Intelligent gateway for heterogeneous maritime affair broadcast protocol

Abstract

The invention discloses an intelligent gateway for a heterogeneous maritime affair broadcast protocol, which comprises a data access layer, an intelligent processing layer and a protocol output layer, and is characterized in that the intelligent processing layer comprises a natural language processing module and is configured to perform semantic analysis on input maritime affair safety information and extract key elements including time, places and event types; the content association engine is configured to establish an association relationship among multi-source information for the same maritime event based on the key elements; the multi-protocol converter is configured to generate broadcast signals conforming to the NAVDAT protocol, the meteorological fax protocol and the NAVTEX protocol according to the incidence relation; and a guidance instruction generator configured to generate a standardized guidance instruction associated across the protocol. According to the intelligent gateway for the heterogeneous maritime affair broadcast protocol, intelligent understanding and automatic association of maritime affair information can be achieved, the problem that multi-system fusion depends on manual operation is solved, and unified, efficient and enhanced information service is provided.

Description

Technical Field

[0001] This invention relates to a gateway, and more particularly to a smart gateway for heterogeneous maritime broadcasting protocols. Background Technology

[0002] With the development of maritime communication technology, a situation has arisen where various heterogeneous broadcasting systems coexist. The NAVTEX system broadcasts using text-encoded protocols on intermediate frequencies (MF) such as 518kHz; weather fax systems broadcast using analog image protocols on multiple high-frequency (HF) bands; and the NAVDAT system, as a new generation of digital broadcasting system, can broadcast at high speeds on intermediate frequencies (such as 500kHz) and multiple high-frequency bands. These systems differ significantly in protocol formats, data structures, and transmission characteristics, leading to information fragmentation, cumbersome operation, and wasted resources. Current technologies lack solutions capable of deeply understanding information content and achieving intelligent cross-protocol association. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide an intelligent gateway for heterogeneous maritime broadcasting protocols, which can realize intelligent understanding and automatic association of maritime information, solve the problem of multi-system integration relying on manual operation, and provide unified, efficient and enhanced information services.

[0004] To address the aforementioned technical problems, this invention provides an intelligent gateway for heterogeneous maritime broadcasting protocols, comprising a data access layer, an intelligent processing layer, and a protocol output layer. The data access layer is responsible for receiving and standardizing multi-source heterogeneous data, receiving text, images, and structured data from different information sources. The protocol output layer is responsible for signal modulation and transmission, ultimately generating and outputting broadcast signals conforming to the NAVDAT, weather fax, and NAVTEX target system standards. The intelligent processing layer includes: a natural language processing module configured to perform semantic analysis on input maritime safety information, extracting key elements including time, location, and event type; a content association engine configured to establish association relationships between information of different formats based on the key elements; a multi-protocol converter configured to generate broadcast signals conforming to the NAVDAT, weather fax, and NAVTEX protocols based on the association relationships; and a guidance instruction generator configured to generate standardized guidance instructions with cross-protocol associations.

[0005] Furthermore, the natural language processing module includes: an element recognition submodule, which uses a pre-trained language model to identify geographical entities, time entities, and event entities in the text; and a semantic understanding submodule, which uses a deep learning model based on an attention mechanism to establish logical relationships between key elements.

[0006] Furthermore, the element recognition submodule employs a Transformer-based pre-trained model to optimize the training of maritime terminology and expressions as follows: Data preparation: Constructing a maritime professional corpus, with sources including: IMO circulars, historical NAVTEX message databases, WMO meteorological coding data, navigation warnings, and maritime accident reports; Preprocessing: Cleaning and denoising the corpus, and standardizing coordinate formats and related expressions; Continued pre-training: Loading a general Chinese BERT model as a foundation, targeting a masked language model, and using maritime corpus to further pre-train the model; During pre-training, by predicting masked words, embedding maritime vocabulary, syntax, and knowledge into the parameters of the maritime corpus; Task fine-tuning: On the pre-trained model, using a maritime text dataset labeled with "time," "location," and "event type," supervised fine-tuning of the named entity recognition task is performed, mapping the input sequence to a structured entity label sequence.

[0007] Furthermore, the content association engine calculates the comprehensive association score of multi-source information using the following formula: Score = α × Ss + β × Sst + γ × St; Ss is semantic similarity, Sst is spatiotemporal overlap, St is timeliness consistency, and α, β, and γ are configurable weight parameters that satisfy α + β + γ = 1; the content association engine sets an association threshold θassoc; when 1 ≥ Score ≥ θassoc, the association relationship between the multi-source information is automatically established.

[0008] Furthermore, the semantic similarity is obtained by calculating cosine similarity using the TF-IDF algorithm combined with text vectors generated from a maritime dictionary. The spatiotemporal overlap is used to measure the degree of overlap between events geographically and temporally, where Sst = Sgeo × Stime, Sgeo is the spatial overlap, and Stime is the temporal overlap. The timeliness consistency is based on the current system time, calculating the freshness of the two pieces of information respectively, and taking the minimum value of the two freshnesses.

[0009] Furthermore, the time overlap is obtained by calculating the overlap ratio of the effective time windows of the two events, or by using the following time decay function: Stime=exp(-∣Δt∣ / τ), where Δt is the difference between the key time points and τ is the decay constant.

[0010] Furthermore, for point locations, the spatial overlap is first calculated by the great circle distance d, and then converted into a value between 0 and 1 using a Gaussian decay function. Sgeo =exp(-d 2 / (σ 2 )), where σ is the attenuation radius; for a region, the spatial overlap is used to calculate its intersection-union ratio.

[0011] Furthermore, the multi-protocol converter includes: a NAVDAT protocol adapter, which encapsulates structured data into digital data packets conforming to the ITU-RM.2012 standard, applies forward error correction coding, and generates an OFDM signal that can be broadcast in the MF / HF band; a weather fax protocol adapter, which converts associated images into analog fax signals conforming to WMO specifications and broadcasts them in the high-frequency band, including generating scan synchronization signals and image modulation; and a NAVTEX protocol adapter, which encodes structured text alarms according to the ITU-RM.2010 standard to generate an FSK modulated signal that is broadcast at an intermediate frequency of 518 kHz.

[0012] Furthermore, the guidance instruction generator generates a guidance instruction format that includes the target system, frequency, time, and unique event identifier; the guidance instruction is embedded in different protocols in the following ways: in NAVTEX messages, the guidance instruction is embedded in the message body; in NAVDAT signals, the guidance instruction is embedded as metadata in the file header; in weather fax signals, the guidance instruction is embedded in the non-core area of ​​the image in a machine-readable graphic encoding form.

[0013] Furthermore, the content association engine evaluates the confidence level of the automatically established association relationships; when the confidence level is lower than the threshold, a manual review process is triggered; the association algorithm parameters are optimized based on the review results; specifically, this includes: mapping the comprehensive association score Score to confidence level through a sigmoid function; or inputting the comprehensive association score Score, semantic similarity Ss, spatiotemporal overlap Sst, and timeliness consistency St into a pre-trained confidence evaluation model to obtain the confidence level; recording multiple sets of manual review results and their corresponding association feature vectors; adjusting the weight parameters α, β, and γ using gradient descent with the goal of minimizing the difference between the manual review results and the results predicted based on the association feature vectors; the objective function for minimizing the prediction error is the following cross-entropy loss function: L=-∑[T·log(Score)+(1-T)·log(1-Score)]; T=1 indicates that the manual review result is correct, and T=0 indicates that the manual review result is incorrect.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The intelligent gateway for heterogeneous maritime broadcasting protocols provided by the present invention can realize intelligent understanding and automatic association of maritime information, solve the problem of multi-system integration relying on manual operation, and provide unified, efficient and enhanced information services. Attached Figure Description

[0015] Figure 1 This is a diagram of the smart gateway architecture for heterogeneous maritime broadcasting protocols according to the present invention. Figure 2 Flowchart of content association processing for the smart gateway of this invention; Figure 3 This is a schematic diagram illustrating the operation of the natural language processing module of the present invention; Figure 4 This is a schematic diagram illustrating the generation of guidance instructions and the embedding of multiple protocols in this invention. Detailed Implementation

[0016] The present invention will now be further described with reference to the accompanying drawings and embodiments.

[0017] Figure 1 This is a diagram of the smart gateway architecture for heterogeneous maritime broadcasting protocols according to the present invention.

[0018] Please see Figure 1 The intelligent gateway for heterogeneous maritime broadcasting protocols provided by this invention is divided into three layers from top to bottom: data access layer, intelligent processing layer and protocol output layer; The data access layer is responsible for receiving and standardizing multi-source heterogeneous data, including text, images, and structured data from different information sources.

[0019] The intelligent processing layer is the core, comprising four modules: a natural language processing module (responsible for entity recognition and semantic understanding), a content association engine (responsible for calculating association scores and establishing mappings), a multi-protocol converter (containing adapters for NAVDAT, weather fax, and NAVTEX protocols), and a guidance instruction generator (responsible for generating unified association identifiers). The natural language processing module uses a Transformer-based pre-trained model optimized for the maritime domain; the content association engine uses a multimodal association algorithm to calculate association scores; the multi-protocol converter ensures that signals conform to the standards of each system; and the guidance instruction generator creates machine-readable unified identifiers.

[0020] The protocol output layer is responsible for signal modulation and transmission, and ultimately generates and outputs broadcast signals that conform to the three target system standards.

[0021] The intelligent gateway of this invention begins by receiving raw maritime information and then performs the following processing steps in sequence: semantic element extraction (identifying geographical, temporal, event, and degree entities) → multi-dimensional association analysis (calculating semantic, spatiotemporal, and timeliness similarity to obtain a comprehensive association score) → confidence assessment → determining whether the confidence score is higher than a threshold; if it is higher than the threshold, an association relationship is automatically established; if it is lower than the threshold, the process proceeds to manual review → generating multi-protocol signals (parallel generation of NAVDAT, weather fax, and NAVTEX signals) → embedding guidance instructions → collaborative broadcast control, at which point the process ends. Key steps include: 1. Semantic element extraction: Utilize NLP technology to deeply analyze text and construct a knowledge representation of information.

[0022] 2. Establishment of association: The association score is quantitatively evaluated using the formula α × semantic similarity + β × spatiotemporal overlap + γ × timeliness consistency, and the association is confirmed based on a threshold or manual review.

[0023] 3. Protocol Adaptive Conversion: Generate corresponding legal signals according to the technical specifications of each system (NAVDAT MF / HF, Weather Fax HF, NAVTEX MF).

[0024] 4. Instruction Embedding: Embed associated identifiers using appropriate methods (plaintext, metadata, graphic encoding) in different protocols.

[0025] Please continue reading Figure 2 The natural language processing module of this invention operates as follows: 1. Receive and parse raw information; The process begins by receiving raw text, images, or data from maritime, meteorological, and other departments to prepare data for subsequent intelligent processing.

[0026] 2. Intelligent understanding and associative decision-making; Step 1: Semantic element extraction - Use NLP technology to enable machines to "understand" the content and identify key entities such as events (e.g., typhoons), locations, times, and intensities.

[0027] Step 2: Multi-dimensional association analysis - Calculate the association degree between new information and existing information (such as cloud maps) based on semantics, spatiotemporal factors, and timeliness to achieve intelligent matching.

[0028] Step 3: Confidence Decision - The system automatically determines whether the association results are reliable.

[0029] High confidence level: Automatically establish associations; Low confidence level: Introduce manual review to ensure accuracy, while the system learns and optimizes the algorithm.

[0030] 3. Multi-protocol generation and embedding; The associated information is converted into protocol signals of the three major systems (NAVDAT, weather fax, and NAVTEX) in parallel and a unified association identifier is embedded; this achieves "process once, distribute multiple times" and embeds the "key" (association ID) for fusion display in each signal.

[0031] Figure 3 This is a schematic diagram illustrating the operation of the natural language processing module of the present invention.

[0032] This diagram uses a specific typhoon warning as an example to illustrate how a natural language processing module gradually "understands" human language: 1. Input raw text: The system receives the raw maritime safety information text; Example: "Typhoon 'Marie' will pass near 25N / 125E at 0800UTC, with a maximum wind speed at its center of 35m / s."

[0033] 2. Chinese word segmentation and part-of-speech tagging: This is a fundamental step in natural language processing; Function: To segment continuous sentences into meaningful lexical units (word segmentation) and determine the part of speech of each word (such as noun, verb, etc.).

[0034] Process: The example sentence is broken down into sequences such as ["typhoon", "Mary", "will", "at", "0800UTC", "at", "25N", " / ", "125E", "nearby", "passing through", "center", "maximum", "wind speed", "35m / s"] to prepare for subsequent analysis.

[0035] 3. Entity recognition and classification (core step); Function: Using a Transformer-based pre-trained model, it accurately identifies and classifies specific types of entities from segmented sequences.

[0036] Process: The model will identify and label: "Typhoon" → Event type; "Mary" → Name; "0800UTC" → Time; "25N / 125E" → Location; "Passed through" → action; "35m / s" → Intensity.

[0037] 4. Structured information output; The identified scattered entities are organized into a standardized format that is easy for machines to process and understand; the entities are filled into structured fields; the final output is a structured data object containing all the key information; this structured data object provides direct and explicit input to the content association engine in the next stage and is the cornerstone of intelligent association.

[0038] Figure 4 This is a schematic diagram illustrating the generation of guidance instructions and the embedding of multiple protocols in this invention.

[0039] This flowchart illustrates the core process of multi-protocol embedding in this invention: Step 1: Input event; The system receives information on maritime safety incidents, such as typhoon warnings; Each event is assigned a unique identifier, such as "TY2024082001" which represents Typhoon Warning No. 1 on August 20, 2024.

[0040] Step 2: Generate a unified identifier; The system creates a unified association identifier for events; • This identifier remains consistent across all protocols and is key to enabling cross-system information association; • The identifier uses a standardized format and includes the event type, date, and sequence number.

[0041] Step 3: Multi-protocol embedding; The unified identifier is embedded in the three major maritime communication systems: NAVTEX: Embeds identifiers and instructions in plaintext within text messages; • NAVDAT: Embeds identifiers in the metadata of digital data files; • Weather Fax: An identifier is embedded in the image edge as a QR code.

[0042] Step 4: Shipborne terminal fusion display; • The shipborne fusion terminal receives signals from the three major systems; • Automatically extracts unified identifiers to associate information of different formats (text, images, data) related to events; • Integrated display within a unified interface, providing crew members with complete safety situational awareness.

[0043] As can be seen, this invention solves the problem of protocol incompatibility in heterogeneous systems by using a unified associated identifier, thereby achieving intelligent integration of the three major maritime communication systems.

[0044] The intelligent gateway of this invention can be deployed on standard server hardware (such as industrial servers equipped with GPU accelerators) and run the Linux operating system. The core software modules can be written in Python and developed using deep learning frameworks such as PyTorch or TensorFlow.

[0045] The natural language processing module of this invention employs a pre-trained language model based on the Transformer architecture, preferably a bidirectional encoder model (such as BERT). To address the problem of general models' insufficient understanding of maritime terminology (such as "poor visibility," "narrow channel," and "tropical cyclone"), maritime-domain optimized training can be performed. This process includes: 1. Data preparation: Construct a professional corpus in the maritime field. The corpus sources include: International Maritime Organization (IMO) circulars, historical NAVTEX message database, World Meteorological Organization (WMO) weather coding data, navigation warnings, maritime accident reports, etc.

[0046] 2. Preprocessing: Clean and denoise the corpus, and standardize key expressions (e.g., unify various coordinate formats to "DD-MM.MM N / S, DDD-MM.MM E / W").

[0047] 3. Continue pre-training: Using the general Chinese BERT model as a base, and targeting the Masked Language Model (MLM), the model is further pre-trained using the aforementioned maritime corpus. During this process, the model embeds maritime vocabulary, syntax, and knowledge into its parameters by predicting masked words in phrases such as "[MASK] central air pressure is 925 hPa".

[0048] 4. Task Fine-tuning: To further improve the accuracy of feature extraction, supervised fine-tuning of the named entity recognition task can be performed on the pre-trained model using a maritime text dataset labeled with entity tags such as "time," "location," and "event type." After fine-tuning, the model can directly map the input sequence to a structured sequence of entity tags.

[0049] Please refer to the above. Figure 3 Taking the input text "Typhoon 'Marie' will pass near 25N / 125E at 0800UTC" as an example, the workflow and output of this invention are as follows: The text is first segmented and tagged with parts of speech; The processed sequence is input into the optimized BERT model described above; The model's feature recognition submodule (usually the classification layer at the top of the model) identifies and labels entities such as "Typhoon" (event type), "Mary" (name), "0800UTC" (time), and "25N / 125E" (location); The semantic understanding submodule of the model (through its internal attention mechanism) analyzes and establishes the logical relationships between these entities, such as "Mary" being an instance of "Typhoon", and "0800UTC" and "25N / 125E" being the "time" and "location" of the event, respectively; the final output is a structured knowledge representation.

[0050] For example, a JSON object: {"event_type":"Typhoon","name":"Mary","time":"0800UTC","location":"25N / 125E"}.

[0051] The content association engine of this invention is the core of the system's "intelligent decision-making," responsible for quantitatively evaluating and establishing the association relationships between information in different formats (text, images, and data).

[0052] 1. Multi-dimensional similarity calculation; The correlation assessment is based on the following three quantifiable dimensions: Semantic similarity (Ss): Measures the degree to which two pieces of information match in describing a specific topic. Multiple implementation methods are available. Method 1 (based on word frequency): Use the TF-IDF algorithm combined with a maritime dictionary to generate text vectors and calculate cosine similarity.

[0053] Method 2 (Deep Learning-Based Optimization): Input the text descriptions of the two pieces of information into the optimization model to obtain sentence vectors (such as vectors labeled [CLS]), and calculate the cosine similarity. For example, the sentence vector similarity between a typhoon warning and a satellite cloud image analysis report can reach 0.85.

[0054] Spatiotemporal overlap (Sst): measures the degree of overlap between events geographically and temporally.

[0055] Spatial overlap: For point locations (such as the center of a typhoon), calculate the great circle distance d (the great circle distance is the shortest actual distance between two points on the Earth's surface), and convert it to a value between 0 and 1 using a Gaussian decay function. Sgeo =exp(-d 2 / (σ 2 )), where σ is the attenuation radius (e.g., 300 km). For regions (e.g., typhoon path polygons), calculate their intersection-union ratio.

[0056] Here, "point location" refers to the location of an event represented by a latitude and longitude coordinate point. For example, the location of the typhoon center is 25.1°N, 125.3°E; the location of the accident report is 34.5°S, 58.0°W. When calculating the great circle distance, d refers to the fact that when the content association engine needs to determine whether two pieces of information (e.g., a typhoon warning and a satellite cloud image report) describe the same event, it will extract the location coordinates (latitude and longitude) of the two events and then calculate the shortest actual distance (kilometers or nautical miles) between the two coordinate points on the Earth's surface.

[0057] Time overlap: Calculate the overlap ratio of the effective time windows of two events, or use the time decay function: Stime=exp(-∣Δt∣ / τ), where Δt is the difference between the key time points and τ is the decay constant (e.g., 6 hours).

[0058] The overall spatiotemporal overlap can be expressed as: Sst = Sgeo × Stime.

[0059] Timeliness consistency (St): measures whether information is within the same decision-making cycle, rather than simply being close in time. For example, real-time observations and outdated forecasts have low timeliness consistency.

[0060] Calculation method: Based on the current system time, calculate the "freshness" of both pieces of information separately, for example, Freshness=max(0, 1-age / expiration period). The timeliness consistency St can be taken as the minimum value of the two freshness values.

[0061] 2. Comprehensive correlation scoring and weighting configuration; The final association score is calculated using a linear weighted formula: Score=α×Ss+β×Sst+γ×St; Here, α, β, γ are configurable weight parameters, and they usually satisfy α+β+γ=1.

[0062] Weight determination strategy: Preset based on experience: This is set according to different event types. For example, for rapidly changing events such as typhoons, emphasis is placed on time, space, and timeliness: α=0.2, β=0.5, γ=0.3; for steady-state information such as NOTAMs, emphasis is placed on semantics: α=0.6, β=0.3, γ=0.1.

[0063] Data-driven optimization: Collect historical correct / incorrect association data as a training set, use human judgment as labels, and optimize α, β, and γ using gradient descent to achieve the highest classification accuracy of the Score formula.

[0064] Dynamic adjustment: The information source can be dynamically adjusted based on its authority, such as increasing the alpha weight of information from authoritative institutions.

[0065] 3. Related decision-making; Set an association threshold θassoc (e.g., 0.7). When 1 ≥ Score ≥ θassoc, the system automatically establishes an association; otherwise, the association is considered invalid or the process proceeds to the verification stage.

[0066] The multi-protocol converter of the present invention comprises three independent physical or logical adapters.

[0067] NAVTEX Protocol Adapter: Encodes structured text alerts according to the ITU-RM.2010 standard and generates FSK modulated signals for broadcast at 518kHz intermediate frequency.

[0068] Weather facsimile protocol adapter: Converts associated images (such as cloud images) into analog facsimile signals compliant with WMO specifications for broadcasting in the high-frequency band, including generating scan synchronization signals and image modulation.

[0069] NAVDAT Protocol Adapter: Encapsulates structured data (such as typhoon path grid data) into digital data packets conforming to the ITU-RM.2012 standard, applies forward error correction coding, and generates OFDM signals that can be broadcast in the MF / HF bands.

[0070] Guidance Instruction Generation and Embedding: Generate a globally unique identifier (such as TY2024082001) for each confirmed associated event cluster and embed it into each signal in a protocol-compatible manner, such as Figure 4 as shown: NAVTEX: Add a plaintext instruction at the end of the message body, such as REF:FAX8460kHzID:TY2024082001.

[0071] NAVDAT: Add a dedicated field to the metadata header (such as JSON or XML header) of the data file.

[0072] Weather Facsimile: Embed the identifier in a machine-readable graphic code (such as a QR code) in the non-information area at the edge of the image (such as the lower right corner).

[0073] To ensure reliability and achieve continuous improvement, the present invention also introduces the following human-machine collaborative verification and associated verification and parameter self-optimization closed-loop process.

[0074] 1. Confidence Evaluation; For each automatic association decision, the system will evaluate its confidence C (0-1). Provide multiple evaluation methods: Method 1 (Based on Score Mapping): Directly map the association score Score to the confidence through the Sigmoid function.

[0075] Method 2 (Based on Independent Model - Preferred): Train a binary classification model (such as logistic regression). This model uses features such as Ss, Sst, St, and data source credibility, and uses historical manual review results as labels for training. The output probability is the confidence C.

[0076] 2. Manual Review Trigger and Threshold Management; Set double thresholds for decision diversion: High confidence threshold θhigh = 0.85: If 1≥C≥θhigh, automatically confirm the association.

[0077] Low confidence threshold θlow = 0.60: If 0≤C≤θlow, automatically reject or must be manually reviewed.

[0078] When θlow < C < θhigh, the system triggers the manual review process and pushes the association details to the operator for final adjudication. The thresholds can be dynamically adjusted according to the event risk.

[0079] The association threshold θ_assoc is used to determine whether there is an association relationship between information, while the confidence threshold θ_high is used to determine whether the association decision is reliable enough for automatic dissemination.

[0080] 3. Parameter Self-Optimization (System Learning); The system uses the results of manual review to evolve itself. Feedback collection: Record the final decision T (1=correct, 0=incorrect) and the corresponding feature vector F (including Ss, Sst, St, etc.) for each manual review.

[0081] Optimization trigger: When the feedback data accumulates to a certain scale (e.g., N=100), the optimization task will be started automatically.

[0082] Optimization execution: With the goal of minimizing prediction error, for example, the cross-entropy loss function L=-∑[T·log(Score)+(1-T)·log(1-Score)] is used, and the weight parameters α, β, γ in the correlation formula are automatically updated using gradient descent.

[0083] Model update: Seamlessly update the optimized parameters to the online system, making the associated decisions increasingly accurate.

[0084] Taking the handling of Typhoon Haiyan warning as an example, the above process is demonstrated in detail: Input and Understanding: Input text alerts, and the NLP module parses out structured elements.

[0085] Intelligent Association: The content association engine calculates the similarity between the alert and a satellite cloud image in the database: Ss=0.88 (calculated via BERT), Sst=0.90 (the path and the cloud image area highly overlap), St=1.0 (both are the latest data). Using typhoon weights (0.3, 0.5, 0.2), the score is 0.3×0.88+0.5×0.90+0.2×1.0=0.914, exceeding the threshold of 0.7, thus triggering automatic association.

[0086] Verification: The confidence model evaluation yielded C=0.92>0.85, which is automatically confirmed.

[0087] Generation and Embedding: Three protocol signals are generated in parallel and embedded with the associated IDTY2024082001 respectively.

[0088] Terminal converged display: After receiving the signal, the shipborne terminal automatically associates and displays text alarms, cloud maps and refined forecast data on the same screen using a unified ID.

[0089] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications and improvements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be defined by the claims.

Claims

1. A smart gateway for heterogeneous maritime broadcast protocols, characterized in that, It includes a data access layer, an intelligent processing layer, and a protocol output layer; the data access layer is responsible for receiving and standardizing multi-source heterogeneous data, receiving text, images, and structured data from different information sources; the protocol output layer is responsible for signal modulation and transmission, and finally generates and outputs broadcast signals that conform to the three target system standards of NAVDAT, weather fax, and NAVTEX. The intelligent processing layer includes: The natural language processing module is configured to perform semantic analysis on the input maritime safety information and extract key elements including time, location, and event type. The content association engine is configured to establish association relationships between multi-source information on the same maritime event based on the aforementioned key elements. A multi-protocol converter configured to generate broadcast signals conforming to NAVDAT, weather fax, and NAVTEX protocols based on the aforementioned association; Instruction generator, configured to generate standardized instruction commands that are associated across protocols.

2. The smart gateway according to claim 1, characterized in that, The natural language processing module includes: The feature recognition submodule uses a pre-trained language model to identify geographic entities, time entities, and event entities in text. The semantic understanding submodule uses a deep learning model based on an attention mechanism to establish logical relationships between key elements.

3. The smart gateway according to claim 2, characterized in that, The element recognition submodule uses a Transformer-based pre-trained model to optimize the training of maritime terminology and expressions as follows: Data preparation: Construct a professional corpus in the maritime field. The corpus sources include: International Maritime Organization circulars, historical NAVTEX message database, World Meteorological Organization weather coding data, navigation warnings and maritime accident reports; Preprocessing: The corpus is cleaned and denoised, and expressions related to coordinate formats and technical terms are standardized. Continue pre-training: Load the general Chinese BERT model as the base, take the masked language model as the target, and use the maritime corpus to continue pre-training the model; and during the pre-training process, by predicting the masked words, the vocabulary, syntax and knowledge of the maritime domain are embedded into the parameters of the maritime corpus. Task fine-tuning: On the pre-trained model, supervised fine-tuning of the named entity recognition task is performed using a maritime text dataset labeled with "time", "location" and "event type" to map the input sequence into a structured entity label sequence.

4. The smart gateway according to claim 1, characterized in that, The content association engine uses the following formula to calculate the comprehensive association score of multi-source information: Score=α×Ss+β×Sst+γ×St; Ss is semantic similarity, Sst is spatiotemporal overlap, St is timeliness consistency, and α, β, and γ are configurable weight parameters that satisfy α+β+γ=1. The content association engine sets an association threshold θassoc; when 1≥Score≥θassoc, the association relationship between the multi-source information is automatically established.

5. The smart gateway according to claim 4, characterized in that, The semantic similarity is obtained by calculating cosine similarity using the TF-IDF algorithm combined with text vectors generated by the maritime dictionary. The spatiotemporal overlap is used to measure the degree of overlap between events geographically and temporally, where Sst = Sgeo × Stime, Sgeo is the spatial overlap, and Stime is the temporal overlap. The timeliness consistency is based on the current system time, and the freshness of the two pieces of information is calculated separately, and the minimum value of the two freshnesses is taken.

6. The smart gateway according to claim 5, characterized in that, The time overlap is obtained by calculating the overlap ratio of the effective time windows of the two events, or by using the following time decay function: Stime=exp(-∣Δt∣ / τ), where Δt is the difference between the key time points and τ is the decay constant.

7. The smart gateway according to claim 5, characterized in that, For point locations, the spatial overlap is first calculated by the great circle distance d, and then converted to a value between 0 and 1 using a Gaussian decay function. Sgeo =exp(-d 2 / (σ 2 )), where σ is the attenuation radius; for a region, the spatial overlap is used to calculate its intersection-union ratio.

8. The smart gateway according to claim 1, characterized in that, The multi-protocol converter includes: The NAVDAT protocol adapter encapsulates structured data into digital data packets conforming to the ITU-RM.2012 standard, applies forward error correction coding, and generates OFDM signals that can be broadcast in the MF / HF bands. The weather facsimile protocol adapter converts associated images into analog facsimile signals compliant with WMO specifications for broadcasting in the high-frequency band, including generating scan synchronization signals and image modulation; The NAVTEX protocol adapter encodes structured text alerts according to the ITU-RM.2010 standard and generates an FSK modulated signal for broadcast at 518kHz intermediate frequency.

9. The smart gateway according to claim 1, characterized in that, The guidance instruction generator generates guidance instructions in a format that includes the target system, frequency, time, and a unique event identifier; the guidance instructions are embedded in different protocols in the following manner: In NAVTEX messages, the instruction is embedded in the message body; In the NAVDAT signal, the instruction is embedded as metadata in the file header; In weather fax signals, guidance instructions are embedded in non-core areas of the image in a machine-readable graphic encoding format.

10. The smart gateway according to claim 4, characterized in that, The content association engine evaluates the confidence level of automatically established associations; when the confidence level is lower than the threshold, a manual review process is triggered. Optimize the correlation algorithm parameters based on the audit results; specifically including: The comprehensive association score (Score) is mapped to confidence level using a sigmoid function; or the comprehensive association score (Score), semantic similarity (Ss), spatiotemporal overlap (Sst), and timeliness consistency (St) are input into a pre-trained confidence evaluation model to obtain the confidence level. Record multiple sets of manual review results and their corresponding associated feature vectors; With the goal of minimizing the difference between the manual review results and the results predicted based on the associated feature vectors, the weight parameters α, β, and γ are adjusted using the gradient descent method. The objective function for minimizing the prediction error is the following cross-entropy loss function: L = -∑[T·log(Score) + (1-T)·log(1-Score)]; T = 1 indicates that the manual review result is correct, and T = 0 indicates that the manual review result is incorrect.

Images