Intelligent marine emergency equipment automatic positioning method and system

Abstract

The invention provides an intelligent marine emergency equipment automatic positioning method and system. Wherein multiple distress signals are received and fused, and optimal geographical location information and timestamps are generated; determining a distress area on a preset sea area electronic map, correcting a distress position, evaluating probability distribution of each potential distress area, and generating a primary search target with the highest probability; a thermal imaging and optical sensing system on the rescue unmanned aerial vehicle is started for scanning, and a high-accuracy potential survivor position is generated; the search and rescue path is dynamically adjusted and optimized, the search strategy is adjusted, the optimal flight path is updated in real time, and an economical and efficient search and rescue path plan is generated; and timely feeding back the accurate coordinates of the position of the potential survivor to the nearest rescue coordination center, and guiding first-aid equipment to go along the optimal path to implement rescue. According to the technical scheme provided by the invention, the search and rescue efficiency and accuracy are improved, and the position of a potential survivor can be quickly and accurately positioned.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of automatic positioning of intelligent marine emergency equipment, and in particular to a method and system for automatic positioning of intelligent marine emergency equipment. Background Art

[0002] Maritime search and rescue missions are complex and urgent, especially in the vast ocean environment, where it is crucial to quickly and accurately locate people and ships in distress. Traditional search and rescue methods rely on a single signal source, which often cannot provide sufficient accuracy and reliability in a changing ocean environment. In addition, factors such as severe weather conditions, complex sea conditions, and drone endurance also increase the difficulty of search and rescue. Therefore, an intelligent automatic positioning method is needed that can integrate multiple distress signals, generate optimal geographic location information, and combine advanced machine learning algorithms to improve the recognition ability of non-fixed objects and path planning efficiency, so as to ensure the rapid and accurate positioning of potential survivors.

[0003] At present, maritime search and rescue mainly relies on the following traditional technical means to receive distress signals and determine the location of the distress through satellites. However, satellite signals alone are easily affected by weather and terrain, resulting in reduced positioning accuracy. Radio waves are used for communication and positioning, but their coverage is limited and they are not effective in remote areas. Relying on eyewitness reports or radio communications provided by nearby ships, but this method has a slow response speed and is limited by the distribution of ships and communication distance.

[0004] However, the existing solutions have the following major defects: the single signal source is easily interfered with and cannot provide high-precision distress locations, especially in complex and changing marine environments. Traditional methods rely on manual scheduling and manual path planning, which makes it difficult to achieve rapid response and delays precious rescue time. The lack of adaptability to different weather conditions and complex sea conditions affects the effectiveness and reliability of search equipment. For example, traditional image recognition technology performs poorly in adverse weather conditions, and the endurance of drones also limits long-term operations.

[0005] In order to overcome the above problems, an intelligent automatic positioning method for maritime emergency equipment is proposed, which improves the search and rescue efficiency and accuracy, and also enhances the robustness and adaptability of the system in complex and changeable marine environments, providing strong technical support for maritime search and rescue. Summary of the invention

[0006] The embodiments of the present application provide a method and system for automatically positioning intelligent marine emergency equipment to solve the problems of low search and rescue efficiency and accuracy in the prior art.

[0007] In a first aspect, an embodiment of the present application provides a method for automatically positioning intelligent marine emergency equipment, comprising:

[0008] Receive and fuse multiple distress signals from satellites, radio relay stations and nearby ships to generate optimal geographic location information and timestamps;

[0009] Based on the optimal geographical location information, the distress area is determined on the preset sea area electronic map, and the signal propagation delay is calculated according to the timestamp to correct the distress position, and the probability distribution of each potential distress area is evaluated by applying the Bayesian network prediction model to generate the primary search target with the highest probability;

[0010] Based on the primary search target with the highest probability, the thermal imaging and optical sensing systems on the rescue drone are activated to scan, and the convolutional neural network is combined to enhance the recognition ability of non-fixed objects. The generative adversarial network is used to simulate the image features under different weather conditions to generate highly accurate potential survivor locations;

[0011] Using ocean current prediction algorithms and wind direction data, the search and rescue path is dynamically adjusted and optimized to ensure rapid coverage of the distress area where the potential survivors are located. The search strategy is adaptively adjusted through a reinforcement learning algorithm, the optimal flight route is updated in real time, and the remaining battery power and endurance of the drone are taken into consideration to generate an economical and efficient search and rescue path plan.

[0012] Based on the search and rescue path planning, the precise coordinates of the potential survivors' locations are fed back to the nearest rescue coordination center in a timely manner, and the emergency equipment is guided along the optimal path to carry out the rescue.

[0013] Optionally, the distress area is determined on a preset sea area electronic map based on the optimal geographical location information, and the signal propagation delay is calculated according to the timestamp to correct the distress position, and a Bayesian network prediction model is applied to evaluate the probability distribution of each potential distress area to generate a primary search target with the highest probability, including:

[0014] Using the optimal geographical location information, matching the initial distress location on a preset sea area electronic map to determine the distress area;

[0015] Calculate the signal propagation delay according to the timestamp, and perform correction processing on the initial distress position to obtain a corrected distress position;

[0016] Based on the corrected distress location, applying a Bayesian network prediction model to evaluate the probability distribution of each distress area to generate an occurrence probability value for each distress area;

[0017] According to the generated occurrence probability value of each distress area, the distress area with the highest probability is selected, and the primary search target with the highest probability is generated to ensure that resources are preferentially invested in the distress area with the highest probability.

[0018] Optionally, based on the corrected distress location, applying a Bayesian network prediction model to evaluate the probability distribution of each distress area to generate an occurrence probability value of each distress area includes:

[0019] The corrected distress location and surrounding environment parameters are used as input variables into the pre-trained Bayesian network prediction model to obtain basic data;

[0020] Based on the basic data and the conditional probability distribution learned from the data set of historical distress events, the possibility of each preset distress area becoming an actual distress location is evaluated and processed, and a preliminary occurrence probability value is output;

[0021] Based on the preliminary occurrence probability value, further adjusting and optimizing the occurrence probability value to ensure accuracy, and obtaining an optimized occurrence probability value;

[0022] According to the optimized occurrence probability value, the occurrence probability value of each distress area is generated to ensure that resources are preferentially invested in the distress area with the highest probability.

[0023] Optionally, according to the primary search target with the highest probability, the thermal imaging and optical sensing systems on the rescue drone are started to scan, a convolutional neural network is combined to enhance the recognition capability of non-fixed form objects, and a generative adversarial network is used to simulate image features under different weather conditions to generate high-accuracy potential survivor locations, including:

[0024] Based on the primary search target with the highest probability, the thermal imaging and optical sensing systems on the rescue drone are activated to scan and obtain preliminary image data;

[0025] Using the preliminary image data in combination with a convolutional neural network, the non-fixed form objects in the preliminary image data are identified and processed to obtain preliminary object recognition results;

[0026] Based on the preliminary object recognition results, applying a generative adversarial network to simulate image features under different weather conditions to generate optimized image features;

[0027] Based on the optimized image features, the precise location of the potential survivors is further analyzed and confirmed to generate a highly accurate potential survivor location.

[0028] Optionally, the using of the preliminary image data in combination with a convolutional neural network to perform recognition processing on non-fixed form objects in the preliminary image data to obtain a preliminary object recognition result includes:

[0029] Using the preliminary image data as input to a pre-trained convolutional neural network, performing feature extraction processing on the preliminary image data to generate detailed feature information;

[0030] Based on the detailed feature information, further refining the feature information through multi-layer convolution and pooling operations of a convolutional neural network, and obtaining an optimized feature representation according to the visual characteristics of shape, texture and color;

[0031] The optimized feature representation is classified through a classification layer of a convolutional neural network to accurately classify irregular or floating objects and generate classification results;

[0032] A preliminary object recognition result is generated according to the classification result, and the preliminary object recognition result includes preliminary position information of potential survivors.

[0033] Optionally, the ocean current prediction algorithm and wind direction data are used to dynamically adjust and optimize the search and rescue path to ensure rapid coverage of the distress range where the potential survivors are located, and the search strategy is adaptively adjusted through a reinforcement learning algorithm to update the best flight route in real time, and the remaining power and endurance of the drone are taken into consideration to generate an economical and efficient search and rescue path planning, including:

[0034] Using the obtained high-accuracy potential survivor positions, combined with the real-time acquired ocean current prediction algorithm and wind direction data, the search and rescue path is preliminarily planned to obtain a preliminary search and rescue path;

[0035] Based on the preliminary search and rescue path, according to the ocean current prediction algorithm and wind direction data, the search and rescue path is dynamically adjusted and optimized to adapt to the changing ocean environment, ensure the effectiveness and timeliness of the path, and generate an optimized search and rescue path;

[0036] Adaptively adjust the optimized search and rescue path through a reinforcement learning algorithm, update the search strategy in real time according to the situations encountered during the actual search and rescue process, and generate a search and rescue path that is updated in real time;

[0037] Taking into account the remaining power and endurance of the drone, the search and rescue path updated in real time is further processed to ensure that the drone can return safely after completing the mission while avoiding unnecessary energy waste, thereby generating a final flight route;

[0038] Based on the highly accurate location of potential survivors, real-time ocean current prediction algorithm and wind direction data, dynamically adjusted and optimized search and rescue paths, adaptively adjusted search strategies, and the remaining battery power and endurance of the drone, the final flight route is evaluated and optimized to generate an economical and efficient search and rescue path plan.

[0039] Optionally, the optimized search and rescue path is adaptively adjusted by a reinforcement learning algorithm, and the search strategy is updated in real time according to situations encountered in the actual search and rescue process to generate a search and rescue path updated in real time, including:

[0040] Using the optimized search and rescue path, combined with the actual search and rescue environment data obtained in real time, to provide a training basis for the reinforcement learning algorithm, and obtain initial training data;

[0041] Based on the initial training data, the state of the current search and rescue environment is represented by the state space in the reinforcement learning algorithm, and the position and state of the drone are taken as part of the environment, so as to ensure that the reinforcement learning algorithm can fully understand the current situation and generate an environmental state representation;

[0042] Generate an adjustment plan based on the environmental state representation and through adjustment operations defined in the action space of the reinforcement learning algorithm, wherein the adjustment operations include changing the flight direction, adjusting the flight altitude, accelerating or decelerating;

[0043] Using the adjustment scheme, the effect of each adjustment operation is evaluated through a reward mechanism in the reinforcement learning algorithm, wherein the reward mechanism takes into account factors such as search and rescue efficiency, safety, and resource consumption, and generates an evaluation result;

[0044] Based on the evaluation results, the reinforcement learning algorithm is applied to perform iterative learning, and the search strategy is continuously optimized according to the results of each adjustment operation, so that the reinforcement learning algorithm can gradually find the optimal path adjustment plan in the actual search and rescue process and generate an optimized search strategy;

[0045] According to the optimized search strategy, the search and rescue path is updated in real time, and a real-time updated search and rescue path is generated to ensure that the UAV can dynamically adjust the flight path according to the latest environmental information, thereby improving the search and rescue efficiency.

[0046] In a second aspect, an embodiment of the present application provides an intelligent marine emergency equipment automatic positioning system, comprising:

[0047] The receiving and fusion module is used to receive and fuse multiple distress signals from satellites, radio relay stations and nearby ships, and estimate the optimal geographic location information and timestamp;

[0048] A determination evaluation module is used to determine the distress area on a preset sea area electronic map based on the optimal geographical location information, calculate the signal propagation delay according to the timestamp to correct the distress position, and apply the Bayesian network prediction model to evaluate the probability distribution of each potential distress area to generate the primary search target with the highest probability;

[0049] A scanning generation module is used to start the thermal imaging and optical sensing systems on the rescue drone to scan according to the primary search target with the highest probability, combine convolutional neural networks to enhance the recognition ability of non-fixed form objects, and use generative adversarial networks to simulate image features under different weather conditions to generate high-accuracy potential survivor locations;

[0050] An optimization and update module is used to dynamically adjust and optimize the search and rescue path using the ocean current prediction algorithm and wind direction data to ensure rapid coverage of the distress range where the potential survivors are located, adaptively adjust the search strategy through the reinforcement learning algorithm, update the best flight route in real time, and consider the remaining power and endurance of the drone to generate an economical and efficient search and rescue path plan;

[0051] The feedback guidance module is used to timely feedback the precise coordinates of the potential survivors' locations to the nearest rescue coordination center based on the search and rescue path planning, and guide the emergency equipment to carry out rescue along the optimal path.

[0052] In a third aspect, an embodiment of the present application provides a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an automatic positioning method for intelligent marine emergency equipment as described in any one of the first aspects.

[0053] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a method for automatic positioning of intelligent marine emergency equipment as described in any one of the first aspects.

[0054] In the embodiment of the present application, multiple distress signals from satellites, radio relay stations and nearby ships are received and integrated to generate optimal geographic location information and timestamps; based on the optimal geographic location information, the distress area is determined on a preset electronic map of the sea area, and the signal propagation delay is calculated according to the timestamp to correct the distress position, and at the same time, the Bayesian network prediction model is applied to evaluate the probability distribution of each potential distress area to generate the primary search target with the highest probability; based on the primary search target with the highest probability, the thermal imaging and optical sensing systems on the rescue drone are started to scan, and the convolutional neural network is combined to enhance the recognition ability of non-fixed objects. The system uses a generative adversarial network to simulate image features under different weather conditions to generate highly accurate potential survivor locations. It uses ocean current prediction algorithms and wind direction data to dynamically adjust and optimize the search and rescue path to ensure rapid coverage of the distress range where the potential survivors are located. It uses a reinforcement learning algorithm to adaptively adjust the search strategy, update the optimal flight route in real time, and consider the remaining power and endurance of the drone to generate an economical and efficient search and rescue path planning. Based on the search and rescue path planning, the precise coordinates of the potential survivors' locations are promptly fed back to the nearest rescue coordination center, and the emergency equipment is guided along the optimal path to carry out rescue.

[0055] The technical solution of this application has the following beneficial effects:

[0056] This application generates optimal geographic location information and timestamps by receiving and fusing multiple distress signals from satellites, radio relay stations and nearby ships, ensuring high-precision correction of the distress location. Based on the optimal geographic location information, the distress area is determined on the preset electronic map of the sea area, and the signal propagation delay is calculated in combination with the timestamp, further improving the accuracy of the distress location. The Bayesian network prediction model is used to evaluate the probability distribution of each potential distress area, generate the primary search target with the highest probability, and thus concentrate resources for efficient search to avoid ineffective coverage. The thermal imaging and optical sensing systems on the rescue drone are started for scanning, and the convolutional neural network is combined to enhance the recognition ability of non-fixed objects to ensure accurate identification of potential survivors in complex environments. Generative adversarial networks are used to simulate image features under different weather conditions, generate highly accurate potential survivor locations, adapt to various harsh environments, and improve the robustness of recognition. Using ocean current prediction algorithms and wind direction data, the search and rescue path is dynamically adjusted and optimized to ensure rapid coverage of the distress range where potential survivors are located. Through the reinforcement learning algorithm, the search strategy is adaptively adjusted, the optimal flight route is updated in real time, and the remaining power and endurance of the drone are taken into account to generate economical and efficient search and rescue path planning, maximize the operation time of the drone, and improve the success rate of search and rescue. The precise coordinates of the potential survivors are fed back to the nearest rescue coordination center in a timely manner to ensure the timeliness and accuracy of information transmission. The emergency equipment is guided along the optimal path to carry out the rescue, ensuring that the rescue operation is carried out quickly, shortening the response time, and increasing the survival rate of the survivors. The entire process fully considers the remaining power and endurance of the drone, ensures the effective use of resources, avoids unnecessary energy waste, and improves the economy and sustainability of the overall search and rescue operation.

[0057] Furthermore, the embodiment of the present application further describes in detail the process of determining the distress area based on the optimal geographical location information, correcting the distress position and generating the primary search target. Specifically, it includes: using the optimal geographical location information to match the initial distress position on the preset sea area electronic map to determine the distress area; calculating the signal propagation delay according to the timestamp, correcting the initial distress position, and obtaining the corrected distress position; based on the corrected distress position, applying the Bayesian network prediction model to evaluate the probability distribution of each potential distress area, and generating the probability value of each distress area; selecting the distress area with the highest probability as the primary search target to ensure that resources are invested in this area first. In addition, according to the primary search target with the highest probability, the thermal imaging and optical sensing systems on the rescue drone are started to scan, and the convolutional neural network is combined to enhance the recognition ability of non-fixed form objects, and the generative adversarial network is used to simulate the image features under different weather conditions, and finally generate a high-accuracy potential survivor position.

[0058] Through the above method, the present application significantly improves the efficiency and accuracy of maritime search and rescue missions. First, through multi-source signal fusion and timestamp correction, the high-precision determination of the distressed location is ensured and the positioning error is reduced. Secondly, the application of the Bayesian network prediction model enables the system to intelligently evaluate the probability distribution of each potential distress area, select the most likely search target, and concentrate resources for efficient search. Finally, by combining thermal imaging, optical sensing systems, convolutional neural networks and generative adversarial networks, the recognition ability of non-fixed objects and the ability to adapt to different weather conditions are enhanced, generating high-accuracy potential survivor locations. Overall, this method not only improves the success rate of search and rescue, but also enhances the robustness and adaptability of the system in complex and changeable marine environments, providing strong technical support for rapid and accurate rescue.

[0059] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0061] Figure 1 A flowchart of a method for automatically positioning intelligent marine emergency equipment provided in an embodiment of the present application;

[0062] Figure 2 A schematic diagram of the structure of an intelligent automatic positioning system for marine emergency equipment provided in an embodiment of the present application;

[0063] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0065] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0066] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0067] Figure 1 A flowchart of a method for automatically positioning an intelligent marine emergency equipment is provided for an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0068] 101. Receive and fuse multiple distress signals from satellites, radio relay stations and nearby ships to generate optimal geographic location information and timestamps;

[0069] It involves multi-source signal fusion technology, which aims to generate optimal geographic location information and timestamps by receiving and integrating distress signals from multiple different sources (such as satellites, radio relay stations and nearby ships). Specifically, satellite signals provide high-precision global positioning data, radio relay stations provide regional communication support, and nearby ships provide instant eyewitness reports or auxiliary communications. These data are used together to determine the distress location, and the time synchronization of all signals is ensured through timestamps, thereby improving the accuracy and reliability of positioning.

[0070] The system first receives multiple distress signals from satellites, radio relay stations and nearby ships. Then, the signal processing algorithm is used to pre-process these signals to remove noise and interference to ensure the integrity and accuracy of the data. Next, the signals from different sources are aligned in time and space to generate a unified timestamp to eliminate the time deviation caused by signal propagation delay. Finally, through the multi-source data fusion algorithm, the weight and reliability of each signal source are comprehensively considered to generate the optimal geographic location information. This process not only improves the positioning accuracy, but also enhances the robustness of the system.

[0071] In a simulated search and rescue mission, when a fishing boat was in distress in a certain sea area, the ship sent a distress signal. After receiving the signal, the satellite provided preliminary geographic coordinates; a nearby radio relay station also captured the signal and supplemented detailed information in the area; at the same time, a nearby merchant ship reported the sighting via VHF radio. The system fused these three signals together to generate precise geographic location information and timestamps. Based on this information, the search and rescue center quickly launched subsequent rescue operations.

[0072] 102. Based on the optimal geographical location information, determine the distress area on the preset sea area electronic map, calculate the signal propagation delay according to the timestamp to correct the distress position, and apply the Bayesian network prediction model to evaluate the probability distribution of each potential distress area to generate the primary search target with the highest probability;

[0073] The core is to use the optimal geographical location information to accurately locate the distress area on the preset sea area electronic map, and calculate the signal propagation delay through the timestamp to correct the initial position. In addition, the Bayesian network prediction model is used to evaluate the probability distribution of each potential distress area and select the most likely search target. This step ensures the effective allocation of resources, avoids blind search, and improves the efficiency of search and rescue.

[0074] Based on the generated optimal geographic location information, the system matches a specific distress area on a preset electronic map of the sea area. Subsequently, the signal propagation delay is calculated according to the timestamp, and the initial distress location is corrected to obtain a more accurate distress location. Next, the Bayesian network prediction model is applied, combined with the data set of historical distress events, to evaluate the probability distribution of each potential distress area, and finally the distress area with the highest probability is selected as the primary search target. In this way, the system can concentrate resources for efficient search and reduce ineffective coverage.

[0075] In the above-mentioned simulated search and rescue mission, the system determined the possible distress area of ​​the fishing boat on the preset electronic map of the sea area based on the generated optimal geographical location information. Taking into account the signal propagation delay, the system corrected the initial position and obtained a more accurate distress location. Then, the Bayesian network prediction model was applied to evaluate the probability distribution of multiple potential distress areas, and the results showed that the probability of a specific area was the highest. Based on this, the search and rescue center gave priority to dispatching drones to search in this area, greatly improving the search and rescue efficiency.

[0076] 103. Based on the primary search target with the highest probability, the thermal imaging and optical sensing systems on the rescue drone are activated to scan, and the convolutional neural network is combined to enhance the recognition capability of non-fixed objects, and the generative adversarial network is used to simulate the image features under different weather conditions to generate the potential survivors’ locations with high accuracy;

[0077] Focusing on launching the thermal imaging and optical sensing systems on rescue drones, combined with advanced machine learning algorithms (such as convolutional neural networks and generative adversarial networks) to enhance the recognition of non-fixed objects and generate highly accurate potential survivor locations. The thermal imaging system can detect human body heat at night or in low visibility conditions, while the optical sensing system is used to capture visual information. The convolutional neural network is used to extract features from the image, and the generative adversarial network simulates image features under different weather conditions to ensure accurate identification of potential survivors in all environments.

[0078] Based on the generated primary search target, the system starts the thermal imaging and optical sensing systems on the rescue drone to scan and obtain preliminary image data. Then, the convolutional neural network is used to extract and classify the features of these image data to identify non-fixed objects (such as floating human bodies). In order to cope with complex and changeable weather conditions, the system uses a generative adversarial network to simulate image features under different weather conditions and further optimize image recognition results. Finally, the system generates highly accurate potential survivor locations, providing a reliable basis for subsequent rescue operations.

[0079] In the above search and rescue mission, the drone went to the designated area to scan according to the primary search target. The thermal imaging system successfully detected several points suspected of human heat at night, while the optical sensing system captured visual information of some floating objects. The convolutional neural network analyzed these image data and identified that one of the floating objects might be a person in distress. The generative adversarial network simulated the image features under the current weather conditions and further confirmed the identity of the object. Based on this highly accurate location information, the search and rescue team quickly launched a rescue operation.

[0080] 104. Using ocean current prediction algorithms and wind direction data, dynamically adjust and optimize the search and rescue path to ensure rapid coverage of the distress area where the potential survivors are located. Adaptively adjust the search strategy through reinforcement learning algorithms, update the best flight route in real time, and consider the remaining power and endurance of the drone to generate economical and efficient search and rescue path planning;

[0081] This involves using ocean current prediction algorithms and wind direction data, combined with reinforcement learning algorithms, to dynamically adjust and optimize the search and rescue path to ensure rapid coverage of areas where potential survivors are located. Ocean current prediction algorithms are used to predict the direction and speed of ocean flow, while wind direction data helps drones adjust their flight attitude. Reinforcement learning algorithms continuously learn and optimize search strategies, update the best flight route in real time, and take into account the remaining battery power and endurance of the drone to generate cost-effective path planning. This step ensures the efficiency and sustainability of the search and rescue operation.

[0082] The system uses ocean current prediction algorithms and wind direction data to predict the direction and speed of ocean currents, as well as the impact of wind on drone flight. On this basis, the system adaptively adjusts the search strategy through a reinforcement learning algorithm, updates the best flight route in real time, and ensures that the drone can quickly cover the area where potential survivors are located. At the same time, the system continuously monitors the remaining battery power and endurance of the drone, dynamically adjusts the path planning, and ensures that the drone completes the mission and returns safely before the battery runs out. This not only improves the efficiency of search and rescue, but also maximizes the operation time of the drone.

[0083] During the search and rescue mission, the drone begins to execute search and rescue path planning based on highly accurate information about the location of potential survivors. The system uses ocean current prediction algorithms and wind direction data to predict the direction and speed of ocean flow and adjust the flight path of the drone. The reinforcement learning algorithm continuously optimizes the search strategy and updates the best flight route in real time to ensure that the drone can quickly cover the distress area. At the same time, the system monitors the remaining battery power of the drone and adjusts the path in time to ensure that the drone successfully completes the mission and returns to the base safely. This dynamic adjustment makes search and rescue operations more efficient and economical.

[0084] 105. Based on the search and rescue path planning, the precise coordinates of the potential survivors' locations are fed back to the nearest rescue coordination center in a timely manner, and the emergency equipment is guided along the optimal path to carry out the rescue.

[0085] The core is to promptly feed back the precise coordinates of potential survivors to the nearest rescue coordination center and guide emergency equipment along the optimal path to carry out rescue. This process ensures the timeliness and accuracy of information transmission, enabling rescue teams to respond quickly, shorten rescue time, and increase the chances of survival of survivors. By integrating communication systems and navigation technology, the system achieves seamless connection from information collection to rescue implementation.

[0086] Based on the generated search and rescue path planning, the system promptly feeds back the precise coordinates of the potential survivors' locations to the nearest rescue coordination center. The rescue coordination center communicates this information to the ground rescue team and other relevant units through the communication system. At the same time, the system guides emergency equipment (such as rescue ships, helicopters, etc.) along the optimal path to carry out rescue. Navigation technology ensures that the equipment can reach the designated location quickly and accurately, improving the overall efficiency of the rescue operation.

[0087] In the above-mentioned search and rescue mission, the system promptly fed back the precise coordinates of the potential survivors found by the drone to the nearest rescue coordination center. The rescue coordination center quickly conveyed this information to the ground rescue team and nearby merchant ships. At the same time, the system guided the rescue ships and helicopters along the optimal path to carry out the rescue. Navigation technology ensured that these devices could reach the designated location quickly and accurately, greatly shortening the rescue time. In the end, the rescue team successfully rescued the fishermen in distress, significantly increasing their chances of survival.

[0088] Through the implementation of steps 101 to 105, the method of the present application significantly improves the efficiency and accuracy of maritime search and rescue missions. First, multi-source signal fusion and timestamp correction ensure high-precision determination of the distressed location and reduce positioning errors. Second, the application of the Bayesian network prediction model enables the system to intelligently evaluate the probability distribution of each potential distress area, select the most likely search target, and concentrate resources for efficient search. Third, by combining thermal imaging, optical sensing systems, convolutional neural networks, and generative adversarial networks, the recognition ability of non-fixed objects and the ability to adapt to different weather conditions are enhanced to generate high-accuracy potential survivor locations. Fourth, by using ocean current prediction algorithms and wind direction data, combined with reinforcement learning algorithms, the search and rescue path is dynamically adjusted and optimized to ensure rapid coverage of the potential survivors' areas, while taking into account the remaining power of the drone to generate economical and efficient path planning. Finally, by integrating communication systems and navigation technologies, seamless connection from information collection to rescue implementation is achieved, ensuring that the rescue team can respond quickly, shorten the rescue time, and increase the survival rate of survivors. Overall, this method not only improves the success rate of search and rescue, but also enhances the robustness and adaptability of the system in complex and changeable marine environments, providing strong technical support for rapid and accurate rescue.

[0089] In order to solve the problem of signal propagation delay and accuracy of distress position correction, in some embodiments, the distress area determined on a preset sea area electronic map based on the optimal geographical location information in step 102, and the signal propagation delay is calculated according to the timestamp to correct the distress position, and the probability distribution of each potential distress area is evaluated by applying a Bayesian network prediction model to generate the primary search target with the highest probability, including:

[0090] By using the optimal geographical location information, the initial distress position is matched on a preset electronic map of the sea area to determine the distress area; the signal propagation delay is calculated according to the timestamp, and the initial distress position is corrected to obtain a corrected distress position; based on the corrected distress position, the probability distribution of each distress area is evaluated using a Bayesian network prediction model to generate an occurrence probability value for each distress area; based on the generated occurrence probability value for each distress area, the distress area with the highest probability is selected to generate a primary search target with the highest probability, thereby ensuring that resources are preferentially invested in the distress area with the highest probability.

[0091] In this embodiment, it involves matching the optimal geographical location information generated by multi-source signal fusion with a preset sea area electronic map. The preset sea area electronic map contains detailed marine geographical information, such as water depth, seabed topography, navigation signs, etc. Through this matching process, the system can accurately locate the specific area where the distress event occurs. The optimal geographical location information includes multiple distress signal data from satellites, radio relay stations and nearby ships, which is used to ensure high accuracy and reliability of positioning. The timestamp is used to record the reception time of each distress signal. Since it takes a certain amount of time for the signal to propagate from the distress location to the receiving point, there is a propagation delay. Through the accurate timestamp, the system can calculate this delay and correct the initial distress location accordingly. The correction process takes into account factors such as signal propagation speed and path loss to ensure that the final distress location is more accurate. The Bayesian network prediction model is a method based on probabilistic reasoning. It combines the conditional probability distribution learned from the data set of historical distress events to evaluate the possibility of each potential distress area becoming an actual distress location. The model takes into account a variety of environmental parameters (such as ocean currents, wind direction, weather conditions, etc.) and generates a probability value for each distress area. This step improves the accuracy and reliability of the prediction. Based on the probability of occurrence values ​​generated by the Bayesian network prediction model, the system selects the distress area with the highest probability as the primary search target. This allows resources to be concentrated on the most likely area for search, avoiding ineffective coverage and improving search and rescue efficiency. In addition, it ensures the effective allocation of resources, making rescue operations more efficient.

[0092] In the embodiment of the present application, the system first uses the optimal geographical location information to match the initial distress position on the preset electronic map of the sea area to determine the specific distress area. Then, the signal propagation delay is calculated according to the timestamp, and the initial distress position is corrected to obtain a more accurate corrected distress position. Then, the Bayesian network prediction model is used to evaluate the probability distribution of each distress area and generate the probability value of each distress area. Finally, based on these probability values, the distress area with the highest probability is selected as the primary search target to ensure that resources are invested in this area first. The whole process not only improves the accuracy of the distress position, but also enhances the pertinence and efficiency of the search and rescue operation.

[0093] Here is a specific example:

[0094] In a specific maritime search and rescue mission, when a fishing boat was in distress in a certain sea area, the system received multiple distress signals and generated the optimal geographical location information through multi-source signal fusion. Next, the system matched this information with the preset electronic map of the sea area to determine the possible distress area of ​​the fishing boat. Taking into account the signal propagation delay, the system corrected the initial position and obtained a more accurate distress location. Then, the Bayesian network prediction model was applied to evaluate the probability distribution of multiple potential distress areas, and the results showed that the probability of a specific area was the highest. Based on this, the search and rescue center gave priority to dispatching drones to the area for search, which greatly improved the search and rescue efficiency. After the drone arrived at the designated area, it successfully found the fishermen in distress, and guided the subsequent rescue forces to quickly launch rescue operations, and finally successfully rescued the people in distress.

[0095] In this way, the system not only improves the accuracy of the distress location, but also enhances the robustness and adaptability of the system in complex and changeable marine environments, providing strong technical support for rapid and accurate rescue.

[0096] In order to solve the accuracy and reliability problems of distress area assessment, in some embodiments, the process of step 102, based on the corrected distress position, applying a Bayesian network prediction model to assess the probability distribution of each distress area to generate an occurrence probability value of each distress area, includes:

[0097] The corrected distress location and surrounding environment parameters are used as input variables to a pre-trained Bayesian network prediction model to obtain basic data; based on the basic data, combined with the conditional probability distribution learned from the data set of historical distress events, the possibility of each preset distress area becoming an actual distress location is evaluated and processed, and a preliminary occurrence probability value is output; based on the preliminary occurrence probability value, the occurrence probability value is further adjusted and optimized to ensure accuracy, and an optimized occurrence probability value is obtained; based on the optimized occurrence probability value, an occurrence probability value for each distress area is generated to ensure that resources are preferentially invested in the distress area with the highest probability.

[0098] In this embodiment, the distress location after signal propagation delay correction and the surrounding environmental parameters (such as ocean currents, wind direction, weather conditions, etc.) are used as input variables and input into a pre-trained Bayesian network prediction model. These environmental parameters provide additional contextual information, which helps to more accurately evaluate the possibility of potential distress areas. Basic data refers to the probability distribution under various conditions preliminarily calculated by the model. The Bayesian network prediction model uses the data set of historical distress events to learn the conditional probability distribution of each distress area becoming an actual distress location under different conditions. By combining the basic data with these conditional probability distributions, the system can evaluate the possibility of each preset distress area becoming an actual distress location and output a preliminary probability value of occurrence. This step improves the accuracy of the evaluation and ensures the reliability of the results. Although the preliminary probability value of occurrence has taken into account multiple factors, in order to further improve the accuracy, the system will adjust and optimize the preliminary probability value of occurrence based on the latest environmental data and other real-time information. For example, if a new distress signal or environmental change is found during the evaluation process, the system will recalculate and update the probability value to ensure that the final result is as accurate as possible. Finally, the system generates the probability of occurrence for each distress area based on the optimized probability of occurrence. The distress area with the highest probability is selected as the primary search target to ensure that resources are invested in this area first. This not only improves the efficiency of search and rescue, but also minimizes invalid coverage and concentrates efforts on searching in the most likely areas.

[0099] In an embodiment of the present application, the system first uses the corrected distress location and surrounding environment parameters as input variables, and inputs them into a pre-trained Bayesian network prediction model to obtain basic data. Then, combined with the conditional probability distribution learned from the data set of historical distress events, the possibility of each preset distress area becoming an actual distress location is evaluated, and a preliminary probability of occurrence value is output. Next, based on the preliminary probability of occurrence values, the system further adjusts and optimizes these probability values ​​to ensure their accuracy. Finally, the system generates a probability of occurrence value for each distress area, and selects the distress area with the highest probability as the primary search target to ensure that resources are invested in this area first. The entire process not only improves the accuracy of the evaluation, but also enhances the robustness and adaptability of the system.

[0100] Here is a specific example:

[0101] In a specific maritime search and rescue mission, when a fishing boat is in distress in a certain sea area, the system receives multiple distress signals and generates the optimal geographical location information through multi-source signal fusion. Taking into account the signal propagation delay, the system corrects the initial position and obtains a more accurate distress location. Then, the system inputs the corrected distress location and surrounding environmental parameters (such as ocean currents, wind direction, weather conditions, etc.) into the pre-trained Bayesian network prediction model to obtain basic data. Then, combined with the conditional probability distribution learned from the data set of historical distress events, the system evaluates the possibility of multiple potential distress areas becoming actual distress locations and outputs preliminary probability values. To further improve accuracy, the system is adjusted and optimized based on the latest environmental data, and finally generates the probability value of each distress area. The results show that the probability of a specific area is the highest, and the search and rescue center gives priority to dispatching drones to search in this area, which greatly improves the efficiency of search and rescue. After the drone arrived at the designated area, it successfully found the fishermen in distress and guided the subsequent rescue forces to quickly launch rescue operations, and finally successfully rescued the people in distress.

[0102] In this way, the system not only improves the accuracy of distress location assessment, but also enhances the robustness and adaptability of the system in complex and changeable marine environments, providing strong technical support for rapid and accurate rescue.

[0103] In order to solve the problem of accuracy in identifying non-fixed-form objects under complex weather conditions, in some embodiments, the thermal imaging and optical sensing systems on the rescue drone are started to scan according to the primary search target with the highest probability in step 103, and the non-fixed-form object recognition capability is enhanced by combining a convolutional neural network, and a generative adversarial network is used to simulate image features under different weather conditions to generate a high-accuracy potential survivor position, including:

[0104] Based on the primary search target with the highest probability, the thermal imaging and optical sensing systems on the rescue drone are activated to scan and obtain preliminary image data; the preliminary image data is used in combination with a convolutional neural network to identify and process non-fixed-form objects in the preliminary image data to obtain preliminary object recognition results; based on the preliminary object recognition results, a generative adversarial network is used to simulate image features under different weather conditions to generate optimized image features; based on the optimized image features, the precise locations of potential survivors are further analyzed and confirmed to generate highly accurate potential survivor locations.

[0105] In this embodiment, it involves starting the thermal imaging and optical sensing systems on the rescue drone to scan according to the primary search target with the highest probability. The thermal imaging system can detect human body heat in low visibility or night conditions, while the optical sensing system is used to capture visual information. The preliminary image data collected by these sensors is the basis for subsequent processing for further analysis and confirmation of the existence of potential survivors. Convolutional neural network is a deep learning model that is good at extracting complex features from images. By inputting the preliminary image data into a pre-trained convolutional neural network, the system is able to identify non-fixed form objects in the image (such as floating human bodies, ship wreckage, etc.). The preliminary object recognition results provide a preliminary judgment on the existence of potential survivors, laying the foundation for further analysis. The generative adversarial network consists of a generator and a discriminator, which can simulate image features under different weather conditions. Based on the preliminary object recognition results, the system uses the generative adversarial network to generate optimized image features, taking into account the impact of various severe weather conditions (such as fog, rain, strong wind, etc.) on the image. This step improves the robustness of image recognition and ensures that potential survivors can be accurately identified even in complex and changing environments. Finally, the system further analyzes and confirms the precise location of potential survivors based on the optimized image features. By integrating multiple image features and environmental parameters, the system is able to generate highly accurate potential survivor locations, providing a reliable basis for subsequent rescue operations. This step not only improves the accuracy of recognition, but also enhances the adaptability and reliability of the system.

[0106] In an embodiment of the present application, the system first starts the thermal imaging and optical sensing system on the rescue drone, scans according to the primary search target with the highest probability, and obtains preliminary image data. Then, these preliminary image data are input into a pre-trained convolutional neural network to identify non-fixed form objects in the image and obtain preliminary object recognition results. Next, the system uses a generative adversarial network to simulate image features under different weather conditions and generate optimized image features. Finally, based on the optimized image features, the system further analyzes and confirms the precise location of potential survivors and generates highly accurate potential survivor locations. The entire process not only improves the accuracy of non-fixed form object recognition, but also enhances the robustness and adaptability of the system under complex and changeable weather conditions.

[0107] Here is a specific example:

[0108] In a specific maritime search and rescue mission, once the system identified a primary search target with the highest probability, it immediately activated the thermal imaging and optical sensing systems on the rescue drone to scan the area. The drone conducted a detailed scan in the designated area and collected preliminary image data. These image data contain the heat distribution captured by the thermal imaging and the visual information captured by the optical sensing system. Next, the system input these preliminary image data into the pre-trained convolutional neural network, identified multiple suspected floating objects, and obtained preliminary object recognition results. In order to cope with the current complex weather conditions (such as dense fog), the system applied the generative adversarial network to generate optimized image features, taking into account the impact of fog on the image. Finally, based on the optimized image features, the system further analyzed and confirmed the precise location of potential survivors and generated highly accurate potential survivor locations. Based on these highly accurate location information, the search and rescue team quickly launched a rescue operation and successfully rescued the people in distress.

[0109] In this way, the system not only improves the accuracy of identifying objects with non-fixed shapes, but also enhances the robustness and adaptability of the system under complex and changeable weather conditions, providing strong technical support for fast and accurate rescue.

[0110] In order to solve the problem of accuracy in identifying non-fixed-form objects under complex weather conditions, in some embodiments, the use of the preliminary image data in step 103 in combination with a convolutional neural network to identify non-fixed-form objects in the preliminary image data to obtain preliminary object recognition results includes:

[0111] The preliminary image data is used as input to a pre-trained convolutional neural network, and feature extraction processing is performed on the preliminary image data to generate detailed feature information; based on the detailed feature information, the feature information is further refined through multi-layer convolution and pooling operations of the convolutional neural network, and an optimized feature representation is obtained according to the visual characteristics of shape, texture and color; the optimized feature representation is classified through the classification layer of the convolutional neural network, and irregular or floating objects are accurately classified to generate classification results; based on the classification results, a preliminary object recognition result is generated, and the preliminary object recognition result includes preliminary location information of potential survivors.

[0112] In this embodiment, the preliminary image data obtained by the drone through thermal imaging and optical sensing systems is fed into a pre-trained convolutional neural network as input. Convolutional neural networks are deep learning models that are good at extracting complex features from images. The preliminary image data includes the heat distribution captured by thermal imaging and the visual information captured by the optical sensing system. These data are used for feature extraction processing to generate detailed feature information, such as edges, textures, and colors. The convolutional neural network gradually refines the feature information through multiple layers of convolution and pooling operations. The convolution operation is used to extract local features, such as edges and textures; the pooling operation is used to reduce the size of the feature map and retain the most important features. Through these operations, the system can generate more optimized feature representations based on visual characteristics such as shape, texture, and color. This process enhances the recognition ability of non-fixed objects (such as floating human bodies or ship wreckage) and improves the accuracy of classification. The classification layer is the last part of the convolutional neural network, which is responsible for mapping the optimized feature representations to different categories. For maritime search and rescue missions, the classification layer focuses on identifying irregular or floating objects, such as human bodies, life jackets, floating objects, etc. By classifying the optimized feature representations, the system can generate accurate classification results and distinguish potential survivors from other objects. Finally, the system generates preliminary object recognition results based on the classification results, which not only include the categories of the objects recognized, but also provide the location information of these objects in the image. In particular, for potential survivors, the system will generate their preliminary location information to provide a reliable basis for subsequent rescue operations. This step ensures the practicality and reliability of the recognition results and improves the efficiency of search and rescue.

[0113] In an embodiment of the present application, the system first uses preliminary image data as input and feeds it into a pre-trained convolutional neural network to perform feature extraction processing and generate detailed feature information. Then, through the multi-layer convolution and pooling operations of the convolutional neural network, the feature information is further refined, and an optimized feature representation is obtained based on the visual characteristics of shape, texture, and color. Next, the system classifies the optimized feature representation through the classification layer of the convolutional neural network, accurately classifies irregular or floating objects, and generates classification results. Finally, based on the classification results, the system generates preliminary object recognition results, which contain preliminary location information of potential survivors. The whole process not only improves the accuracy of non-fixed form object recognition, but also enhances the robustness and adaptability of the system in complex and changing environments.

[0114] Here is a specific example:

[0115] In a specific maritime search and rescue mission, once the system has identified a primary search target with the highest probability, it immediately activates the thermal imaging and optical sensing systems on the rescue drone to scan the area. The drone conducted a detailed scan in the designated area and collected preliminary image data. These image data contain the heat distribution captured by the thermal imaging and the visual information captured by the optical sensing system. Next, the system uses these preliminary image data as input and feeds them into a pre-trained convolutional neural network for feature extraction and generation of detailed feature information. Then, through multi-layer convolution and pooling operations, the system further refines these feature information and obtains optimized feature representations based on the visual characteristics of shape, texture, and color. Next, the system classifies the optimized feature representations through the classification layer of the convolutional neural network, identifies multiple suspected floating objects, and generates classification results. Finally, based on the classification results, the system generates preliminary object recognition results and confirms the preliminary location information of potential survivors. Based on these highly accurate location information, the search and rescue team quickly launched a rescue operation and successfully rescued the people in distress.

[0116] In this way, the system not only improves the accuracy of identifying objects with non-fixed shapes, but also enhances the robustness and adaptability of the system under complex and changeable weather conditions, providing strong technical support for fast and accurate rescue.

[0117] In order to solve the efficiency and reliability problems of search and rescue path planning in complex marine environments, in some embodiments, the use of ocean current prediction algorithms and wind direction data in step 104 is used to dynamically adjust and optimize the search and rescue path to ensure rapid coverage of the distress range where the potential survivors are located, and the search strategy is adaptively adjusted through a reinforcement learning algorithm, the optimal flight route is updated in real time, and the remaining power and endurance of the drone are taken into consideration to generate an economical and efficient search and rescue path planning, including:

[0118] The search and rescue path is preliminarily planned by using the obtained high-accuracy potential survivor positions and combining the real-time acquired ocean current prediction algorithm and wind direction data to obtain a preliminary search and rescue path; based on the preliminary search and rescue path, the search and rescue path is dynamically adjusted and optimized according to the ocean current prediction algorithm and wind direction data to adapt to the changing ocean environment, ensure the effectiveness and timeliness of the path, and generate an optimized search and rescue path; the optimized search and rescue path is adaptively adjusted by the reinforcement learning algorithm, and the search strategy is updated in real time according to the situation encountered in the actual search and rescue process to generate a real-time updated search and rescue path; the real-time updated search and rescue path is further processed by comprehensively considering the remaining power and endurance of the drone to ensure that the drone can return safely after completing the mission, while avoiding unnecessary energy waste, and generating a final flight route; based on the high-accuracy potential survivor positions, the real-time ocean current prediction algorithm and wind direction data, the dynamically adjusted and optimized search and rescue path, the adaptively adjusted search strategy, and the remaining power and endurance of the drone, the final flight route is evaluated and optimized to generate an economical and efficient search and rescue path planning.

[0119] In this embodiment, it involves combining highly accurate potential survivor location information with real-time acquired ocean current prediction algorithms and wind direction data to perform preliminary search and rescue path planning. Ocean current prediction algorithms are used to predict the direction and speed of ocean flow, while wind direction data helps adjust the flight attitude and path of the drone. These data are used together to generate a preliminary search and rescue path to ensure the basic accuracy of path planning. Although the preliminary search and rescue path has taken into account a variety of factors, in order to cope with the ever-changing ocean environment (such as sudden changes in ocean currents or wind direction changes), the system will dynamically adjust and optimize the search and rescue path based on the latest ocean current prediction algorithms and wind direction data. This process ensures the effectiveness and timeliness of the path, allowing the drone to quickly and accurately cover the distress area where potential survivors are located. Reinforcement learning algorithm is a machine learning method that can adaptively adjust behavior strategies in a constantly changing environment. Through the reinforcement learning algorithm, the system updates the search strategy in real time according to the situations encountered during the actual search and rescue process (such as new distress signals, environmental changes, etc.), and generates a real-time updated search and rescue path. This step improves the flexibility and response speed of the system, ensuring that the search and rescue operation can quickly adapt to new situations. To ensure the safety and energy saving of the drone, the system will further process the search and rescue path updated in real time by taking into account the remaining power and endurance of the drone. The system will dynamically adjust the path to ensure that the drone can return to the base safely after completing the mission, while avoiding unnecessary energy waste. This step not only improves the efficiency of search and rescue, but also extends the operation time of the drone. Finally, based on all the above information, the system evaluates and optimizes the final flight route to generate a cost-effective search and rescue path plan. This step ensures the efficiency and sustainability of the entire search and rescue operation and maximizes the success rate of the rescue.

[0120] In the embodiment of the present application, the system first uses the highly accurate location information of potential survivors and the real-time acquired current prediction algorithm and wind direction data to perform preliminary search and rescue path planning. Then, according to the latest current prediction algorithm and wind direction data, the search and rescue path is dynamically adjusted and optimized to ensure the effectiveness and timeliness of the path. Next, through the reinforcement learning algorithm, the optimized search and rescue path is adaptively adjusted and the search strategy is updated in real time. In order to ensure the safety and energy saving of the drone, the system comprehensively considers the remaining power and endurance of the drone, further processes the real-time updated search and rescue path, and generates the final flight route. Finally, the system evaluates and optimizes the final flight route to generate an economical and efficient search and rescue path planning. The whole process not only improves the efficiency and reliability of search and rescue path planning, but also enhances the robustness and adaptability of the system in a complex and changeable marine environment.

[0121] Here is a specific example:

[0122] In a specific maritime search and rescue mission, once the system determined the potential survivors’ positions with high accuracy, it immediately started the search and rescue path planning. First, the system combined the real-time current prediction algorithm and wind direction data to conduct preliminary search and rescue path planning and generate a preliminary search and rescue path. Then, the system dynamically adjusted and optimized the search and rescue path according to the latest current prediction algorithm and wind direction data to ensure the effectiveness and timeliness of the path. During the search and rescue process, the system updated the search strategy in real time and generated a real-time updated search and rescue path through the reinforcement learning algorithm according to the actual situation encountered (such as new distress signals, sudden changes in ocean currents, etc.). In order to ensure the safety and energy saving of the drone, the system comprehensively considered the remaining power and endurance of the drone, further processed the real-time updated search and rescue path, and generated the final flight route. Finally, the system evaluated and optimized the final flight route and generated an economical and efficient search and rescue path planning. The drone quickly launched a rescue operation based on the path and successfully found and rescued the people in distress.

[0123] In this way, the system not only improves the efficiency and reliability of search and rescue path planning, but also enhances the robustness and adaptability of the system in complex and changeable marine environments, providing strong technical support for rapid and accurate rescue.

[0124] In order to solve the flexibility and efficiency problems of dynamic adjustment of the search and rescue path in a complex marine environment, in some embodiments, the optimized search and rescue path is adaptively adjusted by the reinforcement learning algorithm in step 104, and the search strategy is updated in real time according to the situation encountered in the actual search and rescue process to generate a real-time updated search and rescue path, including:

[0125] The optimized search and rescue path is used in combination with the actual search and rescue environment data obtained in real time to provide a training basis for the reinforcement learning algorithm to obtain initial training data; based on the initial training data, the state of the current search and rescue environment is represented by the state space in the reinforcement learning algorithm, and the position and state of the drone are taken as part of the environment to ensure that the reinforcement learning algorithm can fully understand the current situation and generate an environmental state representation; according to the environmental state representation, the adjustment operations defined by the action space in the reinforcement learning algorithm, which include changing the flight direction, adjusting the flight altitude, accelerating or decelerating, generate an adjustment plan; using the adjustment plan, the effect of each adjustment operation is evaluated by the reward mechanism in the reinforcement learning algorithm, and the reward mechanism takes into account the factors of search and rescue efficiency, safety and resource consumption to generate an evaluation result; based on the evaluation result, the reinforcement learning algorithm is applied for iterative learning, and the search strategy is continuously optimized according to the result of each adjustment operation, so that the application of the reinforcement learning algorithm can gradually find the optimal path adjustment plan in the actual search and rescue process and generate an optimized search strategy; according to the optimized search strategy, the search and rescue path is updated in real time, and a real-time updated search and rescue path is generated to ensure that the drone can dynamically adjust the flight route according to the latest environmental information and improve the search and rescue efficiency.

[0126] In this embodiment, it involves combining the optimized search and rescue path with the actual search and rescue environment data (such as ocean current, wind direction, weather conditions, etc.) obtained in real time as the training basis of the reinforcement learning algorithm. The initial training data includes not only the position and state of the drone, but also various parameters of the surrounding environment to ensure that the algorithm can fully understand the current situation. The state space in the reinforcement learning algorithm is used to represent the state of the current search and rescue environment. The state space includes state information such as the position, height, speed, etc. of the drone, as well as various parameters of the surrounding environment (such as ocean current direction and speed, wind direction and intensity, etc.). In this way, the system can generate a detailed representation of the environmental state to ensure that the reinforcement learning algorithm can fully understand the current search and rescue environment. The action space defines all possible operations that the drone can perform, such as changing the flight direction, adjusting the flight altitude, accelerating or decelerating, etc. Based on the environmental state representation, the system generates a series of possible adjustment schemes to cope with the current search and rescue environment. These adjustment schemes are intended to optimize the search and rescue path and improve the efficiency and safety of the search and rescue. The reward mechanism is used to evaluate the effect of each adjustment operation. The reward mechanism takes into account multiple factors, such as search and rescue efficiency (rapid coverage of the distress area), safety (avoiding dangerous areas) and resource consumption (saving power and fuel). Through this evaluation, the system can determine which adjustment operations are most conducive to achieving the goal and generate evaluation results. Based on the evaluation results, the system applies the reinforcement learning algorithm for iterative learning. After each adjustment operation, the system will provide feedback based on the actual effect and gradually optimize the search strategy. Through continuous iterative learning, the system can gradually find the optimal path adjustment plan in the actual search and rescue process, further improving the efficiency and reliability of search and rescue. Finally, the system updates the search and rescue path in real time based on the optimized search strategy and generates the latest flight route. This process ensures that the drone can dynamically adjust the flight route based on the latest environmental information, quickly respond to changing search and rescue conditions, and improve the efficiency and success rate of search and rescue.

[0127] In an embodiment of the present application, the system first combines the optimized search and rescue path with the actual search and rescue environment data obtained in real time as the training basis of the reinforcement learning algorithm to generate initial training data. Then, the state of the current search and rescue environment is represented by the state space, and the position and state of the drone are used as part of the environment to generate an environmental state representation. Next, the system generates a series of possible adjustment schemes based on the environmental state representation through the adjustment operations defined in the action space. Using these adjustment schemes, the system evaluates the effect of each adjustment operation through a reward mechanism and generates an evaluation result. Based on the evaluation results, the system applies a reinforcement learning algorithm for iterative learning and continuously optimizes the search strategy. Finally, the system updates the search and rescue path in real time based on the optimized search strategy and generates the latest flight route. The whole process not only improves the flexibility and efficiency of the dynamic adjustment of the search and rescue path, but also enhances the robustness and adaptability of the system in a complex and changing environment.

[0128] Here is a specific example:

[0129] In a specific maritime search and rescue mission, once the system determined the optimized search and rescue path, it immediately started the reinforcement learning algorithm for adaptive adjustment. First, the system combined the optimized search and rescue path with the actual search and rescue environment data (such as ocean currents, wind direction, weather conditions, etc.) obtained in real time as the training basis to generate initial training data. Then, the state of the current search and rescue environment is represented by the state space, and the position and state of the drone are used as part of the environment to generate a detailed representation of the environment state. Next, based on the environmental state representation, the system generates a series of possible adjustment schemes through adjustment operations defined in the action space (such as changing the flight direction, adjusting the flight altitude, accelerating or decelerating). Using these adjustment schemes, the system evaluates the effect of each adjustment operation through a reward mechanism, taking into account factors such as search and rescue efficiency, safety, and resource consumption, and generates evaluation results. Based on the evaluation results, the system applies the reinforcement learning algorithm for iterative learning and continuously optimizes the search strategy. Finally, the system updates the search and rescue path in real time according to the optimized search strategy and generates the latest flight route. The drone quickly launched a rescue operation based on the path and successfully found and rescued the people in distress.

[0130] In this way, the system not only improves the flexibility and efficiency of dynamic adjustment of search and rescue paths, but also enhances the robustness and adaptability of the system in complex and changeable marine environments, providing strong technical support for rapid and accurate rescue.

[0131] This application takes into account that in maritime search and rescue missions, it is crucial to accurately assess the possibility of each preset distress area becoming an actual distress location. Traditional methods often rely on a single environmental parameter or historical data, and it is difficult to fully consider the impact of multiple factors, resulting in inaccurate evaluation results. In order to improve the accuracy of the evaluation, the R&D team introduced a Bayesian network prediction model, and combined with the conditional probability distribution learned from the data set of historical distress events, proposed an optimized formula for calculating the probability of occurrence. This formula aims to comprehensively consider the impact of the preliminary probability value and the prior probability, and adjust it through weight factors and exponential factors to ensure the high accuracy and reliability of the evaluation results. Therefore, a new optional scheme is proposed, which includes:

[0132] Based on the corrected distress location, the probability distribution of each distress area is evaluated using a Bayesian network prediction model to generate an occurrence probability value for each distress area, including:

[0133] The corrected distress location and surrounding environment parameters are used as input variables X to input into the pre-trained Bayesian network prediction model to obtain basic data D, where X = [x1, x2, ..., x n} represents the set of corrected distress location and surrounding environmental parameters, D represents the basic data obtained after processing by the Bayesian network prediction model, including the preliminary analysis results of the model on environmental parameters;

[0134] Based on the basic data D and the conditional probability distribution P(Y|X) learned from the data set of historical distress events, the probability of each preset distress area becoming an actual distress location is evaluated and processed, and a preliminary occurrence probability value P is output. pre (Y i |D), where Y i represents the i-th preset distress area, P(Y|X) is the conditional probability distribution learned from the data set of historical distress events, which represents the probability of distress occurring in a certain area Y under given environmental parameters X;

[0135] Based on the preliminary probability value P pre (Y i |D), further adjust and optimize the occurrence probability value to ensure accuracy, and obtain the optimized occurrence probability value P opt (Y i ∣D);

[0136] The optimized probability value P is calculated by the following formula: opt (Y i |D):

[0137]

[0138] Among them, P opt (Y i |D) is the optimized probability value, which improves the accuracy of probability estimation after adjustment and optimization; α is the weight factor (0<α<1), which is used to balance the influence of the initial probability value and the prior probability; β and γ are exponential factors, which are used to adjust the importance of the initial probability value and the prior probability respectively; P prior (Y i ) is the prior probability, indicating that in the absence of specific environmental parameters X, a certain area Y i The probability of distress; j is an index variable that traverses all preset distress areas, ranging from 1 to m; m is the number of preset distress areas;

[0139] According to the optimized probability value P opt (Y i |D), generate the probability of occurrence of each distress area, and ensure that resources are invested in the distress areas with the highest probability first.

[0140] The following is a detailed explanation of each parameter:

[0141] X: is the input variable, which represents the set of corrected distress location and surrounding environmental parameters. Through multi-source signal fusion and timestamp correction, the optimal geographical location information is obtained, combined with real-time environmental parameters such as ocean currents, wind direction, and weather conditions. These data together constitute the input variable X.

[0142] D: is the basic data, which is obtained after processing by the Bayesian network prediction model, and contains the preliminary analysis results of the model on environmental parameters. Input variable X is input into the pre-trained Bayesian network prediction model, and the basic data D output by the model contains the preliminary analysis results of environmental parameters.

[0143] P(Y|X): is a conditional probability distribution learned from a data set of historical distress events, which indicates the probability of distress occurring in a certain area Y under given environmental parameters X. Through machine learning methods, a large number of historical distress event data sets are used for training to obtain the conditional probability distribution P(Y|X).

[0144] P pre (Y i |D): is the preliminary probability value. Based on the basic data D and the conditional probability distribution P(Y|X), the possibility of each preset distress area becoming an actual distress location is evaluated and processed, and the preliminary probability value of occurrence is output. Based on the basic data D and the conditional probability distribution P(Y|X), the probability of each preset distress area Y i The initial probability of occurrence.

[0145] P prior (Y i ) : is the prior probability, indicating that in the absence of specific environmental parameters X, a certain area Y i The probability of distress. The prior probability of each area is calculated by statistically analyzing the frequency distribution of historical distress event data sets.

[0146] α: It is a weight factor used to balance the impact of the initial probability value and the prior probability, and its value range is 0<α<1. It is usually set through experiments or expert experience, and can also be optimized through methods such as cross-validation.

[0147] β and γ are exponential factors used to adjust the importance of the initial probability value and the prior probability, respectively. They can be set through experiments or expert experience, and can also be optimized through methods such as cross-validation.

[0148] j: is the index variable. Iterates through all preset distress areas, ranging from 1 to m. It is automatically determined based on the number of preset distress areas m.

[0149] m: is the number of preset distress areas, indicating the total number of distress areas preset by the system. It is determined according to specific mission requirements and geographic information.

[0150] The following is an introduction to the design reasons of each sub-item:

[0151] α·P pre (Y i ∣D) β :This part reflects the initial probability value P pre (Y i |D) on the final optimized probability value. The weight factor α controls its importance, while the exponential factor β is used to further adjust the influence of the preliminary probability value.

[0152] (1-α)·P prior (Y i ) γ :This part reflects the prior probability P prior (Y i ) on the final optimized probability value. The weight factor 1-α controls its importance, while the exponential factor γ is used to further adjust the influence of the prior probability.

[0153] This part is to perform a weighted summation of the preliminary probability values ​​of all preset distress areas to ensure the normalization of the probability values ​​after final optimization.

[0154] This part is to perform a weighted summation of the prior probabilities of all preset distress areas, and also ensure the normalization of the probability values ​​after the final optimization.

[0155] The addition method is used in both the numerator and the denominator to comprehensively consider the impact of the preliminary probability value and the prior probability. By adding, it can ensure that the impact of both can be reasonably reflected while maintaining the normalized nature of the final result.

[0156] The overall design of the formula aims to improve the accuracy and reliability of distress area assessments. By introducing the Bayesian network prediction model and learning from historical distress event data sets, the system is able to more comprehensively consider the impact of multiple factors. Specifically, through the weight factors α and 1-α, the formula is able to find a reasonable balance between the preliminary probability value and the prior probability to avoid over-reliance on a certain factor. Through the exponential factors β and γ, the formula can flexibly adjust the relative importance of the preliminary probability value and the prior probability according to actual conditions, thereby improving the flexibility and adaptability of the assessment. Through the weighted summation of the denominator, the formula ensures the normalization of the probability value after the final optimization, making the results comparable and interpretable.

[0157] In summary, this formula not only improves the accuracy of distress area assessment in complex marine environments, but also enhances the robustness and adaptability of the system, providing strong technical support for rapid and accurate rescue.

[0158] Here is a specific example:

[0159] In a specific maritime search and rescue mission, the system receives multiple distress signals and generates the optimal geographic location information through multi-source signal fusion. Taking into account the signal propagation delay, the system corrects the initial position and obtains a more accurate distress location. Next, the system needs to evaluate the possibility of multiple potential distress areas becoming actual distress locations to determine the primary search target. To achieve this goal, the system applies the above optimized occurrence probability calculation formula.

[0160] Assume that the system has input the corrected distress location and surrounding environmental parameters (such as ocean currents, wind direction, weather conditions, etc.) as input variables X into the pre-trained Bayesian network prediction model to obtain basic data D. These data include:

[0161] Corrected distress location and surrounding environment parameter set: X = {x1, x2, ..., x n}

[0162] Basic data D, including the model's preliminary analysis results of environmental parameters

[0163] The number of preset distress areas m = 5

[0164] Here are the calculation steps:

[0165] Based on the basic data D and the conditional probability distribution P(Y|X) learned from the data set of historical distress events, the probability of each preset distress area becoming an actual distress location is evaluated and processed, and a preliminary occurrence probability value P is output. pre (Y i ∣D).

[0166] Assume that the following preliminary probability values ​​are obtained:

[0167] P pre (Y1|D)=0.85;

[0168] P pre (Y2|D)=0.70;

[0169] P pre (Y3|D)=0.60;

[0170] P pre (Y4|D)=0.45;

[0171] P pre (Y5|D)=0.30;

[0172] Get the prior probability:

[0173] Prior probability P prior (Y i ) means that in the absence of specific environmental parameters X, a certain area Y i The probability of distress. Assume that the following prior probabilities are obtained from historical data:

[0174] P prior (Y1) = 0.20;

[0175] P prior (Y2) = 0.15;

[0176] P prior (Y3) = 0.10;

[0177] P prior (Y4) = 0.08;

[0178] P prior (Y5) = 0.07;

[0179] Setting parameters:

[0180] Set the weight factor α=0.7, the exponential factor β=2, and γ=1.5.

[0181] Use the formula to calculate the optimized occurrence probability value P for each preset distress area opt (Y i |D):

[0182]

[0183] Substitute the numerical calculation:

[0184] For Y1:

[0185]

[0186] P opt (Y1|D)≈0.402

[0187] Similarly, the optimized occurrence probability values ​​of other regions can be calculated:

[0188] P opt (Y2|D)≈0.308;

[0189] P opt (Y3|D)≈0.212;

[0190] P opt (Y4|D)≈0.065;

[0191] P opt (Y5|D)≈0.013;

[0192] Through the above calculations, the optimized occurrence probability values ​​of each preset distress area are obtained. According to these probability values, the system can prioritize resources to the distress area Y1 with the highest probability, and its optimized occurrence probability value is 0.402. This not only improves the efficiency of search and rescue, but also minimizes invalid coverage and concentrates efforts on searching in the most likely areas, thereby significantly improving the success rate of rescue.

[0193] By applying the optimized occurrence probability calculation formula, the system significantly improves the accuracy and reliability of distress area assessment in complex marine environments. Specifically, the formula comprehensively considers the impact of the preliminary probability value and the prior probability, and adjusts it through weight factors and exponential factors to ensure the high accuracy of the assessment results. Ultimately, the system can generate the occurrence probability value for each distress area based on the optimized occurrence probability value, ensuring that resources are invested first in the distress areas with the highest probability, thereby improving the efficiency and success rate of search and rescue. Overall, this method not only improves the flexibility and efficiency of search and rescue, but also enhances the robustness and adaptability of the system in complex and changeable marine environments, providing strong technical support for rapid and accurate rescue.

[0194] This application takes into account that under complex weather conditions, maritime search and rescue missions face many challenges, such as low visibility, strong winds, waves and other environmental factors, which make the identification of non-fixed objects (such as floating human bodies or ship wreckage) extremely difficult. Traditional image processing methods are difficult to effectively cope with these challenges, resulting in low recognition accuracy. In order to solve this problem, the R&D team introduced convolutional neural networks, and combined with specific formulas to extract features, optimize and classify preliminary image data to improve the recognition accuracy of non-fixed objects. This formula scheme aims to gradually refine feature information through multi-layer convolution and pooling operations, and generate high-confidence object recognition results through the classification layer. Therefore, a new optional scheme is proposed, which includes:

[0195] Using the preliminary image data in combination with a convolutional neural network, the non-fixed form objects in the preliminary image data are identified and processed to obtain preliminary object recognition results, including:

[0196] The preliminary image data I is used as input to feed into a pre-trained convolutional neural network to perform feature extraction on the preliminary image data to generate detailed feature information F (l) ;

[0197] Detailed feature information F is extracted through the following formula (l) :

[0198] F (l) =σ(W (l) *I+b (l) )

[0199] Where I is the preliminary image data; W (l) and b (l) are the convolution kernel weight matrix and bias term of the lth layer respectively; * represents the convolution operation; σ is the activation function; F (l) Represents detailed feature information;

[0200] Based on the detailed feature information F (l) Through the multi-layer convolution and pooling operations of the convolutional neural network, the feature information is further refined, and the optimized feature representation is obtained according to the visual characteristics of shape, texture and color.

[0201] The optimized feature representation is obtained through the following formula

[0202]

[0203] in, represents the optimized feature representation of the lth layer, which is a more refined feature obtained after multiple layers of convolution and pooling operations; γ l is the exponential factor of the lth layer, which is used to adjust the importance of each layer output; Pooling represents the pooling operation; σ is the activation function; F (l-1) Represents the detailed feature information of the previous layer;

[0204] Through the classification layer of the convolutional neural network, the optimized feature representation Perform classification processing, accurately classify irregular or floating objects, and generate classification results C i ;

[0205] The classification result C is calculated by the following formula: i :

[0206]

[0207] Among them, C i Represents the classification result, which is the probability distribution calculated by the classification layer and the softmax function; W c and b c are the weight matrix and bias term of the classification layer respectively; N is the number of categories; β' is an exponential factor used to adjust the sharpness of the classification result; i represents the index of the category, which is used to traverse all possible categories; j is an index variable ranging from 1 to N, which is used to traverse all possible categories;

[0208] According to the classification result C i , generating a preliminary object recognition result R, wherein the preliminary object recognition result includes preliminary location information of potential survivors;

[0209] The preliminary object recognition result R is calculated using the following formula:

[0210]

[0211] Where R represents the preliminary object recognition result; arg max(C i) The function returns the classification result C i The index of the category with the highest probability; θ is the confidence threshold, which is used to filter the results with low confidence; max(C i ) represents the classification result C i The maximum probability value in the if conditional statement is used to determine whether the classification result is credible. If the maximum probability value max(C i ) is greater than the set threshold θ, the category with the highest probability is selected as the final recognition result; otherwise, "unrecognized" is output, indicating that there is not enough confidence to determine any category.

[0212] The following is a detailed explanation of each parameter:

[0213] I: It is the preliminary image data, the original image input to the convolutional neural network, usually a two-dimensional matrix, each element represents a pixel value. It is obtained through a camera, drone or other image acquisition equipment.

[0214] W (l) and b (l) : The convolution kernel weight matrix and bias term are the parameters used to extract features in the convolution layer. The weight matrix is ​​used to detect different types of features (such as edges, textures), while the bias term allows the model to better fit the data. They are automatically adjusted by the back-propagation algorithm during training to minimize the loss function.

[0215] σ: is the activation function, which introduces nonlinear factors and enables the model to learn more complex patterns.

[0216] F (l) : It is detailed feature information. The output after each convolution operation represents the degree of existence of certain specific patterns or features in the image. It is generated by the convolution operation and the activation function.

[0217] It is an optimized feature representation, a more abstract and discriminative feature representation obtained after multiple layers of convolution and pooling operations. Pooling operations are applied after each layer of convolution, and an exponential factor may be applied to increase or decrease the importance of the layer.

[0218] γ l: is an exponential factor that adjusts the importance of each layer’s output and can be considered a hyperparameter that affects the composition of the final feature representation. Adjust according to specific task requirements and experimental results to achieve optimal performance.

[0219] W c b c : They are the weight matrix and bias term of the classification layer, which are used to map the last layer features to the parameters of the category space. They are also automatically adjusted during the training process.

[0220] N: is the number of categories, the number of different objects the system tries to identify. It is determined according to the application scenario. For example, in a search and rescue scenario, it may be a human body, a life jacket, and others.

[0221] β': It is an exponential factor, which is used to adjust the sharpness of the classification results. A higher value will make the probability distribution more concentrated in a certain category. Adjust it according to actual needs to obtain more accurate classification results.

[0222] θ: is the confidence threshold, which determines whether to accept the classification result. Results above this value are considered reliable. In order to ensure high-precision recognition results and avoid false positives.

[0223] C i : is the classification result, using the probability distribution calculated by the softmax function, indicating the possibility of each category appearing. It provides a probabilistic classification result, which is convenient for selecting the most likely category.

[0224] R: is the preliminary object recognition result, which contains the location information of potential targets. According to the classification result Ci and the confidence threshold θ, it is determined whether there is a target and its location.

[0225] Here is a specific example:

[0226] In a specific maritime search and rescue mission, the system receives multiple distress signals and generates the optimal geographic location information through multi-source signal fusion. Taking into account the signal propagation delay, the system corrects the initial position and obtains a more accurate distress location. Next, the system needs to process the preliminary image data I captured by the drone to identify the location of potential survivors. To achieve this goal, the system applies the feature extraction and classification formulas detailed above.

[0227] It is assumed that the system has acquired preliminary image data I and processed it through a pre-trained convolutional neural network.

[0228] The specific parameters are as follows:

[0229] Preliminary image data I: grayscale image with a size of 256×256 pixels.

[0230] Convolution kernel weight matrix W (l) and the bias term b (l) : They are the convolution kernel weight matrix and bias term of the lth layer, respectively, which are assumed to have been pre-trained.

[0231] Activation function σ: ReLU activation function is used.

[0232] Exponential factor γ l : Used to adjust the importance of each layer's output and is set to 1.5.

[0233] Classification layer weight matrix W c and the bias term b c : They are the weight matrix and bias term of the classification layer, respectively, which are assumed to have been pre-trained.

[0234] Number of categories N: set to 3 categories (human body, Raw clothing, others).

[0235] Exponential factor β': used to adjust the sharpness of the classification results, set to 0.5.

[0236] Confidence threshold θ: set to 0.7.

[0237] Here are the calculation steps:

[0238] The preliminary image data I is used as input to feed into the pre-trained convolutional neural network to extract features from the preliminary image data and generate detailed feature information F (l) .

[0239] F (l) =σ(W (l) *I+b (l) )

[0240] Assume that the detailed feature information obtained after the first layer of convolution operation is F (1) .

[0241] Based on detailed feature information F (l) Through the multi-layer convolution and pooling operations of the convolutional neural network, the feature information is further refined, and the optimized feature representation is obtained according to the visual characteristics of shape, texture and color.

[0242]

[0243] Assume that the optimized feature representation obtained after the second layer of convolution and pooling operations is

[0244] Through the classification layer of the convolutional neural network, the optimized feature representation Perform classification processing, accurately classify irregular or floating objects, and generate classification results C i ;

[0245]

[0246] Assume that the calculated classification result is:

[0247] C1≈0.85 (human body);

[0248] C2≈0.10 (number of raw clothes);

[0249] C3≈0.05(others);

[0250] Generate preliminary object recognition results:

[0251] According to the classification results C i , generate preliminary object recognition results R.

[0252]

[0253] In this case, the maximum probability value max(C i )=0.85 is greater than the set threshold θ=0.7, so the category with the highest probability is selected as the final recognition result.

[0254] R = arg max (C i )=1(human body)

[0255] Through the above calculations, we get the preliminary object recognition result R, confirming that the object in the image is a "human body" with a corresponding probability of 0.85. This not only improves the accuracy of recognition, but also enhances the robustness and adaptability of the system, ensuring that the location of potential survivors can be accurately identified in complex and changing environments. Ultimately, the system can quickly launch rescue operations based on these high-confidence recognition results, significantly improving the success rate of rescue.

[0256] By applying a detailed feature extraction and classification formula, the system significantly improves the accuracy of non-fixed form object recognition under complex weather conditions. Specifically, the formula scheme achieves improvements through the following points, gradually refining feature information and enhancing the recognition ability of non-fixed form objects (such as floating human bodies or ship wreckage). The probability distribution is calculated through the softmax function to ensure high confidence in the classification results. By setting a threshold θ, low-confidence results are filtered to avoid misjudgment.

[0257] Overall, this method not only improves the efficiency of search and rescue, but also enhances the robustness and adaptability of the system in complex and changeable marine environments, providing strong technical support for rapid and accurate rescue.

[0258] Figure 2 A schematic diagram of the structure of an intelligent automatic positioning system for marine emergency equipment is provided for an embodiment of the present application. Figure 2 As shown, the system includes:

[0259] A receiving and fusion module 21 is used to receive and fuse multiple distress signals from satellites, radio relay stations and nearby ships, and estimate the optimal geographical location information and time stamp;

[0260] Determine the evaluation module 22, which is used to determine the distress area on the preset sea area electronic map based on the optimal geographical location information, calculate the signal propagation delay according to the timestamp to correct the distress position, and apply the Bayesian network prediction model to evaluate the probability distribution of each potential distress area to generate the primary search target with the highest probability;

[0261] The scanning generation module 23 is used to start the thermal imaging and optical sensing systems on the rescue drone to scan according to the primary search target with the highest probability, combine the convolutional neural network to enhance the recognition ability of non-fixed form objects, and use the generative adversarial network to simulate the image features under different weather conditions to generate high-accuracy potential survivor positions;

[0262] The optimization and updating module 24 is used to dynamically adjust and optimize the search and rescue path by using the ocean current prediction algorithm and wind direction data to ensure rapid coverage of the distress range where the potential survivors are located, adaptively adjust the search strategy by using the reinforcement learning algorithm, update the best flight route in real time, and consider the remaining power and endurance of the drone to generate an economical and efficient search and rescue path planning;

[0263] The feedback guidance module 25 is used to timely feedback the precise coordinates of the potential survivors' positions to the nearest rescue coordination center based on the search and rescue path planning, and guide the emergency equipment to carry out rescue along the optimal path.

[0264] Figure 2 The intelligent marine emergency equipment automatic positioning system can perform Figure 1 The implementation principle and technical effect of the automatic positioning method for intelligent marine emergency equipment described in the illustrated embodiment will not be described in detail. The specific manner in which each module and unit performs operations in the automatic positioning system for intelligent marine emergency equipment in the above embodiment has been described in detail in the embodiment of the method, and will not be described in detail here.

[0265] In one possible design, Figure 2 The intelligent marine emergency equipment automatic positioning system of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0266] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0267] The processing component 32 is used to: receive and fuse multiple distress signals from satellites, radio relay stations and nearby ships to generate optimal geographic location information and timestamps; based on the optimal geographic location information, determine the distress area on a preset sea area electronic map, and calculate the signal propagation delay according to the timestamp to correct the distress position, and at the same time apply the Bayesian network prediction model to evaluate the probability distribution of each potential distress area to generate the most likely primary search target; according to the most likely primary search target, start the thermal imaging and optical sensing system on the rescue drone to scan, and combine the convolutional neural network to enhance the recognition of non-fixed form objects The system uses generative adversarial networks to simulate image features under different weather conditions and generate highly accurate potential survivor locations. It uses ocean current prediction algorithms and wind direction data to dynamically adjust and optimize the search and rescue path to ensure rapid coverage of the distress range where the potential survivors are located. It uses a reinforcement learning algorithm to adaptively adjust the search strategy, update the optimal flight route in real time, and consider the remaining power and endurance of the drone to generate an economical and efficient search and rescue path planning. Based on the search and rescue path planning, the precise coordinates of the potential survivors' locations are promptly fed back to the nearest rescue coordination center, and the emergency equipment is guided along the optimal path to carry out rescue.

[0268] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0269] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component 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.

[0270] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0271] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0272] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0273] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0274] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The automatic positioning method of intelligent marine emergency equipment in the illustrated embodiment.

[0275] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0276] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0277] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0278] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for automatically positioning intelligent marine emergency equipment, characterized in that: include: Receive and fuse multiple distress signals from satellites, radio relay stations and nearby ships to generate optimal geographic location information and timestamps; Based on the optimal geographical location information, the distress area is determined on the preset sea area electronic map, and the signal propagation delay is calculated according to the timestamp to correct the distress position, and the probability distribution of each potential distress area is evaluated by applying the Bayesian network prediction model to generate the primary search target with the highest probability; Based on the primary search target with the highest probability, the thermal imaging and optical sensing systems on the rescue drone are activated to scan, and the convolutional neural network is combined to enhance the recognition ability of non-fixed objects. The generative adversarial network is used to simulate the image features under different weather conditions to generate highly accurate potential survivor locations; Using ocean current prediction algorithms and wind direction data, the search and rescue path is dynamically adjusted and optimized to ensure rapid coverage of the distress area where the potential survivors are located. The search strategy is adaptively adjusted through a reinforcement learning algorithm, the optimal flight route is updated in real time, and the remaining battery power and endurance of the drone are taken into consideration to generate an economical and efficient search and rescue path plan. Based on the search and rescue path planning, the precise coordinates of the potential survivors' locations are fed back to the nearest rescue coordination center in a timely manner, and the emergency equipment is guided along the optimal path to carry out the rescue.

2. The method according to claim 1, characterized in that The distress area determined on the preset sea area electronic map based on the optimal geographical location information, and the signal propagation delay calculated according to the timestamp to correct the distress position, while applying the Bayesian network prediction model to evaluate the probability distribution of each potential distress area, and generating the primary search target with the highest probability, including: Using the optimal geographical location information, matching the initial distress location on a preset sea area electronic map to determine the distress area; Calculate the signal propagation delay according to the timestamp, and perform correction processing on the initial distress position to obtain a corrected distress position; Based on the corrected distress location, applying a Bayesian network prediction model to evaluate the probability distribution of each distress area to generate an occurrence probability value for each distress area; According to the generated occurrence probability value of each distress area, the distress area with the highest probability is selected, and the primary search target with the highest probability is generated to ensure that resources are preferentially invested in the distress area with the highest probability.

3. The method according to claim 2, characterized in that Based on the corrected distress location, applying the Bayesian network prediction model to evaluate the probability distribution of each distress area to generate an occurrence probability value of each distress area includes: The corrected distress location and surrounding environment parameters are used as input variables into the pre-trained Bayesian network prediction model to obtain basic data; Based on the basic data and the conditional probability distribution learned from the data set of historical distress events, the possibility of each preset distress area becoming an actual distress location is evaluated and processed, and a preliminary occurrence probability value is output; Based on the preliminary occurrence probability value, further adjusting and optimizing the occurrence probability value to ensure accuracy, and obtaining an optimized occurrence probability value; According to the optimized occurrence probability value, the occurrence probability value of each distress area is generated to ensure that resources are preferentially invested in the distress area with the highest probability.

4. The method according to claim 1, characterized in that: According to the primary search target with the highest probability, the thermal imaging and optical sensing systems on the rescue drone are activated for scanning, and the convolutional neural network is combined to enhance the recognition ability of non-fixed objects, and the generative adversarial network is used to simulate the image features under different weather conditions to generate high-accuracy potential survivor locations, including: Based on the primary search target with the highest probability, the thermal imaging and optical sensing systems on the rescue drone are activated to scan and obtain preliminary image data; Using the preliminary image data in combination with a convolutional neural network, the non-fixed form objects in the preliminary image data are identified and processed to obtain preliminary object recognition results; Based on the preliminary object recognition results, applying a generative adversarial network to simulate image features under different weather conditions to generate optimized image features; Based on the optimized image features, the precise location of the potential survivors is further analyzed and confirmed to generate a highly accurate potential survivor location.

5. The method according to claim 4, characterized in that The method of using the preliminary image data in combination with a convolutional neural network to perform recognition processing on non-fixed form objects in the preliminary image data to obtain preliminary object recognition results includes: Using the preliminary image data as input to a pre-trained convolutional neural network, performing feature extraction processing on the preliminary image data to generate detailed feature information; Based on the detailed feature information, further refining the feature information through multi-layer convolution and pooling operations of a convolutional neural network, and obtaining an optimized feature representation according to the visual characteristics of shape, texture and color; The optimized feature representation is classified through a classification layer of a convolutional neural network to accurately classify irregular or floating objects and generate classification results; A preliminary object recognition result is generated according to the classification result, and the preliminary object recognition result includes preliminary position information of potential survivors.

6. The method according to claim 1, characterized in that The ocean current prediction algorithm and wind direction data are used to dynamically adjust and optimize the search and rescue path to ensure rapid coverage of the distress range where the potential survivors are located. The search strategy is adaptively adjusted through the reinforcement learning algorithm, the optimal flight route is updated in real time, and the remaining power and endurance of the drone are taken into consideration to generate an economical and efficient search and rescue path planning, including: Using the obtained high-accuracy potential survivor positions, combined with the real-time acquired ocean current prediction algorithm and wind direction data, the search and rescue path is preliminarily planned to obtain a preliminary search and rescue path; Based on the preliminary search and rescue path, according to the ocean current prediction algorithm and wind direction data, the search and rescue path is dynamically adjusted and optimized to adapt to the changing ocean environment, ensure the effectiveness and timeliness of the path, and generate an optimized search and rescue path; Adaptively adjust the optimized search and rescue path through a reinforcement learning algorithm, update the search strategy in real time according to the situations encountered during the actual search and rescue process, and generate a search and rescue path that is updated in real time; Taking into account the remaining power and endurance of the drone, the search and rescue path updated in real time is further processed to ensure that the drone can return safely after completing the mission while avoiding unnecessary energy waste, thereby generating a final flight route; Based on the highly accurate location of potential survivors, real-time ocean current prediction algorithm and wind direction data, dynamically adjusted and optimized search and rescue paths, adaptively adjusted search strategies, and the remaining battery power and endurance of the drone, the final flight route is evaluated and optimized to generate an economical and efficient search and rescue path plan.

7. The method according to claim 6, characterized in that The optimized search and rescue path is adaptively adjusted by the reinforcement learning algorithm, and the search strategy is updated in real time according to the situation encountered in the actual search and rescue process to generate a real-time updated search and rescue path, including: Using the optimized search and rescue path, combined with the actual search and rescue environment data obtained in real time, to provide a training basis for the reinforcement learning algorithm, and obtain initial training data; Based on the initial training data, the state of the current search and rescue environment is represented by the state space in the reinforcement learning algorithm, and the position and state of the drone are taken as part of the environment, so as to ensure that the reinforcement learning algorithm can fully understand the current situation and generate an environmental state representation; Generate an adjustment plan based on the environmental state representation and through adjustment operations defined in the action space of the reinforcement learning algorithm, wherein the adjustment operations include changing the flight direction, adjusting the flight altitude, accelerating or decelerating; Using the adjustment scheme, the effect of each adjustment operation is evaluated through a reward mechanism in the reinforcement learning algorithm, wherein the reward mechanism takes into account factors such as search and rescue efficiency, safety, and resource consumption, and generates an evaluation result; Based on the evaluation results, the reinforcement learning algorithm is applied to perform iterative learning, and the search strategy is continuously optimized according to the results of each adjustment operation, so that the reinforcement learning algorithm can gradually find the optimal path adjustment plan in the actual search and rescue process and generate an optimized search strategy; According to the optimized search strategy, the search and rescue path is updated in real time, and a real-time updated search and rescue path is generated to ensure that the UAV can dynamically adjust the flight path according to the latest environmental information, thereby improving the search and rescue efficiency.

8. An intelligent automatic positioning system for marine emergency equipment, characterized in that: include: The receiving and fusion module is used to receive and fuse multiple distress signals from satellites, radio relay stations and nearby ships, and estimate the optimal geographic location information and timestamp; A determination evaluation module is used to determine the distress area on a preset sea area electronic map based on the optimal geographical location information, calculate the signal propagation delay according to the timestamp to correct the distress position, and apply the Bayesian network prediction model to evaluate the probability distribution of each potential distress area to generate the primary search target with the highest probability; A scanning generation module is used to start the thermal imaging and optical sensing systems on the rescue drone to scan according to the primary search target with the highest probability, combine convolutional neural networks to enhance the recognition ability of non-fixed form objects, and use generative adversarial networks to simulate image features under different weather conditions to generate high-accuracy potential survivor locations; An optimization and update module is used to dynamically adjust and optimize the search and rescue path using the ocean current prediction algorithm and wind direction data to ensure rapid coverage of the distress range where the potential survivors are located, adaptively adjust the search strategy through the reinforcement learning algorithm, update the best flight route in real time, and consider the remaining power and endurance of the drone to generate an economical and efficient search and rescue path plan; The feedback guidance module is used to timely feedback the precise coordinates of the potential survivors' locations to the nearest rescue coordination center based on the search and rescue path planning, and guide the emergency equipment to carry out rescue along the optimal path.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an automatic positioning method for intelligent marine emergency equipment as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, an automatic positioning method for intelligent marine emergency equipment as described in any one of claims 1 to 7 is implemented.

Images