Water distress target searching method, device, equipment and medium

Abstract

The invention relates to a water distress target searching method, device and equipment and a medium. The drifting speed of a water distress target at a target time point is intelligently predicted through environment prediction information corresponding to a preset space range around the drifting position of the water distress target at the target time point and the drifting speed of the water distress target before the target time point. And the drifting position, the drifting track and the searching range of the water distress target are predicted, and finally cooperative searching is carried out in the searching range through available searching equipment. The environment prediction information is spatial information, the drift speed is time information, and the nonlinear effect of the environment prediction information is considered, so that the drift speed of the water distress target can be accurately predicted on the basis of combining the spatial information and the time information, and an accurate search range is obtained; therefore, the searching device is used to search the water distress target timely and accurately, and the searching efficiency and success rate of the water distress target are improved.

Description

Technical Field

[0001] The present disclosure relates to the field of maritime technology, and in particular, to a method, device, equipment and medium for searching for waterborne distress targets. Background Art

[0002] When searching for waterborne distress targets, it is necessary to predict the drift of waterborne distress targets (i.e., based on the water environment, predict the future drift speed, drift position, drift trajectory and search range of waterborne distress targets). However, the current prediction accuracy of drift speed is relatively low, which leads to inaccurate prediction of the search range, resulting in low search efficiency and low success rate for waterborne distress targets. Summary of the Invention

[0003] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a method, device, equipment and medium for searching for waterborne distress targets, so as to search for waterborne distress targets in a timely and accurate manner, and improve the search efficiency and success rate of waterborne distress targets.

[0004] In a first aspect, an embodiment of the present disclosure provides a method for searching for waterborne distress targets, the method comprising: Obtaining a starting position and a starting time for predicting the drift of a waterborne distress target, the drift prediction including predictions of drift speed, drift position, drift trajectory, and search range; Obtaining environmental prediction information corresponding to a preset spatial range around the starting position at the starting time; Generating a plurality of initial positions for the drift prediction according to the starting position, generating a plurality of initial times for the drift prediction according to the starting time, generating a plurality of initial environmental information for the drift prediction according to the environmental prediction information corresponding to the starting time, and constructing a plurality of groups of initial conditions according to the plurality of initial positions, the plurality of initial times, and the plurality of initial environmental information, each group of initial conditions including any one of the initial positions, any one of the initial times, and any one of the initial environmental information; Predicting the drift speed and drift position of the waterborne distress target at a plurality of target time points after any one of the initial times according to each group of initial conditions, wherein the drift speed of the waterborne distress target at each target time point is obtained according to the environmental prediction information corresponding to a preset spatial range around the drift position of the waterborne distress target at the target time point, and the drift speed of the waterborne distress target before the target time point, and the drift positions of the waterborne distress target at the plurality of target time points form a drift trajectory; Predicting the search range of the waterborne distress target according to the end point of each drift trajectory among the plurality of drift trajectories corresponding to the plurality of groups of initial conditions; Cooperatively search for the water distress target within the search range by using available search devices.

[0005] In a second aspect, an embodiment of the present disclosure provides a water distress target search device, which includes: A first acquisition module, configured to acquire a starting position and a starting time for predicting the drift of a water distress target, where the drift prediction includes predictions of drift speed, drift position, drift trajectory, and search range; A second acquisition module, configured to acquire environmental prediction information corresponding to a preset spatial range around the starting position at the starting time; A generation module, configured to generate multiple initial positions for the drift prediction according to the starting position, generate multiple initial times for the drift prediction according to the starting time, and generate multiple initial environmental information for the drift prediction according to the environmental prediction information corresponding to the starting time; A first construction module, configured to construct multiple groups of initial conditions according to the multiple initial positions, the multiple initial times, and the multiple initial environmental information, where each group of initial conditions includes any one of the initial positions, any one of the initial times, and any one of the initial environmental information; A prediction module, configured to predict the drift speed and drift position of the water distress target at multiple target time points after any one of the initial times according to each group of initial conditions. The drift speed of the water distress target at each target time point is obtained according to the environmental prediction information corresponding to a preset spatial range around the drift position of the water distress target at the target time point, and the drift speed of the water distress target before the target time point. The drift positions of the water distress target at the multiple target time points form a drift trajectory; predict the search range of the water distress target according to the end point of each drift trajectory among the multiple drift trajectories corresponding to the multiple groups of initial conditions; A cooperative search module, configured to cooperatively search for the water distress target within the search range by using available search devices.

[0006] In a third aspect, an embodiment of the present disclosure provides an electronic device, including: A memory; A processor; and A computer program; wherein, the computer program is stored in the memory and is configured to be executed by the processor to implement the method as described in the first aspect.

[0007] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to implement the method as described in the first aspect.

[0008] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, including a computer program which, when executed by a processor, implements the method described in the first aspect.

[0009] The method, apparatus, device and medium for searching for a waterborne distress target provided by the embodiments of the present disclosure obtain a starting position, a starting time and environmental prediction information for predicting the drift of the waterborne distress target, then generate a plurality of initial positions according to the starting position, generate a plurality of initial times according to the starting time, and generate a plurality of initial environmental information according to the environmental prediction information corresponding to the starting time. Further, according to the plurality of initial positions, the plurality of initial times and the plurality of initial environmental information, a plurality of sets of initial conditions are constructed. For each target time point after any initial time included in each set of initial conditions among the plurality of sets of initial conditions, according to the environmental prediction information corresponding to a preset space range around the drift position of the waterborne distress target at the target time point, and the drift speed of the waterborne distress target before the target time point, the drift speed of the waterborne distress target at the target time point is intelligently predicted, and then the search range of the waterborne distress target is predicted. Finally, the available search devices are used to perform collaborative search within the search range. Since the environmental prediction information is spatial information, the drift speed of the waterborne distress target before the target time point is time information, and the non-linear effect of the environmental prediction information is considered, therefore, based on the combination of spatial information and time information, the drift speed of the waterborne distress target at the target time point can be accurately predicted, so as to accurately predict the drift position of the waterborne distress target at each target time point, obtain an accurate drift trajectory and search range, and then use the available search devices to search for the waterborne distress target in a timely and accurate manner, improving the search efficiency and success rate of the waterborne distress target.

[0010] In addition, by constructing a pre-trained intelligent drift speed prediction model, the non-linear interaction of environmental information such as wind, waves and currents and the non-linear influence on the drift speed of the waterborne distress target are fully considered by using an artificial intelligence model, improving the drift prediction accuracy. In addition, by combining remote sensing satellite observation data, the prediction results such as the drift speed, drift trajectory and search range of the waterborne distress target observed by the remote sensing satellite are updated, and the drift speed, drift trajectory and search range of other unobserved distress targets are also corrected and updated, while improving the drift prediction accuracy of waterborne distress targets that can and cannot be observed by satellite remote sensing. Further, based on deep reinforcement learning, a joint search model using unmanned aerial vehicles and unmanned boats is proposed, improving the search efficiency and success rate of waterborne distress targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present disclosure and used together with the specification to explain the principles of the present disclosure.

[0012] To more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0013] Figure 1 It is a flowchart of a method for searching for waterborne distress targets provided by an embodiment of the present disclosure; Figure 2 It is a schematic diagram of an application scenario provided by an embodiment of the present disclosure; Figure 3 It is a schematic diagram of a preset spatial range provided by an embodiment of the present disclosure; Figure 4 It is a schematic diagram of the probability distribution within the search range provided by an embodiment of the present disclosure; Figure 5 It is a schematic diagram of the probability distribution within the search range provided by an embodiment of the present disclosure; Figure 6 It is a flowchart of a method for searching for waterborne distress targets provided by an embodiment of the present disclosure; Figure 7 It is a flowchart of a method for searching for waterborne distress targets provided by an embodiment of the present disclosure; Figure 8 It is a schematic diagram of an intelligent prediction model for drift speed provided by an embodiment of the present disclosure; Figure 9 It is a flowchart of a method for searching for waterborne distress targets provided by an embodiment of the present disclosure; Figure 10 It is a flowchart of a method for searching for waterborne distress targets provided by an embodiment of the present disclosure; Figure 11 It is a schematic diagram of a prediction method for the search range of small targets provided by an embodiment of the present disclosure; Figure 12 It is a flowchart of a method for searching for waterborne distress targets provided by an embodiment of the present disclosure; Figure 13 It is a schematic diagram of a target detection model provided by an embodiment of the present disclosure; Figure 14 It is a schematic diagram of an empty-spectrum double attention module provided by an embodiment of the present disclosure; Figure 15 It is a flowchart of a method for searching for waterborne distress targets provided by an embodiment of the present disclosure; Figure 16 It is a schematic diagram of a spectral attention mechanism provided by an embodiment of the present disclosure; Figure 17 It is a schematic diagram of an empty-spectrum double attention module provided by an embodiment of the present disclosure; Figure 18 Schematic diagram of the spatial attention mechanism provided by an embodiment of the present disclosure; Figure 19 Schematic diagram of the structure of a water distress target search device provided by an embodiment of the present disclosure; Figure 20 Schematic diagram of the structure of an electronic device embodiment provided by an embodiment of the present disclosure. Detailed implementation manners

[0014] In order to more clearly understand the above objects, features, and advantages of the present disclosure, the solutions of the present disclosure will be further described below. It should be noted that, without conflict, the embodiments of the present disclosure and the features in the embodiments may be combined with each other.

[0015] Many specific details are set forth in the following description to facilitate a thorough understanding of the present disclosure, but the present disclosure may be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some embodiments of the present disclosure, rather than all embodiments.

[0016] Currently, the prediction accuracy of the drift speed is relatively low, which in turn leads to inaccurate prediction of the search range, resulting in low efficiency and low success rate in searching for water distress targets. To address this problem, an embodiment of the present disclosure provides a method for searching for water distress targets, which involves the following term explanations: Distress target: A person falling into the water, a ship (such as a fishing boat, a yacht, a cargo ship, etc.), a marine life raft, etc. that is in distress at sea or on water.

[0017] Remote sensing identification: Identification of distress targets based on remote sensing technology, such as identifying the type and location information of distress targets.

[0018] Drift prediction: Based on environmental information at sea or on water (such as wind field, wave field, current field), predicting the drift speed, drift position, drift trajectory, and search range of a water distress target for a period of time in the future.

[0019] Search range: The possible distribution range of a distress target in the future for a period of time after the distress occurs.

[0020] Cooperative search: Based on the predicted search range, comprehensively carrying out the search for distress targets by combining drones, unmanned boats, etc.

[0021] Figure 1Flowchart of the method for searching for water distress targets provided by the embodiments of the present disclosure. Specifically, this method can be executed by a water distress target search device, which can be implemented in software and / or hardware, and can be configured in a server, a server cluster, or a terminal device. Among them, a server cluster can gather multiple servers to perform the same service together. From the perspective of the client, it seems like there is only one server. A server cluster can use multiple computers for parallel computing to obtain a higher computing speed, or use multiple computers for backup, so that the entire system can still operate normally after any one machine breaks down. Terminal devices include, for example, mobile phones, personal digital assistants, tablet computers, wearable devices with displays, desktop computers, laptop computers, all-in-one computers, smart home devices, etc. Figure 2 The figure shows a schematic diagram of an application scenario applicable to the embodiments of the present disclosure. After the remote sensing satellite 21 observes the sea area or water area where the water distress target 20 is located to obtain a remote sensing image, it sends the remote sensing image to the server 22. The server 22 performs image processing on the remote sensing image to determine whether there is a water distress target in the remote sensing image. When there is a water distress target in the remote sensing image, the server 22 can further identify the type of the water distress target and the position information of the water distress target in the remote sensing image, and calculate the geographical coordinates of the water distress target, that is, the remote sensing observation position, according to the position information of the water distress target in the remote sensing image, so as to achieve remote sensing identification. Or, after the remote sensing satellite 21 observes the sea area or water area where the water distress target 20 is located to obtain a remote sensing image, it first determines whether there is a water distress target in the remote sensing image. When there is a water distress target in the remote sensing image, the remote sensing satellite 21 sends the remote sensing image to the server 22. The server 22 performs image processing on the remote sensing image to determine the type of the water distress target and the position information of the water distress target in the remote sensing image, and calculates the geographical coordinates of the water distress target, that is, the remote sensing observation position, according to the position information of the water distress target in the remote sensing image. The remote sensing observation position can be used as the actual position of the water distress target. The method for searching for water distress targets provided by the embodiments of the present disclosure will be introduced below with reference to the figures. As Figure 1 As shown in the figure, the specific steps of this method are as follows: S101. Obtain the starting position and starting time for predicting the drift of the water distress target. The drift prediction includes predictions of drift speed, drift position, drift trajectory, and search range.

[0022] As Figure 2As shown in the figure, the position of the water distress target 20 may change in real time, resulting in the remote sensing satellite 21 not being able to observe the water distress target 20 in real time. Therefore, the server 22 does not always obtain the actual position of the water distress target 20 in real time. Suppose at a certain moment, for example, at time t1, the remote sensing satellite 21 observes the water distress target 20, then time t1 is recorded as the remote sensing observation time point. In this case, the remote sensing observation position of the water distress target 20 is used as the starting position for drift prediction of the water distress target 20, and this remote sensing observation time point is used as the starting time for drift prediction of the water distress target 20. Specifically, drift prediction includes prediction of drift speed, drift position, drift trajectory, and search range. It can be understood that the starting position and starting time for drift prediction of the water distress target 20 are not limited to the remote sensing observation position and remote sensing observation time point. For example, if other ships sailing in the same sea area or water area observe the water distress target 20, the ship can send the observation time of the water distress target 20, the position of the water distress target 20 relative to the ship at this observation time, and the positioning information of the ship at this observation time to the server 22. The server 22 determines the position of the water distress target 20 at this observation time according to the position of the water distress target 20 relative to the ship at this observation time and the positioning information of the ship at this observation time, and uses this position as the starting position for drift prediction of the water distress target 20, and uses the time when the ship observes the water distress target 20 as the starting time for drift prediction of the water distress target 20. Or, the starting position for drift prediction of the water distress target 20 can be the position when the water distress target 20 encounters distress, and the starting time for drift prediction of the water distress target 20 can be the time when the water distress target 20 encounters distress. That is to say, the starting position and starting time for drift prediction of the water distress target 20 are the actual position and time of the water distress target 20. For example, the actual position and time of the water distress target 20 obtained by the server 22 most recently.

[0023] S102. Obtain the environmental prediction information corresponding to the preset spatial range around the starting position at the starting time.

[0024] For example, after determining the starting position and starting time for drift prediction of the water distress target 20, the server 22 further obtains the environmental prediction information corresponding to the preset spatial range around the starting position at the starting time. As Figure 3 shown in the figure, the water area 30 represents the water area where the water distress target 20 is located, the position 31 represents the starting position, the preset spatial range 32 represents the preset spatial range around the starting position, and the preset spatial range can be a spatial range with a longitude range from lon1 to lon2 and a latitude range from lat1 to lat2 centered on the starting position.

[0025] Specifically, the environmental prediction information corresponding to the starting time within the preset space range may be the wind field prediction result, wave field prediction result, and flow field prediction result at a height of 10 meters above the water surface within the preset space range. Among them, the wind field prediction result includes wind speed and wind direction. The wave field prediction result includes wave height, wave direction, wave length, wave period, etc. The flow field prediction result includes flow velocity and flow direction. For example, in the case where the preset space range is divided into multiple high-resolution grids (for example, the high resolution reaches the order of ten meters), the wind field prediction result, wave field prediction result, and flow field prediction result at a height of 10 meters above the water surface within the preset space range are, in sequence, the wind field prediction result, wave field prediction result, and flow field prediction result at a height of 10 meters at the vertex (or center point) of each high-resolution grid within the preset space range. It can be understood that the wind field prediction result, wave field prediction result, and flow field prediction result at a height of 10 meters at the vertex of the same high-resolution grid change with time.

[0026] S103. Generate multiple initial positions for the drift prediction according to the starting position, generate multiple initial times for the drift prediction according to the starting time, generate multiple initial environmental information for the drift prediction according to the environmental prediction information corresponding to the starting time, and construct multiple sets of initial conditions according to the multiple initial positions, the multiple initial times, and the multiple initial environmental information, where each set of initial conditions includes any one of the initial positions, any one of the initial times, and any one of the initial environmental information.

[0027] For example, a random error with a normal distribution is added to the starting position to obtain multiple initial positions, and this set of multiple initial positions constitutes a set representing the uncertainty of the spatial position of the waterborne distress target, and this set is denoted as . A random error with a normal distribution is added to the starting time to obtain multiple initial times, and this set of multiple initial times constitutes a set representing the uncertainty of the time position of the waterborne distress target, and this set is denoted as . A random error with a normal distribution is added to the environmental prediction information corresponding to the starting time to obtain multiple initial environmental information. For example, the environmental prediction information corresponding to the starting time includes the wind field prediction result, wave field prediction result, and flow field prediction result. Further, a random error with a normal distribution is respectively added to the wind field prediction result, the wave field prediction result, and the flow field prediction result to obtain multiple wind field perturbation results (i.e., the results after the wind field prediction result is perturbed multiple times), multiple wave field perturbation results (i.e., the results after the wave field prediction result is perturbed multiple times), and multiple flow field perturbation results (i.e., the results after the flow field prediction result is perturbed multiple times). Further, the wind field prediction result and the multiple wind field perturbation results constitute a set representing the uncertainty of the final wind field prediction result, and this set is denoted as 。The wave field prediction result and the multiple wave field perturbation results form a set representing the uncertainty of the final wave field prediction result, and this set is denoted as 。The flow field prediction result and the multiple flow field perturbation results form a set representing the uncertainty of the final flow field prediction result, and this set is denoted as 。Each of the initial environmental information as described above includes any element in (such as the wind field prediction result or any wind field perturbation result), any element in (such as the wave field prediction result or any wave field perturbation result), and any element in (such as the flow field prediction result or any flow field perturbation result).

[0028] After obtaining , , , , these sets, a method of multivariate random forest sampling is adopted to randomly select one element from each set to form a combination, and this combination is denoted as a set of initial conditions, that is, each set of initial conditions includes any initial position, any initial time, and any initial environmental information. For example, randomly select N times to obtain N sets of initial conditions.

[0029] S104. According to each set of the initial conditions, predict the drift speed and drift position of the waterborne distress target at multiple target time points after the any initial time. The drift speed of the waterborne distress target at each target time point is obtained based on the environmental prediction information corresponding to a preset spatial range around the drift position of the waterborne distress target at the target time point, and the drift speed of the waterborne distress target before the target time point. The drift positions of the waterborne distress target at the multiple target time points form a drift trajectory.

[0030] For example, for each set of the N sets of initial conditions, each set of initial conditions includes any initial position, any initial time, and any initial environmental information. According to the any initial environmental information, the drift speed of the water distress target 20 at the any initial time is calculated. Assume that what needs to be predicted here is the drift trajectory of the water distress target 20 from the any initial time to a certain future moment. For example, the drift trajectory in the subsequent M hours starting from the any initial time. There are M target time points after the any initial time. The time interval between the first target time point among the M target time points and the any initial time is 1 hour, and the time interval between every two adjacent target time points is 1 hour. Further, according to the drift speed at the any initial time and the time interval between the first target time point and the any initial time, which is 1 hour, the drift position of the water distress target 20 at the first target time point can be calculated. Further, according to the environmental prediction information (such as wind field forecast result, wave field forecast result, current field forecast result) corresponding to the first target time point within the preset space range around the drift position of the water distress target 20 at the first target time point, and the drift speed of the water distress target 20 at the any initial time, the drift speed of the water distress target 20 at the first target time point is calculated. Similarly, according to the drift speed of the water distress target 20 at the first target time point and the time interval between the first target time point and the second target time point, which is 1 hour, the drift position of the water distress target 20 at the second target time point can be calculated. Similarly, according to the environmental prediction information corresponding to the second target time point within the preset space range around the drift position of the water distress target 20 at the second target time point, and the drift speed of the water distress target 20 at the any initial time and the drift speed of the water distress target 20 at the first target time point, the drift speed of the water distress target 20 at the second target time point is calculated. That is to say, the drift speed of the water distress target 20 at each target time point is obtained according to the environmental prediction information corresponding to the target time point and the drift speed of the water distress target 20 before the target time point. In addition, the drift positions of the water distress target 20 at the M target time points form a drift trajectory.

[0031] S105. Predict the search range of the water distress target according to the end point of each drift trajectory among the multiple drift trajectories corresponding to the multiple sets of initial conditions.

[0032] For example, a drift trajectory can be generated according to each set of initial conditions as described above. Therefore, N drift trajectories can be generated according to the N sets of initial conditions. Further, according to the end points of each drift trajectory among the N drift trajectories, that is, according to the N end points, the search range of the water distress target 20 is predicted. The search range can be a range including the N end points. For example, according to the N end points, the minimum bounding rectangle is determined, and the minimum bounding rectangle is denoted as the prediction result of the search range. In addition, in some embodiments, the probability distribution within the search range can also be calculated according to the number of end points included in different sub-regions within the search range. As Figure 4 shown, the range 40 represents the search range. The search range includes 4 sub-regions. The number of end points included in each sub-region divided by the total number of end points within the search range can obtain the proportion of the end points in that sub-region. The proportions corresponding to the 4 sub-regions respectively constitute the probability distribution within the search range. It can be understood that the shape of the search range and the shape of each sub-region are not limited in this embodiment. For example, in other embodiments, the shape of the search range and the shape of each sub-region can be circular. As Figure 5 shown, the range 50 represents the search range, and the sub-regions 51 and 52 respectively represent the sub-regions within the search range. Similarly, according to the proportion of the end points in each sub-region, the probability distribution within the search range can be obtained. In addition, the number of sub-regions within the search range is not limited in this embodiment.

[0033] S106. Cooperatively search for the water distress target within the search range by using available search devices.

[0034] For example, the available search devices include drones and unmanned boats. Based on the search range of the water distress target 20, the drone and the unmanned boat are used to conduct a joint search for the water distress target 20 within the search range.

[0035] The disclosed embodiment obtains the starting position, starting time and environmental prediction information for drift prediction of the target in distress on the water, and then generates multiple initial positions according to the starting position, generates multiple initial times according to the starting time, and generates multiple initial environmental information according to the environmental prediction information corresponding to the starting time. Further, multiple groups of initial conditions are constructed based on the multiple initial positions, the multiple initial times and the multiple initial environmental information. For each target time point after any initial time included in each group of initial conditions in the multiple groups of initial conditions, the drift speed of the target in distress on the water at the target time point is intelligently predicted based on the environmental prediction information corresponding to the preset spatial range around the drift position of the target in distress on the water at the target time point, and the drift speed of the target in distress on the water before the target time point, and then the search range of the target in distress on the water is predicted, and finally a collaborative search is performed within the search range through the available search equipment. Since the environmental prediction information is spatial information, the drift speed of the target in distress on the water before the target time point is time information, and the nonlinear effect of the environmental prediction information is taken into account, the drift speed of the target in distress on the water at the target time point can be accurately predicted on the basis of combining the spatial information and the time information, thereby accurately predicting the drift position of the target in distress on the water at each target time point, obtaining a precise drift trajectory and a search range, and then using the available search equipment to search for the target in distress on the water in a timely and accurate manner, thereby improving the search efficiency and success rate of the target in distress on the water.

[0036] Optionally, obtaining environmental prediction information corresponding to a preset spatial range around the starting position at the starting time includes: inputting position information corresponding to multiple first-resolution grids contained in the preset spatial range around the starting position, environmental prediction information corresponding to the multiple first-resolution grids at the starting time, and position information corresponding to multiple second-resolution grids contained in the preset spatial range into a downscaling deep learning model, and using the environmental prediction information corresponding to the multiple second-resolution grids output by the downscaling deep learning model at the starting time as the environmental prediction information corresponding to the preset spatial range around the starting position at the starting time, wherein the second resolution is greater than the first resolution.

[0037] Specifically, the embodiment of the present disclosure uses the first resolution grid as the low-resolution grid and the second resolution grid as the high-resolution grid. Figure 3Taking the preset spatial range 32 around the shown starting position as an example, the preset spatial range 32 can be divided into multiple low-resolution grids or multiple high-resolution grids, and the number of high-resolution grids into which the preset spatial range 32 is divided is greater than the number of low-resolution grids. Since the number of low-resolution grids is smaller than the number of high-resolution grids within the preset spatial range 32, the storage space required to store the position information of the low-resolution grids (such as the longitude and latitude of the vertices of the low-resolution grids) and the environmental prediction information corresponding to the low-resolution grids (such as the wind field forecast results, wave field forecast results, and flow field forecast results at a height of 10 meters at the vertices of the low-resolution grids) is smaller, and the calculation time required to predict the environmental prediction information corresponding to the low-resolution grids is shorter. The environmental prediction information corresponding to the high-resolution grids (such as the wind field forecast results, wave field forecast results, and flow field forecast results at a height of 10 meters at the vertices of the high-resolution grids) is more refined than the environmental prediction information corresponding to the low-resolution grids. Therefore, when a maritime or waterborne distress accident occurs, the environmental prediction information corresponding to the high-resolution grids can be predicted based on the environmental prediction information corresponding to the low-resolution grids and the position information of the low-resolution grids, so as to predict the drift of the waterborne distress target based on the environmental prediction information corresponding to the high-resolution grids. Specifically, the position information corresponding to the multiple low-resolution grids included in the preset spatial range 32, the environmental prediction information corresponding to the multiple low-resolution grids at the starting time, and the position information corresponding to the multiple high-resolution grids included in the preset spatial range 32 (such as the longitude and latitude of the vertices of the high-resolution grids) are input into the downscaling deep learning model, and the downscaling deep learning model outputs the environmental prediction information corresponding to the multiple high-resolution grids at the starting time. Here, the environmental prediction information corresponding to the multiple high-resolution grids output by the downscaling deep learning model at the starting time is used as the environmental prediction information corresponding to the preset spatial range 32 around the starting position at the starting time. That is to say, after a maritime or waterborne distress accident occurs and subsequent drift prediction needs to be carried out, the downscaling deep learning model is used to quickly and intelligently downscale the low-resolution environmental prediction information for the required time period to obtain the high-resolution wind field forecast results, wave field forecast results, and flow field forecast results at a height of 10 meters on the water surface for the required time period.

[0038] Optionally, the environmental prediction information corresponding to a preset spatial range around the drift position of the water distress target at the target time point is obtained in the following manner: input the position information corresponding to a plurality of first-resolution grids included in the preset spatial range around the drift position of the water distress target at the target time point, the environmental prediction information corresponding to the plurality of first-resolution grids at the target time point, and the position information corresponding to a plurality of second-resolution grids included in the preset spatial range into a downscaling deep learning model, and use the environmental prediction information corresponding to the plurality of second-resolution grids at the target time point output by the downscaling deep learning model as the environmental prediction information corresponding to the preset spatial range around the drift position of the water distress target at the target time point, where the second resolution is greater than the first resolution.

[0039] For example, input the position information corresponding to a plurality of low-resolution grids included in the preset spatial range around the drift position of the water distress target 20 at any target time point, the environmental prediction information corresponding to the plurality of low-resolution grids at the target time point, and the position information corresponding to a plurality of high-resolution grids included in the preset spatial range (such as the longitude and latitude of the vertices of the high-resolution grids) into a downscaling deep learning model, and the downscaling deep learning model outputs the environmental prediction information corresponding to the plurality of high-resolution grids at the target time point. Here, use the environmental prediction information corresponding to the plurality of high-resolution grids at the target time point output by the downscaling deep learning model as the environmental prediction information corresponding to the preset spatial range around the drift position of the water distress target 20 at the target time point.

[0040] Optionally, the environmental prediction information includes a wind field forecast result, a wave field forecast result, and a current field forecast result; in the embodiments of the present disclosure, the downscaling deep learning model may specifically be a wind-wave-current coupled downscaling deep learning model based on a physics-informed convolutional long short-term memory network. Specifically, the process of constructing the convolutional long short-term memory network may be as follows: construct an encoder-decoder network, where the encoder extracts the features of the low-resolution input data and the decoder generates the high-resolution output. Use convolutional layers (Conv2D) in the encoder and decoder to extract spatial features, use transposed convolutional layers (Conv2DTranspose) in the decoder to perform feature upsampling to generate the high-resolution output, and add a convolutional long short-term memory (ConvLSTM) module to the encoder-decoder structure to capture the dependencies between adjacent time steps in the time series. The downscaling deep learning model is trained according to the Figure 6 steps shown below: S601. Calculate the environmental prediction information of multiple first-resolution grids in the target water area at preset time intervals within a preset future duration according to the wind-wave-current coupled prediction model, where the wind-wave-current coupled prediction model is a coupled model of an atmospheric numerical model, a wave numerical model, and a hydrodynamic numerical model.

[0041] In the embodiments of the present disclosure, the atmospheric numerical model can output a wind field prediction result, the wave numerical model can output a wave field prediction result, and the hydrodynamic numerical model can output a flow field prediction result. Specifically, the atmospheric numerical model can specifically be the Weather Research & Forecasting Model (WRF). The hydrodynamic numerical model can be the Finite-Volume Community Ocean Model (FVCOM), the Regional Ocean Modeling System (ROMS), etc. The wave numerical model can be the Simulating WAves Nearshore (SWAN), the WAVEWATCH, etc. In the embodiments of the present disclosure, the atmospheric numerical model, the wave numerical model, and the hydrodynamic numerical model can be coupled through a coupler to obtain a wind-wave-current coupled prediction model, enabling the exchange of relevant variables between the atmospheric numerical model, the wave numerical model, and the hydrodynamic numerical model. For example, the atmospheric numerical model provides the wind field and air pressure field at a height of 10 meters above the water surface to the wave numerical model and the hydrodynamic numerical model. The wave numerical model provides the effective wave height, wave direction, wavelength, and spectral peak period to the hydrodynamic numerical model. The hydrodynamic numerical model provides the flow velocity and water level to the wave numerical model.

[0042] Specifically, according to this wind-wave-current coupled prediction model, simulations are carried out to obtain the environmental prediction information of multiple low-resolution grids in the target water area at preset time intervals within a preset future duration. For example, at a fixed time every day, the environmental prediction information of each low-resolution grid in the target water area for every 1 hour within the next 7 days is obtained. This environmental prediction information is recorded as the first simulation result, and this first simulation result is recorded as , the wind field prediction result included in this first simulation result is recorded as , the wave field prediction result included in this first simulation result is recorded as , the flow field prediction result included in this first simulation result is recorded as .

[0043] S602. Calculate the environmental prediction information of multiple second-resolution grids in the target water area at preset time intervals within the preset future duration according to the wind-wave-current coupled prediction model.

[0044] For example, the coupled wind-wave-current prediction model can also be used to calculate the environmental prediction information of each high-resolution grid in the target water area every hour within the next 7 days. This environmental prediction information is denoted as the second simulation result, and the second simulation result is denoted as , and the wind field prediction result included in the second simulation result is denoted as , and the wave field prediction result included in the second simulation result is denoted as , and the current field prediction result included in the second simulation result is denoted as . It can be understood that the time granularity of the second simulation result and the first simulation result is the same, but the environmental prediction information in the second simulation result is more than that in the first simulation result in terms of space.

[0045] S603. Input the position information corresponding to the multiple first-resolution grids, the environmental prediction information of the multiple first-resolution grids at each preset time interval within the future preset duration, and the position information corresponding to the multiple second-resolution grids into the to-be-trained downscaling deep learning model, and the downscaling deep learning model outputs the environmental prediction information of the multiple second-resolution grids at each preset time interval within the future preset duration.

[0046] For example, input the position information corresponding to the multiple low-resolution grids included in the target water area, the environmental prediction information of the multiple low-resolution grids every hour within the next 7 days, and the position information corresponding to the multiple high-resolution grids included in the target water area into the to-be-trained downscaling deep learning model, so that the downscaling deep learning model outputs the environmental prediction information of the multiple high-resolution grids every hour within the next 7 days.

[0047] S604. Train the downscaling deep learning model according to the environmental prediction information of the multiple second-resolution grids at each preset time interval within the future preset duration output by the downscaling deep learning model and the environmental prediction information of the multiple second-resolution grids at each preset time interval within the future preset duration calculated according to the coupled wind-wave-current prediction model.

[0048] For example, denote the difference between the environmental prediction information of the multiple high-resolution grids every hour within the next 7 days output by the downscaling deep learning model and as . In addition, denote the cost function terms composed of the control equations of the atmospheric numerical model, the hydrodynamic numerical model, and the wave numerical model as , , . , , used to constrain the downscaled wind field prediction results, wave field prediction results, and flow field prediction results to satisfy physical constraints. Further, a loss function is constructed, and the downscaled deep learning model is trained according to the loss function . The loss function is expressed as the following formula (1): (1) where , , represent coefficients respectively.

[0049] In addition, in some other embodiments, the wind field prediction results, wave field prediction results, and flow field prediction results at a water surface height of 10 meters in , as well as the longitude and latitude, time information of the low-resolution grid vertices, and the longitude and latitude of the high-resolution grid vertices can also be used as the inputs of the downscaled deep learning model, so that the downscaled deep learning model outputs the environmental prediction information corresponding to the high-resolution grid. Further, according to the difference between the environmental prediction information corresponding to the high-resolution grid output by the downscaled deep learning model and the corresponding to the time information, the downscaled deep learning model is trained.

[0050] Optionally, according to each set of initial conditions, the drift speed and drift position of the waterborne distress target at multiple target time points after any initial time are predicted, including the following steps as shown in Figure 7 : S701. Input any initial environmental information into the drift speed intelligent prediction model corresponding to the waterborne distress target, and the drift speed intelligent prediction model outputs the drift speed of the waterborne distress target at any initial time.

[0051] For example, for each set of initial conditions in the above-mentioned N sets of initial conditions, each set of initial conditions includes any initial position, any initial time, and any initial environmental information. The any initial environmental information can be regarded as the environmental prediction information corresponding to a preset spatial range around any initial position at any initial time. Input the any initial environmental information into the drift speed intelligent prediction model corresponding to the waterborne distress target 20, and the drift speed intelligent prediction model outputs the drift speed of the waterborne distress target 20 at any initial time. Among them, the drift speed intelligent prediction model corresponding to the waterborne distress target 20 can be a drift speed intelligent prediction model corresponding to the type of the waterborne distress target 20. For example, when the waterborne distress target 20 is a ship, the drift speed intelligent prediction model corresponding to the waterborne distress target 20 is the drift speed intelligent prediction model of the ship.

[0052] S702. For each target time point after any one of the initial times, calculate the drift position of the water distress target at the target time point according to the drift speed and drift position of the water distress target at the previous target time point.

[0053] For example, there are M target time points after any one of the initial times. The time interval between the first target time point among the M target time points and any one of the initial times is 1 hour, and the time interval between every two adjacent target time points is 1 hour. According to the drift speed of the water distress target 20 at any one of the initial times and the time interval between the first target time point and any one of the initial times, which is 1 hour, the displacement of the water distress target 20 within this 1-hour period can be calculated. Further, according to any one of the initial positions included in this set of initial conditions (i.e., the drift position of the water distress target 20 at any one of the initial times) and this displacement, the drift position of the water distress target 20 at the first target time point can be calculated.

[0054] S703. Input the environmental prediction information corresponding to the preset spatial range around the drift position of the water distress target at the target time point and the drift speed of the water distress target before the target time point into the intelligent drift speed prediction model, and the intelligent drift speed prediction model outputs the drift speed of the water distress target at the target time point.

[0055] For example, the environmental prediction information corresponding to the preset spatial range around the drifting position of the water distress target 20 at the first target time point, and the drifting speed of the water distress target 20 at any initial time are input into the drifting speed intelligent prediction model, and the drifting speed intelligent prediction model outputs the drifting speed of the water distress target 20 at the first target time point. Similarly, based on the drifting speed and drifting position of the water distress target 20 at the first target time point, and the time interval between the first target time point and the second target time point, which is 1 hour, the drifting position of the water distress target 20 at the second target time point is calculated. Further, the environmental prediction information corresponding to the preset spatial range around the drifting position of the water distress target 20 at the second target time point, the drifting speed of the water distress target 20 at any initial time, and the drifting speed of the water distress target 20 at the first target time point are input into the drifting speed intelligent prediction model, and the drifting speed intelligent prediction model outputs the drifting speed of the water distress target 20 at the second target time point, and so on. That is to say, for each target time point after the any initial time, the environmental prediction information corresponding to the preset spatial range around the drifting position of the water distress target 20 at the target time point, and the drifting speed of the water distress target 20 before the target time point are input into the drifting speed intelligent prediction model, and the drifting speed intelligent prediction model outputs the drifting speed of the water distress target 20 at the target time point. That is, after the environmental prediction information corresponding to the preset spatial range around the drifting position of the water distress target 20 at the current moment and the historical drifting speed of the water distress target 20 are input into the drifting speed intelligent prediction model, the drifting speed intelligent prediction model can output the drifting speed of the water distress target 20 at the current moment.

[0056] For example, the drifting speed of the water distress target 20 at any initial time as described above is denoted as V(0), and the drifting speed of the water distress target 20 at the k-th target time point after any initial time is denoted as V(k). As Figure 8As shown, the environmental prediction information corresponding to the preset spatial range around the drifting position of the water distress target 20 at the k-th target time point, and the drifting speeds of the water distress target 20 before the k-th target time point, namely V(0), V(1), …, V(k−1), are input into the drifting speed intelligent prediction model, and the drifting speed intelligent prediction model outputs the component of V(k) in the east-west direction and the component of V(k) in the north-south direction. Among them, V(0), V(1), …, V(k−1) can be recorded as the time series of historical drifting speeds. For each drifting speed in V(0), V(1), …, V(k−1), the drifting speed obtained by dividing the remotely sensed observation displacement by the remotely sensed observation time interval is preferably used. If there is no corresponding remotely sensed observation displacement, the speed predicted by the drifting speed intelligent prediction model is used. For example, if the remotely sensed satellite observes the water distress target 20 at the (k−1)-th target time point and the k-th target time point respectively, the remotely sensed observation displacement of the water distress target 20 between the (k−1)-th target time point and the k-th target time point is obtained according to the remotely sensed observation position of the water distress target 20 at the (k−1)-th target time point and the remotely sensed observation position of the water distress target 20 at the k-th target time point. Further, the remotely sensed observation displacement is divided by the remotely sensed observation time interval (i.e., the time interval between the (k−1)-th target time point and the k-th target time point) to obtain V(k−1). If there is no corresponding remotely sensed observation position at the (k−1)-th target time point and the k-th target time point, the speed predicted by the drifting speed intelligent prediction model is used for V(k−1). Similarly, when the drifting speed intelligent prediction model predicts V(k−1), the input of the drifting speed intelligent prediction model is the environmental prediction information corresponding to the preset spatial range around the drifting position of the water distress target 20 at the (k−1)-th target time point, and the drifting speeds of the water distress target 20 before the (k−1)-th target time point, namely V(0), V(1), …, V(k−2). In addition, in other embodiments, the number of historical drifting speeds input into the drifting speed intelligent prediction model can also be limited to not more than a preset number, such as 3. For example, the environmental prediction information corresponding to the preset spatial range around the drifting position of the water distress target 20 at the k-th target time point, and the 3 drifting speeds of the water distress target 20 before the k-th target time point, namely V(k−3), V(k−2), V(k−1), are input into the drifting speed intelligent prediction model, and the drifting speed intelligent prediction model outputs the component of V(k) in the east-west direction and the component of V(k) in the north-south direction. Further, V(k) is obtained according to the component in the east-west direction and the component in the north-south direction.

[0057] Specifically, the intelligent prediction model for drift velocity can be an intelligent prediction model for the drift velocity of waterborne distress targets that combines a Convolutional Neural Network (CNN) and spatio-temporal feature splicing. That is to say, this intelligent prediction model for drift velocity is based on the convolutional neural network CNN, which includes an input layer, a convolutional layer, an activation function, a pooling layer, a fully connected layer, and an output layer. Among them, the input data of the input layer is the environmental prediction information input into this intelligent prediction model for drift velocity, and this environmental prediction information is input data with a two-dimensional spatial distribution. The convolutional layer uses a convolutional kernel to perform a convolutional operation on this environmental prediction information to obtain the spatial features of this environmental prediction information. In addition, the convolutional layer can also have a residual connection to solve the problems of gradient disappearance and degradation. Further, multiple types of activation functions such as the Rectified Linear Unit (ReLU), Sigmoid growth curve (Sigmoid), and hyperbolic tangent function (tanh) are used to perform a non-linear transformation on the output of the convolutional layer, so as to reflect the non-linear interaction between the wind field, current field, and wave field. In addition, the pooling layer is used to reduce the size of the spatial features, thereby enhancing the robustness and generalization ability of this intelligent prediction model for drift velocity. Before the fully connected layer of the CNN, the drift velocity of the waterborne distress target after the distress is introduced, that is, the time series of the historical drift velocity as described above. Further, the spatial features of this environmental prediction information and the time series of this historical drift velocity are jointly used as the input of the fully connected layer, so that this intelligent prediction model for drift velocity predicts the component of the drift velocity of the current moment in the east-west direction and the component in the north-south direction.

[0058] Optionally, the intelligent prediction model for drift velocity corresponding to the waterborne distress target is trained according to the steps as shown in Figure 9 below: S901. Obtain the drift positions corresponding to multiple preset time points during the free drift of the experimental target representing the waterborne distress target in the target water area through the Beidou positioning system.

[0059] The intelligent prediction model for drift velocity corresponding to the waterborne distress target as described above is trained according to water drift experiment data. In the water drift experiment, an experimental target representing the waterborne distress target (such as a life raft, lifeboat, dummy, wreckage, etc.) is thrown into the target water area and allowed to drift freely. If the experimental target is a ship, the ship's power is turned off and it is allowed to drift freely. During the experiment, the drift position of this experimental target during the free drift process is obtained through the Beidou positioning system. For example, the drift positions corresponding to multiple preset time points during the free drift process.

[0060] S902. Calculate the drift velocities corresponding to the experimental target at the multiple preset time points according to the drift positions corresponding to the multiple preset time points.

[0061] For example, according to the drift positions respectively corresponding to multiple preset time points in accordance with the experimental objective, and the time intervals between every two adjacent preset time points, the drift speeds respectively corresponding to the experimental objective at the multiple preset time points are calculated.

[0062] S903. For each of the multiple preset time points, input the environmental prediction information corresponding to a preset spatial range around the drift position corresponding to the experimental objective at the preset time point, and the drift speed of the experimental objective before the preset time point into the drift speed intelligent prediction model, and the drift speed intelligent prediction model outputs the drift speed corresponding to the experimental objective at the preset time point.

[0063] For example, for each of the multiple preset time points, input the environmental prediction information corresponding to a preset spatial range around the drift position corresponding to the experimental objective at the preset time point, and the drift speed of the experimental objective before the preset time point into the drift speed intelligent prediction model to be trained, and the drift speed intelligent prediction model outputs the drift speed corresponding to the experimental objective at the preset time point.

[0064] S904. Train the drift speed intelligent prediction model according to the calculated drift speed corresponding to the experimental objective at the preset time point and the drift speed corresponding to the experimental objective at the preset time point output by the drift speed intelligent prediction model.

[0065] For example, for each of the multiple preset time points, calculate the difference between the drift speed corresponding to the experimental objective at the preset time point output by the drift speed intelligent prediction model and the drift speed corresponding to the experimental objective at the preset time point calculated in S902 above, and train the drift speed intelligent prediction model according to the difference.

[0066] In the embodiments of the present disclosure, for each type of water distress target, a water drift experiment needs to be carried out, and the drift speed intelligent prediction model corresponding to the water distress target is trained specifically by using the water drift experiment data. For example, the drift speed intelligent prediction model corresponding to a ship, the drift speed intelligent prediction model corresponding to a person falling into the water, the drift speed intelligent prediction model corresponding to a life raft, the drift speed intelligent prediction model corresponding to wreckage, etc. Without special instructions, they are collectively referred to as the drift speed intelligent prediction model corresponding to the water distress target.

[0067] Since the embodiment of the present disclosure inputs the environmental prediction information (i.e., the output of the downscaled deep learning model) within a certain spatial range around the location point of the experimental target into the drift speed intelligent prediction model to be trained, instead of inputting the actual data of the wind field, wave field, flow field, etc. at the location point of the experimental target into the drift speed intelligent prediction model to be trained. Therefore, compared with the traditional water drift experiment, the water drift experiment described in the embodiment of the present disclosure does not require synchronous follow-up observation of the actual data of the wind field, wave field, flow field, etc. at the location point of the experimental target, thereby greatly reducing the experimental cost. In addition, since it is not necessary to synchronously follow-up observation of the actual data of the wind field, wave field, flow field, etc. at the location point of the experimental target, the number of experiments and experimental data can be increased when conducting water drift experiments in bad weather. In addition, since the embodiment of the present disclosure uses the output of the downscaled deep learning model as a part of the input of the drift speed intelligent prediction model to be trained, the system error of the downscaled deep learning model can be optimized during the training of the drift speed intelligent prediction model, so that the trained drift speed intelligent prediction model does not need to consider the influence of the system error of the downscaled deep learning model during the subsequent reasoning process or application process, that is, when predicting the drift of actual distress targets on the water, thereby improving the accuracy of drift prediction, that is, improving the prediction accuracy of the drift trajectory and search range of the distress targets on the water.

[0068] Optionally, after predicting the search range of the target in distress on water, the method further comprises: guiding the remote sensing satellite to observe within the search range.

[0069] For example, if the remote sensing satellite fails to observe the target in distress on the water, the search range of the target in distress on the water can be used as the reference area for the remote sensing satellite to observe, thereby guiding the remote sensing satellite to conduct on-site observation within the search range. In addition, since the search range is an area where the target in distress on the water may arrive in the future, if there are multiple remote sensing satellites, the server 22 can ask the remote sensing satellite 21 whether it can reach the sky above the search range in the future. If the remote sensing satellite 21 can arrive on time, the remote sensing satellite 21 is guided to conduct on-site observation within the search range. If the remote sensing satellite 21 cannot arrive on time, other remote sensing satellites that can arrive on time are guided to conduct on-site observation within the search range.

[0070] Optionally, the method also includes: when the remote sensing satellite observes the distressed target on the water, based on the remote sensing observation position and remote sensing observation time point of the distressed target on the water, using the drift speed intelligent prediction model corresponding to the distressed target on the water to perform a re-drift prediction on the distressed target on the water, so as to obtain a new search range of the distressed target on the water, the remote sensing observation position is the starting position for the re-drift prediction, and the remote sensing observation time point is the starting time for the re-drift prediction.

[0071] For example, when the remote sensing satellite 21 observes the water distress target 20 within the search range, record the remote sensing observation position and the remote sensing observation time point of the remote sensing satellite 21 for the water distress target 20. Further, use this remote sensing observation position as the starting position for predicting the next drift of the water distress target 20, and use this remote sensing observation time point as the starting time for predicting the next drift of the water distress target 20, and predict the next drift of the water distress target 20. That is, use the remote sensing observation position of the remote sensing satellite 21 for the water distress target 20 as the precise position of the water distress target 20 at this remote sensing observation time point, and use this remote sensing observation position and this remote sensing observation time point as the starting position and starting time for the next drift prediction by the drift speed intelligent prediction model respectively. The process of the next drift prediction refers to the process of the drift prediction described above, which will not be elaborated here.

[0072] Optionally, the method further includes the following steps as Figure 10 shown: S1001. When the remote sensing satellite observes the water distress target within the search range, select a preset number of end points around the water distress target within the search range.

[0073] For example, there are 6 groups of initial conditions for predicting the drift of the water distress target 20, and these 6 groups of initial conditions are sequentially recorded as initial condition 1, initial condition 2,..., initial condition 6. The initial positions included in initial condition 1, initial condition 2,..., initial condition 6 are the initial positions 1, initial positions 2,..., initial positions 6 as Figure 11 shown. Six drift trajectories can be generated according to these 6 groups of initial conditions, thereby obtaining 6 end points, and these 6 end points are the end points 7 - end points 12 as Figure 11 shown. According to end points 7 - end points 12, obtain the search range 110 of the water distress target 20. Assume that the remote sensing satellite observes the water distress target 20 within the search range 110. Further, select a preset number of end points around the water distress target 20 within the search range 110. For example, select 3 end points around the water distress target 20, which are end points 7 - end points 9 respectively. Or, select the end points within the preset range 111 around the water distress target 20, such as end points 7 - end points 9.

[0074] S1002. According to the initial conditions corresponding to the preset number of end points respectively, and the drift speed intelligent prediction model corresponding to other distress targets that are in distress simultaneously with the water distress target, predict the drift speed, drift position, drift trajectory and search range of the other distress targets, where the sizes of the other distress targets are smaller than the size of the water distress target.

[0075] For example Figure 11As shown, based on end points 7, 8, and 9, trace back to the initial conditions corresponding to end points 7, 8, and 9 respectively, such as initial condition 1, initial condition 2, and initial condition 3. Further, based on initial condition 1, initial condition 2, initial condition 3, and the drift speed intelligent prediction model corresponding to other distress targets that are in distress simultaneously with the water distress target 20, predict the drift trajectories and search ranges of other distress targets. The specific prediction process refers to the prediction process described in the above embodiments and will not be elaborated here. For example, the water distress target 20 is a distress target with a relatively large size such as a fishing boat, a yacht, or a cargo ship, which is easily observable by remote sensing satellites. Other distress targets are distress targets with a relatively small size such as people and wreckage, which are not easily observable by remote sensing satellites.

[0076] For example, the drift trajectory of other distress targets predicted according to initial condition 1 is b, the drift trajectory of other distress targets predicted according to initial condition 2 is a, and the drift trajectory of other distress targets predicted according to initial condition 3 is c. End point 14 is the end point of drift trajectory a, end point 13 is the end point of drift trajectory b, and end point 15 is the end point of drift trajectory c. Further, based on end points 13, 14, and 15, obtain the search range of other distress targets, such as search range 112.

[0077] In a water distress accident, there are often medium and large-sized targets such as ships and life rafts, as well as small-sized targets such as people and wreckage. However, medium and large-sized targets may be identified and located by remote sensing satellites, while small-sized targets may not be identified and located by remote sensing satellites. Through the embodiments of the present disclosure, not only can the search range of water distress targets with a relatively large size be predicted, but also based on the remote sensing observation position of the water distress target in the search range, the drift trajectories and search ranges of other small-sized distress targets that are in distress simultaneously with the water distress target can be deduced, thus making up for the defect that small-sized targets cannot be identified by remote sensing satellites and improving the timeliness and accuracy of searching for small-sized targets.

[0078] In the embodiments of the present disclosure, the process in which the server 22 receives the remote sensing image sent by the remote sensing satellite 21 and the server 22 performs remote sensing identification on the water distress target may be as follows Figure 12 shown in the following several steps: S1201. Input the remote sensing image captured by the remote sensing satellite into a target detection model. The target detection model includes a backbone network, a spatial-spectral dual attention module, and a detection head. The backbone network is used to extract features from the remote sensing image to obtain a feature map.

[0079] For example, a target detection model is deployed on the server 22, and this target detection model 130 is as Figure 13As shown in the figure. Specifically, the target detection model includes a backbone network, a spatial-spectral dual attention module, and a detection head. Among them, the backbone network can be a high-performance real-time target detector (YOLOX). The server 22 inputs the received remote sensing image into the target detection model. The target detection model first extracts features from the remote sensing image through the backbone network to obtain a feature map. For example, the backbone network includes multiple feature layers that sequentially extract features from the remote sensing image, and each feature layer outputs a feature map. For example, the backbone network includes 5 feature layers, and the output of the previous feature layer is the input of the next feature layer. For example Figure 13 The feature map 131 shown in the figure is the output of the first feature layer, the feature map 132 is the output of the second feature layer, the feature map 133 is the output of the third feature layer, the feature map 134 is the output of the fourth feature layer, and the feature map 135 is the output of the fifth feature layer.

[0080] S1202. Process the feature map through the spatial-spectral dual attention module. The detection head is used to predict the type of the water distress target in the remote sensing image and the position information of the water distress target in the remote sensing image according to the output information of the spatial-spectral dual attention module.

[0081] For example, the result obtained after the feature map 135 is processed by the spatial-spectral dual attention module is fused with the feature map 134 to obtain the feature map 136. The result obtained after the feature map 136 is processed by the spatial-spectral dual attention module is fused with the feature map 133 to obtain the feature map 137. The result obtained after the feature map 137 is processed by the spatial-spectral dual attention module is fused with the feature map 136 to obtain the feature map 138. The result obtained after the feature map 138 is processed by the spatial-spectral dual attention module is fused with the feature map 135 to obtain the feature map 139. Then, the feature map 137 is input into the detection head 1310, the feature map 138 is input into the detection head 1311, and the feature map 139 is input into the detection head 1312. The detection head 1310 outputs the type of the water distress target in the remote sensing image, such as a ship, a fishing boat, etc. The detection head 1311 outputs the position information of the water distress target in the remote sensing image. The detection head 1312 outputs the confidence levels of the first two results. The confidence level can be the confidence levels corresponding to the first two results respectively, or can be the total confidence level calculated comprehensively according to the confidence levels corresponding to the first two results.

[0082] S1203. Determine the remote sensing observation position of the water distress target according to the position information of the water distress target in the remote sensing image.

[0083] Since each pixel in the remote sensing image corresponds to a geographic coordinate, therefore, according to the position information of the water distress target in the remote sensing image, the geographic coordinate of the water distress target, that is, the remote sensing observation position, can be determined.

[0084] Optionally, the hyperspectral dual attention module includes a spectral attention mechanism and a spatial attention mechanism. As Figure 14 shown, the feature map 140 represents the feature map input to the hyperspectral dual attention module, and the spectral attention mechanism and the spatial attention mechanism are connected in series. The processing of the feature map by the hyperspectral dual attention module includes the following steps as Figure 15 shown: S1501. Extract spectral features from the feature map through the spectral attention mechanism to obtain a one-dimensional spectral attention map.

[0085] For example, the feature map 140 is represented as . The spectral features of the feature map 140 are extracted through the spectral attention mechanism (CA) to obtain a one-dimensional spectral attention map, which is denoted as . This one-dimensional spectral attention map is used to regulate the importance of different channels in the feature map, further improving the robustness and expression ability of the features. As Figure 16 shown is a schematic diagram of the spectral attention mechanism, where 160 represents the input of the spectral attention mechanism. For example, . In this spectral attention mechanism, first, average pooling and max pooling operations are used to aggregate the spatial information of the feature map, thereby generating two different spatial context descriptors. For example, the spatial context descriptor obtained by average pooling of is represented as , and the spatial context descriptor obtained by max pooling of is represented as . Further, and are passed into the shared network to obtain the feature vectors and . The shared network uses a multi-layer perceptron (MLP) with one hidden layer to reduce the overhead of network parameters, while effectively improving the feature representation ability and extraction ability. Then, in an element-wise summation manner, and are combined to obtain a combined result, which is represented as . Further, the combined result is processed using the Sigmoid function to obtain . can be expressed as the following formula (2): (2) S1502. Multiply the feature map by the spectral attention map and then add it to the feature map to obtain an intermediate result.

[0086] To obtain rich spectral features, the embodiments of the present disclosure adopt a combination of a spectral attention mechanism and residual learning. For example Figure 14 as shown, the feature map and the spectral attention map are multiplied element by element and then added to to obtain an intermediate result, which is denoted as , and can be expressed as formula (3) shown below: (3) wherein, can be the intermediate result 141 as Figure 14 shown.

[0087] In addition, in the embodiments of the present disclosure, the empty-spectrum dual attention module may include multiple (e.g., 5) spectrally-attentive mechanisms in series, as Figure 17 shown. Through these multiple spectrally-attentive mechanisms in series, deep spectral features can be extracted. In this case, as Figure 17 shown, the intermediate result 171 is the intermediate result obtained by multiplying the output of the last spectrally-attentive mechanism by its input and then adding the input.

[0088] S1503. Extract spatial features from the intermediate result through the spatial attention mechanism to obtain a two-dimensional spatial attention map.

[0089] As Figure 14 shown, the spatial attention mechanism extracts spatial features from the intermediate result 141 to obtain a two-dimensional spatial attention map. Or as Figure 17 shown, the spatial attention mechanism extracts spatial features from the intermediate result 171 to obtain a two-dimensional spatial attention map. As Figure 18 shown is a schematic diagram of the spatial attention mechanism, wherein 180 represents the input of the spatial attention mechanism. For example, this input can be Figure 14 the intermediate result 141 as Figure 17 shown or the intermediate result 171 as shown. Here, this input is denoted as and . In this spatial attention mechanism (SA), first, average pooling and max pooling are used to process and It can be regarded as context information for guiding the generation of attention. After obtaining these context descriptors, these context descriptors are concatenated along the channel dimension, and a convolutional layer is applied to process them. Then, the output of the convolutional layer is processed using the Sigmoid function to obtain a two-dimensional spatial attention map, which is represented as . It is represented by the following formula (4): (4) where represents a convolutional operation with a kernel size of 7×7.

[0090] S1504. Multiply the intermediate result by the spatial attention map and then add it to the intermediate result to obtain the output information of the spatial-spectral dual attention module.

[0091] For example Figure 14 or Figure 17 as shown, the intermediate result and the spatial attention map are multiplied element by element and then added to to obtain the output information of the spatial-spectral dual attention module, which is represented as . It can be represented by the following formula (5): (5) In the embodiments of the present disclosure, by adjusting the weights of different regions in the spatial attention map, the spatial attention mechanism can extract spatial features more precisely, thereby improving the performance of the target detection model. In addition, the spatial-spectral dual attention module in the embodiments of the present disclosure is composed of a spectral attention mechanism and a spatial attention mechanism in series. By outputting a one-dimensional spectral attention map through the spectral attention mechanism and a two-dimensional spatial attention map through the spatial attention mechanism, the spectral attention mechanism is used to reallocate each weight and accordingly adjust the contribution degree of each channel feature, so that the network pays more attention to useful features. Then, the spatial attention mechanism is used to identify the local regions crucial for the current task in the image, enabling the target detection model to focus on the most important visual information when processing complex scenes, ignoring irrelevant backgrounds or interferences, thereby greatly improving the accuracy of the image processing task.

[0092] Optionally, the available search devices include drones and unmanned boats; the method further includes: based on deep reinforcement learning, constructing a joint search model for water distress targets for the drones and the unmanned boats. In the joint search model, the water area environment field and the distribution of obstacles are used as the environment, and the drones and the unmanned boats are used as agents to construct a Markov decision process to realize the autonomous control of the search process of the drones and the unmanned boats.

[0093] For example, the search devices that can be used include drones and unmanned boats. Based on deep reinforcement learning, a joint search model for water distress targets for drones and unmanned boats is constructed. In this model, the water area environment field and the distribution of obstacles are regarded as the environment, and drones and unmanned boats are regarded as agents to construct a Markov decision process to realize the autonomous control of the search process of drones and unmanned boats. The state space includes the wind, wave, and current environments where agents such as drones and unmanned boats are located, the obstacles (fixed obstacles and moving drones, unmanned boats, etc.) around drones and unmanned boats, etc., while the action space includes the heading control and speed adjustment of drones and unmanned boats themselves; in the design of the reward function, a multi-layer reward mechanism is designed, including: 1) Target discovery reward, that is, when a drone or an unmanned boat discovers a target, a relatively high positive reward is given to encourage quick target positioning; 2) Proximity to target reward, that is, when a drone or an unmanned boat approaches a target, a certain positive reward is given to guide the movement towards the target; 3) Search efficiency reward, that is, a reward is given according to the search efficiency of drones and unmanned boats. For example, when a drone or an unmanned boat covers a new search area, a positive reward is given; when an unmanned boat repeatedly searches an area that has been covered, a negative reward is given; 4) Safety reward, that is, when a drone or an unmanned boat avoids collisions or avoids obstacles, a positive reward is given to ensure the safety of drones and unmanned boats; 5) Overall optimal reward, that is, when the total reward value of the search process of drones and unmanned boats for multiple targets becomes larger, a positive reward is given, otherwise, a negative reward is given.

[0094] In the training stage, all drones and unmanned boats share global information to improve the learning efficiency through centralized training. In the execution stage, each drone and unmanned boat makes decisions based on its own local observations to achieve distributed execution; the experience replay mechanism is used to store and replay historical experiences to help unmanned boats learn more complex strategies.

[0095] Figure 19 It is a schematic structural diagram of the water distress target search device provided by the embodiments of the present disclosure. The water distress target search device provided by the embodiments of the present disclosure can execute the processing flow provided by the embodiments of the water distress target search method, as Figure 19 shown, the water distress target search device 190 includes: A first acquisition module 191, configured to acquire a starting position and a starting time for drift prediction of a water distress target, where the drift prediction includes predictions of drift speed, drift position, drift trajectory, and search range; A second acquisition module 192, configured to acquire environmental prediction information corresponding to a preset spatial range around the starting position at the starting time; A generation module 193 is configured to generate a plurality of initial positions for the drift prediction according to the starting position, generate a plurality of initial times for the drift prediction according to the starting time, and generate a plurality of initial environment information for the drift prediction according to the environmental prediction information corresponding to the starting time; A first construction module 194 is configured to construct multiple sets of initial conditions according to the plurality of initial positions, the plurality of initial times, and the plurality of initial environment information, where each set of initial conditions includes any one of the initial positions, any one of the initial times, and any one of the initial environment information; A prediction module 195 is configured to predict the drift speed and drift position of the waterborne distress target at a plurality of target time points after any one of the initial times according to each set of initial conditions. The drift speed of the waterborne distress target at each target time point is obtained according to the environmental prediction information corresponding to a preset spatial range around the drift position of the waterborne distress target at the target time point, and the drift speed of the waterborne distress target before the target time point. The drift positions of the waterborne distress target at the plurality of target time points form a drift trajectory; according to the end points of each drift trajectory among the plurality of drift trajectories corresponding to the multiple sets of initial conditions, predict the search range of the waterborne distress target; A collaborative search module 196 is configured to perform a collaborative search for the waterborne distress target within the search range by using available search devices.

[0096] Optionally, when the second acquisition module 192 acquires the environmental prediction information corresponding to the preset spatial range around the starting position at the starting time, it is specifically configured to: Input the position information corresponding to a plurality of first-resolution grids included in the preset spatial range around the starting position, the environmental prediction information corresponding to the plurality of first-resolution grids at the starting time, and the position information corresponding to a plurality of second-resolution grids included in the preset spatial range into a downscaling deep learning model, and use the environmental prediction information corresponding to the plurality of second-resolution grids at the starting time output by the downscaling deep learning model as the environmental prediction information corresponding to the preset spatial range around the starting position at the starting time, where the second resolution is greater than the first resolution.

[0097] Optionally, the environmental prediction information corresponding to the preset spatial range around the drift position of the waterborne distress target at the target time point is obtained in the following manner: Input the position information corresponding to each of the multiple first-resolution grids included within a preset space around the drifting position of the waterborne distress target at the target time point, the environmental prediction information corresponding to each of the multiple first-resolution grids at the target time point, and the position information corresponding to each of the multiple second-resolution grids included within the preset space into a downscaling deep learning model, and use the environmental prediction information corresponding to each of the multiple second-resolution grids at the target time point output by the downscaling deep learning model as the environmental prediction information corresponding to the preset space around the drifting position of the waterborne distress target at the target time point, where the second resolution is greater than the first resolution.

[0098] Optionally, the environmental prediction information includes a wind field forecast result, a wave field forecast result, and a current field forecast result; The downscaling deep learning model is trained according to the following steps: According to a wind-wave-current coupled forecast model, calculate the environmental prediction information for each of the multiple first-resolution grids in a target water area at each preset time interval within a preset future time period. The wind-wave-current coupled forecast model is a coupled model of an atmospheric numerical model, a wave numerical model, and a hydrodynamic numerical model; According to the wind-wave-current coupled forecast model, calculate the environmental prediction information for each of the multiple second-resolution grids in the target water area at each preset time interval within the preset future time period; Input the position information corresponding to each of the multiple first-resolution grids, the environmental prediction information for each of the multiple first-resolution grids at each preset time interval within the preset future time period, and the position information corresponding to each of the multiple second-resolution grids into a downscaling deep learning model to be trained. The downscaling deep learning model outputs the environmental prediction information for each of the multiple second-resolution grids at each preset time interval within the preset future time period; Train the downscaling deep learning model according to the environmental prediction information for each of the multiple second-resolution grids at each preset time interval within the preset future time period output by the downscaling deep learning model and the environmental prediction information for each of the multiple second-resolution grids at each preset time interval within the preset future time period calculated according to the wind-wave-current coupled forecast model.

[0099] Optionally, when the prediction module 195 predicts the drift speed and drift position of the waterborne distress target at multiple water time points after any initial time according to each set of initial conditions, it is specifically configured to: Input the any initial environmental information into the drift speed intelligent prediction model corresponding to the waterborne distress target, and the drift speed intelligent prediction model outputs the drift speed of the waterborne distress target at the any initial time; For each target time point after any of the initial time points, calculate the drift position of the water distress target at the target time point according to the drift speed and drift position of the water distress target at the previous target time point. Input the environmental prediction information corresponding to the preset space range around the drift position of the water distress target at the target time point, and the drift speed of the water distress target before the target time point into the intelligent drift speed prediction model, and the intelligent drift speed prediction model outputs the drift speed of the water distress target at the target time point.

[0100] Optionally, the intelligent drift speed prediction model corresponding to the water distress target is trained according to the following steps: Obtain the drift positions corresponding to multiple preset time points during the free drift of the experimental target representing the water distress target in the target water area through the Beidou positioning system. Calculate the drift speeds of the experimental target corresponding to the multiple preset time points according to the drift positions corresponding to the multiple preset time points. For each preset time point among the multiple preset time points, input the environmental prediction information corresponding to the preset space range around the drift position of the experimental target at the preset time point, and the drift speed of the experimental target before the preset time point into the intelligent drift speed prediction model, and the intelligent drift speed prediction model outputs the drift speed of the experimental target at the preset time point. Train the intelligent drift speed prediction model according to the calculated drift speed of the experimental target corresponding to the preset time point and the drift speed of the experimental target corresponding to the preset time point output by the intelligent drift speed prediction model.

[0101] Optionally, the water distress target search device 190 further includes a guidance module 197, configured to guide the remote sensing satellite to observe within the search range after the prediction module 195 predicts the search range of the water distress target.

[0102] Optionally, the prediction module 195 is further configured to: when the remote sensing satellite observes the water distress target, perform a re-drift prediction on the water distress target by using the intelligent drift speed prediction model corresponding to the water distress target according to the remote sensing observation position and remote sensing observation time point of the water distress target, and obtain a new search range of the water distress target, where the remote sensing observation position is the starting position for the re-drift prediction, and the remote sensing observation time point is the starting time for the re-drift prediction.

[0103] Optionally, the water distress target search device 190 further includes a selection module 198, configured to select a preset number of end points around the water distress target within the search range when the remote sensing satellite observes the water distress target within the search range; the prediction module 195 is further configured to: based on the initial conditions corresponding to the preset number of end points respectively, and the drift speed intelligent prediction model corresponding to other distress targets that are in distress simultaneously with the water distress target, predict the drift speed, drift position, drift trajectory and search range of the other distress targets, where the size of the other distress targets is smaller than the size of the water distress target.

[0104] Optionally, the water distress target search device 190 further includes: An input module 199, configured to input the remote sensing image captured by the remote sensing satellite into a target detection model, where the target detection model includes a backbone network, a spectral-spatial dual attention module and a detection head, and the backbone network is configured to perform feature extraction on the remote sensing image to obtain a feature map; A processing module 1910, configured to process the feature map through the spectral-spatial dual attention module, and the detection head is configured to predict the type of the water distress target in the remote sensing image and the position information of the water distress target in the remote sensing image according to the output information of the spectral-spatial dual attention module; A determination module 1911, configured to determine the remote sensing observation position of the water distress target according to the position information of the water distress target in the remote sensing image.

[0105] Optionally, the spectral-spatial dual attention module includes a spectral attention mechanism and a spatial attention mechanism. When the spectral-spatial dual attention module processes the feature map, it is specifically configured to: Perform spectral feature extraction on the feature map through the spectral attention mechanism to obtain a one-dimensional spectral attention map; Multiply the feature map and the spectral attention map and then add the result to the feature map to obtain an intermediate result; Perform spatial feature extraction on the intermediate result through the spatial attention mechanism to obtain a two-dimensional spatial attention map; Multiply the intermediate result and the spatial attention map and then add the result to the intermediate result to obtain the output information of the spectral-spatial dual attention module.

[0106] Optionally, the available search devices include drones and unmanned boats; the waterborne distress target search device 190 further includes: a second construction module 1912, configured to construct a joint search model for waterborne distress targets for the drones and the unmanned boats based on deep reinforcement learning. In the joint search model, the water area environment field and the distribution of obstacles are used as the environment, and the drones and the unmanned boats are used as agents to construct a Markov decision process, so as to realize the autonomous control of the search processes of the drones and the unmanned boats.

[0107] Figure 19 The waterborne distress target search device in the illustrated embodiment can be used to execute the technical solutions of the above method embodiment, and its implementation principle and technical effects are similar, which will not be elaborated here.

[0108] The internal functions and structures of the waterborne distress target search device are described above, and the device can be implemented as an electronic device. Figure 20 The structural schematic diagram of the electronic device embodiment provided by the present disclosure is shown as Figure 20 As shown, the electronic device includes a memory 201 and a processor 202.

[0109] The memory 201 is used to store programs. In addition to the above programs, the memory 201 can also be configured to store various other data to support operations on the electronic device. Examples of these data include instructions for any application program or method for operating on the electronic device, contact data, phone book data, messages, pictures, videos, etc.

[0110] The memory 201 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0111] The processor 202 is coupled to the memory 201 and executes the programs stored in the memory 201 to execute the technical solutions of the above method embodiment.

[0112] Furthermore, as Figure 20 shown, the electronic device may further include: other components such as a communication component 203, a power supply component 204, an audio component 205, a display 206, etc. Figure 20 Only some components are schematically shown in Figure 20 and it does not mean that the electronic device only includes

[0113] The communication component 203 is configured to facilitate communication between the electronic device and other devices in a wired or wireless manner. The electronic device can access a wireless network based on communication standards, such as WiFi, 2G, or 3G, or a combination thereof. In an exemplary embodiment, the communication component 203 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 203 further includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0114] The power supply component 204 provides power to various components of the electronic device. The power supply component 204 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device.

[0115] The audio component 205 is configured to output and / or input audio signals. For example, the audio component 205 includes a microphone (MIC). When the electronic device is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive external audio signals. The received audio signals can be further stored in the memory 201 or transmitted via the communication component 203. In some embodiments, the audio component 205 further includes a speaker for outputting audio signals.

[0116] The display 206 includes a screen, and the screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can not only sense the boundaries of touch or swipe actions but also detect the duration and pressure associated with the touch or swipe operation.

[0117] In addition, an embodiment of the present disclosure further provides a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to implement the method described in the above embodiments.

[0118] An exemplary embodiment of the present disclosure further provides a computer program product, including a computer program, wherein the computer program, when executed by a processor of a computer, is used to cause the computer to execute to implement the method described in the above embodiments.

[0119] It should be noted that in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the said element.

[0120] The above are only specific embodiments of the present disclosure, enabling those skilled in the art to understand or implement the present disclosure. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to the embodiments described herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for searching for water distress targets, characterized in that, The method includes: Obtaining a starting position and a starting time for predicting the drift of a water distress target, where the drift prediction includes predictions of drift speed, drift position, drift trajectory, and search range; Obtaining the environmental prediction information corresponding to the starting time within a preset spatial range around the starting position; Generating multiple initial positions for the drift prediction according to the starting position, generating multiple initial times for the drift prediction according to the starting time, generating multiple initial environmental information for the drift prediction according to the environmental prediction information corresponding to the starting time, and constructing multiple sets of initial conditions based on the multiple initial positions, the multiple initial times, and the multiple initial environmental information, where each set of initial conditions includes any one of the initial positions, any one of the initial times, and any one of the initial environmental information; According to each set of initial conditions, predicting the drift speed and drift position of the water distress target at multiple target time points after the any one initial time, where the drift speed of the water distress target at each target time point is obtained based on the environmental prediction information corresponding to a preset spatial range around the drift position of the water distress target at the target time point, and the drift speed of the water distress target before the target time point, and the drift positions of the water distress target at the multiple target time points form a drift trajectory; Predicting the search range of the water distress target according to the end point of each drift trajectory among the multiple drift trajectories corresponding to the multiple sets of initial conditions; Conducting a collaborative search for the water distress target within the search range by available search devices.

2. The method according to claim 1, characterized in that, The environmental prediction information corresponding to a preset spatial range around the drift position of the water distress target at the target time point is obtained in the following manner: Inputting the position information corresponding to multiple first-resolution grids included within a preset spatial range around the drift position of the water distress target at the target time point, the environmental prediction information corresponding to the multiple first-resolution grids at the target time point respectively, and the position information corresponding to multiple second-resolution grids included within the preset spatial range into a downscaling deep learning model, and taking the environmental prediction information corresponding to the multiple second-resolution grids at the target time point output by the downscaling deep learning model as the environmental prediction information corresponding to a preset spatial range around the drift position of the water distress target at the target time point, where the second resolution is greater than the first resolution.

3. The method according to claim 2, wherein The environmental prediction information includes wind field forecast results, wave field forecast results, and current field forecast results; The downscaling deep learning model is trained according to the following steps: Calculating the environmental prediction information of multiple first-resolution grids in a target water area at each preset time interval within a preset future duration according to a wind-wave-current coupled forecast model, where the wind-wave-current coupled forecast model is a coupled model of an atmospheric numerical model, a wave numerical model, and a hydrodynamic numerical model; According to the coupled wind-wave-current prediction model, calculate the environmental prediction information of multiple second-resolution grids in the target water area at each preset time interval within the preset future duration; Input the position information corresponding to the multiple first-resolution grids, the environmental prediction information of the multiple first-resolution grids at each preset time interval within the preset future duration, and the position information corresponding to the multiple second-resolution grids into the to-be-trained downscaling deep learning model, and the downscaling deep learning model outputs the environmental prediction information of the multiple second-resolution grids at each preset time interval within the preset future duration; Train the downscaling deep learning model according to the environmental prediction information of the multiple second-resolution grids output by the downscaling deep learning model at each preset time interval within the preset future duration, and the environmental prediction information of the multiple second-resolution grids calculated according to the coupled wind-wave-current prediction model at each preset time interval within the preset future duration.

4. The method according to claim 1, wherein According to each set of initial conditions, predict the drift speed and drift position of the waterborne distress target at multiple target time points after any initial time, including: Input the any initial environmental information into the drift speed intelligent prediction model corresponding to the waterborne distress target, and the drift speed intelligent prediction model outputs the drift speed of the waterborne distress target at the any initial time; For each target time point after the any initial time, calculate the drift position of the waterborne distress target at the target time point according to the drift speed and drift position of the waterborne distress target at the previous target time point; Input the environmental prediction information corresponding to the preset spatial range around the drift position of the waterborne distress target at the target time point, and the drift speed of the waterborne distress target before the target time point into the drift speed intelligent prediction model, and the drift speed intelligent prediction model outputs the drift speed of the waterborne distress target at the target time point.

5. The method according to claim 4, wherein The drift speed intelligent prediction model corresponding to the waterborne distress target is trained according to the following steps: Obtain the drift positions corresponding to multiple preset time points during the free drift process of the experimental target representing the waterborne distress target in the target water area through the Beidou positioning system; Calculate the drift speeds of the experimental target at the multiple preset time points according to the drift positions corresponding to the multiple preset time points; For each preset time point among the multiple preset time points, input the environmental prediction information corresponding to the preset spatial range around the drift position of the experimental target at the preset time point, and the drift speed of the experimental target before the preset time point into the drift speed intelligent prediction model, and the drift speed intelligent prediction model outputs the drift speed of the experimental target at the preset time point; Train the intelligent drift speed prediction model based on the drift speed corresponding to the experimental target calculated according to the above calculation and the drift speed corresponding to the experimental target at the preset time point output by the intelligent drift speed prediction model.

6. The method according to claim 1, wherein After predicting the search range of the water distress target, the method further includes: Guiding the remote sensing satellite to observe within the search range.

7. The method according to claim 1, characterized in that, The method further includes: When the remote sensing satellite observes the water distress target, based on the remote sensing observation position and the remote sensing observation time point of the water distress target, use the intelligent drift speed prediction model corresponding to the water distress target to perform a re-drift prediction on the water distress target to obtain a new search range for the water distress target. The remote sensing observation position is the starting position for the re-drift prediction, and the remote sensing observation time point is the starting time for the re-drift prediction.

8. The method according to claim 1, wherein The method further includes: When the remote sensing satellite observes the water distress target within the search range, select a preset number of end points around the water distress target within the search range; Based on the initial conditions corresponding to the preset number of end points respectively and the intelligent drift speed prediction model corresponding to other distress targets that are in distress simultaneously with the water distress target, predict the drift speed, drift position, drift trajectory and search range of the other distress targets. The size of the other distress targets is smaller than the size of the water distress target.

9. The method according to any one of claims 6-8, characterized in that, The method further includes: Input the remote sensing image captured by the remote sensing satellite into the target detection model. The target detection model includes a backbone network, a spectral-spatial dual attention module and a detection head. The backbone network is used to extract features from the remote sensing image to obtain a feature map; Process the feature map through the spectral-spatial dual attention module. The detection head is used to predict the type of the water distress target in the remote sensing image and the position information of the water distress target in the remote sensing image according to the output information of the spectral-spatial dual attention module; Determine the remote sensing observation position of the water distress target according to the position information of the water distress target in the remote sensing image.

10. The method according to claim 9, wherein The spectral-spatial dual attention module includes a spectral attention mechanism and a spatial attention mechanism. The processing of the feature map by the spectral-spatial dual attention module includes: Perform spectral feature extraction on the feature map through the spectral attention mechanism to obtain a one-dimensional spectral attention map; Multiply the feature map and the spectral attention map and then add them to the feature map to obtain an intermediate result; Perform spatial feature extraction on the intermediate result through the spatial attention mechanism to obtain a two-dimensional spatial attention map; Multiply the intermediate result and the spatial attention map and then add them to the intermediate result to obtain the output information of the spectral-spatial dual attention module.

11. The method according to claim 1, characterized in that, The available search devices include unmanned aerial vehicles and unmanned boats; the method further includes: Based on deep reinforcement learning, a joint search model for waterborne distress targets of the unmanned aerial vehicle and the unmanned surface vehicle is constructed. In the joint search model, the water area environment field and the distribution of obstacles are regarded as the environment, and the unmanned aerial vehicle and the unmanned surface vehicle are regarded as agents to construct a Markov decision process, so as to realize the autonomous control of the search process of the unmanned aerial vehicle and the unmanned surface vehicle.

12. A device for searching for targets in distress on water, characterized in that, The device includes: A first acquisition module, configured to acquire a starting position and a starting time for drift prediction of a waterborne distress target, where the drift prediction includes predictions of drift speed, drift position, drift trajectory, and search range; A second acquisition module, configured to acquire environmental prediction information corresponding to a preset spatial range around the starting position at the starting time; A generation module, configured to generate multiple initial positions for the drift prediction according to the starting position, generate multiple initial times for the drift prediction according to the starting time, and generate multiple initial environmental information for the drift prediction according to the environmental prediction information corresponding to the starting time; A first construction module, configured to construct multiple groups of initial conditions according to the multiple initial positions, the multiple initial times, and the multiple initial environmental information, where each group of initial conditions includes any one of the initial positions, any one of the initial times, and any one of the initial environmental information; A prediction module, configured to predict the drift speed and drift position of the waterborne distress target at multiple target time points after any one of the initial times according to each group of initial conditions. The drift speed of the waterborne distress target at each target time point is obtained according to the environmental prediction information corresponding to a preset spatial range around the drift position of the waterborne distress target at the target time point, and the drift speed of the waterborne distress target before the target time point. The drift positions of the waterborne distress target at the multiple target time points form a drift trajectory; predict the search range of the waterborne distress target according to the end point of each drift trajectory among the multiple drift trajectories corresponding to the multiple groups of initial conditions; A collaborative search module, configured to perform collaborative search for the waterborne distress target within the search range by available search devices.

13. An electronic device, characterized in that, It includes: A memory; A processor; And A computer program; Wherein, the computer program is stored in the memory and is configured to be executed by the processor to implement the method according to any one of claims 1-11.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method according to any one of claims 1-11.

Images