Maritime search and rescue scheme generation method and device

Abstract

The invention discloses a maritime search and rescue scheme generation method and device. The method comprises the following steps: acquiring multi-modal data information and a historical case data set; processing the multi-modal data information to obtain a multi-modal vector; processing the historical case data set to obtain a historical case vector set; the historical case vector set comprises a first case vector set, a second case vector set and a third case vector set; and carrying out fusion processing on the multi-modal vector and the historical case vector set to obtain maritime search and rescue scheme result information. Visibly, according to the embodiment, by processing the dynamic first-line rescue information and the historical rescue information on the sea, a more accurate and optimized search and rescue scheme can be obtained, the efficiency and the success rate of the maritime search and rescue action can be improved, then maritime rescue is supported quickly and accurately, the rescue time is shortened, and the risk is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of maritime search and rescue data processing, and in particular to a method and device for generating a maritime search and rescue plan. Background Art

[0002] In recent years, with the rapid development of artificial intelligence (AI), particularly deep learning, breakthroughs have been made. Large language models (LLMs) based on the Transformer architecture have made significant breakthroughs in natural language processing (NLP). Their powerful language understanding and generation capabilities are changing the way humans interact with machines at an unprecedented rate, gaining widespread application across various fields and becoming a current research hotspot. While these models have demonstrated impressive performance in general domains, enabling them to accomplish tasks ranging from simple text classification and sentiment analysis to complex question-answering systems and text creation, they still face numerous challenges when applied to highly specialized fields such as maritime search and rescue.

[0003] In current maritime search and rescue plan development, operators' personal experience plays a significant role. However, differences in experience and knowledge among different operators can lead to biased assessments of the same maritime search and rescue incident, thus impacting the consistency and accuracy of plan development. While maritime search and rescue is a niche field, the underlying expertise is vast and constantly evolving, encompassing a wide range of fundamental theories and practical experience, encompassing fields like oceanography, meteorology, medicine, mechanical engineering, and even cutting-edge complex knowledge like artificial intelligence. Faced with such a vast amount of maritime search and rescue information, it's difficult for average operators to fully grasp it, and they may even miss something. Furthermore, the maritime search and rescue field is rife with specialized terminology and abbreviations, posing a significant challenge to the understanding capabilities of large language models. Currently, some international maritime search and rescue agencies have introduced intelligent search and rescue plan generation systems based on rule engines or knowledge graphs, but these systems have limitations. These include poor scalability, high knowledge base maintenance costs, slow updates, opaque reasoning chains, and poor interpretability, leading to inability, fear, and inability to use them. In recent years, with the rapid development of large language models (LLMs), more and more organizations have attempted to apply LLMs to intelligent solution generation. However, pure LLM models have shortcomings in generating maritime search and rescue solutions, such as a lack of a verification mechanism for search and rescue-related professional knowledge, a broken reasoning chain, and insufficient support for multimodal data. Finally, maritime search and rescue decisions require logical reasoning and judgment based on complex on-site information, including the location of the drowning point, the weather on the day, the on-site ocean currents, the physical condition of the drowning person, and real-time on-site images. However, existing large language models still have deficiencies in logical reasoning and contextual understanding, making it difficult for them to comprehensively consider multiple factors and make accurate solution recommendations like professional maritime search and rescue plan developers. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and device for generating a maritime search and rescue plan. By processing dynamic front-line rescue information and historical rescue information at sea, a more accurate and optimized search and rescue plan can be obtained, which is conducive to improving the efficiency and success rate of maritime search and rescue operations, and then quickly and accurately supporting maritime rescue, reducing rescue time and reducing risks.

[0005] In order to solve the above technical problems, a first aspect of an embodiment of the present invention discloses a method for generating a maritime search and rescue plan, the method comprising:

[0006] S1, obtaining multimodal data information and a historical case data set; the multimodal data information includes text information, audio information and video information; the historical case data set includes a historical search and rescue scene data set, a historical search and rescue equipment data set and a historical search and rescue plan data set; the historical search and rescue scene data set includes a historical search and rescue audio and video data set;

[0007] S2, processing the multimodal data information to obtain a multimodal vector;

[0008] S3, processing the historical case data set to obtain a historical case vector set; the historical case vector set includes a first case vector set, a second case vector set, and a third case vector set;

[0009] S4, fusing the multimodal vector and the historical case vector set to obtain maritime search and rescue solution result information.

[0010] As an optional implementation manner, in the first aspect of the embodiments of the present invention, processing the multimodal data information to obtain a multimodal vector includes:

[0011] S21, performing data cleaning processing on the multimodal data information to obtain first multimodal data information;

[0012] S22, performing denoising processing on the first multimodal data information to obtain second multimodal data information;

[0013] S23, performing deduplication processing on the second multimodal data information to obtain third multimodal data information;

[0014] S24, performing normalization processing on the third multimodal data information to obtain fourth multimodal data information; the fourth multimodal data information includes fourth text information, fourth audio information, and fourth video information;

[0015] S25: Process the fourth multimodal data information to obtain a multimodal vector.

[0016] As an optional implementation manner, in the first aspect of the embodiments of the present invention, processing the fourth multimodal data information to obtain a multimodal vector includes:

[0017] S251, performing named entity recognition processing on the fourth text information to obtain named entity information;

[0018] S252, performing audio recognition processing on the fourth audio information to obtain audio semantic information;

[0019] S253, performing video recognition processing on the fourth video information to obtain video semantic information;

[0020] S254: Fusing the named entity information, the audio semantic information, and the video semantic information to obtain a multimodal vector.

[0021] As an optional implementation manner, in the first aspect of the embodiment of the present invention, processing the historical case dataset to obtain the historical case vector set includes:

[0022] S31, processing the historical case data set to obtain maritime search and rescue knowledge graph information;

[0023] S32: Process the historical case data set and the maritime search and rescue knowledge graph information to obtain a historical case vector set.

[0024] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the fusing of the multimodal vector and the historical case vector set to obtain the maritime search and rescue solution result information includes:

[0025] S41, processing the multimodal vector and the first case vector set to obtain a preliminary search and rescue solution vector;

[0026] S42, processing the preliminary search and rescue solution vector and the second case vector set to obtain an intermediate search and rescue solution vector;

[0027] S43: Process the intermediate search and rescue solution vector and the third case vector set to obtain maritime search and rescue solution result information.

[0028] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the processing of the multimodal vector and the first case vector set to obtain a preliminary search and rescue solution vector includes:

[0029] S411, using a first maritime search and rescue calculation model, performing calculation processing on the multimodal vector and the first case vector set to obtain first similarity information;

[0030] Wherein, the first maritime search and rescue calculation model is:

[0031]

[0032] Where, SY is the first similarity information, SY i is the i-th first similarity value in the first similarity information, MT is the multimodal vector, DY i is the first case vector in the first case vector set, DY k is the kth first case vector in the first case vector set, N is the number of the first case vectors in the first case vector set, |·| is the modulus of the orientation quantity, and δ1 is the first weight parameter;

[0033] S412, preset s=1;

[0034] S413, determining whether the sth first similarity value in the first similarity information is greater than the preset similarity threshold, and obtaining a first determination result;

[0035] When the first judgment result is yes, the sth first case vector in the first case vector set is added to the preliminary solution vector set, and S414 is executed;

[0036] When the first judgment result is no, executing S414;

[0037] S414, determining whether s is greater than the number of the first case vectors in the first case vector set, and obtaining a second determination result;

[0038] When the second judgment result is no, increment s by 1 and execute S412;

[0039] When the second judgment result is yes, execute S415;

[0040] S415, using a second maritime search and rescue calculation model, performing calculation processing on the multimodal vector and the preliminary solution vector set to obtain preliminary solution weight information;

[0041] Among them, the second maritime search and rescue calculation model is:

[0042]

[0043]

[0044] Where, QZY is the weight information of the preliminary plan, QZY i1 is the weight value of the i1th preliminary solution in the preliminary solution weight information, CB i1 is the i1th first case vector in the preliminary solution vector set, N1 is the number of the first case vectors in the preliminary solution vector set, δ2 and δ3 are the second weight parameter and the third weight parameter respectively;

[0045] S416, fusing the preliminary solution vector set and the preliminary solution weight information to obtain a preliminary solution vector;

[0046] S417: Process the multimodal vector and the preliminary solution vector to obtain a preliminary search and rescue solution vector.

[0047] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the processing the preliminary search and rescue solution vector and the second case vector set to obtain an intermediate search and rescue solution vector includes:

[0048] S421, using a fourth maritime search and rescue calculation model, performing calculation processing on the preliminary search and rescue solution vector and the second case vector set to obtain a similarity vector set;

[0049] Wherein, the fourth maritime search and rescue calculation model is:

[0050]

[0051] Where, SSE is the similarity vector set, SSE i2 is the i2th similarity vector in the similarity vector set, CBS is the preliminary search and rescue solution vector, DE i2 is the i2-th second case vector in the second case vector set, N2 represents the number of the second case vectors in the second case vector set, ‖·‖2 represents the L2 norm, <·> represents the inner product of two vectors, and θ represents the equipment weight factor;

[0052] S422, performing calculation processing on the preliminary search and rescue solution vector and the similarity vector set to obtain an intermediate fusion vector set;

[0053] S423, performing splicing processing on the intermediate fusion vector set to obtain an intermediate solution fusion vector;

[0054] S424: Process the intermediate solution fusion vector to obtain an intermediate search and rescue solution vector.

[0055] A second aspect of an embodiment of the present invention discloses a device for generating a maritime search and rescue plan, the device comprising:

[0056] An acquisition module is configured to acquire multimodal data information and a historical case data set; the multimodal data information includes text information, audio information, and video information; the historical case data set includes a historical search and rescue scene data set, a historical search and rescue equipment data set, and a historical search and rescue plan data set; the historical search and rescue scene data set includes a historical search and rescue audio and video data set;

[0057] a first computing module, configured to process the multimodal data information to obtain a multimodal vector;

[0058] A second computing module is configured to process the historical case data set to obtain a historical case vector set; the historical case vector set includes a first case vector set, a second case vector set, and a third case vector set;

[0059] The third calculation module is used to fuse the multimodal vector and the historical case vector set to obtain maritime search and rescue plan result information.

[0060] A third aspect of an embodiment of the present invention discloses another device for generating a maritime search and rescue plan, the device comprising:

[0061] processor;

[0062] a memory coupled to the processor and storing executable program code;

[0063] The processor calls the executable program code stored in the memory to execute part or all of the steps of the method for generating a maritime search and rescue plan disclosed in the first aspect of the embodiment of the present invention.

[0064] The fourth aspect of an embodiment of the present invention discloses a computer-readable storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute part or all of the steps of the maritime search and rescue plan generation method disclosed in the first aspect of the embodiment of the present invention.

[0065] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0066] In an embodiment of the present invention, multimodal data information and a historical case data set are obtained; the multimodal data information includes text information, audio information, and video information; the historical case data set includes a historical search and rescue scene data set, a historical search and rescue equipment data set, and a historical search and rescue plan data set; the historical search and rescue scene data set includes a historical search and rescue audio and video data set; the multimodal data information is processed to obtain a multimodal vector; the historical case data set is processed to obtain a historical case vector set; the historical case vector set includes a first case vector set, a second case vector set, and a third case vector set; the multimodal vector and the historical case vector set are fused to obtain maritime search and rescue plan result information. It can be seen that this embodiment can obtain a more accurate and optimized search and rescue plan by processing the dynamic front-line rescue information and historical rescue information at sea, which is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0068] Figure 1 A schematic flow chart of a method for generating a maritime search and rescue plan disclosed in an embodiment of the present invention;

[0069] Figure 2 This is a schematic structural diagram of a device for generating a maritime search and rescue plan disclosed in an embodiment of the present invention;

[0070] Figure 3 This is a structural diagram of another device for generating a maritime search and rescue plan disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0071] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0072] The terms "first," "second," and so on, in the description and claims of the present invention and the accompanying drawings are used to distinguish between different objects, not to describe a specific order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or device comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or device.

[0073] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0074] The present invention discloses a method and device for generating a maritime search and rescue plan. By processing dynamic frontline rescue information and historical rescue information at sea, a more accurate and optimized search and rescue plan can be obtained, which is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time and mitigating risks. A detailed description of each method is provided below.

[0075] Example 1

[0076] See also Figure 1 , Figure 1 This is a flow chart of a method for generating a maritime search and rescue plan disclosed in an embodiment of the present invention. Figure 1 The method for generating a maritime search and rescue plan is applied to a maritime search and rescue plan generating device, such as a local server or cloud server for optimizing and managing the generation of a maritime search and rescue plan, and is not limited in the embodiment of the present invention. Figure 1As shown, the method for generating a maritime search and rescue plan may include the following operations:

[0077] S1, obtaining multimodal data information and a historical case data set; the multimodal data information includes text information, audio information and video information; the historical case data set includes a historical search and rescue scene data set, a historical search and rescue equipment data set and a historical search and rescue plan data set; the historical search and rescue scene data set includes a historical search and rescue audio and video data set;

[0078] It should be noted that the multimodal data information is obtained by real-time collection of rescue data at the current offshore scene through wireless, wired, satellite signals, sensors, etc. For example, text information is obtained through the search and rescue logs and on-site notifications at the offshore search and rescue scene, through the offshore text message system, emergency platform input and other input systems, and is obtained locally by logging into the system. The audio information is the microphone of the on-site unmanned boat / rescue equipment, telephone distress audio, and radio communication records. The video information is obtained by shooting through unmanned aerial vehicles (UAVs), satellite remote sensing, shore-based cameras, ship-borne monitoring, etc. Specifically, the present invention does not limit this.

[0079] It should be noted that the text information also includes marine environmental data (ocean currents, waves, sea surface temperature, salinity, sea ice, seawater density, rainfall, etc.) at the search and rescue site, underwater terrain data (underwater terrain data of islands and reefs, ship-borne sonar data, airborne laser detection point cloud, etc.), and basic geographic information (OSM, coastline, GRanD, GOOD2, etc.); audio data includes on-site distress audio, radio communication records, voiceprint recognition data of ships or rescue equipment, and environmental background sounds (wind, explosion, water sounds), etc.; video information includes images taken by drones at the search and rescue site, satellite remote sensing image sequences, ship-borne monitoring videos, and target tracking images, etc., and the specific details are not limited in the embodiments of the present invention.

[0080] It should be noted that the historical case data set can be obtained through domestic official databases such as the China Maritime Search and Rescue Center data statistics, local maritime bureau statistics, etc., and can also be obtained through the SeaDronesSee data set, SAR-Ship-Dataset or SSDD. Specifically, the embodiments of the present invention do not limit this.

[0081] It should be noted that the historical search and rescue scene data set covers scene data information of previous maritime search and rescue, including the time, location, marine environment (such as ocean currents, waves, sea surface temperature, salinity, rainfall, wind speed), meteorological information, basic geographic data (such as OSM, coastline, islands and reefs, nautical charts, etc.), underwater topography (such as ship-borne sonar, airborne laser point cloud), and on-site audio and video records (such as distress audio, drone aerial photography, SAR images, etc.); the historical search and rescue equipment data set records the types, performance parameters, usage and deployment time and space of ships, aircraft, unmanned systems and other equipment used in previous maritime search and rescue missions; the historical search and rescue plan data set contains the command structure, resource deployment, communication links, mission timeline, path planning and mission evaluation information of each search and rescue mission in previous maritime search and rescue missions. Specifically, the embodiments of the present invention are not limited.

[0082] S2, processing the multimodal data information to obtain a multimodal vector;

[0083] S3, processing the historical case data set to obtain a historical case vector set; the historical case vector set includes a first case vector set, a second case vector set, and a third case vector set;

[0084] S4, fusing the multimodal vector and the historical case vector set to obtain maritime search and rescue solution result information.

[0085] It can be seen that implementing the method for generating a maritime search and rescue plan described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0086] In an optional embodiment, the processing the multimodal data information to obtain a multimodal vector includes:

[0087] S21, performing data cleaning processing on the multimodal data information to obtain first multimodal data information;

[0088] S22, performing denoising processing on the first multimodal data information to obtain second multimodal data information;

[0089] S23, performing deduplication processing on the second multimodal data information to obtain third multimodal data information;

[0090] It should be noted that for the above-mentioned data cleaning, denoising and deduplication processing, for text information, stop word removal, spelling correction, TF-IDF and cosine similarity can be used for processing; for audio information, background noise suppression, Wiener filtering, wavelet transform, audio fingerprint, etc. can be used; for video information, frame difference, video fingerprint, deep denoising network and other methods can be used for processing. Since data cleaning, denoising and deduplication processing are all conventional technical means, the specific embodiments of the present invention do not limit them.

[0091] S24, performing normalization processing on the third multimodal data information to obtain fourth multimodal data information; the fourth multimodal data information includes fourth text information, fourth audio information, and fourth video information;

[0092] It should be noted that the above-mentioned normalization process converts data of different scales into a unified scale for effective comparison and further processing. Specifically, for text information, TF-IDF normalization is used, and for audio information and video information, Z-score normalization is used. The specific embodiment of the present invention does not limit this.

[0093] S25: Process the fourth multimodal data information to obtain a multimodal vector.

[0094] It can be seen that implementing the method for generating a maritime search and rescue plan described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0095] In another optional embodiment, the processing the fourth multimodal data information to obtain a multimodal vector includes:

[0096] S251, performing named entity recognition processing on the fourth text information to obtain named entity information;

[0097] It should be noted that the above-mentioned named entity recognition processing can be processed using a pre-trained BERT model, and the specific embodiment of the present invention does not limit it.

[0098] For example, the following method can be used:

[0099]

[0100] Among them, x text Indicates the fourth text information entered, E t The text data is passed through the BERT model to obtain a 768-dimensional high-dimensional feature representation of named entity information. [CLS] is a special token used to represent the semantic representation of the entire input sequence.

[0101] S252, performing audio recognition processing on the fourth audio information to obtain audio semantic information;

[0102] It should be noted that the above-mentioned audio recognition processing can be processed using the Whisper speech model, and the specific embodiment of the present invention does not limit it.

[0103] For example, the following method can be used:

[0104]

[0105] Among them, x audio Indicates the fourth audio information input, E a It means that the speech data passes through the Whisper speech model to obtain a 1280-dimensional high-dimensional feature representation of audio semantic information.

[0106] S253, performing video recognition processing on the fourth video information to obtain video semantic information;

[0107] It should be noted that the above-mentioned audio recognition processing can be processed using a pre-trained visual Transformer model (ViT), and the specific embodiment of the present invention does not limit this.

[0108] For example, the following method can be used:

[0109]

[0110] Among them, x image Indicates the fourth video information input, E i It means that the image data passes through the ViT model to obtain a 2048-dimensional high-dimensional feature representation of video semantic information.

[0111] S254: Fusing the named entity information, the audio semantic information, and the video semantic information to obtain a multimodal vector.

[0112] It can be seen that implementing the method for generating a maritime search and rescue plan described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0113] In yet another optional embodiment, the fusing the named entity information, the audio semantic information, and the video semantic information to obtain a multimodal vector includes:

[0114] S2541, performing embedding coding processing on the named entity information to obtain a text semantic vector;

[0115] It should be noted that the above-mentioned embedded coding process can be processed using pre-trained language models such as BERT and RoBERTa, and the specific embodiment of the present invention does not limit this.

[0116] Through embedded coding processing, the named entity information in the text can be converted into a fixed-dimensional semantic vector representation, so that the text information can be mapped into a unified semantic space.

[0117] S2542, processing the audio semantic information to obtain an audio semantic vector;

[0118] It should be noted that the above-mentioned audio semantic information processing can be performed using audio pre-training models such as Wav2Vec and HuBERT, and the specific embodiment of the present invention does not limit this.

[0119] Through audio semantic information processing, the audio signal can be converted into a fixed-dimensional semantic vector representation, so that the audio information can be mapped into a unified semantic space.

[0120] S2543, processing the video semantic information to obtain a visual semantic vector;

[0121] It should be noted that the above-mentioned video semantic information processing can be performed using a visual pre-training model such as ViT or ResNet, and the specific embodiment of the present invention does not limit this.

[0122] Through video semantic information processing, video frames can be converted into fixed-dimensional semantic vector representations, so that visual information can be mapped into a unified semantic space.

[0123] S2544: performing cross-modal alignment processing on the text semantic vector, the audio semantic vector, and the visual semantic vector to obtain preliminary fused semantic information;

[0124] It should be noted that the above cross-modal alignment processing can be performed using cross-modal alignment models such as CLIP and ALIGN, and the specific embodiment of the present invention does not limit this.

[0125] Through cross-modal alignment processing, the semantic vectors of different modalities can be aligned to the same semantic space, making the information of different modalities comparable in the semantic space.

[0126] S2545, performing multimodal semantic construction processing on the preliminary fused semantic information to obtain fused semantic information;

[0127] It should be noted that the above-mentioned multimodal semantic construction process can be processed using models such as Transformer and Cross-modal Transformer, and the specific embodiments of the present invention do not limit this.

[0128] Through multimodal semantic construction processing, semantic information of different modalities can be integrated to construct a unified semantic representation, thereby obtaining a unified semantic representation containing multimodal information.

[0129] S2546: Process the fused semantic information to obtain a multimodal vector.

[0130] It should be noted that the above-mentioned fusion semantic information processing can be performed using models such as MLP and Transformer, and the specific embodiment of the present invention is not limited thereto. Through the fusion semantic information processing, the fused semantic information can be converted into a final multimodal vector, thereby obtaining a unified multimodal feature representation that can be used in subsequent steps.

[0131] It can be seen that implementing the method for generating a maritime search and rescue plan described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0132] In an optional embodiment, the processing of the historical case data set to obtain the historical case vector set includes:

[0133] S31, processing the historical case data set to obtain maritime search and rescue knowledge graph information;

[0134] The above processing, by constructing a maritime search and rescue knowledge graph, can structure the key information in historical cases, establish the relationship between different cases, and form a complete knowledge network, which is convenient for subsequent knowledge reasoning and case retrieval, helps to discover the laws and patterns in historical cases, and improve the decision-making efficiency of maritime search and rescue.

[0135] The maritime search and rescue knowledge graph information is a structured knowledge network, in which nodes represent entities in the case (such as accident type, search and rescue method, rescue equipment, etc.), and edges represent the relationships between entities (such as "use", "lead to", "applicable to", etc.). It also includes the attributes of the entities (such as time, location, weather conditions, etc.) and the weights of the relationships, forming a complete maritime search and rescue knowledge system.

[0136] S32: Process the historical case data set and the maritime search and rescue knowledge graph information to obtain a historical case vector set.

[0137] It can be seen that implementing the method for generating a maritime search and rescue plan described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0138] In an optional embodiment, the processing of the historical case data set to obtain maritime search and rescue knowledge graph information includes:

[0139] S311, processing the historical search and rescue scene data set to obtain a historical search and rescue scene entity set;

[0140] It should be noted that the above-mentioned historical search and rescue scene data processing can use BERT-CRF, BiLSTM-CRF, SpanBERT and other named entity recognition models, or use a rule-based regular expression matching method for processing. Specifically, the embodiment of the present invention does not limit this.

[0141] By processing historical search and rescue scene data, entity information related to the search and rescue scene can be extracted from historical cases, such as accident type, weather conditions, geographical location, etc., thereby forming a structured scene entity set.

[0142] S312, processing the historical search and rescue equipment data set to obtain a historical search and rescue equipment entity set;

[0143] It should be noted that the above-mentioned historical search and rescue equipment data processing can use entity recognition models such as BERT-NER, RoBERTa-CRF, or use dictionary-based equipment classification methods, such as TF-IDF, Word2Vec and other word vector models for processing. Specifically, the embodiments of the present invention do not limit this.

[0144] By processing historical search and rescue equipment data, entity information related to search and rescue equipment can be extracted from historical cases, such as rescue ships, search and rescue equipment, communication equipment, etc., thus forming a structured equipment entity set.

[0145] S313, processing the historical search and rescue plan data set to obtain a historical search and rescue plan entity set;

[0146] It should be noted that the above-mentioned historical search and rescue plan data processing can use sequence labeling models such as BERT-Span, BiLSTM-Attention, or use template-based solution parsing methods such as regular expressions, rule engines, etc. for processing. Specifically, the embodiments of the present invention do not limit this.

[0147] By processing historical search and rescue plan data, entity information related to the search and rescue plan can be extracted from historical cases, such as rescue steps, personnel configuration, time schedule, etc., thereby forming a structured plan entity set.

[0148] S314: Perform relationship extraction on the historical search and rescue scene entity set, the historical search and rescue equipment entity set, and the historical search and rescue plan entity set to obtain maritime search and rescue knowledge graph information.

[0149] It should be noted that the above-mentioned relationship extraction processing can use relationship extraction models such as BERT-RE, PCNN, R-BERT, or use rule-based relationship matching methods such as dependency syntactic analysis, semantic role labeling, etc. for processing. Specifically, the embodiments of the present invention do not limit this.

[0150] Through relationship extraction processing, it is possible to establish association relationships between different entities, such as the "scenario-equipment" relationship, the "scenario-solution" relationship, the "equipment-solution" relationship, etc., thereby constructing a complete maritime search and rescue knowledge graph.

[0151] It can be seen that implementing the method for generating a maritime search and rescue plan described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0152] In an optional embodiment, the processing of the historical case dataset and the maritime search and rescue knowledge graph information to obtain a historical case vector set includes:

[0153] S321, processing the historical search and rescue scene dataset, the historical search and rescue plan dataset, and the maritime search and rescue knowledge graph information to obtain a first case vector set;

[0154] It should be noted that the above processing can use pre-trained language models such as BERT and RoBERTa for text encoding, and combine graph neural networks such as GCN and GAT to process knowledge graph information. Specifically, the embodiments of the present invention do not limit this.

[0155] Through this processing, the scene data, solution data and knowledge graph information can be fused and converted into a unified vector representation, thereby forming a first case vector set containing scene and solution features.

[0156] S322, obtaining a historical equipment usage information set;

[0157] It should be noted that the above-mentioned historical equipment usage information set can be obtained in the following ways: extracting equipment usage logs from historical search and rescue records; obtaining usage frequency and duration from equipment maintenance records; obtaining equipment usage effects from search and rescue effect evaluation reports; and obtaining standard usage procedures from equipment operation manuals. Specifically, the embodiments of the present invention do not limit this.

[0158] It should be noted that the historical equipment usage information set includes equipment usage time, usage scenarios, usage effects, operators, maintenance records, etc., which are used for subsequent analysis of equipment usage patterns and effects.

[0159] S323, processing the historical equipment usage information set and the historical search and rescue equipment data set to obtain a second case vector set;

[0160] It should be noted that the above processing can use word vector models such as Word2Vec and FastText to encode equipment descriptions, and combine sequence models such as LSTM and Transformer to process usage records. The specific details are not limited in the embodiments of the present invention.

[0161] Through this processing, the equipment information and usage records can be converted into a unified vector representation, thereby forming a second case vector set containing equipment features and usage pattern features.

[0162] S324: Process the historical search and rescue audio and video dataset to obtain a third case vector set.

[0163] It should be noted that the above processing can use visual models such as ResNet and ViT to process video frames, and use audio models such as Wav2Vec and HuBERT to process audio data. Specifically, the embodiments of the present invention do not limit this.

[0164] Through this processing, the audio and video data can be converted into a unified vector representation, thereby forming a third case vector set containing visual and auditory features.

[0165] It can be seen that implementing the method for generating a maritime search and rescue plan described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0166] In an optional embodiment, the fusing of the multimodal vector and the historical case vector set to obtain the maritime search and rescue solution result information includes:

[0167] S41, processing the multimodal vector and the first case vector set to obtain a preliminary search and rescue solution vector;

[0168] S42, processing the preliminary search and rescue solution vector and the second case vector set to obtain an intermediate search and rescue solution vector;

[0169] S43: Process the intermediate search and rescue solution vector and the third case vector set to obtain maritime search and rescue solution result information.

[0170] It can be seen that implementing the method for generating a maritime search and rescue plan described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0171] In an optional embodiment, the processing of the multimodal vector and the first case vector set to obtain a preliminary search and rescue solution vector includes:

[0172] S411, using a first maritime search and rescue calculation model, performing calculation processing on the multimodal vector and the first case vector set to obtain first similarity information;

[0173] Wherein, the first maritime search and rescue calculation model is:

[0174]

[0175] Where, SY is the first similarity information, SY i is the i-th first similarity value in the first similarity information, MT is the multimodal vector, DY i is the first case vector in the first case vector set, DY k is the kth first case vector in the first case vector set, N is the number of the first case vectors in the first case vector set, |·| is the modulus of the orientation quantity, and δ1 is the first weight parameter;

[0176] It should be noted that the first maritime search and rescue calculation model calculates the distance between the multimodal vector and the historical case vector, combined with sin(δ1·|MT-DY i |) item captures the periodic characteristics of the marine environment (such as tides, weather changes, etc.), using As a weight, it ensures that historical cases that are more similar to the current situation receive higher weights. At the same time, the normalization of the denominator ensures that the similarity calculation is not affected by the vector dimension. It can accurately process multimodal vectors and multimodal features in the first case vector set, and can accurately process maritime search and rescue scenarios. The first maritime search and rescue calculation model can simultaneously consider the similarity of multiple dimensions such as geographical location, weather conditions, accident type, and search and rescue difficulty. The strategy of similarity calculation is dynamically adjusted through the δ1 parameter to adapt to the needs of different search and rescue scenarios. The calculation process is stable and reliable, the results are highly interpretable, and are convenient for practical application and adjustment, providing a more reliable reference for the subsequent generation of search and rescue plans.

[0177] It should be noted that the first weight parameter may be set by a user or obtained based on historical data, and the embodiment of the present invention does not limit this.

[0178] It should be noted that the value range of the first weight parameter is [0.1, 2]. When the periodicity feature needs to be emphasized, a larger value (1.5-2.0) is taken, when the periodicity feature needs to be weakened, a smaller value (0.1-0.5) is taken, and when the periodicity and distance features need to be balanced, an intermediate value (0.5-1.5) is taken. This parameter is adjusted by adjusting sin(δ1·|MT-DY i |) affects the sensitivity of the similarity calculation: when the first weight parameter is large, the period becomes shorter, making it more sensitive to similarity changes and suitable for search and rescue scenarios requiring precise matching; when the first weight parameter is small, the period becomes longer, making it less sensitive to similarity changes and suitable for search and rescue scenarios requiring fuzzy matching. This design allows the model to flexibly adjust the similarity calculation strategy based on the characteristics of different sea areas (for example, using a larger first weight parameter in areas with significant tidal fluctuations and a smaller first weight parameter in areas with frequent weather changes), balance the importance of different features, and improve the accuracy and stability of the similarity calculation, thereby enhancing the model's adaptability and flexibility and providing a more reliable reference for generating search and rescue plans.

[0179] S412, preset s=1;

[0180] S413, determining whether the sth first similarity value in the first similarity information is greater than the preset similarity threshold, and obtaining a first determination result;

[0181] It should be noted that the preset similarity threshold value range is between [0.6, 0.9]. When strict screening is required, a larger value (such as 0.8-0.9) is taken. When loose screening is required, a smaller value (such as 0.6-0.7) is taken. When a balanced screening strictness is required, an intermediate value (such as 0.7-0.8) is taken.

[0182] When the first judgment result is yes, the sth first case vector in the first case vector set is added to the preliminary solution vector set, and S414 is executed;

[0183] When the first judgment result is no, executing S414;

[0184] S414, determining whether s is greater than the number of the first case vectors in the first case vector set, and obtaining a second determination result;

[0185] When the second judgment result is no, increment s by 1 and execute S412;

[0186] When the second judgment result is yes, execute S415;

[0187] S415, using a second maritime search and rescue calculation model, performing calculation processing on the multimodal vector and the preliminary solution vector set to obtain preliminary solution weight information;

[0188] Among them, the second maritime search and rescue calculation model is:

[0189]

[0190] Where, QZY is the weight information of the preliminary plan, QZY i1 is the weight value of the i1th preliminary solution in the preliminary solution weight information, CB i1 is the i1th first case vector in the preliminary solution vector set, N1 is the number of the first case vectors in the preliminary solution vector set, δ2 and δ3 are the second weight parameter and the third weight parameter respectively;

[0191] It should be noted that the second weight parameter and the third weight parameter may be set by the user or obtained based on historical data, and the embodiment of the present invention does not limit this.

[0192] It should be noted that the value ranges of the second weight parameter and the third weight parameter are [0.5, 2.0] and [0.1, 1.0] respectively. When it is necessary to emphasize the directional similarity, a larger second weight parameter value is taken, and when it is necessary to emphasize the influence of the vector size, a larger third weight parameter value is taken. Through this setting, both the vector directional similarity (reflecting the similarity of the search and rescue scenes) and the vector size (reflecting the scale difference of the search and rescue scenes) can be considered at the same time. The weight calculation is dynamically adjusted through the second weight parameter and the third weight parameter, which can adapt to the needs of different search and rescue scenes, facilitate practical application and adjustment, and provide a more reliable weight reference for the subsequent search and rescue plan generation.

[0193] It's important to note that the second maritime search and rescue calculation model calculates the similarity and importance between the current search and rescue scenario vector and the historical case vector, assigning a scientifically sound weight to each historical case. This ensures that more similar and important historical cases have greater influence in the subsequent solution integration. This weighting mechanism ensures that the resulting search and rescue solution fully incorporates the advantages of the most relevant historical experience while adapting to the specific needs of the current search and rescue scenario. This improves the solution's relevance, applicability, and effectiveness, and is crucial for increasing the success rate of maritime search and rescue.

[0194] S416, fusing the preliminary solution vector set and the preliminary solution weight information to obtain a preliminary solution vector;

[0195] It should be noted that the above-mentioned fusion processing can use a weighted average fusion algorithm, an attention mechanism fusion model, a high-order tensor decomposition method, and can also use a third maritime search and rescue calculation model for fusion processing. Specifically, the embodiment of the present invention does not limit this.

[0196] Wherein, the third maritime search and rescue calculation model is:

[0197]

[0198] Wherein, CBF is the preliminary solution vector, QZP is the average value of all the preliminary solution weight values ​​in the preliminary solution weight information, γ1 and γ2 are the first scaling factor and the second scaling factor respectively, and tanh(·) is a nonlinear activation function;

[0199] It should be noted that the first scaling factor and the second scaling factor may be set by a user or acquired based on historical data, and the embodiment of the present invention does not limit this.

[0200] It should be noted that the third maritime search and rescue calculation model generates a high-quality preliminary solution vector by fusing the preliminary solution vector set and weight information. By using the tanh nonlinear activation function to process the weighted historical case vector, not only the characteristics of the case itself γ1·CB are considered i1 , and also introduces the weight bias term γ2·(QZY i1 -QZP), so that more important case features are strengthened and general case features are suppressed.

[0201] It should be noted that the value ranges of the first scaling factor and the second scaling factor are [0.5, 2.0] and [0.1, 1.0] respectively, and can be flexibly adjusted according to the requirements of the search and rescue mission. When it is necessary to retain the integrity of the original features, increase the first scaling factor, and when it is necessary to highlight high-weight cases, increase the second scaling factor. In this way, it is possible to adaptively handle a variety of complex search and rescue scenarios, balance historical experience and current characteristics, eliminate the influence of outliers, and generate more stable and reliable solution vectors. Through this intelligent fusion mechanism, the model can extract the most valuable information from multiple relevant historical cases, provide high-quality comprehensive vector representations for subsequent search and rescue solution generation, and significantly improve the accuracy and applicability of search and rescue solutions.

[0202] S417: Process the multimodal vector and the preliminary solution vector to obtain a preliminary search and rescue solution vector.

[0203] It should be noted that the above processing is to first convert the multimodal vector and the preliminary plan vector into structured prompt words through a feature decoding algorithm (such as PCA), input the structured prompt words into the pre-trained LLM large language model, and the LLM generates a formatted search and rescue plan text based on the input information reasoning; then the search and rescue plan text is converted into a vector through pre-trained language models such as BERT / RoBERTa and text vectorization algorithms such as Sentence-BERT, and merged into a unified representation using a feature fusion algorithm (such as weighted average, attention mechanism); finally, a vector enhancement algorithm (knowledge distillation, vector calibration, anomaly detection) is applied for optimization to generate the final preliminary search and rescue plan vector.

[0204] It should be noted that the preliminary search and rescue plan vector plays a critical connecting role in this application: as an integration point for pre-processing, it fuses multimodal perception data (including text, images, audio, and other information about the current search and rescue scenario) with historical experience (similar cases extracted through the knowledge graph), unifying complex and heterogeneous information into a structured vector representation through the intelligent processing of the LLM. As the fundamental input for subsequent optimization, it provides framework guidance for intermediate and final plans, ensuring a clear starting point and direction for the development of search and rescue plans. The value of the preliminary plan vector lies in: on the one hand, it preserves the specificity and urgent needs of the current search and rescue scenario; on the other hand, it incorporates the successful experience and patterns of historical cases, providing a high-quality prototype of the search and rescue strategy while ensuring responsiveness. Compared with intermediate and final plans, the preliminary plan focuses more on the overall picture and framework construction, laying the foundation for subsequent detailed refinement and professional tuning. In time-sensitive maritime search and rescue scenarios, it can quickly provide feasible search and rescue ideas, buying valuable time for the entire search and rescue process and improving the search and rescue success rate.

[0205] It can be seen that implementing the method for generating a maritime search and rescue plan described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0206] In an optional embodiment, processing the preliminary search and rescue solution vector and the second case vector set to obtain an intermediate search and rescue solution vector includes:

[0207] S421, using a fourth maritime search and rescue calculation model, performing calculation processing on the preliminary search and rescue solution vector and the second case vector set to obtain a similarity vector set;

[0208] Wherein, the fourth maritime search and rescue calculation model is:

[0209]

[0210] Where, SSE is the similarity vector set, SSEi2 is the i2th similarity vector in the similarity vector set, CBS is the preliminary search and rescue solution vector, DE i2 is the i2-th second case vector in the second case vector set, N2 represents the number of the second case vectors in the second case vector set, ‖·‖2 represents the L2 norm, <·> represents the inner product of two vectors, and θ represents the equipment weight factor;

[0211] It should be noted that the equipment weight factor may be set by the user or obtained based on historical data, and the embodiment of the present invention does not limit this.

[0212] It should be noted that the fourth maritime search and rescue calculation model generates a similarity vector set by calculating the similarity between the preliminary search and rescue plan vector and the second case vector set (mainly containing equipment usage information). This model introduces an equipment weight factor to adjust the sensitivity of the similarity calculation, using the vector inner product <CBS,DE i2 >Reflects the direction similarity and is expressed by the L2 norm ‖CBS‖2·‖DE i2 ‖2 is normalized to ensure similarity values ​​are between [-1, 1]. This model, used in the maritime search and rescue equipment matching phase, accurately assesses the compatibility of preliminary plans with historical equipment use cases and identifies the most suitable equipment combination for the current search and rescue scenario. The equipment weighting factor can be adjusted based on different search and rescue missions, improving the model's adaptability, especially in complex search and rescue scenarios where equipment selection is critical.

[0213] In this application, the model acts as a bridge connecting abstract search and rescue strategies with specific implementation methods. Through scientific similarity calculations, it ensures that theoretical plans can be implemented, provides a basis for equipment selection for subsequent intermediate and final plans, improves search and rescue efficiency, and reduces resource waste. It is a key support for the implementation of search and rescue plans.

[0214] It should be noted that the equipment weight factor, which ranges from 0.5 to 3.0, adjusts the sensitivity and accuracy of matching the preliminary search and rescue plan with historical equipment cases. When the equipment weight factor is large (2.0-3.0), the similarity calculation is more sensitive to equipment matching, which is suitable for equipment-dependent missions such as deep-sea rescue, ensuring the selection of the most appropriate specialized equipment. When the equipment weight factor is medium (1.0-2.0), it balances the accuracy and flexibility of equipment selection, which is suitable for routine search and rescue missions. When the equipment weight factor is small (0.5-1.0), it allows for more equipment cases to be considered, which is suitable for emergency and simple search and rescue, expands the range of available equipment, and improves response speed. By flexibly adjusting the equipment weight factor, it is possible to optimize equipment resource allocation, balance equipment accuracy and availability, improve the cost-effectiveness of search and rescue missions, adapt to search and rescue missions of varying scale and complexity, ensure that theoretical plans can be effectively translated into actual operations, and significantly improve search and rescue efficiency and success rates.

[0215] S422, performing calculation processing on the preliminary search and rescue solution vector and the similarity vector set to obtain an intermediate fusion vector set;

[0216] It should be noted that the above-mentioned calculation processing can be performed by conventional technical processing through cross-attention mechanism, feature map fusion, kernel method mapping and other technologies, or it can be processed through the fifth maritime search and rescue calculation model. Specifically, the embodiment of the present invention does not limit it.

[0217] Among them, the fifth maritime search and rescue calculation model is:

[0218]

[0219] Where RH is the intermediate fusion vector set, RH i2 is the i2th intermediate fusion vector in the intermediate fusion vector set, μ1, μ2 and μ3 represent the first adjustment factor, the second adjustment factor and the third adjustment factor respectively.

[0220] It should be noted that the first adjustment factor, the second adjustment factor, and the third adjustment factor may be set by a user or acquired based on historical data, and the embodiment of the present invention does not limit this.

[0221] It is important to note that the fifth maritime search and rescue computational model achieves a sophisticated fusion of preliminary plans and equipment matching, organically integrating abstract search and rescue strategies with specific equipment selection. The intermediate fusion vector set, a key component in forming the search and rescue plan, not only incorporates equipment selection information but also retains the strategic guidance of the preliminary plan, forming a more specific and actionable search and rescue plan framework. Compared to the preliminary plan, the intermediate fusion vector focuses more on practical feasibility; compared to the final plan, it focuses more on equipment and resource matching. This multi-dimensional and multi-level fusion design effectively addresses the problem of "theoretical solutions failing to be implemented" in maritime search and rescue, significantly improving the relevance and practicality of the search and rescue plan, and laying a solid foundation for the final, efficient and scientific search and rescue plan.

[0222] It should be noted that the value ranges of the first adjustment factor, the second adjustment factor, and the third adjustment factor are [0.2, 1.0], [0.5, 2.0], and [0.1, 0.8], respectively. When the accuracy of equipment matching is important, the first adjustment factor should be increased; when the consistency of the solution is emphasized, the second adjustment factor should be increased; when the overall similarity distribution needs to be considered, the third adjustment factor should be increased.

[0223] S423, performing splicing processing on the intermediate fusion vector set to obtain an intermediate solution fusion vector;

[0224] It's important to note that the aforementioned splicing process uses vector concatenation technology to organically integrate multiple intermediate fusion vectors into a unified intermediate solution fusion vector, achieving the transition from "multiple alternative solutions" to "a single comprehensive solution." This concatenation process isn't a simple concatenation of vectors; instead, it employs optimized concatenation algorithms, such as dimensional alignment, feature selection, and weighted concatenation, to ensure that critical information is preserved and redundant information is compressed. This process effectively addresses the complexities of multiple equipment and strategies coexisting in maritime search and rescue, integrating dispersed equipment options into a coordinated system and forming a more systematic and comprehensive search and rescue strategy framework.

[0225] S424: Process the intermediate solution fusion vector to obtain an intermediate search and rescue solution vector.

[0226] It should be noted that the intermediate solution fusion vector is fine-tuned through vector optimization technology, including dimensional normalization, feature enhancement, noise filtering, and other processing, to generate a more refined and efficient intermediate search and rescue solution vector. This process can use conventional technologies such as deep learning models or expert rule systems to semantically optimize the vector to ensure the logic, completeness, and operability of the solution. The intermediate search and rescue solution vector is more specific and feasible than the preliminary solution, including detailed equipment configuration, personnel arrangements, and operational procedures. However, compared to the final solution, it still needs to be further improved in combination with the third case vector set (audio and video data).

[0227] It can be seen that implementing the method for generating a maritime search and rescue plan described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0228] In an optional embodiment, the processing of the intermediate search and rescue solution vector and the third case vector set to obtain maritime search and rescue solution result information includes:

[0229] S431, processing the intermediate search and rescue solution vector and the third case vector set to obtain third similarity information;

[0230] It should be noted that the above processing may use an improved cosine similarity, a vector space model or a deep learning similarity model to ensure effective matching of text strategies with audio and video cases, which is not limited in the embodiments of the present invention.

[0231] S432: Filter the third case vector set according to the third similarity information to obtain a third case filtered vector set;

[0232] It should be noted that the above-mentioned screening process is performed using techniques such as threshold screening, Top-K screening or dynamic threshold adjustment, which are not limited in specific embodiments of the present invention. Through the screening process, only audio and video information that is highly relevant to the intermediate search and rescue plan is retained. Through the screening process, irrelevant or low-quality audio and video cases are excluded, thereby improving the efficiency and quality of subsequent fusion. The filtered third case screening vector set contains the most valuable audio and video experience, which can provide intuitive operational guidance, environmental response strategies and solutions to common problems for the search and rescue plan.

[0233] S433, processing the intermediate search and rescue solution vector and the third case screening vector set to obtain third solution weight information;

[0234] It should be noted that the above-mentioned calculation and processing can be performed by conventional technical processing through cosine similarity calculation, bilinear mapping model, multi-head attention mechanism, deep similarity network, graph neural network similarity propagation and other technologies, or can be processed through the sixth maritime search and rescue calculation model. Specifically, the embodiments of the present invention do not limit this.

[0235] Among them, the sixth maritime search and rescue calculation model is:

[0236]

[0237] Where, QZS is the weight information of the third solution, QZS i3 is the i3th third solution weight value in the third solution weight information, ZSJ is the intermediate search and rescue solution vector, DS i3is the i3th third case vector in the third case screening vector set, |·| represents the modulus of the vector, N3 is the number of the third case vectors in the third case screening vector set, δ4 and δ5 represent the fourth weight parameter and the fifth weight parameter, respectively;

[0238] It should be noted that the fourth weight parameter and the fifth weight parameter may be set by the user or obtained based on historical data, and the embodiment of the present invention does not limit this.

[0239] It should be noted that the sixth maritime search and rescue calculation model assigns scientific and reasonable weights to audio and video cases, ensuring that the most relevant and valuable cases play a greater role in subsequent integration. By considering directional similarity and scale matching, the weight distribution is more comprehensive and precise. This information directly affects the quality and reliability of the final plan and serves as a critical bridge between intermediate and final plans. Compared with other models, the sixth maritime search and rescue calculation model focuses more on the precise weighting of audio and video cases, ensuring that search and rescue plans fully incorporate intuitive and vivid practical experience, improving their practicality and operability.

[0240] It should be noted that the value ranges of the fourth weight parameter and the fifth weight parameter are [0.5, 2.0] and [0.1, 1.0], respectively. The importance of the overall sensitivity and the modulus length ratio can be flexibly adjusted through the fourth weight parameter and the fifth weight parameter.

[0241] S434, using the seventh maritime search and rescue calculation model, fusing the third case screening vector set and the third solution weight information to obtain a third solution vector;

[0242] Among them, the seventh maritime search and rescue calculation model is:

[0243]

[0244] Wherein, DSS is the third solution vector;

[0245] It should be noted that the seventh maritime search and rescue computational model integrates empirical knowledge from multiple relevant audio and video cases, forming a rich library of audiovisual experience. As a vector representation of audio and video information, it overcomes the limitations of text-only solutions and adds more intuitive and practical operational guidance to the search and rescue plan. Together with the intermediate search and rescue plan vectors, it constitutes a key input to the final search and rescue plan, jointly influencing the quality and reliability of the final plan. Compared with other complex models, the seventh maritime search and rescue computational model, after meticulously calculating similarity and assigning weights in the early stages, uses a simple and intuitive weighted summation for the final fusion, ensuring both result quality and computational efficiency. This design is particularly well-suited for time-sensitive scenarios such as maritime search and rescue, enabling rapid generation of high-quality fusion vectors, providing strong support for subsequent final plan formulation.

[0246] S435 , processing the intermediate search and rescue solution vector and the third solution vector to obtain maritime search and rescue solution result information.

[0247] It should be noted that the above processing is to input the intermediate search and rescue plan vector (including framework strategy and equipment selection) and the third plan vector (integrating audio and video experience) into the pre-trained large language model (LLM) for intelligent reasoning, and finally generate the maritime search and rescue plan result information.

[0248] Exemplarily, the above processing first uses a feature dimensionality reduction algorithm (PCA) and a semantic mapping network to convert the intermediate search and rescue plan vector into a strategy framework description, and applies a multimodal decoder (such as CLIP decoder) and a case extraction algorithm to convert the third plan vector into an audio and video experience summary; then, the prompt engineering technology is applied to construct a professional prompt template containing scene description, plan framework and historical experience, and a constraint template based on domain knowledge is designed; then, the prompt is input into a large language model (such as GPT-4 / Claude) that has been fine-tuned with search and rescue domain knowledge for multi-step reasoning, and CoT (Chain-of-Thought) and ReAct (Reasoning+Acting) technology are used to fuse text strategies and audio and video experience; finally, a structured extraction algorithm is used to organize the plans generated by LLM, and a professional terminology corrector and a visualization enhancement engine (such as a roadmap generator) are applied to form the final maritime search and rescue plan result information. This design combines technologies such as vector semantic processing, prompt engineering, large model reasoning and post-processing optimization, fully leveraging the LLM's intelligent fusion capabilities, professional knowledge application capabilities and situational adaptability, and transforming the results of early data processing and case matching into professional plans that directly guide actual search and rescue operations, achieving a key shift from data to decision-making, and significantly improving the quality, efficiency and feasibility of maritime search and rescue plans.

[0249] It can be seen that implementing the method for generating a maritime search and rescue plan described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0250] Example 2

[0251] See also Figure 2 , Figure 2 This is a schematic diagram of the structure of a device for generating a maritime search and rescue plan disclosed in an embodiment of the present invention. Figure 2 The described maritime search and rescue plan generation device is applied to a maritime search and rescue plan generation optimization system, such as a local server or cloud server for maritime search and rescue plan generation, and the embodiment of the present invention does not limit this. Figure 2 As shown, the maritime search and rescue plan generating device includes:

[0252] Acquisition module 201, for acquiring multimodal data information and historical case data sets; the multimodal data information includes text information, audio information and video information; the historical case data sets include historical search and rescue scene data sets, historical search and rescue equipment data sets and historical search and rescue plan data sets; the historical search and rescue scene data sets include historical search and rescue audio and video data sets;

[0253] A first calculation module 202 is configured to process the multimodal data information to obtain a multimodal vector;

[0254] A second calculation module 203 is configured to process the historical case data set to obtain a historical case vector set; the historical case vector set includes a first case vector set, a second case vector set, and a third case vector set;

[0255] The third calculation module 204 is configured to perform a fusion process on the multimodal vector and the historical case vector set to obtain maritime search and rescue solution result information.

[0256] It can be seen that the implementation of the maritime search and rescue plan generation device described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0257] Example 3

[0258] See also Figure 3 , Figure 3 This is a schematic diagram of the structure of another device for generating a maritime search and rescue plan disclosed in an embodiment of the present invention. Figure 3 The described maritime search and rescue plan generation device is applied to a maritime search and rescue plan generation optimization system, such as a local server or cloud server for maritime search and rescue plan generation, and the embodiment of the present invention does not limit this. Figure 3As shown, the maritime search and rescue plan generating device includes:

[0259] Processor 301;

[0260] A memory 302 coupled to the processor 301 and storing executable program code;

[0261] The processor 301 calls the executable program code stored in the memory 302 to execute part or all of the steps of the method for generating a maritime search and rescue plan in the first embodiment.

[0262] It can be seen that the implementation of the maritime search and rescue plan generation device described in the embodiment of the present invention is conducive to improving the efficiency and success rate of maritime search and rescue operations, thereby quickly and accurately supporting maritime rescue, reducing rescue time, and reducing risks.

[0263] Example 4

[0264] An embodiment of the present invention discloses a computer-readable storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute some or all steps of the method for generating a maritime search and rescue plan in embodiment 1.

[0265] Example 5

[0266] An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to enable a computer to execute some or all of the steps in the maritime search and rescue plan generation method described in Example 1.

[0267] The system embodiments described above are merely illustrative. Modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules. They may be located in one place or distributed across multiple network modules. Some or all of these modules may be selected based on actual needs to achieve the objectives of this embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0268] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform, or of course, by means of hardware. Based on this understanding, the above technical solution, in essence, or the portion that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0269] Finally, it should be noted that the method and device for generating a maritime search and rescue plan disclosed in the embodiments of the present invention only disclose a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. 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 various embodiments of the present invention.

Claims

1. A method for generating a maritime search and rescue plan, characterized in that: The method comprises: S1, obtaining multimodal data information and a historical case data set; the multimodal data information includes text information, audio information and video information; the historical case data set includes a historical search and rescue scene data set, a historical search and rescue equipment data set and a historical search and rescue plan data set; the historical search and rescue scene data set includes a historical search and rescue audio and video data set; S2, processing the multimodal data information to obtain a multimodal vector; S3, processing the historical case data set to obtain a historical case vector set; the historical case vector set includes a first case vector set, a second case vector set, and a third case vector set; S4, fusing the multimodal vector and the historical case vector set to obtain maritime search and rescue solution result information.

2. The method for generating a maritime search and rescue plan according to claim 1, wherein: The processing of the multimodal data information to obtain a multimodal vector includes: S21, performing data cleaning processing on the multimodal data information to obtain first multimodal data information; S22, performing denoising processing on the first multimodal data information to obtain second multimodal data information; S23, performing deduplication processing on the second multimodal data information to obtain third multimodal data information; S24, performing normalization processing on the third multimodal data information to obtain fourth multimodal data information; the fourth multimodal data information includes fourth text information, fourth audio information, and fourth video information; S25: Process the fourth multimodal data information to obtain a multimodal vector.

3. The method for generating a maritime search and rescue plan according to claim 2, wherein: The processing of the fourth multimodal data information to obtain a multimodal vector includes: S251, performing named entity recognition processing on the fourth text information to obtain named entity information; S252, performing audio recognition processing on the fourth audio information to obtain audio semantic information; S253, performing video recognition processing on the fourth video information to obtain video semantic information; S254: Fusing the named entity information, the audio semantic information, and the video semantic information to obtain a multimodal vector.

4. The method for generating a maritime search and rescue plan according to claim 1, wherein: The processing of the historical case data set to obtain a historical case vector set includes: S31, processing the historical case data set to obtain maritime search and rescue knowledge graph information; S32: Process the historical case data set and the maritime search and rescue knowledge graph information to obtain a historical case vector set.

5. The method for generating a maritime search and rescue plan according to claim 1, wherein: The fusing of the multimodal vector and the historical case vector set to obtain maritime search and rescue solution result information includes: S41, processing the multimodal vector and the first case vector set to obtain a preliminary search and rescue solution vector; S42, processing the preliminary search and rescue solution vector and the second case vector set to obtain an intermediate search and rescue solution vector; S43: Process the intermediate search and rescue solution vector and the third case vector set to obtain maritime search and rescue solution result information.

6. The method for generating a maritime search and rescue plan according to claim 5, wherein: The processing of the multimodal vector and the first case vector set to obtain a preliminary search and rescue solution vector includes: S411, using a first maritime search and rescue calculation model, performing calculation processing on the multimodal vector and the first case vector set to obtain first similarity information; Wherein, the first maritime search and rescue calculation model is: Where, SY is the first similarity information, SY i is the i-th first similarity value in the first similarity information, MT is the multimodal vector, DY i is the first case vector in the first case vector set, DY k is the kth first case vector in the first case vector set, N is the number of the first case vectors in the first case vector set, |·| is the modulus of the orientation quantity, and δ1 is the first weight parameter; S412, preset s=1; S413, determining whether the sth first similarity value in the first similarity information is greater than the preset similarity threshold, and obtaining a first determination result; When the first judgment result is yes, the sth first case vector in the first case vector set is added to the preliminary solution vector set, and S414 is executed; When the first judgment result is no, executing S414; S414, determining whether s is greater than the number of the first case vectors in the first case vector set, and obtaining a second determination result; When the second judgment result is no, increment s by 1 and execute S412; When the second judgment result is yes, execute S415; S415, using a second maritime search and rescue calculation model, performing calculation processing on the multimodal vector and the preliminary solution vector set to obtain preliminary solution weight information; Among them, the second maritime search and rescue calculation model is: Where, QZY is the weight information of the preliminary plan, QZY i1 is the weight value of the i1th preliminary solution in the preliminary solution weight information, CB i1 is the i1th first case vector in the preliminary solution vector set, N1 is the number of the first case vectors in the preliminary solution vector set, δ2 and δ3 are the second weight parameter and the third weight parameter respectively; S416, fusing the preliminary solution vector set and the preliminary solution weight information to obtain a preliminary solution vector; S417: Process the multimodal vector and the preliminary solution vector to obtain a preliminary search and rescue solution vector.

7. The method for generating a maritime search and rescue plan according to claim 5, wherein: The processing of the preliminary search and rescue solution vector and the second case vector set to obtain an intermediate search and rescue solution vector includes: S421, using a fourth maritime search and rescue calculation model, performing calculation processing on the preliminary search and rescue solution vector and the second case vector set to obtain a similarity vector set; Wherein, the fourth maritime search and rescue calculation model is: Where, SSE is the similarity vector set, SSE i2 is the i2th similarity vector in the similarity vector set, CBS is the preliminary search and rescue solution vector, DE i2 is the i2-th second case vector in the second case vector set, N2 represents the number of the second case vectors in the second case vector set, ‖·‖2 represents the L2 norm, <·> represents the inner product of two vectors, and θ represents the equipment weight factor; S422, performing calculation processing on the preliminary search and rescue solution vector and the similarity vector set to obtain an intermediate fusion vector set; S423, performing splicing processing on the intermediate fusion vector set to obtain an intermediate solution fusion vector; S424: Process the intermediate solution fusion vector to obtain an intermediate search and rescue solution vector.

8. A device for generating a maritime search and rescue plan, characterized in that: The device comprises: An acquisition module is configured to acquire multimodal data information and a historical case data set; the multimodal data information includes text information, audio information, and video information; the historical case data set includes a historical search and rescue scene data set, a historical search and rescue equipment data set, and a historical search and rescue plan data set; the historical search and rescue scene data set includes a historical search and rescue audio and video data set; a first computing module, configured to process the multimodal data information to obtain a multimodal vector; A second computing module is configured to process the historical case data set to obtain a historical case vector set; the historical case vector set includes a first case vector set, a second case vector set, and a third case vector set; The third calculation module is used to fuse the multimodal vector and the historical case vector set to obtain maritime search and rescue plan result information.

9. A device for generating a maritime search and rescue plan, characterized in that: The device comprises: processor; a memory coupled to the processor and storing executable program code; The processor calls the executable program code stored in the memory to execute the maritime search and rescue plan generation method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and when the computer instructions are called, they are used to execute the method for generating a maritime search and rescue plan according to any one of claims 1 to 7.

Images