An information extraction method and device for the maritime search and rescue field

Abstract

This invention discloses an information extraction method and apparatus for the maritime search and rescue field. The method includes: pre-setting a set of search and rescue case texts and a marine environmental spatiotemporal dataset; performing semantic information extraction processing on the search and rescue case text set to obtain a search and rescue case set; and constructing a search and rescue case atlas based on the marine environmental spatiotemporal dataset and the search and rescue case set. This invention can construct a maritime search and rescue case atlas with clear semantic relationships, unified description standards, and high completeness.

Description

Technical Field

[0001] This invention belongs to the field of text data processing technology, specifically an information extraction method and apparatus for the maritime search and rescue field. Background Technology

[0002] Information extraction (IE) refers to the process of identifying and extracting valuable information from natural language text and representing and storing it in a structured form. It mainly includes the extraction of elements such as named entities, relationships, and events. Since the Message Understanding Conference (MUC) in the late 1980s, this technology has received widespread attention from international scholars and has developed rapidly.

[0003] In comparison, research on information extraction in China started relatively late. Limited by the lack of clear separating features between words in Chinese text, extraction is relatively difficult, and research development has been relatively slow. Early research on Chinese information extraction mainly focused on basic aspects such as Chinese word segmentation, lexical analysis, and named entity recognition. Currently, research directions in Chinese information extraction have gradually expanded to areas such as relation extraction, event extraction, and coreference resolution, but most of these are still geared towards specific scenarios and have not yet been widely applied to general information extraction systems.

[0004] In recent years, with the continuous evolution of deep learning technology, information extraction methods based on deep neural networks have become a research hotspot. In this type of method, words are semantically modeled in vector form, which effectively alleviates the problems caused by data sparsity and high-dimensional features in traditional representation methods, and is superior to manual selection methods in terms of feature representation capabilities.

[0005] Information is extracted from existing maritime search and rescue case texts and supplemented with relevant marine environmental information to obtain standardized case information. This facilitates comprehensive analysis of historical accidents, assessment of the risk level of specific operations or routes, and provides decision support for risk management and resource allocation. It also helps identify key factors of similar accidents, thereby enabling preventive measures to be taken to reduce the occurrence of similar accidents in the future and improve the safety of maritime operations.

[0006] Although information extraction technology has been widely used in many fields, it is mostly concentrated in specific vertical fields and lacks information extraction methods adapted to the characteristics of the maritime accident search and rescue field. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide an information extraction method and apparatus for the field of maritime search and rescue, which can construct a case map of the maritime search and rescue field with clear semantic relationships, unified description standards and high completeness.

[0008] To address the aforementioned technical problems, a first aspect of this invention discloses an information extraction method for the maritime search and rescue field, comprising:

[0009] S1. A pre-defined set of search and rescue case texts and a marine environmental spatiotemporal dataset; the set of search and rescue case texts includes N maritime search and rescue case texts; N is an integer greater than 1;

[0010] S2. Perform semantic information extraction processing on the search and rescue case text set to obtain the search and rescue case set;

[0011] The search and rescue case set includes N search and rescue case entities; each search and rescue case entity includes the time of the accident, the coordinates of the accident, and a set of event entities.

[0012] S3. Based on the marine environment spatiotemporal dataset and the search and rescue case set, a search and rescue case map is constructed;

[0013] The search and rescue case map includes N extended search and rescue case entities and several edges; the extended search and rescue case entities include the time of the accident, the coordinates of the accident, the set of event entities, and marine environmental information; the marine environmental information includes a first velocity, a second velocity, average direction, average wave height, ice coverage, sea surface temperature, precipitation, and cloud cover.

[0014] As an optional implementation, in the first aspect of the present invention, the semantic information extraction processing of the search and rescue case text set to obtain the search and rescue case set includes:

[0015] S21. Perform time extraction processing on the N maritime search and rescue case texts respectively to obtain the N accident times;

[0016] S22. Using a pre-trained named entity extraction model, process the N maritime search and rescue case texts respectively to obtain N named entity sets;

[0017] The named entity set includes spatial location entities, accident type entities, accident result entities, damage status entities, and several accident target entities;

[0018] S23. Process the spatial location entities of the N named entity sets to obtain the N crash coordinates;

[0019] S24. Combine the N accident times, N accident coordinates and N named entity sets to obtain N search and rescue case entities.

[0020] As an optional implementation, in the first aspect of the present invention, the above-described processing of the spatial location entities of the N named entity sets to obtain the N crash coordinates includes:

[0021] S231. Perform word segmentation on each of the N spatial location entities to obtain N location word sets; the location word sets include a subset of place names and / or a subset of relation words; the subset of place names includes several place names; the subset of relation words includes directional words, numerals, and quantifiers;

[0022] S232. Process the N subsets of place names respectively to obtain N initial coordinates;

[0023] S233. Determine whether each of the N location word sets contains the relation word subset to obtain N spatial judgment results;

[0024] When the spatial judgment result is yes, the corresponding initial coordinates are updated using the subset of relation words in the corresponding set of position words;

[0025] When the spatial determination result is negative, the initial coordinates remain unchanged;

[0026] S234. Determine N crash coordinates, which are N initial coordinates respectively.

[0027] As an optional implementation, in the first aspect of the present invention, the above-described processing of the N subsets of place names to obtain N initial coordinates includes:

[0028] S2321. For each subset of place names, execute S2322 to S2328 respectively;

[0029] S2322. Perform a zoning query on the subset of place names to obtain a corresponding zoning sequence; the zoning sequence includes several zoning names and one place name in sequence;

[0030] S2323. Perform multi-source place name retrieval processing on the location names in the zoning sequence to obtain a set of optional results; the set of optional results includes several optional results; the optional results sequentially include several zoning names and one location name;

[0031] S2324. Based on the zoning sequence, perform matching and filtering processing on the set of optional results to obtain a set of filtered results; the set of filtered results includes several of the optional results;

[0032] S2325. For each of the optional results in the set of filtering results, calculate the similarity between the location name and the location name in the zoning sequence to obtain the corresponding similarity value;

[0033] S2326. Set the query result as the optional result corresponding to the highest similarity value;

[0034] S2327. Perform a multi-source coordinate query on the query results to obtain a set of optional coordinates; the set of optional coordinates includes several optional coordinates;

[0035] S2328. Determine the initial coordinates as the average of all the optional coordinates in the optional coordinate set.

[0036] As an optional implementation, in the first aspect of the present invention, the above-mentioned construction of a search and rescue case atlas based on the marine environmental spatiotemporal dataset and the search and rescue case set includes:

[0037] S31. Based on the marine environment spatiotemporal dataset, the search and rescue case set is expanded to obtain an expanded search and rescue case set; the expanded search and rescue case set includes N expanded search and rescue case entities;

[0038] S32. Using the entity correlation degree calculation model, process the extended search and rescue case set to obtain the correlation degree matrix;

[0039] S33. Process the extended search and rescue case set and the correlation matrix to obtain the search and rescue case map.

[0040] As an optional implementation, in the first aspect of the present invention, the search and rescue case set is expanded based on the aforementioned marine environmental spatiotemporal dataset to obtain an expanded search and rescue case set, including:

[0041] S311. Perform S312 to S315 on each of the search and rescue case entities in the search and rescue case set to obtain the extended search and rescue case entity corresponding to each of the search and rescue case entities;

[0042] S312. Using the accident coordinates of the search and rescue case entity as the center, construct a first grid coordinate set and a second grid coordinate set;

[0043] The first set of grid coordinates includes a first coordinate, a second coordinate, a third coordinate, and a fourth coordinate; the second set of grid coordinates includes a fifth coordinate, a sixth coordinate, a seventh coordinate, and an eighth coordinate.

[0044] S313. Based on the accident time and accident coordinates of the search and rescue case entity, and the corresponding first grid coordinate set and second grid coordinate set, query the marine environment spatiotemporal dataset to obtain central environmental information, first environmental information set and second environmental information set;

[0045] Both the first environmental information set and the second environmental information set include four local environmental information items; both the central environmental information and the local environmental information include the first velocity, the second velocity, the average direction, the average wave height, the ice coverage, the sea surface temperature, the precipitation, and the cloud coverage.

[0046] S314. Using a marine environmental information calculation model, process the central environmental information, the first environmental information set, and the second environmental information set to obtain the marine environmental information corresponding to the search and rescue case entity;

[0047] S315. Combine the search and rescue case entity with the corresponding marine environmental information to obtain the corresponding extended search and rescue case entity.

[0048] As an optional implementation, in the first aspect of the present invention, the expression of the above-mentioned marine environmental information calculation model is:

[0049]

[0050] In the formula, ω1 and ω2 are the preset first weight and second weight, respectively; V1, V2, D, H, R1, T, Y, and R2 are the first velocity, second velocity, average direction, average wave height, ice coverage, sea surface temperature, precipitation, and cloud coverage of the marine environmental information, respectively; VA1 n VA2 n DA n HA n RA1 n TA n YA n and RA2 n These are, respectively, the first velocity, the second velocity, the average direction, the average wave height, the ice coverage, the sea surface temperature, the precipitation, and the cloud cover of the nth local environmental information in the first environmental information set; VB1 n VB2 n DB n HB n RB1 n TB n YB n and RB2 nThe first velocity, second velocity, average direction, average wave height, ice coverage, sea surface temperature, precipitation, and cloud coverage of the nth local environmental information in the second environmental information set are respectively; VC1, VC2, DC, HC, RC1, TC, YC, and RC2 are the first velocity, second velocity, average direction, average wave height, ice coverage, sea surface temperature, precipitation, and cloud coverage of the central environmental information, respectively.

[0051] A second aspect of this invention discloses an information extraction device for the field of maritime search and rescue, the device comprising a data storage module, an information extraction module, and a map construction module;

[0052] The data storage module is used to store a preset set of search and rescue case texts and a marine environmental spatiotemporal dataset;

[0053] The information extraction module is used to perform semantic information extraction processing on the search and rescue case text set to obtain the search and rescue case set;

[0054] The map construction module is used to construct a search and rescue case map based on the marine environment spatiotemporal dataset and the search and rescue case set.

[0055] A third aspect of this invention discloses another information extraction device for the maritime search and rescue field, the device comprising:

[0056] Memory containing executable program code;

[0057] A processor coupled to the memory;

[0058] The processor calls the executable program code stored in the memory to execute some or all of the steps in the information extraction method for the maritime search and rescue field disclosed in the first aspect of the present invention.

[0059] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the information extraction method for the maritime search and rescue field disclosed in the first aspect of the present invention.

[0060] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0061] This invention extracts semantic information from a collection of search and rescue case texts to obtain a search and rescue case set. Based on a marine environmental spatiotemporal dataset and the search and rescue case set, a search and rescue case atlas is constructed. This atlas can describe search and rescue cases with clear semantics and unified standards, improving the completeness of search and rescue cases, enhancing the convenience and scientific nature of case analysis, and providing a solid data foundation for subsequent decision analysis. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 This is a flowchart illustrating an information extraction method for the maritime search and rescue field disclosed in an embodiment of the present invention.

[0064] Figure 2 This is a schematic diagram of the structure of an information extraction device for the field of maritime search and rescue disclosed in an embodiment of the present invention.

[0065] Figure 3 This is a schematic diagram of another information extraction device for the field of maritime search and rescue disclosed in an embodiment of the present invention. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0068] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0069] Example

[0070] Please see Figure 1 . Figure 1 This is a flowchart illustrating an information extraction method for the maritime search and rescue field disclosed in an embodiment of the present invention; wherein, Figure 1 The described information extraction method for the maritime search and rescue field is applied to the field of text data processing, such as information extraction in the maritime search and rescue field; however, this embodiment of the invention is not limited to this specific application. Figure 1 As shown, the method includes:

[0071] S1. Preset search and rescue case text set and marine environmental spatiotemporal dataset; the above search and rescue case text set includes N maritime search and rescue case texts; N is an integer greater than 1.

[0072] Optionally, the aforementioned collection of search and rescue case texts comprises accident investigation reports published by the Maritime Safety Administration of the People's Republic of China over ten years (2014-2023). These accident investigation reports range in length from dozens to hundreds of pages, comprehensively presenting the entire process of maritime accidents, search and rescue operations, and accident handling. The reports consist of several standardized sections, including an accident overview, accident details, vessel information, personnel information, search and rescue process, liability determination, and management recommendations, providing significant reference value for navigation safety management, accident cause analysis, and search and rescue decision support.

[0073] Optionally, the aforementioned marine environmental spatiotemporal dataset is the ERA5 dataset released by the European Centre for Medium-Range Weather Forecasts (ECMWF). As shown in Table 1, this dataset selects six marine environmental factors considered to play an important role in the occurrence of maritime safety accidents and the implementation of search and rescue cases. This dataset is gridded data with a time resolution of 1 hour, providing numerical values ​​of meteorological parameters on a fixed latitude and longitude grid. Its original grid is defined on an N320 Gaussian grid with a grid spacing of approximately 0.25° × 0.25°. For ease of use, ERA5 data is usually interpolated onto a regular latitude and longitude grid. For example, in some studies, the data is interpolated onto a regular spherical grid with a spacing of 0.3°. Furthermore, the ERA5-Land dataset has a higher grid resolution of 0.1° × 0.1°.

[0074] Table 1. Description of the ERA5 dataset

[0075]

[0076]

[0077] S2. Semantic information extraction processing is performed on the above search and rescue case text set to obtain the search and rescue case set;

[0078] The aforementioned search and rescue case set includes N search and rescue case entities; the aforementioned search and rescue case entities include the time of the accident, the coordinates of the accident, and the event entity set.

[0079] S3. Based on the above-mentioned marine environment spatiotemporal dataset and the above-mentioned search and rescue case set, a search and rescue case map is constructed.

[0080] The aforementioned search and rescue case map includes N extended search and rescue case entities and several edges; the aforementioned extended search and rescue case entities include the aforementioned accident time, the aforementioned accident coordinates, the aforementioned event entity set, and marine environmental information; the aforementioned marine environmental information includes the first velocity, the second velocity, the average direction, the average wave height, the ice coverage, the sea surface temperature, the precipitation, and the cloud cover.

[0081] It should be noted that the edges in the search and rescue case graph above represent the relationships between two search and rescue case entities. The two search and rescue case entities connected by each edge correspond to similar marine environments.

[0082] It is evident that implementing the information extraction method for the maritime search and rescue field disclosed in this embodiment of the invention can construct a search and rescue case map with clear semantic relationships, unified description standards, and high completeness, thereby improving the convenience and scientific nature of case analysis and providing a solid data foundation for subsequent decision analysis.

[0083] In an optional embodiment, the semantic information extraction processing of the search and rescue case text set described above, to obtain the search and rescue case set, includes:

[0084] S21. For the N above-mentioned maritime search and rescue case texts, perform time extraction processing to obtain the N above-mentioned accident times.

[0085] S22. Using a pre-trained named entity extraction model, process the texts of the above-mentioned maritime search and rescue cases to obtain N sets of named entities.

[0086] The aforementioned set of named entities includes entities for spatial location, accident type, accident result, damage status, and several accident target entities.

[0087] It should be noted that the aforementioned spatial location entity, accident type entity, accident result entity, damage status entity, and accident target entity are respectively the accident location, accident type, accident result, damage status, and the target lost in the accident described in text form.

[0088] It should be noted that the above-mentioned target entities are used to describe the type and name of the corresponding target in text form, such as "oil tanker 'Ningda 10'".

[0089] S23. Process the spatial location entities of the N named entity sets to obtain the N crash coordinates.

[0090] S24. Combine the N accident times, N accident coordinates and N named entity sets mentioned above to obtain N search and rescue case entities mentioned above.

[0091] It should be noted that the above corresponding combination is to combine the accident time, accident coordinates and named entity set corresponding to the same maritime search and rescue case text to obtain the corresponding search and rescue case entity. The event entity set of the search and rescue case entity includes the accident type entity, accident result entity, damage status entity and several accident target entities in the corresponding named entity set.

[0092] It is evident that the search and rescue case entities fully cover the time, location, and semantic information of various named entities in the case.

[0093] In another optional embodiment, the above-mentioned search and rescue case text set is subjected to time extraction processing to obtain a time set, including:

[0094] S211. Using preset regular expressions, process the N texts of the above-mentioned maritime search and rescue cases to obtain N initial accident times.

[0095] Optionally, the above regular expression is:

[0096] (Year\d{1,2}Month\d{1,2}Day, approximately *d{4}Hour*)|(Year\d{1,2}Month\d{1,2}Day)

[0097] For example, if a maritime search and rescue case text contains the following content: "At approximately 15:51 on September 22, 2018, the Panamanian-flagged container ship 'APL LOS ANGELES' was en route from Fuzhou Port to Shantou...", then the initial time of the accident obtained by the above regular expression matching is "approximately 15:51 on September 22, 2018".

[0098] S212. Normalize the N initial accident times described above to obtain N accident times described above.

[0099] It should be noted that the above standardization process is to convert the initial crash time into a unified crash time format, such as converting "approximately 15:51 on September 22, 2018" into "2018-09-22 15:51:00".

[0100] In another optional embodiment, the named entity extraction model described above includes a feature extraction module, a feature optimization module, and a decoding module connected in sequence.

[0101] Preferably, the feature extraction module described above is built based on the BERT model.

[0102] It should be noted that BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained language representation model based on Transformers. It is pre-trained on a large amount of text data and learns rich language features. A key feature of BERT is its ability to understand contextual information, which is crucial for sequence labeling tasks. BERT learns language representations through the following two pre-training tasks: (1) Masked Language Model (MLM): randomly masking some words in the input and then predicting these masked words; (2) Next Sentence Prediction (NSP): predicting whether two sentences are sequentially related.

[0103] Preferably, the feature optimization module described above is constructed based on a BiLSTM network.

[0104] It's important to note that BiLSTM is a special type of recurrent neural network (RNN) that processes sequence data using two LSTM (Long Short-Term Memory) layers, one from left to right and the other from right to left. This bidirectional processing allows the model to consider contextual information simultaneously, thus enabling a better understanding of each element in the sequence.

[0105] Preferably, the above decoding module is constructed based on the Conditional Random Field (CRF) model.

[0106] It's important to note that CRF is a probabilistic graphical model used for labeling and segmenting sequence data. It predicts the most likely label sequence for the entire sequence by considering the dependencies between labels. CRF takes the output of a BiLSTM layer and predicts the label for each word, while taking into account constraints between entity labels (e.g., the start label of one entity cannot immediately follow the end label of another entity).

[0107] It should be noted that the named entity extraction model described above was trained using a pre-defined labeled dataset, which includes 200 labeled search and rescue case texts. The text content corresponding to spatial location entities, accident type entities, accident result entities, damage status entities, and accident target entities was manually labeled.

[0108] As can be seen, the aforementioned named entity extraction model combines BERT's powerful contextual understanding capabilities, BiLSTM's ability to capture long-distance dependencies in sequences, and CRF's ability to model dependencies between labels, thereby improving the efficiency and accuracy of named entity recognition.

[0109] In yet another optional embodiment, the above-described processing of the spatial location entities of the N named entity sets to obtain N crash coordinates includes:

[0110] S231. Perform word segmentation on each of the N spatial location entities to obtain N sets of location words; the set of location words includes a subset of place names and / or a subset of relation words; the subset of place names includes several place names; the subset of relation words includes directional words, numerals and quantifiers.

[0111] It should be noted that the place names in the above subset of place names reflect the different administrative levels to which the accident coordinates belong.

[0112] Optionally, the above word segmentation process involves using jieba to segment spatial location entities and tagging them with part-of-speech tags. Then, words with the part-of-speech tag "ns" are treated as place names, and a subset of place names is obtained. Words with the part-of-speech tags "f," "m," or "q" are treated as directional words, numerals, and quantifiers, respectively, and a subset of relational words is obtained. The corresponding code is as follows, where `text` represents the spatial location entity, `location_names` is the subset of place names, and `directions` is the subset of relational words: `import jieba.posseg as pseg`

[0113] words = pseg.cut(text)

[0114] location_names=[word forword,flag inwords ifflag=='ns']

[0115] directions=[word forword,flag in words ifflag=='f'or flag=='m'orflag=='q']

[0116] Taking the spatial location entity "about 0.5 nautical miles east of the mouth of the Taoer River in Binzhou, 38°15′177N, 118°09′217E" as an example, the place name subset obtained by the above code includes "Binzhou" and "Taoer River", and the locative word, numeral and quantifier in the relational word subset are "east", "0.5" and "nautical miles" respectively.

[0117] S232. Process the N subsets of the above place names respectively to obtain N initial coordinates.

[0118] S233. Determine whether each of the N sets of position words contains the subset of relation words to obtain N spatial judgment results;

[0119] When the above spatial judgment result is yes, the corresponding initial coordinates are updated using the above relational word subset of the corresponding position word set;

[0120] When the spatial judgment result is negative, the initial coordinates remain unchanged.

[0121] S234. Determine N of the above-mentioned accident coordinates, which are respectively N of the above-mentioned initial coordinates.

[0122] In another optional embodiment, the above processing of the N subsets of place names yields N initial coordinates, including:

[0123] S2321. For each of the above subsets of place names, execute S2322 to S2328 respectively.

[0124] S2322. Perform a zoning query on the above subset of place names to obtain the corresponding zoning sequence; the above zoning sequence includes several zoning names and one place name in sequence.

[0125] Optionally, the above-mentioned administrative division query is performed by using a preset administrative division database to query each place name in the place name subset, returning the corresponding administrative division level, and then sorting the place name subset in descending order of administrative division level to obtain the corresponding administrative division sequence.

[0126] It should be noted that the aforementioned administrative division database includes name information from the provincial level to the township level nationwide. Place names with the lowest administrative division level in the above subset, which are below the township level, are not included in the aforementioned administrative division database; these place names are listed as location names at the end of the division sequence.

[0127] S2323. Perform multi-source place name retrieval processing on the above-mentioned location names in the above-mentioned administrative division sequence to obtain a set of optional results; the above-mentioned set of optional results includes several optional results; the above-mentioned optional results include several of the above-mentioned administrative division names and one of the above-mentioned location names in sequence.

[0128] Optionally, the above multi-source place name retrieval process involves querying the above place names using Baidu Maps, Gaode Maps, and Tianditu respectively, and combining the optional results obtained from the three into an optional result set; the administrative division names in the optional results are arranged in descending order of administrative level, with the place name of the lowest administrative level placed last.

[0129] As can be seen, multi-source place name retrieval can expand the sources of queries and avoid situations where a single platform cannot find the above place names or only returns inaccurate place name options.

[0130] S2324. Based on the above-mentioned zoning sequence, the above-mentioned optional result set is matched and filtered to obtain a filtered result set; the above-mentioned filtered result set includes several of the above-mentioned optional results.

[0131] S2325. For each of the above optional results in the above filtering result set, calculate the similarity between the above location name and the above location name in the above zoning sequence to obtain the corresponding similarity value.

[0132] S2326. Set the query result to the optional result corresponding to the highest similarity value mentioned above.

[0133] S2327. Perform a multi-source coordinate query on the above query results to obtain a set of optional coordinates; the above set of optional coordinates includes several optional coordinates.

[0134] Optionally, the above multi-source coordinate query involves using the geocoding services of Baidu Maps, Gaode Maps, and Tianditu to query the coordinates corresponding to the query results, and converting them to CGCS2000 format to obtain the corresponding optional coordinates.

[0135] It should be noted that when performing coordinate queries, one or two of Baidu Maps, Gaode Maps, and Tianditu may not yield any results, as the corresponding selectable coordinates may not be included in the available coordinate set.

[0136] S2328. Determine the above initial coordinates as the average of all the above optional coordinates in the optional coordinate set.

[0137] It should be noted that the longitude and latitude values ​​of the initial coordinates mentioned above are the average longitude and latitude values ​​of all selectable coordinates, respectively.

[0138] In another optional embodiment, based on the above-mentioned zoning sequence, the optional result set is subjected to matching and filtering processing to obtain a filtered result set, including:

[0139] S23241. Determine whether the above-mentioned zoning sequence matches each of the above-mentioned optional results, and obtain the matching judgment result corresponding to each of the above-mentioned optional results;

[0140] When the above matching judgment result is yes, the above set of optional results remains unchanged;

[0141] If the above matching result is negative, the corresponding optional result will be removed from the optional result set.

[0142] It should be noted that when the zoning sequence is exactly the same as the zoning name of the optional result (i.e., the number of zoning names is the same, and the zoning names with the same sequence number are the same), then the zoning sequence matches the optional result; when the zoning sequence is not exactly the same as the zoning name of the optional result, then the zoning sequence does not match the optional result.

[0143] S23242. Determine the above-mentioned set of filtered results as the above-mentioned set of optional results.

[0144] Optionally, the above similarity calculation involves using the BERT model to process the location names of the optional results and the zoning sequence separately, obtaining two feature vectors. Then, a cosine similarity calculation is performed on the two feature vectors to obtain the corresponding similarity value. The relevant code is as follows, where text1 and text2 are the location names of the optional results and the zoning sequence, respectively:

[0145]

[0146]

[0147] In yet another optional embodiment, updating the corresponding initial coordinates using the subset of relational words from the corresponding set of positional words includes:

[0148] S2331. Based on the above quantifiers in the above relational word subset, the above numerals are converted to units to obtain the offset distance.

[0149] S2332. Determine the azimuth angle based on the directional words in the above-mentioned relational word subset.

[0150] It should be noted that the above azimuth angles are usually measured in degrees, starting from due north and proceeding clockwise. Due north is 0 degrees, due east is 90 degrees, due south is 180 degrees, due west is 270 degrees, northeast is 45 degrees, southeast is 135 degrees, southwest is 225 degrees, and northwest is 315 degrees.

[0151] S2333. Using the GDAL library, based on the above offset distance, azimuth angle, and initial coordinates, the updated initial coordinates are obtained. The corresponding code is as follows, where new_lon and new_lat are the precision and dimension values ​​of the updated initial coordinates.

[0152]

[0153]

[0154] In another optional embodiment, the search and rescue case atlas constructed based on the aforementioned marine environmental spatiotemporal dataset and the aforementioned search and rescue case set includes:

[0155] S31. Based on the above-mentioned marine environment spatiotemporal dataset, the above-mentioned search and rescue case set is expanded to obtain an expanded search and rescue case set; the above-mentioned expanded search and rescue case set includes N above-mentioned expanded search and rescue case entities.

[0156] S32. Using the entity correlation calculation model, the above extended search and rescue case set is processed to obtain the correlation matrix.

[0157] S33. Process the above-mentioned extended search and rescue case set and the above-mentioned correlation matrix to obtain the above-mentioned search and rescue case map.

[0158] It is evident that by leveraging open-source marine environmental spatiotemporal datasets, supplementing marine environmental information at the time of search and rescue incidents, and further calculating correlation matrices, the resulting search and rescue incident map can fully reflect the event and environmental information of each incident, as well as the relationships between them.

[0159] In another optional embodiment, the search and rescue case set is expanded based on the aforementioned marine environmental spatiotemporal dataset to obtain an expanded search and rescue case set, including:

[0160] S311. Perform S312 to S315 on each of the above-mentioned search and rescue case entities in the above-mentioned search and rescue case set to obtain the above-mentioned extended search and rescue case entities corresponding to each of the above-mentioned search and rescue case entities.

[0161] S312. Using the aforementioned accident coordinates of the above-mentioned search and rescue case entity as the center, construct the first grid coordinate set and the second grid coordinate set.

[0162] The aforementioned first grid coordinate set includes the first coordinate, the second coordinate, the third coordinate, and the fourth coordinate; the aforementioned second grid coordinate set includes the fifth coordinate, the sixth coordinate, the seventh coordinate, and the eighth coordinate.

[0163] Preferably, the first and second coordinates are the coordinates at which the accident coordinates deviate from latitude by 0.25° along the due south and due north directions, respectively; the third and fourth coordinates are the coordinates at which the accident coordinates deviate from latitude by 0.25° along the due east and due west directions, respectively; the fifth and sixth coordinates are the coordinates at which the second coordinates deviate from latitude by 0.25° along the due east and due west directions, respectively; and the seventh and eighth coordinates are the coordinates at which the first coordinates deviate from latitude by 0.25° along the due east and due west directions, respectively.

[0164] S313. Based on the above-mentioned accident time and above-mentioned accident coordinates of the above-mentioned search and rescue case entities, as well as the corresponding above-mentioned first grid coordinate set and above-mentioned second grid coordinate set, query the above-mentioned marine environment spatiotemporal dataset to obtain the central environmental information, the first environmental information set and the second environmental information set.

[0165] Both the aforementioned first environmental information set and the aforementioned second environmental information set include four local environmental information items; both the aforementioned central environmental information and the aforementioned local environmental information include the aforementioned first velocity, the aforementioned second velocity, the aforementioned average direction, the aforementioned average wave height, the aforementioned ice coverage, the aforementioned sea surface temperature, the aforementioned precipitation, and the aforementioned cloud coverage.

[0166] It should be noted that the four local environmental information pieces in the aforementioned first environmental information set correspond to the first coordinate, second coordinate, third coordinate, and fourth coordinate, respectively. By combining the first coordinate, second coordinate, third coordinate, and fourth coordinate with the time of the accident, the corresponding environmental information such as the first velocity, second velocity, average direction, average wave height, ice coverage, sea surface temperature, precipitation, and cloud cover can be queried from the marine environmental spatiotemporal dataset.

[0167] It should be noted that the four local environmental information pieces in the aforementioned second environmental information set correspond to the fifth, sixth, seventh, and eighth coordinates, respectively. By combining the fifth, sixth, seventh, and eighth coordinates with the time of the accident, the corresponding environmental information such as the first velocity, second velocity, average direction, average wave height, ice coverage, sea surface temperature, precipitation, and cloud cover can be queried from the marine environmental spatiotemporal dataset.

[0168] It should be noted that the above query of the marine environmental spatiotemporal dataset involves inputting query coordinates and the time of the accident. The dataset then searches for environmental information at several grid points closest to the time of the accident and adjacent to the query coordinates. Finally, bilinear spatial interpolation is performed on these environmental information points based on the corresponding grid point coordinates and the query coordinates to obtain the desired environmental information. The query coordinates can be the accident coordinates, the first coordinate, the second coordinate, the third coordinate, the fourth coordinate, the fifth coordinate, the sixth coordinate, the seventh coordinate, or the eighth coordinate.

[0169] S314. Using a marine environmental information calculation model, process the above-mentioned central environmental information, the above-mentioned first environmental information set, and the above-mentioned second environmental information set to obtain the above-mentioned marine environmental information corresponding to the above-mentioned search and rescue case entity.

[0170] S315. Combine the above-mentioned search and rescue case entities with the corresponding marine environmental information to obtain the corresponding extended search and rescue case entities.

[0171] In yet another optional embodiment, the expression for the above-mentioned marine environmental information calculation model is:

[0172]

[0173] In the formula, ω1 and ω2 are the preset first weight and second weight, respectively; V1, V2, D, H, R1, T, Y, and R2 are the first velocity, the second velocity, the average direction, the average wave height, the ice coverage, the sea surface temperature, the precipitation, and the cloud coverage, respectively, of the above-mentioned marine environmental information; VA1 n VA2 n DA n HA n RA1 n TA n YA n and RA2 n The first velocity, the second velocity, the average direction, the average wave height, the ice coverage, the sea surface temperature, the precipitation, and the cloud cover are respectively the nth local environmental information in the first environmental information set mentioned above; VB1 n VB2 n DB n HB n RB1 n TB n YB n and RB2 n The first velocity, the second velocity, the average direction, the average wave height, the ice coverage, the sea surface temperature, the precipitation, and the cloud coverage are respectively the nth local environmental information in the second environmental information set; VC1, VC2, DC, HC, RC1, TC, YC, and RC2 are respectively the first velocity, the second velocity, the average direction, the average wave height, the ice coverage, the sea surface temperature, the precipitation, and the cloud coverage of the central environmental information.

[0174] Preferably, ω1 and ω2 are 1 / 2 and 1 / 2 respectively.

[0175] It is evident that the above marine environmental information comprehensively reflects the environmental information of the sea area surrounding the accident location.

[0176] In yet another optional embodiment, the expression for the above entity association calculation model is:

[0177]

[0178] In the formula, S i,j and S j,i S represents the element values ​​in the i-th row and j-th column and the j-th row and i-th column of the aforementioned correlation matrix, respectively. i,j and S j,i All of these represent the marine environmental information of the i-th entity in the aforementioned extended search and rescue case set. The marine environmental information of the j-th extended search and rescue case entity mentioned above. The correlation degree mentioned above; 1≤i≤N, 1≤j≤N, and i and j are both integers; a, b and c are the preset first coefficient, second coefficient and third coefficient respectively; They are respectively The k-th, l-th, and 8th components; They are respectively The k-th, l-th, and 8th components; The marine environmental information of the p-th entity in the aforementioned extended search and rescue case set. The kth, lth, and 8th components; 1≤k≤5 and k is an integer, l is 6 or 7.

[0179] It should be noted that the first to eighth components of each marine environmental information are the first velocity, the second velocity, the average direction, the average wave height, the ice coverage, the sea surface temperature, the precipitation, and the cloud cover, respectively.

[0180] It should be noted that, based on the degree of impact on navigation safety, marine environmental information can be divided into the following three categories: first, high-impact factors, including first speed, second speed, average direction, average wave height, and ice coverage; second, medium-impact factors, including sea surface temperature and precipitation; and third, low-impact factors, including only cloud cover.

[0181] Preferably, the first coefficient, the second coefficient, and the third coefficient are 0.2, 0.15, and 0.1, respectively.

[0182] It is evident that the aforementioned entity correlation can reflect the impact of different levels of factors on navigation safety, thus improving the scientific rigor of correlation analysis.

[0183] In another optional embodiment, the above-mentioned processing of the extended search and rescue case set and the above-mentioned correlation matrix to obtain the above-mentioned search and rescue case map includes:

[0184] S331. Initialize the initial case map to include only N of the above-mentioned search and rescue case entities.

[0185] S332. The above correlation matrix is ​​split to obtain N initial correlation vectors.

[0186] S333. Sort the N initial correlation vectors in descending order to obtain N sorted correlation vectors.

[0187] S334. For the N sorted correlation vectors mentioned above, perform filtering processing to obtain N filtered correlation vectors; the filtered correlation vectors include M of the aforementioned correlations; M is an integer greater than 1, and M... <N。

[0188] It should be noted that the above filtering process retains the first M components of the sorted correlation vector to obtain the corresponding filtered correlation vector.

[0189] S335. Based on the above-mentioned filtering correlation vector, the above-mentioned initial case map is processed to obtain the above-mentioned search and rescue case map.

[0190] As can be seen, each search and rescue case entity in the above search and rescue case map is connected to the M search and rescue case entities with the most similar marine environments.

[0191] In another optional embodiment, the initial case map is processed based on the aforementioned filtering correlation vector to obtain the aforementioned search and rescue case map, including:

[0192] S3351. Initialize the first number ii to 1.

[0193] S3352. Set the current correlation vector to the i-th of the above-mentioned filtered correlation vectors.

[0194] S3353. Find the index of each of the above-mentioned correlation degrees in the above-mentioned current correlation degree vector in the above-mentioned initial correlation degree vector to obtain the current index set; the above-mentioned current index set includes M index values.

[0195] S3354. Initialize the first number jj to 1.

[0196] S3355. In the initial case graph above, establish an edge connecting the ii-th search and rescue case entity and the jj-th search and rescue case entity to obtain the updated initial case graph above; increment the value of jj by 1.

[0197] S3356, Repeat S3355 until jj is greater than M; Increment the value of ii by 1.

[0198] S3357, Repeat S3352 to S3356 until the value of ii is greater than N.

[0199] It is evident that implementing the information extraction method for the maritime search and rescue field disclosed in this embodiment of the invention can construct a search and rescue case map with clear semantic relationships, unified description standards, and high completeness, thereby improving the convenience and scientific nature of case analysis and providing a solid data foundation for subsequent decision analysis.

[0200] Example 2

[0201] Please see Figure 2 . Figure 2 This is a schematic diagram of the structure of an information extraction device for the maritime search and rescue field disclosed in an embodiment of the present invention. Figure 2 The described information extraction device for the maritime search and rescue field can be applied to text data processing fields, such as information extraction in the maritime search and rescue field; however, the embodiments of this invention are not limited thereto. Figure 2 As shown, the device may include a data storage module 201, an information extraction module 202, and a map construction module 203.

[0202] The aforementioned data storage module 201 is used to store a preset collection of search and rescue case texts and a marine environmental spatiotemporal dataset.

[0203] The aforementioned information extraction module 202 is used to perform semantic information extraction processing on the aforementioned search and rescue case text set to obtain the search and rescue case set.

[0204] The above-mentioned map construction module 203 is used to construct a search and rescue case map based on the above-mentioned marine environment spatiotemporal dataset and the above-mentioned search and rescue case set.

[0205] It is evident that implementing the information extraction device for the maritime search and rescue field described in the embodiments of the present invention can construct a search and rescue case map with clear semantic relationships, unified description standards, and high completeness, thereby improving the convenience and scientific nature of case analysis and providing a solid data foundation for subsequent decision analysis.

[0206] Example 3

[0207] Please see Figure 3 , Figure 3 This is a schematic diagram of another information extraction device for the maritime search and rescue field disclosed in an embodiment of the present invention. Figure 3 The described information extraction device for the maritime search and rescue field can be applied to text data processing fields, such as information extraction in the maritime search and rescue field; however, the embodiments of this invention are not limited thereto. Figure 3 As shown, the information extraction device for the maritime search and rescue field may include the following parts:

[0208] Memory 301 storing executable program code;

[0209] Processor 302 coupled to memory 301;

[0210] The processor 302 calls the executable program code stored in the memory 301 to execute the steps in the information extraction method for the maritime search and rescue field described in Embodiment 1.

[0211] Example 4

[0212] This invention discloses a computer read storage medium that stores a computer program for electronic data interchange, wherein the computer program causes a computer to execute the steps in the information extraction method for the maritime search and rescue field described in Embodiment 1.

[0213] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0214] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0215] Finally, it should be noted that the information extraction method and apparatus for the maritime search and rescue field disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention, and are only used to illustrate the technical solutions of the present invention, not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for information extraction in the field of maritime search and rescue, characterized in that, include: S1. A pre-set set of search and rescue case texts and a marine environmental spatiotemporal dataset; the set of search and rescue case texts includes N maritime search and rescue case texts; N is an integer greater than 1; S2. Perform semantic information extraction processing on the search and rescue case text set to obtain the search and rescue case set; The search and rescue case set includes N search and rescue case entities; each search and rescue case entity includes the time of the accident, the coordinates of the accident, and a set of event entities. S3. Based on the marine environment spatiotemporal dataset and the search and rescue case set, a search and rescue case map is constructed; The search and rescue case map includes N extended search and rescue case entities and several edges; the extended search and rescue case entities include the time of the accident, the coordinates of the accident, the set of event entities, and marine environmental information; the marine environmental information includes a first velocity, a second velocity, average direction, average wave height, ice coverage, sea surface temperature, precipitation, and cloud cover.

2. The information extraction method for the maritime search and rescue field according to claim 1, characterized in that, The semantic information extraction process performed on the search and rescue case text set to obtain the search and rescue case set includes: S21. Perform time extraction processing on the N maritime search and rescue case texts respectively to obtain the N accident times; S22. Using a pre-trained named entity extraction model, process the N maritime search and rescue case texts respectively to obtain N named entity sets; The named entity set includes spatial location entities, accident type entities, accident result entities, damage status entities, and several accident target entities; S23. Process the spatial location entities of the N named entity sets to obtain the N crash coordinates; S24. Combine the N accident times, N accident coordinates and N named entity sets to obtain N search and rescue case entities.

3. The information extraction method for the maritime search and rescue field according to claim 2, characterized in that, The process of processing the spatial location entities of the N named entity sets to obtain the N crash coordinates includes: S231. Perform word segmentation on each of the N spatial location entities to obtain N location word sets; the location word sets include a subset of place names and / or a subset of relation words; the subset of place names includes several place names; the subset of relation words includes directional words, numerals, and quantifiers; S232. Process the N subsets of place names respectively to obtain N initial coordinates; S233. Determine whether each of the N location word sets contains the relation word subset to obtain N spatial judgment results; When the spatial judgment result is yes, the corresponding initial coordinates are updated using the subset of relation words in the corresponding set of position words; When the spatial determination result is negative, the initial coordinates remain unchanged; S234. Determine N crash coordinates, which are N initial coordinates respectively.

4. The information extraction method for the maritime search and rescue field according to claim 3, characterized in that, The process of processing each of the N subsets of place names yields N initial coordinates, including: S2321. For each subset of place names, execute S2322 to S2328 respectively; S2322. Perform a zoning query on the subset of place names to obtain a corresponding zoning sequence; the zoning sequence includes several zoning names and one place name in sequence; S2323. Perform multi-source place name retrieval processing on the location names in the zoning sequence to obtain a set of optional results; the set of optional results includes several optional results; the optional results sequentially include several zoning names and one location name; S2324. Based on the zoning sequence, perform matching and filtering processing on the set of optional results to obtain a set of filtered results; the set of filtered results includes several of the optional results; S2325. For each of the optional results in the set of filtering results, calculate the similarity between the location name and the location name in the zoning sequence to obtain the corresponding similarity value; S2326. Set the query result as the optional result corresponding to the highest similarity value; S2327. Perform a multi-source coordinate query on the query results to obtain a set of optional coordinates; the set of optional coordinates includes several optional coordinates; S2328. Determine the initial coordinates as the average of all the optional coordinates in the optional coordinate set.

5. The information extraction method for the maritime search and rescue field according to claim 1, characterized in that, The search and rescue case atlas, constructed based on the marine environmental spatiotemporal dataset and the search and rescue case set, includes: S31. Based on the marine environment spatiotemporal dataset, the search and rescue case set is expanded to obtain an expanded search and rescue case set; the expanded search and rescue case set includes N expanded search and rescue case entities; S32. Using the entity correlation degree calculation model, process the extended search and rescue case set to obtain the correlation degree matrix; S33. Process the extended search and rescue case set and the correlation matrix to obtain the search and rescue case map.

6. The information extraction method for the maritime search and rescue field according to claim 5, characterized in that, The search and rescue case set is expanded based on the marine environmental spatiotemporal dataset to obtain an expanded search and rescue case set, including: S311. Perform S312 to S315 on each of the search and rescue case entities in the search and rescue case set to obtain the extended search and rescue case entity corresponding to each of the search and rescue case entities; S312. Using the accident coordinates of the search and rescue case entity as the center, construct a first grid coordinate set and a second grid coordinate set; The first set of grid coordinates includes a first coordinate, a second coordinate, a third coordinate, and a fourth coordinate; the second set of grid coordinates includes a fifth coordinate, a sixth coordinate, a seventh coordinate, and an eighth coordinate. S313. Based on the accident time and accident coordinates of the search and rescue case entity, and the corresponding first grid coordinate set and second grid coordinate set, query the marine environment spatiotemporal dataset to obtain central environmental information, first environmental information set and second environmental information set; Both the first environmental information set and the second environmental information set include four local environmental information items; both the central environmental information and the local environmental information include the first velocity, the second velocity, the average direction, the average wave height, the ice coverage, the sea surface temperature, the precipitation, and the cloud coverage. S314. Using a marine environmental information calculation model, process the central environmental information, the first environmental information set, and the second environmental information set to obtain the marine environmental information corresponding to the search and rescue case entity; S315. Combine the search and rescue case entity with the corresponding marine environmental information to obtain the corresponding extended search and rescue case entity.

7. The information extraction method for the maritime search and rescue field according to claim 5, characterized in that, The expression for the marine environmental information calculation model is: where ω1 and ω2 are preset first and second weights, respectively; V1, V2, D, H, R1, T, Y, and R2 are the first speed, the second speed, the mean direction, the mean wave height, the ice cover, the sea surface temperature, the precipitation, and the cloud cover of the marine environmental information, respectively; VA1 n , VA2 n , DA n , HA n , RA1 n , TA n , YA n , and RA2 n are the first speed, the second speed, the mean direction, the mean wave height, the ice cover, the sea surface temperature, the precipitation, and the cloud cover of the n-th local environmental information of the first environmental information set, respectively; VB1 n , VB2 n , DB n , HB n , RB1 n , TB n , YB n , and RB2 n are the first speed, the second speed, the mean direction, the mean wave height, the ice cover, the sea surface temperature, the precipitation, and the cloud cover of the n-th local environmental information of the second environmental information set, respectively. VC1, VC2, DC, HC, RC1, TC, YC, and RC2 are respectively the first velocity, the second velocity, the average direction, the average wave height, the ice coverage, the sea surface temperature, the precipitation, and the cloud coverage of the central environmental information.

8. An information extraction device for the field of maritime search and rescue, characterized in that, The device includes a data storage module, an information extraction module, and a map construction module; The data storage module is used to store a preset set of search and rescue case texts and a marine environmental spatiotemporal dataset; The information extraction module is used to perform semantic information extraction processing on the search and rescue case text set to obtain the search and rescue case set; The map construction module is used to construct a search and rescue case map based on the marine environment spatiotemporal dataset and the search and rescue case set.

9. An information extraction device for the field of maritime search and rescue, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the information extraction method for the maritime search and rescue field as described in any one of claims 1-7.

10. A computer-storable medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked, are used to execute the information extraction method for the maritime search and rescue field as described in any one of claims 1-7.

Images