimental results show that the typhoon tracking speed based on the typhoon detection method described in this embodiment is approximately 6.9 times that based on the point-by-point detection method. The above experiments demonstrate that the typhoon detection method described in this embodiment can significantly improve typhoon tracking speed.

[0111] Through the above embodiments of this disclosure, a typhoon detection method based on SSD is proposed. This method breaks through the technical bottleneck of previous methods, improves the shortcomings of previous typhoon tracking methods, and enhances typhoon tracking skills.

[0112] Figure 7 A block diagram of a training apparatus for a typhoon detection model according to an embodiment of the present disclosure is shown schematically.

[0113] like Figure 7 As shown, the training device 700 for the typhoon detection model includes a first acquisition module 710, a recording module 720, a testing module 730, and a first determination module 740.

[0114] The first acquisition module 710 is used to acquire multiple sets of samples representing typhoon flow field information. Each set of samples includes a sample flow field bitmap and corresponding annotation information. The sample flow field bitmap represents a bitmap including cyclone information located in the middle of the sample flow field bitmap. The annotation information represents the location information corresponding to the cyclone information. The multiple sets of samples include training set samples and test set samples.

[0115] The recording module 720 is used to perform multiple rounds of iterative training on the target detection model using the training set samples, and to record the first target loss function value and the first target network weight corresponding to the target detection model after the completion of the target round of iterative training.

[0116] The testing module 730 is used to test the target detection model with network weights equal to the first target network weights using the test set samples, and to obtain the average accuracy.

[0117] The first determining module 740 is used to determine the trained typhoon detection model based on at least one of the average accuracy, the first target loss function value, and the first target network weights.

[0118] According to embodiments of this disclosure, the target rounds include multiple rounds, the first target loss function value includes multiple loss function values ​​corresponding one-to-one with the multiple rounds, the first target network weights include multiple network weights corresponding one-to-one with the multiple rounds, and the average accuracy includes multiple average accuracy rates corresponding one-to-one with the multiple network weights. The aforementioned first determining module may include a first determining unit, a second determining unit, a third determining unit, a fourth determining unit, and a fifth determining unit.

[0119] The first determining unit is used to determine a target average accuracy rate that is greater than a first preset value from the plurality of average accuracy rates.

[0120] The second determining unit is used to determine a second target loss function value from the plurality of loss function values ​​that is less than a second preset value.

[0121] The third determining unit is used to determine the second target network weights that correspond to both the target average accuracy and the second target loss function value.

[0122] The fourth determining unit is used to determine the third target network weights corresponding to the minimum third target loss function value based on the third target loss function value corresponding to the second target network weights.

[0123] The fifth determining unit is used to determine the target detection model whose network weight is the same as the third target network weight as the typhoon detection model.

[0124] According to embodiments of this disclosure, the sample flow field bitmap may include at least one of the following: a bitmap determined based on typhoon observation data and observation grid data, and a bitmap determined based on typhoon observation data and reanalysis grid data.

[0125] Figure 8 A block diagram of a typhoon detection device according to an embodiment of the present disclosure is shown schematically.

[0126] like Figure 8 As shown, the typhoon detection device 800 includes a second acquisition module 810, an acquisition module 820, a second determination module 830, and a third determination module 840.

[0127] The second acquisition module 810 is used to acquire the first flow field bitmap corresponding to the region to be detected.

[0128] The module 820 is used to input the first flow field bitmap into the typhoon detection model to obtain the target bitmap coordinates of the typhoon identification area, wherein the typhoon detection model is trained based on the training device of the above-mentioned typhoon detection model.

[0129] The second determining module 830 is used to determine the initial latitude and longitude coordinates of the typhoon identification area based on the target bitmap coordinates.

[0130] The third determining module 840 is used to determine the typhoon center area based on the typhoon characteristics and the initial latitude and longitude coordinates.

[0131] According to embodiments of this disclosure, the above-mentioned obtaining module includes an obtaining unit and a sixth determining unit.

[0132] The acquisition unit is used to input the first flow field bitmap into the typhoon detection model to obtain the initial bitmap coordinates of the typhoon identification area and the confidence information corresponding to the initial bitmap coordinates.

[0133] The sixth determining unit is used to determine the initial bitmap coordinates corresponding to the target bitmap coordinates as the target bitmap coordinates, where the value of the confidence information is greater than the third preset value.

[0134] According to embodiments of this disclosure, the second determining module includes a seventh determining unit, an eighth determining unit, and a ninth determining unit.

[0135] The seventh determining unit is used to determine the latitude and longitude information corresponding to the area to be detected represented by the first flow field bitmap, wherein the latitude and longitude information includes starting latitude information, starting latitude information, ending latitude information and ending latitude information.

[0136] The eighth determining unit is used to determine the conversion rate between the pixels corresponding to the first flow field bitmap and the latitude and longitude information based on the latitude and longitude information and the pixel information corresponding to the first flow field bitmap.

[0137] The ninth determining unit is used to determine the initial latitude and longitude coordinates based on the target bitmap coordinates, the latitude and longitude information, and the conversion rate.

[0138] According to embodiments of this disclosure, the third determining module includes a tenth determining unit and an eleventh determining unit.

[0139] The tenth determining unit is configured to, in response to determining that at least one of the initial longitude interval and the initial latitude interval corresponding to the initial longitude and latitude coordinates is less than a fourth preset value, determine the target longitude and latitude coordinates corresponding to the typhoon identification zone corresponding to the initial longitude and latitude coordinates based on the fourth preset value, the initial longitude and latitude coordinates, the initial longitude interval, and the initial latitude interval, wherein the target longitude interval corresponding to the target longitude and latitude coordinates is closer to the fourth preset value than the initial longitude interval, and the target latitude interval corresponding to the target longitude and latitude coordinates is closer to the fourth preset value than the initial latitude interval.

[0140] The eleventh determining unit is used to determine the area corresponding to the target latitude and longitude coordinates as the typhoon center area when it detects that there are grid points with typhoon characteristics in the area corresponding to the target latitude and longitude coordinates.

[0141] According to embodiments of this disclosure, the first flow field bitmap may include at least one of the following: a bitmap determined based on observed grid data, or a bitmap determined based on reanalysis grid data.

[0142] Figure 9 A block diagram of a typhoon tracking device according to an embodiment of the present disclosure is shown schematically.

[0143] like Figure 9 As shown, the typhoon tracking device 900 includes a third acquisition module 910, a fourth determination module 920, and a fifth determination module 930.

[0144] The third acquisition module 910 is used to acquire the second flow field bitmap corresponding to the region to be detected at multiple detection times.

[0145] The fourth determining module 920 is used to determine the target typhoon center region in the second flow field bitmap based on the typhoon detection device described above.

[0146] The fifth determining module 930 is used to determine the typhoon path based on the typhoon characteristics and the target typhoon center area.

[0147] Any one or more of the modules or units according to embodiments of this disclosure, or at least a portion thereof, may be implemented in a single module. Any one or more of the modules or units according to embodiments of this disclosure may be implemented by dividing them into multiple modules. Any one or more of the modules or units according to embodiments of this disclosure may be at least partially implemented as hardware circuitry, such as a Field-Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a System-on-Chip, a System-on-Substrate, a System-on-Package, an Application-Specific Integrated Circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three methods. Alternatively, one or more of the modules or units according to embodiments of this disclosure may be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0148] For example, any and more of the following modules can be implemented in one module / unit: the first acquisition module 710, the recording module 720, the testing module 730, and the first determination module 740; or the second acquisition module 810, the acquisition module 820, the second determination module 830, and the third determination module 840; or the third acquisition module 910, the fourth determination module 920, and the fifth determination module 930. Alternatively, any one of these modules / units can be split into multiple modules / units. Or, at least some of the functionality of one or more of these modules / units can be combined with at least some of the functionality of other modules / units and implemented in one module / unit. According to embodiments of this disclosure, at least one of the following can be implemented at least partially as hardware circuitry: a first acquisition module 710, a recording module 720, a testing module 730, and a first determination module 740; a second acquisition module 810, an acquisition module 820, a second determination module 830, and a third determination module 840; or a third acquisition module 910, a fourth determination module 920, and a fifth determination module 930. This can be implemented as hardware or firmware, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable method of integrating or packaging the circuitry. Alternatively, it can be implemented in any one of the three methods—software, hardware, and firmware—or in a suitable combination of any of these methods. Alternatively, at least one of the following modules can be implemented, at least partially, as a computer program module: the first acquisition module 710, the recording module 720, the testing module 730, and the first determination module 740; or the second acquisition module 810, the acquisition module 820, the second determination module 830, and the third determination module 840; or the third acquisition module 910, the fourth determination module 920, and the fifth determination module 930. When the computer program module is run, it can perform the corresponding function.

[0149] It should be noted that the training device, typhoon detection device, and typhoon tracking device of the typhoon detection model in the embodiments of this disclosure correspond to the training method, typhoon detection method, and typhoon tracking method of the typhoon detection model in the embodiments of this disclosure. For a detailed description of the training device, typhoon detection device, and typhoon tracking device of the typhoon detection model, please refer to the training method, typhoon detection method, and typhoon tracking method of the typhoon detection model, which will not be repeated here.

[0150] Figure 10 A block diagram of a computer system suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Figure 10 The computer system shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0151] like Figure 10 As shown, a computer system 1000 according to an embodiment of the present disclosure includes a processor 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage portion 1008 into a random access memory (RAM) 1003. The processor 1001 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1001 may also include onboard memory for caching purposes. The processor 1001 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0152] RAM 1003 stores various programs and data required for the operation of system 1000. Processor 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. Processor 1001 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 1002 and / or RAM 1003. It should be noted that the programs may also be stored in one or more memories other than ROM 1002 and RAM 1003. Processor 1001 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0153] According to embodiments of this disclosure, system 1000 may further include an input / output (I / O) interface 1005, which is also connected to bus 1004. System 1000 may also include one or more of the following components connected to I / O interface 1005: an input section 1006 including a keyboard, mouse, etc.; an output section 1007 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to I / O interface 1005 as needed. A removable medium 1011, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 1010 as needed so that computer programs read from it can be installed into storage section 1008 as needed.

[0154] According to embodiments of this disclosure, the method flow according to embodiments of this disclosure can be implemented as a computer software program. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1009, and / or installed from removable medium 1011. When the computer program is executed by processor 1001, it performs the functions defined in the system of embodiments of this disclosure. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0155] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0156] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0157] For example, according to embodiments of this disclosure, a computer-readable storage medium may include the ROM 1002 and / or RAM 1003 described above and / or one or more memories other than ROM 1002 and RAM 1003.

[0158] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this disclosure. When the computer program product is run on an electronic device, the program code enables the electronic device to implement the training method, typhoon detection method, and typhoon tracking method of the typhoon detection model provided in the embodiments of this disclosure.

[0159] When the computer program is executed by the processor 1001, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0160] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 1009, and / or installed from a removable medium 1011. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0161] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0162] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features recited in the various embodiments and / or claims of this disclosure can be combined and / or combined in various ways, even if such combinations or combinations are not expressly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure may be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0163] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A training method for a typhoon detection model, comprising: Multiple sets of samples characterizing typhoon flow field information are obtained, wherein each set of samples includes a sample flow field bitmap and corresponding annotation information. The sample flow field bitmap characterizes a bitmap including cyclone information, the cyclone information being located in the middle of the sample flow field bitmap, and the annotation information characterizing the location information corresponding to the cyclone information. The multiple sets of samples include training set samples and test set samples. The target detection model is trained iteratively for multiple rounds using the training set samples, and the first target loss function value and the first target network weight are recorded after the target detection model has completed the target round of iterative training. The target detection model with network weights equal to the first target network weights was tested using the test set samples to obtain the average accuracy. The trained typhoon detection model is determined based on at least one of the average accuracy, the first target loss function value, and the first target network weights. Wherein, the target rounds include multiple rounds, the first target loss function value includes multiple loss function values ​​corresponding to the multiple rounds, the first target network weights include multiple network weights corresponding to the multiple rounds, and the average accuracy includes multiple average accuracy corresponding to the multiple network weights. The step of determining the trained typhoon detection model based on at least one of the average accuracy, the first target loss function value, and the first target network weights includes: Determine a target average accuracy rate that is greater than a first preset value from the plurality of average accuracy rates; Determine a second target loss function value from the plurality of loss function values ​​that is less than a second preset value; Determine the second target network weights that correspond to both the target average accuracy and the second target loss function value; Based on the third target loss function value corresponding to the second target network weight, determine the third target network weight corresponding to the minimum third target loss function value; The target detection model whose network weights are equal to the network weights of the third target is determined as the typhoon detection model.

2. The method according to claim 1, wherein, The sample flow field bitmap includes at least one of the following: a bitmap determined based on typhoon observation data and observation grid data, and a bitmap determined based on the typhoon observation data and reanalysis grid data.

3. A typhoon detection method, comprising: Obtain the first flow field bitmap corresponding to the region to be detected; The first flow field bitmap is input into the typhoon detection model to obtain the target bitmap coordinates of the typhoon identification area, wherein the typhoon detection model is trained based on the method of any one of claims 1-2; Based on the target bitmap coordinates, determine the initial latitude and longitude coordinates of the typhoon identification zone; Based on the typhoon characteristics and the initial latitude and longitude coordinates, the typhoon center area is determined.

4. The method according to claim 3, wherein, The step of inputting the first flow field bitmap into the typhoon detection model to obtain the target bitmap coordinates of the typhoon identification area includes: The first flow field bitmap is input into the typhoon detection model to obtain the initial bitmap coordinates of the typhoon identification area and the confidence information corresponding to the initial bitmap coordinates. The initial bitmap coordinates corresponding to the target bitmap coordinates are determined by the value of the confidence information being greater than the third preset value.

5. The method according to claim 3, wherein, Determining the initial latitude and longitude coordinates of the typhoon identification zone based on the target bitmap coordinates includes: Determine the latitude and longitude information corresponding to the region to be detected represented by the first flow field bitmap, wherein the latitude and longitude information includes starting longitude information, starting latitude information, ending longitude information, and ending latitude information; Based on the latitude and longitude information and the pixel information corresponding to the first flow field bitmap, determine the conversion rate between the pixels corresponding to the first flow field bitmap and the latitude and longitude information; The initial latitude and longitude coordinates are determined based on the target bitmap coordinates, the latitude and longitude information, and the conversion rate.

6. The method according to claim 3, wherein, The process of determining the typhoon center region based on typhoon characteristics and the initial latitude and longitude coordinates includes: In response to determining that at least one of the initial longitude interval and initial latitude interval corresponding to the initial latitude and longitude coordinates is less than a fourth preset value, the target latitude and longitude coordinates corresponding to the typhoon identification zone corresponding to the initial latitude and longitude coordinates are determined based on the fourth preset value, the initial latitude and longitude coordinates, the initial longitude interval, and the initial latitude interval. The target longitude interval corresponding to the target latitude and longitude coordinates is closer to the fourth preset value than the initial longitude interval, and the target latitude interval corresponding to the target latitude and longitude coordinates is closer to the fourth preset value than the initial latitude interval. If grid points with typhoon characteristics are detected within the region corresponding to the target latitude and longitude coordinates, the region corresponding to the target latitude and longitude coordinates is determined as the typhoon center region.

7. The method according to any one of claims 3-5, wherein, The first flow field bitmap includes at least one of the following: a bitmap determined based on observed grid data, and a bitmap determined based on reanalysis grid data.

8. A typhoon tracking method, comprising: Obtain the second flow field bitmap of the region to be detected at multiple detection times; Based on the method described in any one of claims 3-7, the target typhoon center region in the second flow field bitmap is determined; The typhoon path is determined based on the typhoon characteristics and the target typhoon's central area.

9. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-2, 3-7, or 8.

Images