Ship encounter semantic template matching method

Abstract

This invention belongs to the field of data processing technology, specifically relating to a method for matching semantic templates for ship encounters. The method includes: acquiring ship encounter intention and encounter semantic constraint information; constructing encounter semantic primitives based on the ship encounter intention, and generating understanding constraints to limit correct understanding and prohibited understanding based on the encounter semantic primitives; determining candidate visual semantic templates associated with the encounter semantic primitives from a visual semantic template library; constructing template matching constraints based on the understanding constraints and encounter semantic constraint information, wherein the template matching constraints are used to constrain the misunderstanding suppression relationship, visual differentiation relationship, and output adaptation relationship of the candidate visual semantic templates; determining the target visual semantic template based on the template matching constraints, and generating a visual communication output sequence to output a visual semantic signal. This invention enables the template selection process to simultaneously consider misunderstanding suppression, visual differentiation, and output adaptation relationships.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method for semantic template matching of ship encounters. Background Technology

[0002] During vessel encounters, visual communication is typically used to convey semantic information such as avoidance, warnings, confirmations, and requests for assistance in situations involving close proximity, complex waterways, port operations, or insufficient radio communication. Related technologies usually employ a combination of pre-set messages, fixed light rhythms, icon prompts, or directional indicators for output. The processing largely relies on the message type selected by the crew or a single display template, directly sending preset text, symbols, or flashing patterns to the target vessel. While this method can provide basic prompts, when encounter semantics are influenced by directional references, the recipient, urgency, and prior communication status, fixed templates struggle to distinguish between the correct understanding the recipient should form and the erroneous understanding that needs to be rejected.

[0003] Especially in scenarios where semantics are similar but the consequences of actions differ, such as "this vessel passes by on its starboard side," "the target vessel passes by on its starboard side," "stop approaching," and "maintain distance," relying solely on text or icon similarity to select templates can easily overlook factors such as directional reference rereading, residual preceding prompts, unbound receiving objects, and excessive visual load in low visibility conditions. The technical reason for this is that existing template selection processes typically lack unified modeling for correct understanding, prohibited understanding, easily confused semantic relationships, and output environment constraints. Candidate templates are only called according to preset categories or static mapping relationships, failing to jointly constrain misunderstanding suppression, visual differentiation, and output adaptation within the same processing chain. This can lead to unstable candidate template recall, template ranking lacking semantic exclusion criteria, and misunderstanding discrepancies between the output signal and the vessel's encounter intention. Summary of the Invention

[0004] This invention provides a semantic template matching method for ship encounters, which at least solves the problem of how to perform semantically constrained matching of ship encounter intentions in visual communication of ship encounters and output visual semantic signals that are compatible with the correct understanding of the receiver.

[0005] This invention provides a method for semantic template matching of ship encounters, the method comprising: Obtain information on the ship's encounter intentions and semantic constraints. Encounter semantic primitives are constructed based on the ship's encounter intention, and understanding constraints are generated based on the encounter semantic primitives to limit correct understanding and prohibition of understanding. Identify candidate visual semantic templates associated with encounter semantic primitives from the visual semantic template library; Template matching constraints are constructed based on understanding constraints and encounter semantic constraints. These constraints are used to constrain the misunderstanding suppression relationship, visual distinction relationship, and output adaptation relationship of candidate visual semantic templates. The target visual semantic template is determined from the candidate visual semantic templates based on the template matching constraint, and a visual communication output sequence is generated based on the target visual semantic template, outputting a visual semantic signal that matches the ship's meeting intention.

[0006] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows: By constructing the ship's encounter intention as an encounter semantic primitive, the association and organization of communication objects, risk types, expected actions, direction references, urgency levels, and prohibited misunderstandings are realized in the same structured object, so that subsequent template matching no longer depends on a single text or a single icon call.

[0007] By generating understanding constraints based on the semantic primitives of the encounter to limit correct understanding and prohibited understanding, the appropriate action understanding and the incorrect understanding that the receiver should form can be included in the template selection criteria at the same time, so that the candidate visual semantic templates have clear semantic boundaries before entering the matching stage.

[0008] By constructing template matching constraints based on understanding constraints and encounter semantic constraints, the misunderstanding suppression relationship, visual differentiation relationship and output adaptation relationship jointly constrain the screening and sorting of candidate visual semantic templates, which can reduce the matching bias caused by calling templates based solely on preset categories or single similarity.

[0009] By generating visual communication output sequences based on target visual semantic templates, object binding signals, semantic expression signals, and repetitive output control can be generated according to the same matching result, thereby ensuring that the output visual semantic signals are consistent with the ship's encounter intention, target direction, and recognition conditions. Attached Figure Description

[0010] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a schematic diagram illustrating the generation of semantic constraint information and constraint quantities in an embodiment of the present invention; Figure 3 This is a schematic diagram of the template matching constraint matrix in an embodiment of the present invention; Figure 4 This is a diagram showing the comprehensive matching score and filtering results of candidate visual semantic templates in an embodiment of the present invention; Figure 5 This is a timing diagram of the visual communication output sequence in an embodiment of the present invention. Detailed Implementation

[0011] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of one or more embodiments. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this invention.

[0012] Semantic template matching refers to the process of converting the business semantics to be expressed into a structured semantic object that can be computed and processed, and then searching for an expression template that matches the semantic object in a preset template set. Unlike simple keyword matching or fixed template invocation, semantic template matching not only focuses on the surface correspondence between the input intent and the template text, but also needs to consider the compositional relationship between semantic fields, the possible understanding boundaries formed by the recipient, the confusion relationship between adjacent semantics, and the environmental constraints when the template is output.

[0013] For visual communication of ship encounters, the content to be expressed often involves the communication object, risk scenario, expected action, directional reference, and urgency level simultaneously. The same text or directional symbol may be interpreted differently under different reference systems. Therefore, transforming the ship's encounter intention into encounter semantic primitives and matching visual semantic templates using understanding constraints and encounter semantic constraint information can change the template selection process from "finding displayable content" to "selecting visual expression content that supports correct understanding and eliminates erroneous understanding." Based on this processing approach, this invention establishes a processing chain around ship encounter semantic template matching, from constructing encounter semantic primitives, generating understanding constraints, screening candidate visual semantic templates, constructing template matching constraints, to generating visual communication output sequences.

[0014] like Figure 1 As shown, a semantic template matching method for ship encounters includes the following processing steps.

[0015] Marine visual communication equipment receives the ship's encounter intention input by the crew and simultaneously acquires the semantic constraint information of the encounter. The ship's encounter intention indicates the communication content that this visual communication aims to convey, such as distress, warning, avoidance, maintaining distance, or confirmation, and can be formed by button selection, touch selection, voice recognition results, or shipboard alarm trigger results.

[0016] Encounter semantic constraint information is used to limit the directional conditions, recognition conditions, and preceding semantic environment of this template matching. The processor transmits the ship encounter intention to the encounter semantic primitive construction stage and the encounter semantic constraint information to the template matching constraint construction stage, so that the selection of subsequent candidate visual semantic templates is simultaneously influenced and constrained by the sending intention, target direction, recognition conditions, and preceding output.

[0017] In one embodiment, the processor in the marine visual communication device organizes encounter semantic constraint information into data records bound to the current visual communication task. The encounter semantic constraint information consists of target direction information, identification condition level, and encounter semantic state. Each component enters a different matching constraint generation path and collectively restricts the scope of use of candidate visual semantic templates in the subsequent template matching stage.

[0018] Target direction information is used to describe the directional basis when a visual semantic signal points to a target vessel. Target direction information can come from manual pointing by the crew, equipment attitude sensors, turntable angle feedback, electronic compass heading, radar bearing, or shipboard integrated navigation data. When receiving multiple direction sources, the processor can select the appropriate source according to the factory-defined priority of the equipment. The priority of direction sources can be set based on the stability of direction data, acquisition latency, and equipment access reliability. For example, fixed-installation equipment can prioritize turntable angle feedback, while handheld equipment can prioritize the directional basis formed by the equipment attitude sensor and the crew's continuous pointing time.

[0019] After acquiring the target direction information, the processor converts it into a direction lock level, direction holding time, and direction jitter, which can be used for subsequent calculations. The direction lock level indicates whether the target direction remains consistent over a short period; the direction holding time indicates the duration of the pointing action; and the direction jitter indicates the magnitude of the direction deviation caused by ship rolling, equipment vibration, or changes in target bearing. These parameters do not directly determine the output but are incorporated into the subsequent process of determining the credibility of the communication object binding, used to judge whether the target vessel has been effectively pointed to in this visual communication. If target direction information is missing, the processor can set the direction lock level to a low level and restrict the participation of subsequent complex direction-related templates in matching.

[0020] The recognition condition settings describe the identifiable environment for the target crew member when receiving visual semantic signals. These settings can be selected by the crew member through the operating interface or generated by the ambient light sensor, visibility input, navigation area mode, or nighttime operating mode. Recognition condition settings can be categorized into types such as near distance, long distance, complex background light, low visibility due to rain or fog, nighttime, and non-professional receivers. The settings can be based on equipment factory calibration, historical usage records, or vessel operator configuration files. The processor transmits the recognition condition settings to the recognition load budget generation process to limit the amount of text, the number of icons, the complexity of directional symbols, and the complexity of flashing rhythms. When environmental data is unavailable, the processor uses a conservative setting, reducing long text content in subsequent template matching and prioritizing the preservation of symbol and rhythmic expressions.

[0021] The encounter semantic state records the visual semantic output environment that has occurred prior to the current output. The encounter semantic state can store preceding encounter semantic primitives, preceding output time, preceding output template type, and preceding output end marker. Preceding encounter semantic primitives represent the meaning carried by one or more previous visual semantic signals. The preceding output time is used to determine whether the preceding meaning may still affect the target crew's understanding. The preceding output template type is used to identify differences in states such as ordinary communication, avoidance negotiation, danger warning, or distress call. When the device is first started or has not outputted visual semantic signals for a long time, the processor sets the encounter semantic state to an empty state. After completing a visual semantic signal output, the processor updates the encounter semantic state and writes it to the time record. The encounter semantic state then enters the process of determining the preceding semantic residual, used to constrain the conflict relationship between subsequent candidate visual semantic templates and the preceding output meaning.

[0022] After receiving a ship's intention to meet, the ship's visual communication equipment uses a processor to convert the communication content generated by the crew's actions into meeting semantic primitives. Meeting semantic primitives are structured semantic objects oriented towards template matching, used to carry the recipient, risk scenario, expected action, direction reference, priority, and misunderstanding exclusion boundaries in a ship meeting communication.

[0023] After the processor completes the construction of the encounter semantic primitives, it extracts content from these primitives that can limit the receiver's understanding direction, generating understanding constraints. Understanding constraints are formed jointly by a set of correct understandings and a set of prohibited understandings. The set of correct understandings is used to limit the action understanding that the target vessel should form, while the set of prohibited understandings is used to limit the erroneous understandings that the target vessel should not form. Understanding constraints then proceed to the subsequent template matching constraint construction stage, used to constrain the misunderstanding suppression relationships of candidate visual semantic templates.

[0024] In one embodiment, the processor generates encounter semantic primitives based on the communication purpose, action instructions, and encounter background in the vessel's encounter intention. The vessel's encounter intention can originate from encounter messages selected by the crew in the operating interface, or from shortcut keys, voice-recognized text, shipboard collision avoidance alarms, or manually entered text. After receiving the vessel's encounter intention, the processor performs normalization processing, merging synonymous expressions into the same category of vessel encounter intentions. For example, "Please keep away," "Maintain a safe distance," and "Do not approach" can be merged into the "Maintain distance" category; "Danger," "Stop approaching," and "Immediately avoid" can be merged into the "Danger warning" category. Normalization processing can be jointly completed by a preset intention vocabulary and manual confirmation records. The preset intention vocabulary is configured by the equipment manufacturer or ship operator according to commonly used encounter expressions, and the manual confirmation records are used to store the crew's historical selection results for the same expression.

[0025] The processor determines the communication target, risk type, expected action, direction reference, urgency level, and prohibited misunderstandings based on the standardized vessel encounter intent. The communication target refers to the recipient indicated by the visual semantic signal, which can be generated from target direction information, the target record selected by the crew, or the default target of the current encounter scenario. The risk type indicates the nature of the scenario involved in this communication, which can be categorized as too close, crossing risk, collision risk, channel occupancy, avoidance negotiation, distress request, or general confirmation. The expected action indicates the behavior desired from the target vessel, which can be categorized as stopping approach, maintaining distance, waiting, changing course, confirming receipt, or requesting assistance. The direction reference indicates the reference system used for directional intents, which can be categorized as the vessel's port side, the vessel's starboard side, the target vessel's port side, the target vessel's starboard side, the target bearing, or the channel lateral direction.

[0026] The urgency level is used to prioritize the semantics of this encounter in subsequent matching and output, and can be determined based on the risk type, crew triggering method, and alarm source. Danger warnings and distress requests have a higher urgency level than ordinary confirmations and avoidance negotiations; ship encounter intentions generated by shortcut triggers have a higher priority than those generated by menu browsing selections; ship encounter intentions triggered by onboard collision avoidance alarms can have their urgency level set according to the alarm level. Prohibited misunderstandings are used to record misunderstandings that the recipient should not form, and their source is a pre-configured encounter misunderstanding rule table. The encounter misunderstanding rule table records the correspondence between risk type, expected action, directional reference, and common misunderstandings. For example, the prohibited misunderstanding for "stop approaching" could be "allow approaching" or "ordinary greeting"; the prohibited misunderstanding for "this vessel passes on its starboard side" could be "the target vessel passes on its starboard side" or "passes in the opposite direction".

[0027] The processor writes the communication object, risk type, expected action, direction reference, urgency level, and prohibited misunderstandings into the same encounter semantic primitive record. Each field is stored using an associative storage method; any change in any field triggers the regeneration of the encounter semantic primitive, preventing the direction reference and expected action from becoming disconnected. If the ship's encounter intention lacks a direction reference, the processor records the direction reference as an empty reference and prohibits direction-type templates from directly participating in matching in subsequent stages. If the communication object cannot be determined, the processor records the communication object as a ship within the target direction and requires subsequent stages to further constrain the output through communication object binding credibility. The completed encounter semantic primitive enters the understanding constraint generation stage and serves as the semantic basis for selecting candidate visual semantic templates.

[0028] In one embodiment, when the processor generates understanding constraints based on encounter semantic primitives, it does not directly use the ship's encounter intention as the displayed content. Instead, it breaks down the encounter semantic primitives into correct understanding items and prohibited understanding items. Correct understanding items represent the understanding that the target ship should form after receiving the visual semantic signal, while prohibited understanding items represent the understanding that the target ship should not form after receiving the visual semantic signal. Both correct and prohibited understanding items are stored as structured entries. Each entry records the action meaning, direction reference, urgency level, and source field, enabling subsequent template matching to distinguish encounter semantics with similar text but different action meanings.

[0029] The correct understanding item is determined by the expected action, direction reference, and urgency level. The expected action provides the basic behavior that the target vessel should perform, the direction reference defines the reference object for directional behaviors, and the urgency level determines the priority of the correct understanding item in subsequent template matching. For vessel encounter intentions that do not involve direction, the processor can generate correct understanding items based on the expected action and urgency level. For example, for intentions related to maintaining distance, the correct understanding item "the target vessel stops approaching" can be generated. For vessel encounter intentions that involve direction, the processor must write the direction reference into the correct understanding item. For example, for intentions related to passing on the starboard side, "the starboard side of the vessel" needs to be bound to "passing action" to prevent the target crew from interpreting the direction as their own starboard side if the subsequent template only displays an arrow.

[0030] Prohibited interpretations are determined jointly by risk type, direction reference, and prohibited misunderstandings. Risk type defines the boundaries of scenarios prone to misunderstanding, direction reference identifies directional misreading or subject confusion, and prohibited misunderstandings directly list interpretations that need to be excluded. For example, if the risk type is collision risk and the expected action is to stop approaching, the processor can generate prohibited interpretations such as "allow approach," "general prompt," or "no action required." If the risk type is avoidance negotiation and the direction reference is the starboard side of the vessel, the processor can generate prohibited interpretations such as "target vessel passes on starboard side," "passes in the opposite direction," and "passes in unknown direction." Prohibited interpretations are not displayed externally but are incorporated into the subsequent misunderstanding suppression relationship determination process to filter out candidate visual semantic templates that are prone to causing misunderstandings.

[0031] The processor can generate a set of correct interpretations and a set of prohibited interpretations from the encounter semantic rule table. The encounter semantic rule table is pre-configured by the equipment manufacturer or ship operator, recording the mapping relationship between risk type, expected action, direction reference, urgency level, and correct and prohibited interpretations. The setup of the encounter semantic rule table can be based on ship collision avoidance operation habits, maritime visual communication experience, and the visual expression capabilities supported by the equipment. After querying the encounter semantic rule table, the processor merges duplicate entries and prioritizes conflicting entries. If the same encounter semantic primitive simultaneously generates both a general confirmation type correct interpretation and a danger warning type correct interpretation, the processor retains the correct interpretation with higher urgency and moves lower-priority entries that may cause misjudgment to the prohibited interpretation set or deletes them from the current match.

[0032] After the correct understanding set and the prohibited understanding set are generated, the processor uses both to determine the understanding constraints. These constraints are written into the data record of this visual communication task and passed to the template matching constraint construction stage. If the correct understanding set is empty, the processor does not enter the template matching stage but instead requires a reselection of the ship encounter intent or invocation of the default intent of the safety prompt class. If the prohibited understanding set is empty, the processor retains the correct understanding set and continues subsequent processing, while recording the prohibited understanding set as an empty set. Through this process, the ship encounter intent is transformed into a semantic boundary that can participate in misunderstanding suppression and template matching, allowing subsequent candidate visual semantic templates to be filtered around the correct and prohibited understandings.

[0033] After constructing the encounter semantic primitives, the processor accesses the visual semantic template library and filters out visual semantic templates that have a semantic mapping relationship with the encounter semantic primitives, forming candidate visual semantic templates. The visual semantic template library stores combinations of text, icons, directional cues, and flashing rhythms that the device can output. Each visual semantic template has a matching semantic field. The processor compares the communication object, risk type, expected action, directional reference, urgency level, and prohibited misunderstanding items in the encounter semantic primitives with the semantic fields in the template library. Visual semantic templates that match at least one semantic field and are not excluded by the basic disabling rules are written into the candidate visual semantic template set. This candidate visual semantic template set is passed to the template matching constraint construction stage for further processing such as misunderstanding suppression, visual differentiation, and output adaptation.

[0034] In one embodiment, the visual semantic template library is stored in a processor-accessible non-volatile storage area, or it can be loaded into the runtime memory from a higher-level configuration file by the shipboard visual communication device at startup. The visual semantic template library stores multiple visual semantic templates, each of which is a composite expression unit, rather than a single string or icon. The text layer stores directly displayable phrases, abbreviations, or fixed prompts; the icon layer stores visual symbols such as warnings, distress calls, confirmations, distance maintenance, or avoidance negotiations; the direction layer stores arrows, hull markings, target bearing markings, or lateral navigation markings; and the flashing layer stores flashing frequency, flashing duration, interval pattern, and repetition rhythm. The text layer, icon layer, direction layer, and flashing layer together form visually expressive content recognizable by the target crew member. In subsequent output stages, the corresponding luminous or display components can be called according to the target visual semantic template for combined output.

[0035] Each visual semantic template also stores semantic fields. These semantic fields describe the encounter semantic content that the visual semantic template can express or adapt to, and can be mapped to the communication object, risk type, expected action, directional reference, urgency level, and prohibited misunderstandings in the encounter semantic primitives. For example, when the risk type is "too close," it can be mapped to the "maintain distance" text layer, distance icon layer, and intermittent flashing layer; when the expected action is "stop approaching," it can be mapped to the "stop" text layer, warning icon layer, and rapid flashing layer; when the directional reference is the starboard side of the ship, it can be mapped to the starboard markings, starboard arrows, and sideboard text. These mapping relationships can be pre-configured by the equipment manufacturer or maintained by the ship operator based on the equipment's display capabilities and commonly used encounter expressions. The principle for setting these mapping relationships is to ensure a verifiable content association between the template's semantic fields and the encounter semantic primitive fields, rather than simply matching based on text similarity.

[0036] When filtering candidate visual semantic templates, the processor reads each semantic field in the current encounter semantic primitive and sequentially queries the mapping relationships in the visual semantic template library. If any visual semantic template has a mapping relationship with at least one semantic field in the encounter semantic primitive, the processor includes it in the initial candidate range. To avoid an overly broad candidate range, the processor also checks basic disabling rules. Basic disabling rules are used to exclude obviously unsuitable templates. For example, when the encounter semantic primitive lacks a direction reference, visual semantic templates containing strong directional instructions but unable to independently express safety prompts are not included in the candidate range; when the urgency level is distress, visual semantic templates used only for general confirmation are not included in the candidate range; when the risk type is collision risk, visual semantic templates containing the meaning of allowing passage are not included in the candidate range. Basic disabling rules can be pre-set based on the ship's encounter safety boundaries, template semantic fields, and output device capabilities, and are updated synchronously when the template library is updated.

[0037] After forming a candidate visual semantic template set, the processor retains the matched semantic fields, template hierarchy composition, and basic disabling rule verification record for each candidate visual semantic template. The matched semantic fields are used to determine whether the template covers correct understanding or triggers prohibited understanding when subsequently determining misunderstanding suppression relationships; the template hierarchy composition is used to compare the differences between text layers, icon layers, direction layers, and flashing layers when subsequently calculating visual distinction relationships; the basic disabling rule verification record is used for subsequent anomaly tracing and candidate template removal explanations. If no visual semantic template establishes a mapping relationship with the encounter semantic primitive, the processor does not generate blank output, but marks the matching task as a candidate missing state and calls the safety prompt template matching the risk type to proceed to the next stage; if the basic disabling rules exclude all initial candidate templates, the processor retains the safety prompt template with the highest urgency as a candidate visual semantic template. Through this processing, the visual expression content in the visual semantic template library is converted into a candidate template set associated with the encounter semantic primitive, providing clear, filterable, and traceable processing objects for subsequent template matching constraints.

[0038] After obtaining the understanding constraints and encounter semantic constraints, the marine visual communication equipment uses a processor to construct template matching constraints. Template matching constraints are used to restrict whether candidate visual semantic templates can enter the subsequent target template selection process, and their content is jointly formed by misunderstanding suppression relations, visual discrimination relations, and output adaptation relations.

[0039] The processor uses understanding constraints to determine the ability of candidate visual semantic templates to distinguish between correct and prohibited understandings, uses preset easily confused semantic relationships to determine the discriminable distance between candidate visual semantic templates and adjacent templates, and uses target direction, recognition conditions, and encounter semantic states to limit the output environment. After the constraint construction is completed, the template matching constraints are passed to the target visual semantic template determination stage as the basis for candidate template ranking, filtering, and downgrading.

[0040] In one embodiment, the processor determines the misunderstanding suppression relationship of candidate visual semantic templates based on understanding constraints. Understanding constraints are formed by a correct understanding set and a prohibited understanding set. The correct understanding set stores the understanding content that the target ship should form, while the prohibited understanding set stores the understanding content that the target ship should not form. For each candidate visual semantic template, the processor reads the calibration association records between the text layer, icon layer, direction layer, and flashing layer and the understanding items from the template library, and calculates the first association strength of the candidate visual semantic template with the correct understanding set and the second association strength with the prohibited understanding set, respectively.

[0041] The first association strength represents the degree to which a candidate visual semantic template enables the target ship to form a correct understanding, while the second association strength represents the degree to which a candidate visual semantic template may trigger a prohibited understanding. Both types of association strength are determined by an association table of understanding items in the template library. This table records at least the template number, template level, correct understanding item number, prohibited understanding item number, and association level. The processor matches the text layer, icon layer, direction layer, and flashing layer of the candidate visual semantic template with the correct understanding set and the prohibited understanding set, respectively, and calculates the first and second association strengths according to the association level of the hit. The calibration principle is: the association strength is higher when the text layer directly hits a correct understanding item; the association strength is lower when the icon layer and flashing layer only express attention prompts; the second association strength increases when the candidate visual semantic template contains directional symbols similar to prohibited understanding items, permissive text, or a low warning rhythm. The association level can be converted into a normalized value between 0 and 1; the larger the value, the stronger the semantic association between the template level and the understanding item. When the same candidate visual semantic template hits multiple understanding items, the processor takes the weighted sum or maximum value of the hit association values ​​as the corresponding association strength.

[0042] The misunderstanding potential can be determined using the following calculation relationship:

[0043] in, For the first Misunderstanding potential of candidate visual semantic templates For the first The first association strength between each candidate visual semantic template and the set of correct understandings For the first The second association strength between each candidate visual semantic template and the prohibited understanding set. This calculated relationship is used to distinguish the relative tendency of the candidate visual semantic template to correct understanding and prohibited understanding; the greater the misinterpretation potential, the more the candidate visual semantic template tends to reinforce correct understanding and suppress prohibited understanding.

[0044] The processor writes the misunderstanding potential into the misunderstanding suppression relation in the template matching constraints. If the misunderstanding potential is lower than the misunderstanding suppression lower limit configured by the device, the candidate visual semantic template is marked as a misunderstanding risk template and is not included in the regular sorting. If the current ship encounter intention is a distress call or danger warning, the processor can retain the safety prompt template in the misunderstanding risk template, but visual differentiation constraints or output adaptation constraints need to be added in subsequent stages. The misunderstanding suppression lower limit can be determined based on the misunderstanding potential distribution of confirmed valid templates in the template library, for example, using the low quantile value of the misunderstanding potential distribution of valid templates or the device safety configuration value as the lower limit. The processor marks candidate visual semantic templates lower than the misunderstanding suppression lower limit as misunderstanding risk templates, preventing them from entering the regular sorting, or only using them as safety prompt templates in distress or danger warning scenarios for degraded output.

[0045] In one embodiment, the processor determines the visual distinguishability relationship of candidate visual semantic templates based on preset easily confused semantic relationships. The preset easily confused semantic relationships store the adjacency relationships between encounter semantics that are prone to misreading in the same encounter scenario, cause confusion in actions when appearing consecutively, or conflict in directional references. The processor uses the current encounter semantic primitive as the retrieval entry point, searches for adjacent encounter semantic primitives with confusing relationships with the current encounter semantic primitive in the preset easily confused semantic relationships, and then reads the adjacent visual semantic templates associated with the adjacent encounter semantic primitives from the visual semantic template library.

[0046] The visual code distance margin is used to represent the degree of visual difference between candidate visual semantic templates and adjacent visual semantic templates. The processor compares the differences between candidate visual semantic templates and adjacent visual semantic templates at the text layer, icon layer, direction layer, and flashing layer. Text layer differences can be determined based on whether the text content is synonymous, whether it has opposite action meanings, and whether the text length is similar; icon layer differences can be determined based on icon category, graphic outline, and warning attributes; direction layer differences can be determined based on arrow direction, hull markings, and reference objects; flashing layer differences can be determined based on flashing frequency, on / off interval, and repetition rhythm. The differences at each layer are determined by the difference calibration table in the template library. The difference calibration table records at least the candidate template number, adjacent template number, text layer difference, icon layer difference, direction layer difference, and flashing layer difference; each layer difference is represented by a normalized value between 0 and 1. The larger the value, the greater the identifiable difference between the two at the corresponding layer. Alternatively, it can be manually reviewed and entered during template configuration.

[0047] The visual distance margin can be determined using the following calculation relationship:

[0048] in, For the first Visual code distance margin of each candidate visual semantic template In order to be with the first The set of adjacent visual semantic templates to which each candidate visual semantic template belongs has a confusing relationship with the encounter semantic primitive. For any adjacency visual semantic template in the set of adjacency visual semantic templates, For the first The candidate visual semantic template and the first Textual layer differences between adjacent visual semantic templates The difference in icon layers between the two, The difference in direction between the two, The difference in the flash layer between the two, , , , The weights for the text layer, icon layer, direction layer, and flashing layer are respectively.

[0049] The difference weights can be set according to the recognition conditions and template types. In directional encounter semantics, the difference weight of the directional layer is higher than that of the text layer; in low visibility conditions, the difference weight of the flash layer is higher than that of the text layer; in close-range port area communication conditions, the difference weights of the text layer and icon layer can be increased. The processor writes the visual code distance margin into the visual distinction relationship in the template matching constraint. If the visual code distance margin is lower than the visual distinction lower limit, the processor removes candidate visual semantic templates, or requires the superposition of more obvious directional or flash layer content in subsequent output adaptation processing. The visual distinction lower limit can be calibrated based on the minimum identifiable difference between opposite semantic templates and easily confused semantic templates in the template library; when the visual code distance margin is lower than the visual distinction lower limit, the processor removes the corresponding candidate visual semantic template, or triggers directional layer enhancement, flash layer enhancement, or text layer supplementation processing when retaining the candidate visual semantic template, to prevent visual semantic templates such as stop approaching, allow passage, wait, and confirm from being too similar from the perspective of the target crew member.

[0050] In one embodiment, the processor determines the output adaptation relationship based on encounter semantic constraint information. The output adaptation relationship is used to limit whether the candidate visual semantic template is suitable for output under the current direction binding degree, recognition conditions, and preceding semantic environment. The encounter semantic state in the encounter semantic constraint information provides the preceding encounter semantic primitives and preceding output time; the target direction information provides the direction locking degree, direction holding time, and direction jitter amount; and the recognition condition level provides the upper limit of text load, upper limit of icon load, upper limit of direction symbol load, and upper limit of flash rhythm load.

[0051] The preceding semantic residual is determined based on the preceding encounter semantic primitives and the preceding output time. The processor reads the preceding encounter semantic primitives and the preceding output time from the encounter semantic state. Based on the risk type, urgency level, output template type, and output frequency in the preceding encounter semantic primitives, it queries the semantic strength table to obtain the pre-calibrated semantic strength. Then, based on the time interval between the preceding output time and the current time, it queries the time decay table or decay function to obtain the time decay coefficient. The preceding semantic residual can be determined using the following calculation relationship:

[0052] in, This is a semantic residue from the preceding sequence. The pre-defined semantic strength corresponds to the semantic primitive of the preceding encounter. This is the time decay coefficient determined based on the preceding output time. Semantic strength is determined by the urgency of the preceding encounter semantic primitives, the number of outputs, and the template type; the time decay coefficient decreases as the interval between the preceding output time and the current time increases. The preceding semantic residual is used to determine whether the current candidate visual semantic template will conflict with the preceding understanding that the receiver may still retain.

[0053] After the preceding semantic residue is written into the output adaptation relationship, the processor generates a residual inverse vector based on the conflict rules between the preceding encountered semantic primitives and the semantic categories of the candidate visual semantic templates. The residual inverse vector is used to participate in the calculation of the comprehensive matching score, or to trigger the elimination or downgrading of the candidate visual semantic templates when the conflict level is high and the preceding semantic residue is high.

[0054] The credibility of communication object binding is determined based on the target direction information. The processor reads the direction lock degree, direction holding time, and direction jitter, and calculates the credibility of communication object binding according to the direction parameter weights. The direction parameter weights are determined by the device pointing stability calibration record, and the direction lock degree, direction holding time, and direction jitter are all converted to normalized values ​​between 0 and 1. The higher the direction lock degree and the longer the direction holding time, the higher the credibility of communication object binding; the greater the direction jitter, the lower the credibility of communication object binding. The credibility of communication object binding is used to determine whether complex direction class templates can be output. The output result of the credibility of communication object binding simultaneously enters the template matching constraint and visual communication output sequence generation stage; the processor compares the credibility of communication object binding with the lower and upper bounds of binding to determine whether the candidate visual semantic templates of the direction class participate in the target template sorting and to determine the duration of the object binding signal. If the credibility of communication object binding is lower than the lower bound, the processor does not allow the complex direction layer in the candidate visual semantic template to directly participate in the output, and only retains warning or general prompt expressions.

[0055] The recognition load budget is determined based on the recognition condition level. Each recognition condition level is associated with the upper limits of text load, icon load, directional symbol load, and flashing rhythm load through a load mapping table. The load mapping table can be determined based on the device display area, illumination duration, background light complexity, visibility level, and ship operator configuration. After reading the current recognition condition level, the processor extracts each load upper limit from the load mapping table and combines them into the recognition load budget. The recognition load budget is used to limit the expressive complexity of candidate visual semantic templates. The processor compares the text length, icon quantity, directional symbol complexity, and flashing rhythm complexity of candidate visual semantic templates with their corresponding load upper limits. If any load upper limit is exceeded, the processor marks the candidate visual semantic template as an output overload template and removes it, compresses the text layer content, or increases the number of repeated outputs in the subsequent target template determination stage.

[0056] The processor incorporates the pre-sequence semantic residuals, the credibility of the communication object binding, and the recognition load budget into the output adaptation relation. The output adaptation relation, along with the misunderstanding suppression relation and the visual discrimination relation, constitutes the template matching constraint. After construction, the template matching constraint is passed to the target visual semantic template determination stage for comprehensive scoring, conditional filtering, and necessary downgrading of candidate visual semantic templates.

[0057] After receiving template matching constraints and a set of candidate visual semantic templates, the processor performs conditional filtering and sorting on each candidate visual semantic template. Template matching constraints record misunderstanding suppression relationships, visual distinction relationships, and output adaptation relationships, used to determine whether a candidate visual semantic template is suitable for the ship's intended encounter. The processor includes candidate visual semantic templates that meet preset matching conditions in the sorting range and determines the target visual semantic template based on the comprehensive matching score. After the target visual semantic template is determined, the processor generates a visual communication output sequence, whereby the object binding signal first establishes communication direction, and then the semantic expression signal outputs the specific meaning, controlling the visual semantic signal output according to the number of repetitions. The visual communication output sequence is then passed to the illumination control or display control stage for execution.

[0058] In one embodiment, the processor determines the comprehensive matching score of each candidate visual semantic template based on template matching constraints. The comprehensive matching score is used to establish a comparable selection criterion among multiple candidate visual semantic templates, enabling misunderstanding suppression relations, visual discrimination relations, and output adaptation relations to jointly participate in the determination of the target visual semantic template. The comprehensive matching score can be determined using the following calculation relationship:

[0059] in, For the first The overall matching score of the candidate visual semantic templates. For the first Misunderstanding potential of candidate visual semantic templates For the first Visual code distance margin of each candidate visual semantic template For the first The residual inverse vectors of candidate visual semantic templates, Bind trustworthiness to communication objects. For the first The recognition load margin of each candidate visual semantic template , , , , These are the matching weights for the corresponding items. The residual inverse vector is calculated based on the conflict rules between the pre-sequential semantic residual and the semantic category of the candidate visual semantic template, and is used to represent the degree of fit between the candidate visual semantic template and the influence of the pre-sequential semantic residual; the recognition load margin is determined by the difference between the recognition load budget and the actual visual load required by the candidate visual semantic template.

[0060] The conflict rules record the correspondence between the semantic categories of the preceding semantic primitives, the semantic categories of the candidate visual semantic templates, and the conflict level. The processor determines the residual inverse vector according to the conflict level and the amount of preceding semantic residue; the higher the conflict level and the larger the amount of preceding semantic residue, the lower the residual inverse vector. The actual visual load required by the candidate visual semantic template is determined by the text length, the number of icons, the complexity of directional symbols, and the complexity of flashing rhythm according to the load mapping table corresponding to the recognition condition level. The recognition load margin is used to represent the remaining carrying capacity of the recognition load budget relative to the actual required visual load.

[0061] The matching weights are determined by the matching weight table in the template library configuration file. This table is indexed by risk type, directional marker, recognition condition level, and urgency, and records the weights for misunderstanding potential difference, visual code distance margin, residual inverse vector, communication object binding credibility, and recognition load margin. The weight setting principle prioritizes misunderstanding potential difference and visual code distance margin in high-risk encounter scenarios, communication object binding credibility in directional encounter scenarios, and recognition load margin and flashing rhythm load in low-visibility scenarios. Before calculating the overall matching score, the processor performs a preset matching condition check on the candidate visual semantic templates. The preset matching conditions can consist of one or more of the following: misunderstanding suppression conditions, visual discrimination conditions, and output adaptation conditions. Candidate visual semantic templates that do not meet the misunderstanding suppression conditions, visual discrimination conditions, or output adaptation conditions are not included in the regular sorting; candidate visual semantic templates that meet the preset matching conditions are included in the sorting range. The misunderstanding suppression condition is used to limit the misunderstanding potential difference from being lower than the device calibration lower limit. The visual discrimination condition is used to limit the visual code distance margin from being lower than the adjacent template discrimination lower limit. The output adaptation condition is used to limit the communication object binding credibility and the recognition load budget to support the output of candidate visual semantic templates.

[0062] Candidate visual semantic templates that meet preset matching conditions are entered into the sorting range. The processor determines the candidate visual semantic template with the highest comprehensive matching score as the target visual semantic template and transmits the text layer, icon layer, direction layer, flashing layer, and matching record of the target visual semantic template to the visual communication output sequence generation stage. If no candidate visual semantic template meets the preset matching conditions, the processor enters the safety processing path according to the risk type of the ship's encounter intention. When the risk type is a danger warning or distress request, the processor selects the safety prompt template with the highest misunderstanding potential difference under the same risk type in the template library and restricts the output of complex direction layers; when the risk type is a normal confirmation or avoidance negotiation, the processor stops automatic output and generates a reconfirmation prompt. This process makes the selection process of the target visual semantic template have clear screening criteria and abnormal destinations.

[0063] In one embodiment, the processor generates a visual communication output sequence based on a target visual semantic template. The visual communication output sequence describes the arrangement order and execution conditions of different signal segments during a single visual semantic signal output. The processor reads the communication object binding confidence level from the template matching constraints and generates an object binding signal based on the confidence level. The object binding signal is used to establish the directional relationship of this communication before the semantic content is output, and can be represented as a short-duration directional flash, a wide-angle cue light, a target orientation cue, or a fixed-rhythm attention cue. When the communication object binding confidence level is higher than the binding upper limit, the processor allows the object binding signal to last for a shorter period; when the communication object binding confidence level is lower than the binding lower limit, the processor extends the duration of the object binding signal and prohibits the direct output of semantic expression signals containing complex directional instructions. The binding upper and lower limits can be set according to the device's directional stability calibration value to limit the output complexity when directional binding is insufficient.

[0064] The processor generates semantic expression signals based on the target visual semantic template. These signals are formed by combining text, icon, direction, and flashing layers from the template. When the target visual semantic template includes a direction layer, the processor synchronizes the direction layer with the text or icon layer to avoid unclear reference objects caused by outputting directional arrows alone. When the target visual semantic template is a hazard warning, the processor retains the warning icon and flashing layers and keeps the text layer within the allowable range of the recognition load budget. When the target visual semantic template is an confirmation, the processor uses a combination of low-conflict confirmation symbols and short text to avoid confusion with permission-based semantics. After the semantic expression signal is generated, the processor places it after the object binding signal, ensuring that the target crew member first obtains the communication direction before receiving the specific semantic content.

[0065] The number of repetitions is determined by the recognition load budget. The recognition load budget records the maximum text load, icon load, directional symbol load, and flashing rhythm load that can be carried under the current recognition conditions. The processor compares the visual load value of the target visual semantic template with the recognition load budget. When the recognition load margin is large, the number of repetitions is reduced; when the recognition load margin is small, the number of repetitions is increased or the text content in the semantic expression signal is compressed. In low visibility conditions such as rain and fog, complex background lighting, or at non-professional receiver settings, the processor prioritizes increasing rhythmic repetitions rather than increasing text content to maintain output recognizability. The value of the number of repetitions can be given by a recognition condition setting mapping table, which can be determined based on the device's display capabilities, illumination duration limitations, and the ship operator's configuration.

[0066] The processor writes the object binding signal, semantic expression signal, and number of repeated outputs into the visual communication output sequence. The visual communication output sequence can also record the duration, interval, and end marker of each signal segment. After generating the visual communication output sequence, the processor sends it to the illumination control circuit or display control circuit, where the corresponding components drive the lights, display screen, or combined illumination unit to output visual semantic signals according to the sequence. After output is complete, the processor writes the target visual semantic template, output time, and output result into the encounter semantic state for use in the subsequent determination of preceding semantic residue. If the illumination control circuit reports an output failure, the processor retains the target visual semantic template and the unexecuted visual communication output sequence, and decides whether to retry or switch to a safety prompt template based on the risk type of the current vessel encounter intention.

[0067] In this embodiment, the marine visual communication equipment is installed outside the bridge of the working vessel. The equipment includes an operation input component, a direction acquisition component, a template memory, a processor, and a visual output component. The processor receives a communication task for a potential encounter, and the crew selects "Please pass through the starboard side of this vessel" as the intended encounter on the operation interface. The direction acquisition component records the target bearing as 074°, the pointing duration as 5.8s, and the direction jitter as 3.2°. The operation interface simultaneously records the recognition condition as "complex background lighting in the port area," the previous output saved in the encounter semantic state as "KEEP CLEAR," and the interval between the end of the previous output and the current time as 8s.

[0068] like Figure 2 As shown, Figure 2This diagram is formed from the input records and constraint generation records of this encounter communication task. The left side of the diagram shows a specific encounter communication task, the ship's encounter intention, target direction information, identification condition level, and encounter semantic state; the middle section shows the ship's encounter intention and encounter semantic constraint information; the right side shows the communication object binding credibility, identification load budget, and preceding semantic residuals generated from the encounter semantic constraint information; and the bottom is the entry point for subsequent template matching constraint construction. The target direction information, identification condition level, and encounter semantic state do not directly determine the final output content but serve as the environmental input for subsequent template matching constraint construction.

[0069] The processor normalizes "Please pass through the starboard side of this vessel" into a starboard passage intent and constructs an encounter semantic primitive. The communication target is identified as a vessel in the target direction, the risk type is identified as avoidance negotiation, the expected action is identified as starboard passage, the direction reference is identified as the starboard side of this vessel, the urgency level is identified as negotiation level, and the prohibited misunderstandings are identified as reverse passage and target vessel starboard passage. The processor writes these fields into the same encounter semantic primitive record, which is used to generate understanding constraints and filter candidate visual semantic templates.

[0070] The processor generates understanding constraints based on the encounter semantic primitives. Expected action, direction reference, and urgency level jointly generate a correct understanding set, which records "safe passage via the starboard side of this vessel". Risk type, direction reference, and prohibited misunderstandings jointly generate a prohibited understanding set, which records "passing via the starboard side of the target vessel" and "passing in the opposite direction". The direction reference is a mandatory understanding boundary in this encounter. If the direction reference is missing, although the candidate template may retain the meaning of passage, it is easily misinterpreted as a general avoidance prompt during output. Therefore, the direction reference must be included in subsequent template matching constraints.

[0071] The visual semantic template library stores four visual semantic templates related to the starboard passage-type encounter intent. Template T01 is the starboard passage template, with the text layer "PASS STBD," the icon layer "Avoidance symbol," the direction layer "Starboard arrow of the ship," and the flash layer "Medium-speed repetition." Template T02 is the maintain distance template, with the text layer "KEEP CLEAR," the icon layer "Distance warning symbol," the direction layer "Empty," and the flash layer "Intermittent flash." Template T03 is the stop approach template, with the text layer "STOP," the icon layer "Warning triangle symbol," the direction layer "Empty," and the flash layer "Rapid flash." Template T04 is the normal confirmation template, with the text layer "ROGER," the icon layer "Confirmation symbol," the direction layer "Empty," and the flash layer "Low-frequency flash." After querying the mapping relationships in the template library, the processor includes templates that have a mapping relationship with at least one semantic field of the encounter semantic primitive in the candidate range, resulting in candidate visual semantic templates T01, T02, T03, and T04.

[0072] Template matching constraints consist of misunderstanding suppression relations, visual discrimination relations, and output adaptation relations. The processor reads the first association strength between candidate visual semantic templates and the set of correct understandings, and the second association strength between candidate visual semantic templates and the set of prohibited understandings, and calculates the misunderstanding potential. The misunderstanding potential is... Calculate, where, For the first Misunderstanding potential of candidate visual semantic templates For the first The first association strength between each candidate visual semantic template and the set of correct understandings For the first The second association strength between each candidate visual semantic template and the prohibited understanding set. The first association strength of T01 is 0.84, the second association strength is 0.08, and the misinterpretation potential is 0.76; the first association strength of T02 is 0.61, the second association strength is 0.09, and the misinterpretation potential is 0.52; the first association strength of T03 is 0.56, the second association strength is 0.15, and the misinterpretation potential is 0.41; the first association strength of T04 is 0.30, the second association strength is 0.12, and the misinterpretation potential is 0.18. In this embodiment, the lower limit of the misinterpretation potential is set to 0.50. This lower limit is determined based on the distribution of the passed and misread templates in historical valid samples, and is used to eliminate candidate templates that are prone to deviating from correct understanding.

[0073] The processor determines the visual code distance margin based on preset easily confused semantic relationships. For starboard passage intents, the adjacent encounter semantic primitives recorded in the preset easily confused semantic relationships include maintaining distance, stopping approach, and normal confirmation. The processor compares the differences between candidate visual semantic templates and adjacent visual semantic templates in the text layer, icon layer, direction layer, and flashing layer, obtaining visual code distance margins of 0.68, 0.59, 0.66, and 0.35 for T01, T02, T03, and T04, respectively. In this embodiment, the visual discrimination lower limit is set to 0.55. This lower limit is calibrated based on the minimum distinguishable difference between easily confused templates and is used to limit the visual proximity between the direction indication template and the normal confirmation template or stop-type template.

[0074] The output adaptation relationship is determined by the preceding semantic residual, the communication object binding trustworthiness, and the recognition load budget. The processor determines the preceding semantic residual based on the preceding encounter semantic primitive KEEP CLEAR in the encounter semantic state and the preceding output time of 8 seconds. The preceding semantic residual uses... Calculate, where, This is a semantic residue from the preceding sequence. This represents the semantic strength corresponding to the preceding semantic primitive. This is the time decay factor. The semantic strength corresponding to KEEP CLEAR is set to 0.80, and the time decay factor is set to 0.40, resulting in a calculated pre-sequence semantic residual of 0.32. This value is related to... Figure 2 The "preceding semantic residual of 0.32" on the right side is consistent and is used to characterize the residual effect of the preceding distance class output on the current starboard passage class output.

[0075] To facilitate their integration into the template matching matrix along with other constraints, the processor further generates residual inverse vectors. These residual inverse vectors represent the degree of fit between the current candidate template and the residual influence of preceding semantics; a larger value indicates that the current candidate template is less likely to conflict with the preceding semantics. The residual inverse vectors for T01, T02, T03, and T04 are 0.82, 0.74, 0.58, and 0.80, respectively.

[0076] The reliability of communication object binding is determined based on target direction information. The direction lock degree in the target direction information is determined to be 0.74 based on the stability of the target azimuth within the continuous sampling window, the hold value corresponding to a direction holding time of 5.8s is determined to be 0.78, and the jitter deduction value corresponding to a direction jitter of 3.2° is determined to be 0.21. Based on the comprehensive calibration results of the direction lock degree, direction holding time, and direction jitter, the processor determines the reliability of communication object binding to be 0.71. This value is consistent with... Figure 2 right side and Figure 5 The "Binding Trustworthiness 0.71" above is consistent and is used to constrain the usage conditions of directional templates. When the binding trustworthiness is below 0.60, direct output of directional commands is not allowed. This encounter task meets the directional binding requirements.

[0077] The recognition load budget is determined by the complexity of the port area background light setting. This setting corresponds to a maximum load of 8 characters for text, 2 characters for icons, 1 character for directional symbols, and a maximum flash rhythm load of medium-speed repetition. Figure 2 The recognition load budget is represented by "text ≤ 8 characters, icons ≤ 2, direction symbols ≤ 1, rhythm ≤ medium speed"; Figure 5 The main restrictions for entering the output sequence control are displayed as "text ≤ 8 characters, icons ≤ 2, direction symbols ≤ 1". The processor compares the actual visual load of the candidate template with this budget to obtain the recognition load margin. The recognition load margin for T01 is 0.63, for T02 it is 0.78, for T03 it is 0.69, and for T04 it is 0.86. Although the ordinary confirmation template has a high recognition load margin, its misinterpretation potential and visual code distance margin are insufficient, therefore it cannot be used as a target visual semantic template.

[0078] like Figure 3 As shown, Figure 3It is generated from the calculation results of four candidate visual semantic templates on five types of template matching constraints. The horizontal axis represents the misunderstanding potential, visual code distance margin, residual inverse vector, binding confidence and load margin, and the vertical axis represents the four candidate templates T01 to T04. Figure 3 The threshold conditions are listed at the bottom: misinterpretation potential difference is not less than 0.50, visual code distance margin is not less than 0.55, and load margin is not less than 0.50. Figure 3 The values ​​in the table correspond to the constraint results that actually participate in the matching calculation in this embodiment, and are used to support the subsequent calculation of the comprehensive matching score.

[0079] The processor calculates the overall matching score based on template matching constraints. The overall matching score is obtained by weighting the misunderstanding potential, visual code distance margin, residual inverse vector, binding confidence, and load margin, and is verified and eliminated when candidate templates do not meet preset matching conditions. Since this encounter task is an avoidance negotiation with a clear directional reference, the weights of the misunderstanding potential and visual code distance margin are higher than other constraints. In this embodiment, the overall matching score is calculated using matching weight reassembly corresponding to the avoidance negotiation direction scenario. The matching weight reassembly is pre-recorded in the template library configuration file. Before calculation, the processor reads the matching weight reassembly and normalizes and weights the misunderstanding potential, visual code distance margin, residual inverse vector, binding confidence, and load margin. After calculation, the overall matching score is 0.717 for T01, 0.629 for T02, 0.514 for T03, and 0.420 for T04. The preset matching conditions are set as follows: misinterpretation potential difference not less than 0.50, visual code distance margin not less than 0.55, load margin not less than 0.50, and overall matching score not less than 0.60. T01 and T02 meet the preset matching conditions. T03 is eliminated due to insufficient misinterpretation potential difference, and T04 is eliminated due to weak semantic field hit and insufficient visual code distance margin. T01 has the highest overall matching score among the candidate templates that meet the preset matching conditions and is determined as the target visual semantic template.

[0080] like Figure 4 As shown, Figure 4 The result is generated from the combined matching scores of four candidate templates and the filtering results. The horizontal axis represents the candidate visual semantic templates, the vertical axis represents the combined matching score, and the dashed line represents the threshold of 0.60. Figure 4 In the test, T01 scored 0.717 and was marked as the target template; T02 scored 0.629 and was marked as the alternative template; T03 scored 0.514 and T04 scored 0.420, both of which were below the threshold and were eliminated. Figure 4 The filtering results and Figure 3 The constraint matrix in the model supports the determination process of the target visual semantic template T01.

[0081] The processor generates a visual communication output sequence based on the target visual semantic template T01. The communication object binding confidence level of 0.71 is higher than the lower binding limit of 0.60, so the processor generates an object binding signal, manifested as a directional flash pointing to the target bearing of 074°, lasting for 1.8 seconds. The target visual semantic template T01 generates a semantic expression signal, consisting of the text layer "PASS STBD" and the starboard arrow direction layer, combined with the corresponding icon layer and medium-speed flash layer. The recognition load budget shows a maximum text load of 8 characters, an icon load of 2 characters, and a direction symbol load of 1 character. T01 does not exceed any of these load limits, so the processor determines the number of repeated outputs to be 2.

[0082] like Figure 5 As shown, Figure 5 The output control record is generated by the processor. The timeline in the figure represents a complete visual communication output sequence: the object binding signal is output from 0.0s to 1.8s; the semantic expression signal is output from 1.8s to 4.8s, containing PASS STBD and a starboard arrow; there is an interval from 4.8s to 5.4s; and there are two medium-speed flashes repeated from 5.4s to 8.2s. Figure 5 The labels above, indicating a binding confidence level of 0.71, a recognition load budget, and a target template T01, signify that the output sequence matches the aforementioned template. After output, the processor writes the target template T01, output time 8.2s, and an output completion marker into the encounter semantic state for subsequent tasks to calculate the preceding semantic residuals. The final output visual semantic signal consists of an object binding signal, a starboard passage semantic expression signal, and a repetitive rhythm signal, consistent with the ship's encounter intention "Please pass through the starboard side of this vessel," and avoiding misinterpretation as a normal confirmation or reverse passage.

[0083] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0084] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

Claims

1. A method for semantic template matching of ship encounters, characterized in that, The method includes: Obtain information on the ship's encounter intentions and semantic constraints. Based on the ship's encounter intention, an encounter semantic primitive is constructed, and an understanding constraint is generated based on the encounter semantic primitive to limit the correct understanding and prohibit the understanding. Candidate visual semantic templates associated with the encounter semantic primitives are determined from the visual semantic template library; Based on the understanding constraints and the encounter semantic constraints, template matching constraints are constructed. The template matching constraints are used to constrain the misunderstanding suppression relationship, visual distinction relationship and output adaptation relationship of the candidate visual semantic templates. The target visual semantic template is determined from the candidate visual semantic templates according to the template matching constraints, and a visual communication output sequence is generated based on the target visual semantic template to output a visual semantic signal that matches the ship's meeting intention.

2. The method according to claim 1, characterized in that, The encounter semantic constraint information includes target direction information, identification condition level, and encounter semantic state. The target direction information is used to determine the trustworthiness of the communication object binding, the identification condition level is used to determine the identification load budget, and the encounter semantic state is used to determine the preceding semantic residual amount.

3. The method according to claim 1, characterized in that, The construction of encounter semantic primitives based on the ship's encounter intention includes: Based on the vessel's intended encounter, determine the communication target, risk type, expected action, direction reference, urgency level, and prohibited misunderstandings; The communication object, the risk type, the expected action, the direction reference, the urgency level, and the prohibited misunderstanding items are associated and stored to obtain the encounter semantic primitive.

4. The method according to claim 3, characterized in that, The step of generating understanding constraints based on the encounter semantic primitives to limit correct understanding and prohibited understanding includes: Based on the expected action, the direction reference, and the urgency level, at least one correct understanding item is determined, and a set of correct understanding items is generated based on the at least one correct understanding item; Based on the risk type, the direction reference, and the prohibited misunderstanding item, at least one prohibited comprehension item is determined, and a prohibited comprehension set is generated based on the at least one prohibited comprehension item; The set of correct understandings and the set of prohibited understandings are defined as the understanding constraints.

5. The method according to claim 1, characterized in that, The visual semantic template library includes multiple visual semantic templates, and any one of the visual semantic templates includes a text layer, an icon layer, a direction layer, and a flashing layer; The step of determining candidate visual semantic templates associated with the encounter semantic primitives from the visual semantic template library includes: The visual semantic template that has a mapping relationship with at least one semantic field in the encounter semantic primitive is determined as the candidate visual semantic template.

6. The method according to claim 4, characterized in that, The misunderstanding suppression relationship is determined based on the misunderstanding potential difference; The misunderstanding potential difference is determined based on the following steps: Determine the first association strength between the candidate visual semantic template and the set of correct understandings, and the second association strength between the candidate visual semantic template and the set of prohibited understandings; The misunderstanding potential difference is determined based on the difference between the first association strength and the second association strength.

7. The method according to claim 5, characterized in that, The visual distinction relationship is determined based on the visual code distance margin; The visual code distance margin is determined based on the following steps: Based on the preset easily confused semantic relationships, the adjacent encounter semantic primitives that have a confusing relationship with the encounter semantic primitives are determined; Determine the adjacency visual semantic template associated with the adjacency encounter semantic primitive from the visual semantic template library; The visual code distance margin is determined based on the differences between the candidate visual semantic template and the adjacent visual semantic template in the text layer, the icon layer, the direction layer, and the flashing layer.

8. The method according to claim 2, characterized in that, The output adaptation relationship is determined based on the pre-sequence semantic residual, the credibility of the communication object binding, and the identification load budget. The preceding semantic residual is determined based on the preceding encounter semantic primitives and preceding output time in the encounter semantic state. The communication object binding credibility is determined based on the direction locking degree, direction holding time and direction jitter in the target direction information. The recognition load budget is determined based on the text load limit, icon load limit, direction symbol load limit and flash rhythm load limit corresponding to the recognition condition level.

9. The method according to claim 1, characterized in that, The step of determining the target visual semantic template from the candidate visual semantic templates according to the template matching constraints includes: The comprehensive matching score of each candidate visual semantic template is determined based on the template matching constraints. The candidate visual semantic template that meets the preset matching conditions and has the highest comprehensive matching score is determined as the target visual semantic template. The preset matching conditions are used to define the misunderstanding suppression relationship, the visual differentiation relationship and the output adaptation relationship respectively.

10. The method according to claim 8, characterized in that, The step of generating a visual communication output sequence based on the target visual semantic template includes: Generate an object binding signal based on the credibility of the communication object binding; Generate semantic expression signals based on the target visual semantic template; The number of repeated outputs is determined based on the identified load budget; The visual communication output sequence is generated based on the object binding signal, the semantic expression signal, and the number of repeated outputs.

Images