Intelligent aviation carrying compliance determination method and system

Abstract

The invention discloses an intelligent aviation carrying compliance determination method and system, and the method comprises the steps: responding to a user operation to obtain the context information of a current journey, and receiving an article image of a to-be-queried article inputted by a user; a pre-trained multi-modal model is utilized to analyze the article image, a structured article description object is generated, and the structured article description object comprises article visual attributes analyzed from the article image; retrieving related information from a pre-constructed knowledge database based on the context information, and performing logic matching and reasoning on the structured article description object and the related information to generate a carrying state judgment result; and converting the carrying state judgment result into an output natural language conclusion. According to the method and the system, by fusing multi-modal perception and logical reasoning, the problems that a traditional method is low in intelligent degree and cannot process composite articles and fuzzy scenes are solved, and the judgment accuracy and the user experience are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of aviation transportation safety, and in particular to an intelligent aviation carrying compliance determination method and system. BACKGROUND

[0002] Before passengers travel by air, they need to accurately understand whether the carried articles meet the aviation safety regulations. At present, the main query methods mainly include: self-checking the lengthy and professional text clauses issued by airlines; accessing the static article list of the airline website or third-party platform; or using a simple question and answer robot based on keyword matching.

[0003] However, these existing technical solutions have obvious limitations and insufficient intelligence: Single perception dimension, unable to understand the real object: the existing solution only supports text input and cannot process the real object image directly taken by the user. Therefore, the system cannot automatically extract the key visual attributes of the article, such as identifying the article category by appearance, reading the capacity identification by OCR, judging the article state, or analyzing the internal structure of the composite article, resulting in a lack of real object determination capability.

[0004] Lack of dynamic scene association, and the result is not accurate enough: the query process is usually disconnected from the specific travel context of the user. Since there are differences in rules of different airlines and different routes, static rule lists or general question and answer cannot provide personalized determination results deeply bound to the current trip, which may lead to inaccurate information.

[0005] Weak logical reasoning ability, difficult to handle complex situations: for the complex logical conditions (such as "and", "or") in the aviation rules, exception clauses, or attribute ambiguous articles, the system based on simple keyword matching cannot perform effective logical reasoning and interaction clarification, and often can only give "cannot be determined" or general, even incorrect results, resulting in poor user experience. SUMMARY

[0006] To solve the technical problems in the background art, the present application provides an intelligent aviation carrying compliance determination method and system.

[0007] The intelligent aviation carrying compliance determination method provided by the present application comprises the following steps: S1, responding to user operation to obtain context information of the current trip, and receiving an article image of a to-be-queried article input by the user; S2, analyzing the article image using a pre-trained multi-modal model to generate a structured article description object, the structured article description object containing article visual attributes parsed from the article image; S3, retrieving relevant information from a pre-constructed knowledge database based on the context information, and logically matching and reasoning the structured item description object with the relevant information to generate a state determination result; S4, converting the state determination result into an outputable natural language conclusion.

[0008] Preferably, in step S1, a user-input supplementary text description is also received; in step S2, the pre-trained multi-modal model cooperatively analyzes the item image and the supplementary text description to generate the structured item description object.

[0009] Preferably, step S2 specifically comprises: analyzing the item image by using a pre-trained multi-modal large model to identify a main category of the item, at least one component, text identification information on the surface of the item obtained by optical character recognition, and an estimated value of at least one key physical attribute; organizing and encapsulating the main category, the component, the text identification information, and the estimated value of the key physical attribute obtained by analysis according to a pre-set machine-readable data structure to generate the structured item description object; wherein the component, the text identification information, and the estimated value of the key physical attribute constitute item visual attributes parsed from the item image.

[0010] Preferably, step S3 specifically comprises: retrieving associated structured compliance knowledge items from a pre-constructed knowledge database based on airline and route information in the context information; logically operating and matching the key physical attribute estimate, the component, and the text identification information contained in the structured item description object with corresponding condition parameters defined in the structured compliance knowledge items to obtain a matching result; inputting the matching result into a pre-set reasoning engine, and outputting a state determination result by the reasoning engine according to a mapping relationship between the matching result and pre-defined compliance actions in the structured compliance knowledge items.

[0011] Preferably, the state determination result comprises at least one of the following categories: allowed carrying state, prohibited carrying state, or restrictive allowed carrying state; wherein the restrictive allowed carrying state is associated with at least one specific operation restriction condition.

[0012] Preferably, step S3 further comprises: When the estimated value of a key physical attribute in the structured object of the article description is missing or in a fuzzy numerical range, resulting in the logical operation and matching failing to produce a unique matching result, an interactive request is initiated through the user interface, the interactive request being used to guide the user to supplement or confirm the estimated value of the key physical attribute; The response information of the user to the interactive request is received, and the corresponding estimated value of the key physical attribute in the structured object of the article description is updated based on the response information; Based on the updated structured object of the article description, the logical operation and matching and subsequent steps are re-executed.

[0013] Preferably, in step S3, for a composite article including multiple components in the structured object of the article description, independent matching and reasoning are performed on each component, and then overall compliance is judged according to the association relationship between the components.

[0014] Preferably, the pre-constructed knowledge database stores a plurality of structured compliance knowledge entries, each structured compliance knowledge entry being composed of a constraint condition entity, a constraint numerical range and a corresponding compliance action parsed from the aviation carrying compliance text according to a preset logical format.

[0015] Preferably, step S4 specifically includes: The carrying state judgment result is parsed to extract at least one carrying state category identifier and at least one operation restriction condition associated with the carrying state category identifier when the carrying state category identifier is a restrictive allowed carrying state; According to the carrying state category identifier and the operation restriction condition, a corresponding target text template is selected or combined from a pre-set natural language template library; The key physical attribute estimate value or component in the structured object of the article description and the travel information in the context information are filled as variable values into corresponding placeholders in the target text template to generate an outputable natural language conclusion.

[0016] The present application provides an intelligent aviation carrying compliance judgment system, comprising: An information input module is configured to obtain context information of a current trip in response to a user operation and receive an article image of an article to be queried input by the user; A multi-modal perception module is configured to analyze the article image by using a pre-trained multi-modal model to generate a structured object of the article, the structured object of the article including visual attributes of the article parsed from the article image; A decision module is configured to retrieve relevant information from a pre-constructed knowledge database based on the context information and perform logical matching and reasoning on the structured object of the article and the relevant information to generate a carrying state judgment result; A result generation and output module is configured to convert the carrying state determination result into an outputable natural language conclusion.

[0017] In the present application, the proposed intelligent aviation carrying compliance determination method and system improves the intelligent level of article compliance determination by introducing a collaborative mechanism of multi-modal perception and logical reasoning, realizes the leap from simple matching relying on text keywords to comprehensive determination integrating visual understanding and rule reasoning, can process real object recognition, attribute extraction and complex rule logical application like a human expert, and improves the accuracy and reliability of determination. Secondly, by dynamically binding the determination process with the user's real-time itinerary and searching and reasoning based on the knowledge database, the misjudgment problem caused by rule differences is solved. In addition, the whole process converts the tedious autonomous inquiry into intuitive interaction of "one-key photographing, instant result", actively guides clarification when information is insufficient, provides basis and explanation when outputting conclusions, reduces the user's use threshold and cognitive burden, and enhances the service credibility. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A workflow schematic diagram of an intelligent aviation carrying compliance determination method proposed in the present application; Figure 2 A system architecture schematic diagram of an intelligent aviation carrying compliance determination system proposed in the present application. DETAILED DESCRIPTION

[0019] With reference to Figure 1 and Figure 2 , the present application proposes an intelligent aviation carrying compliance determination method, which comprises the following steps: S1, responding to a user operation to obtain context information of a current itinerary, and receiving an article image of an article to be queried input by the user.

[0020] In the present embodiment, in step S1, a supplementary text description input by the user is also received; in step S2, the pre-trained multi-modal model cooperatively analyzes the article image and the supplementary text description to generate a structured article description object.

[0021] Specifically, the operation of the system corresponding to the method in the present embodiment is usually triggered by the user in a specific interactive interface of an aviation service application, wherein the specific interactive interface is, for example, a ticket order detail page or a luggage query dedicated entrance. After triggering, the system automatically binds and extracts the context information C of the current itinerary from the user's itinerary data. The context information C is a structured data object, and the core fields at least include: airline, route (including departure place and destination), etc.

[0022] Subsequently, the system calls the device camera interface, guiding the user to take an image I of the item to be queried. At the same time, the system provides an optional text input box for the user to input a supplementary text description T, such as: brand new and unopened, containing lithium batteries, etc., to make up for the possible deficiencies of visual information or emphasize specific intentions.

[0023] S2, analyzing the image of the item using a pre-trained multi-modal model to generate a structured item description object, which includes the visual attributes of the item parsed from the image of the item.

[0024] In this embodiment, step S2 specifically includes: The pre-trained multi-modal large model is used to analyze the image of the item to identify the main category of the item, at least one component, text identification information on the surface of the item obtained through optical character recognition, and an estimated value of at least one key physical attribute; The parsed main category, component, text identification information, and estimated value of the key physical attribute are organized and encapsulated according to a pre-set machine-readable data structure to generate a structured item description object; wherein the component, text identification information, and estimated value of the key physical attribute constitute the visual attributes of the item parsed from the image of the item.

[0025] In this embodiment, the pre-trained multi-modal large model is a visual language model based on the Transformer architecture.

[0026] Specifically, the training process of the multi-modal large model includes: First, pre-training is performed on a data set containing a large amount of Internet text-image pairs, and through a self-supervised learning objective (such as text-image contrast learning, mask modeling, etc.), the model learns to establish a general association between visual features and language semantics, and obtains basic cross-modal understanding and representation capabilities.

[0027] Subsequently, for the specific task of aviation item compliance determination, a high-quality domain fine-tuning data set is collected and constructed, which contains images of common items in the aviation scene and their corresponding fine-grained structured attribute annotations. The pre-set multi-modal large model is supervised and task-adapted fine-tuned using this domain fine-tuning data set to optimize the performance and robustness of the model in downstream tasks such as item classification, component recognition, and attribute (such as capacity, material) extraction.

[0028] During training, the total loss is the weighted sum of multiple task losses such as classification, attribute prediction, and compliance determination, to balance the learning of different tasks.

[0029] Through the above training paradigm combining pre-training and domain fine-tuning, a trained multi-modal large model is finally obtained, which is specialized in aviation item understanding and can accurately generate structured item description objects from images and texts.

[0030] Specifically, in the application process of the pre-trained multi-modal large model, the system input is the image and / or related text description (such as passenger declaration information) of the item to be inspected. The image and text are input into the respective encoders after preprocessing. The output of the pre-trained multi-modal large model is a structured judgment result object, usually presented in JSON format, containing key fields such as item category, predicted attribute, compliance status, and judgment basis. The pre-trained multi-modal large model internally establishes a direct mapping from visual features to domain attributes. For example, the model learns to recognize the scale on the bottle body and associate it with the "capacity" attribute; then, combined with the "liquid" category and the capacity attribute of "more than 100 ml", the built-in rule logic is activated, and finally the "non-compliance" judgment and specific reasons are output.

[0031] It should be noted that the pre-trained multi-modal large model cooperatively analyzes and deeply understands the input item image I and text description T, which specifically includes: Visual analysis: the pre-trained multi-modal large model performs end-to-end analysis on the item image I, completing multiple tasks: Main category identification: determine the primary classification of the item, such as power bank.

[0032] Component segmentation and identification: analyze the constituent parts of the item, such as identifying lithium-ion battery core, plastic shell, circuit board, etc.

[0033] Key visual attribute extraction: through the integration of an optical character recognition module, locate and read the text labels on the surface of the item (such as rated capacity: 20000mAh); through regression analysis to estimate physical quantities (such as estimating the volume of liquid by the size of the bottle body); and evaluate the state of the item (such as intact packaging, worn label).

[0034] Text understanding: perform natural language processing on the user input supplementary text description T, extracting explicit or implicit attributes (such as new), states (such as unopened) or user concerns.

[0035] Information fusion and structuring: the system aligns, verifies, and fuses the visual analysis results and text understanding results. For example, if the text emphasizes unopened and the visual recognition also supports intact packaging, the attribute confidence is enhanced. Finally, all information is organized and encapsulated into a machine-readable, queryable structured item description object O. The structured item description object O adopts a lightweight data structure (such as JSON) in the form of key-value pairs, an example is as follows: {"subject": "Power bank", "material / component": ["lithium ion battery"], "key attributes": {"rated energy": "74 Wh", "label clarity": "high"}, "user supplemental context": "brand new, unopened"}; This structured item description object provides accurate and structured input for subsequent rule inference.

[0036] S3, retrieve relevant information from the pre-constructed knowledge database based on the context information, and logically match and infer the structured item description object with the relevant information to generate a state determination result.

[0037] In this embodiment, step S3 specifically includes: Based on the airline and route information in the context information, retrieve the associated structured compliance knowledge items from the pre-constructed knowledge database; Logically operate and match the key physical attribute estimate values, component parts, and text identification information contained in the structured item description object with the corresponding condition parameters defined in the structured compliance knowledge items to obtain a matching result; Input the matching result into the pre-set inference engine, and output a state determination result according to the mapping relationship between the matching result and the pre-defined compliance actions in the structured compliance knowledge items.

[0038] Specifically, the carry-on state determination result includes at least one of the following categories: allowed carry-on state, prohibited carry-on state, or restricted allowed carry-on state; wherein the restricted allowed carry-on state is associated with at least one specific operation restriction condition.

[0039] In this embodiment, step S3 further includes: When the key physical attribute estimate values in the structured item description object are missing or in a fuzzy numerical range, resulting in that the logical operation and matching cannot produce a unique matching result, initiate an interaction request through a user interface, and the interaction request is used to guide the user to supplement or confirm the key physical attribute estimate values; Receive the user's response information to the interaction request, and update the corresponding key physical attribute estimate values in the structured item description object based on the response information; Based on the updated structured item description object, re-execute the logical operation and matching and the subsequent steps.

[0040] Specifically, in step S3, for a composite item containing multiple components in the structured item description object, first perform independent matching and inference on each component, and then perform overall compliance judgment according to the association relationship between the components.

[0041] It should be noted that the pre-constructed knowledge database stores a plurality of structured compliance knowledge entries, each structured compliance knowledge entry is composed of constraint condition entities, constraint numerical ranges and corresponding compliance actions parsed from the aviation carrying compliance text according to a preset logical format.

[0042] Specifically, step S3 realizes the leap from data matching to knowledge reasoning. In this step, instead of accessing a static rule list, conditional retrieval is performed from a structured, dynamically updatable knowledge database according to the context information C (especially the key fields such as airline and route type). Each rule in the knowledge database is stored in the form of explicit "condition action" logic, ensuring machine-parsable and executable. For example, a typical rule logic is expressed as: IF (item type ∈ ['power bank', 'backup lithium battery']) AND (rated energy ≤ 100Wh) THEN (action = 'allow to carry, prohibit to ship').

[0043] The system retrieves all applicable rules according to the context information C, forming a relevant rule subset R for the current judgment scenario. The reasoning engine receives the structured object description object O and the rule subset R, and performs a logical calculation process. It compares and logically operates each attribute value in the structured object description object O with the condition part of each rule in the rule subset R. This process can handle complex condition combinations containing "AND" (and), "OR" (or) and other logical connectors, as well as numerical range comparisons (such as "≤ 100Wh") and set membership relations (such as "∈ [...]"). Only when the attributes of the structured object description object O satisfy all the premise conditions of a rule, will the "action" of the rule conclusion part be triggered, thus generating a definite carrying status judgment result. This way effectively solves the problem of weak handling of complex logic and exception clauses in traditional solutions.

[0044] S4, converting the carrying status judgment result into an outputable natural language conclusion.

[0045] In this embodiment, step S4 specifically includes: parsing the carrying status judgment result to extract at least one carrying status category identifier, and at least one operation restriction condition associated with the carrying status category identifier when the carrying status category identifier is a restrictive allowed carrying status; selecting or combining a corresponding target text template from a pre-set natural language template library according to the carrying status category identifier and the operation restriction condition; filling the key physical property estimate value or component in the structured object description object and the flight information in the context information into the corresponding placeholders in the target text template as variable values to generate an outputable natural language conclusion.

[0046] Example 1: Precise matching determination of standard items This embodiment uses the example of a passenger carrying a power bank on an international flight to demonstrate the end-to-end automated judgment process for standard items in this application.

[0047] (1) Scene and Input Passenger Mr. Zhao planned to take a Star Airlines international flight from Beijing to Berlin. On the baggage claim page, he used his phone's camera to take a clear photo of his power bank and uploaded it. The system automatically linked it to his itinerary and generated contextual information. {Airline: "Star Airlines", Route Type: "International", Departure City: "Beijing", Destination: "Berlin"}. The obtained item image is denoted as... .

[0048] (2) Multimodal perception and structured generation The system calls a pre-trained multimodal large model to process the object image. Analysis: The main category of the identified item is "power bank".

[0049] Using optical character recognition technology, the text markings on the casing were located and read: "Rated capacity: 20000mAh" and "Rated voltage: 3.7V".

[0050] It automatically calculates its rated energy (20Ah 3.7V=74Wh).

[0051] Finally, a structured item description object is generated. As shown below: {"Main Category": "Power Bank", "Components": ["Lithium-ion Battery"], "Key Attributes": {"Rated Energy": 74, "Unit": "Wh", "Identification Clarity": "Clear"}}.

[0052] (3) Intelligent reasoning and decision making Information retrieval: The inference engine uses contextual information. (Especially for the airline "Star Airlines" and the route type "International"), retrieve all applicable structured compliance knowledge entries from the knowledge database.

[0053] Attribute matching: The engine will structure the item description object. The attributes in the data are logically matched with the retrieved structured compliance knowledge entries. For example, a key structured compliance knowledge entry is matched, with the following condition: the subject category belongs to ["power bank", "backup lithium battery"] and the rated energy is ≤100Wh.

[0054] Result generation: Due to the structured item description object Key attributes: rated energy = 74 Wh. The item satisfies all conditions of this entry, and the reasoning engine triggers its conclusion action, generating a state determination result: {Carry-on status: "Allowed to carry", operation requirement: "Allowed to carry-on, prohibited to ship", basis entry ID: "STAR_AIR_001"}.

[0055] (4) Result generation and feedback After the system parses the above determination result, it selects the corresponding template from the natural language template library, and fills the key attributes (rated energy 74 Wh) in the structured item description object and the trip information (Star Space International Flight) in the context information into the template variables to generate the final user conclusion: "Your power bank (rated energy 74 Wh) can be carried on, but not shipped. Basis: Star Space regulations allow lithium battery devices with rated energy not exceeding 100 Wh to be carried on."

[0056] Example 2: Deep reasoning and interaction of complex composite items This example takes a passenger querying a gift set containing perfume as an example to demonstrate the deep reasoning ability of the invention for complex composite items and the active interaction mechanism when information is ambiguous.

[0057] (1) Scene and input Passenger Ms. Li received an unopened gift set, with opaque paper box, containing a bottle of perfume and a decorative item. She is uncertain about whether it can be carried on, so she takes a photo of the outside of the gift box and uploads it. The system obtains the context information of the current trip (any domestic flight), and the item image is recorded as .

[0058] (2) Multi-modal perception and structured generation The system analyzes the item image : Identifies the main category as "gift box".

[0059] Concludes that it contains "glass bottle (liquid)" and "paper decorative item".

[0060] Estimates the volume of the liquid in the glass bottle to be about 30 ml, and identifies the outer packaging material as "opaque paper box".

[0061] Generate structured item description object : {"Main Category": "Gift Box", "Component": [{"Item": "Glass Bottle", "Attribute": {"Contents": "Liquid", "Estimated Volume per Bottle": 30", "Unit": "ml"}}, {"Item": "Paper Decoration", "Attribute": {}}], "Key Attribute": {"Outer Packaging Material": "Non-transparent Cardboard Box"}}.

[0062] (3) Intelligent reasoning and decision making Rule retrieval: The system retrieves rules based on context information. Search for structured compliance knowledge entries regarding "carrying liquid items" on domestic flights.

[0063] Component-level reasoning: The reasoning engine describes structured item objects. Each component was analyzed independently. For the "glass bottle" component, its attribute (approximately 30ml of liquid per bottle) matched the condition of the entry "single liquid container volume ≤ 100ml", thus determining that the component itself was compliant.

[0064] Overall Association Reasoning: The engine further performed association analysis and found that the "glass bottle" component was placed in a "non-transparent cardboard box". This violates another entry's packaging condition: "Liquid items carried on your person must be placed in a transparent, resealable 1-liter plastic bag".

[0065] Comprehensive decision: Therefore, the system generates a composite carrying status determination result: {Carrying status: "Restricted permission", Restriction conditions: ["Must be removed from the original packaging", "Place in a transparent plastic bag"], Recommendation: "Outer packaging box can be checked in"}.

[0066] (4) Results generation and feedback The system transforms the composite judgment results into specific, actionable suggestions: "Your gift contains liquid (approximately 30ml). The liquid itself meets the carry-on volume requirements, but according to security regulations, it must be removed from its current opaque packaging and placed in a transparent 1-liter plastic bag for inspection. It is recommended that the outer packaging be checked in."

[0067] Example 3: Proactive Interactive Clarification When Information is Ambiguous This embodiment demonstrates how to guide users to make accurate judgments through proactive interaction when the system lacks sufficient perceived information.

[0068] (1) Scene and Input Passenger Mr. Zhang's power bank was worn and the capacity markings were illegible. He took a photo and uploaded it, which was then linked to his trip's context information by the system. .

[0069] (2) Multimodal perception and structured generation The system identifies the item as a "power bank", but cannot read the exact capacity through OCR. Based on its physical dimensions, the model outputs a fuzzy estimated range, generating a structured item description object with the key attributes being: {“Nominal Energy Estimate”: “50-120”, “Unit”: “Wh”, “Confidence”: “Low”}.

[0070] (3) Intelligent reasoning and interaction The reasoning engine retrieves relevant rules and finds that the fuzzy range (50-120 Wh) intersects with the conditions of both “Nominal Energy ≤ 100 Wh” (portable) and “Nominal Energy 100 Wh” (approval required), which cannot be directly matched.

[0071] The system initiates an interactive request, pushing options through the user interface: “Cannot accurately identify capacity. Please select: ① Confirm Nominal Energy ≤ 100 Wh; ② Nominal Energy 100 Wh; ③ Re-upload clear photo.” (4) User feedback and final determination: Mr. Zhang chooses ①.

[0072] The system updates the key attributes in the structured item description object to {“Nominal Energy”: 100, “Unit”: “Wh”, “Confidence”: “User Confirmation”}.

[0073] Based on the updated structured item description object , the reasoning is re-performed, clearly matching the “portable” rule, generating the final determination result and outputting it.

[0074] This embodiment reflects the intelligence of the system when it is aware of insufficient information, guiding the user to supplement key judgment information, converting the ambiguous scenario that cannot be processed into a process that can be accurately determined, improving the reliability and practicality of the system.

[0075] From the above embodiments, it can be seen that the present application not only realizes efficient and accurate determination in standard scenarios, but also can handle complex situations such as composite items and ambiguous information through deep reasoning, significantly improving the level of intelligence and user experience.

[0076] Referring to Figure 1 and Figure 2 , the intelligent aviation carrying compliance determination system proposed by the present application comprises: An information input module for responding to user operations to obtain context information of the current trip and receiving an item image of a queried item input by the user; a multi-modal perception module configured to analyze the item image using a pre-trained multi-modal model to generate a structured item description object including item visual attributes parsed from the item image; a decision module configured to retrieve relevant information from a pre-built knowledge database based on the context information, and perform logical matching and reasoning between the structured item description object and the relevant information to generate a state determination result; a result generation and output module configured to convert the state determination result into an outputable natural language conclusion.

[0077] The above merely provides the preferred embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, according to the technical solution and the inventive concept of the present application, can make equivalent replacements or changes within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for determining the compliance of intelligent aviation carry-on equipment, characterized in that, Includes the following steps: S1. Respond to user actions to obtain context information of the current trip, and receive the item image of the item to be queried input by the user; S2. Analyze the object image using a pre-trained multimodal model to generate a structured object description object, which contains the object visual attributes parsed from the object image. S3. Based on the context information, retrieve relevant information from the pre-built knowledge database, and perform logical matching and reasoning between the structured item description object and the relevant information to generate a carrying status determination result; S4. Convert the carrying state determination result into an outputtable natural language conclusion.

2. The method for determining the compliance of intelligent aviation carry-on equipment according to claim 1, characterized in that, In step S1, supplementary text descriptions input by the user are also received; in step S2, the pre-trained multimodal model collaboratively analyzes the item image and the supplementary text descriptions to generate a structured item description object.

3. The method for determining the compliance of intelligent aviation carry-on equipment according to claim 1 or 2, characterized in that, Step S2 specifically includes: The object image is analyzed using a pre-trained multimodal large model to identify the main category of the object, at least one component, text identification information obtained from optical character recognition on the object surface, and an estimated value of at least one key physical attribute. The parsed main category, components, text identification information, and key physical attribute estimates are organized and encapsulated according to a preset machine-readable data structure to generate a structured item description object; wherein, the components, text identification information, and key physical attribute estimates constitute the item visual attributes parsed from the item image.

4. The method for determining the compliance of intelligent aviation carry-on equipment according to claim 1, characterized in that, Step S3 specifically includes: Based on the airline and route information in the context information, relevant structured compliance knowledge entries are retrieved from the pre-built knowledge database; The estimated values ​​of key physical attributes, components, and text identification information contained in the structured item description object are logically operated and matched with the corresponding condition parameters defined in the structured compliance knowledge item to obtain the matching result; The matching results are input into a pre-defined inference engine, which outputs a status determination result based on the mapping relationship between the matching results and the predefined compliance actions in the structured compliance knowledge entries.

5. The method for determining the compliance of intelligent aviation carry-on equipment according to claim 1, characterized in that, The carrying status determination result includes at least one of the following categories: allowed carrying status, prohibited carrying status, or restricted allowed carrying status; wherein the restricted allowed carrying status is associated with at least one specific operational restriction condition.

6. The method for determining the compliance of intelligent aviation carry-on equipment according to claim 4, characterized in that, Step S3 also includes: When the estimated value of a key physical attribute in a structured item description object is missing or is in a fuzzy range, causing logical operations and matching to fail to produce a unique matching result, an interaction request is initiated through the user interface. The interaction request is used to guide the user to supplement or confirm the estimated value of the key physical attribute. Receive the user's response information to the interaction request, and update the corresponding key physical attribute estimates in the structured item description object based on the response information; Based on the updated structured item description object, re-execute logical operations and matching, as well as subsequent steps.

7. The method for determining the compliance of intelligent aviation carry-on equipment according to claim 1, characterized in that, In step S3, for complex items whose structured item description objects contain multiple components, each component is first matched and reasoned independently, and then the overall compliance is judged based on the relationship between the components.

8. The method for determining the compliance of intelligent aviation carry-on equipment according to claim 1, characterized in that, The pre-built knowledge database stores multiple structured compliance knowledge entries. Each structured compliance knowledge entry consists of constraint entities, constraint value ranges, and corresponding compliance actions parsed from aviation compliance texts, arranged according to a preset logical format.

9. The method for determining the compliance of intelligent aviation carry-on equipment according to claim 5, characterized in that, Step S4 specifically includes: The carrying status determination result is parsed to extract at least one carrying status category identifier and at least one operation restriction condition associated with the carrying status category identifier being a restricted carrying status. Based on the carrying status category identifier and operation restrictions, the corresponding target text template is selected or combined from the preset natural language template library to generate the corresponding target text template. The estimated values ​​of key physical attributes or components in the structured item description object, as well as the travel information in the context information, are used as variable values ​​to fill the corresponding placeholders in the target text template, generating an outputtable natural language conclusion.

10. An intelligent aviation compliance determination system, characterized in that, include: The information input module is used to respond to user operations to obtain contextual information of the current trip and to receive the item image of the item to be queried by the user. The multimodal perception module is used to analyze the object image using a pre-trained multimodal model and generate a structured object description object, which contains the object visual attributes parsed from the object image. The decision module is used to retrieve relevant information from a pre-built knowledge database based on the context information, and to perform logical matching and reasoning between the structured item description object and the relevant information to generate a carrying status determination result; The result generation and output module is used to convert the carried state determination result into an outputtable natural language conclusion.

Images