A method and system for multimodal information semantic anchoring and dynamic incremental update of star map based on star map coordinates

Abstract

This invention provides a method and system for multimodal information semantic anchoring and dynamic incremental updating of star maps based on star map coordinates. It solves problems such as deep semantic alignment and includes: S1: multimodal semantic anchoring; S2: incremental update triggering; S3: local impact assessment; S4: atomized incremental operation; S5: consistency maintenance. This invention has advantages such as deep semantic alignment and good learning stability.

Description

Technical Field

[0001] This invention belongs to the field of mechanical automation equipment technology, specifically relating to a method and system for multimodal information semantic anchoring and dynamic incremental updating of star maps based on star map coordinates. Background Technology

[0002] With the development of general artificial intelligence systems, one of their core requirements is to achieve a unified understanding and continuous learning of multi-source heterogeneous information from the real world. This involves two closely related key technical challenges: the semantic unification of multimodal information and the dynamic evolution of core knowledge models. In multimodal information processing, current mainstream technologies typically focus on mapping data from different modalities, such as images, text, and speech, into a shared latent space through large-scale alignment training. While this approach achieves preliminary correlations between modalities, it has fundamental limitations: the latent space representation is essentially a high-dimensional vector, lacking explicit, structured semantic relationships and logical frameworks, making it difficult for the system to perform complex symbolic reasoning and causal inference. Its alignment process is a data-driven statistical association, unable to clearly trace which feature of which modality a judgment originates from, posing risks in high-reliability fields such as medicine and finance. Furthermore, existing methods typically employ fixed fusion networks or attention mechanisms, making it difficult to dynamically adjust the credibility and contribution weights of different modalities based on the specific task context. In addition, introducing new modalities often requires redesigning the model architecture and performing costly re-alignment training, resulting in poor scalability.

[0003] Regarding the continuous updating of core knowledge models, traditional incremental learning methods in machine learning primarily target parametric models such as neural networks, adapting to new data by adjusting parameters. However, when faced with structured knowledge models like star graph coordinates, which are based on graph structures and rich in symbolic logic and complex relationships, directly adding new knowledge can easily disrupt the original logical structure and semantic consistency of the star graph, leading to a degradation of existing reasoning capabilities. Furthermore, new information may logically contradict existing knowledge, lacking effective online arbitration and resolution mechanisms. The introduction of new concepts and relationships often requires adjustments to the topological structure of the knowledge graph itself, not just parameter fine-tuning. Existing methods lack controllable primitives for such structured changes. Additionally, AGI systems require real-time or near-real-time learning during service, but global knowledge graph reconstruction incurs enormous computational overhead, making it difficult to meet online requirements, and the update process may cause instability in the system's reasoning service.

[0004] To address the aforementioned technical issues, this application proposes a method and system for semantic anchoring of multimodal information and dynamic incremental updating of star map based on star map coordinates. This method and system can deeply and interpretably anchor multimodal information into a structured knowledge model and support the safe, efficient, and continuous dynamic evolution of the model. Summary of the Invention

[0005] The purpose of this invention is to address the above-mentioned problems by providing a reasonably designed method for multimodal information semantic anchoring and dynamic incremental update of star map based on star map coordinates, which achieves deep semantic alignment.

[0006] Another objective of this invention is to address the aforementioned problems by providing a sustainable learning system for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates, comprising the following steps: S1: Multimodal semantic anchoring, receiving input information from at least two different modalities, mapping each modal information to corresponding semantic nodes or relationships in a unified star map coordinate model, and completing semantic anchoring; the semantic anchoring includes similarity-based retrieval anchoring and / or generative anchoring that creates new nodes; First, a dedicated multimodal encoder array transforms the raw data into feature vectors and candidate semantic labels. Then, the semantic anchoring engine, referencing a star-map coordinate knowledge base, performs two types of operations: retrieval-based anchoring calculates the similarity between candidate semantics and existing node vectors in the base, achieving precise association; generative anchoring, upon identifying a new concept, creates a new node in the knowledge base and links it to a higher-level concept, expanding knowledge. This process transforms unstructured multimodal data into structured, interpretable knowledge nodes.

[0008] S2: Incremental update triggered, the semantic anchoring result is encapsulated into a structured knowledge increment unit; The encapsulation process packages the newly generated nodes, relationships, weight adjustment signals, and other discrete results into a self-contained knowledge increment unit according to a preset format specification. This unit not only contains entity content but also includes metadata such as information source, confidence level, and timestamp, providing a complete context for subsequent evaluation and decision-making.

[0009] S3: Local impact assessment: In the star map coordinate model, determine the local sub-graph region that the knowledge increment unit intends to influence, and assess its consistency, complementarity, and structural impact with the existing knowledge in that region. The system first uses a graph traversal algorithm to locate the nodes and edges directly associated with the knowledge increment units, delineating the local influence domain. The evaluation engine then performs multi-dimensional analysis within this domain: consistency metrics check for logical contradictions; complementarity metrics distinguish whether the new information is entirely new, refined, or redundant; and structural impact assessment simulates the topological changes after its introduction, predicting whether it will lead to logical loops, node semantic overload, or network imbalance. This is equivalent to a pre-update simulation.

[0010] S4: Atomized incremental operation: Based on the evaluation results, select and execute one or more operations from a predefined set of atomic incremental operations to update the star map coordinate model; the set of atomic incremental operations includes operations for creating, strengthening, weakening, merging, and splitting nodes, as well as operations for adding, adjusting weights, changing types, and deleting relationships between nodes; The system abandons global adjustments and employs a set of well-defined, single-function atomic operations as its surgical tools. Based on the evaluation results of step S3, the system invokes the tool combination following the principle of minimal intervention. For example, for entirely new and consistent information, it performs node creation and relationship addition; for cognitive refinement, it may perform node splitting; and for information updates, it may perform relationship weight adjustment. This atomic operation minimizes disturbance to the overall structure of the knowledge base.

[0011] S5: Consistency maintenance. After the incremental operation is executed, the logical consistency of the local subgraph region is verified. If a conflict is detected, the conflict resolution mechanism is triggered.

[0012] After the atomic operation is executed, the system immediately performs a rule scan of the affected domain to check for any newly generated logical conflicts. Once a conflict is detected, a resolution mechanism is triggered. For example, the priority of new and old knowledge can be arbitrated through confidence competition, or context-bound nodes can be added to ensure that contradictory knowledge holds true under specific conditions, thereby achieving non-destructive conflict resolution.

[0013] In the above-mentioned method for semantic anchoring of multimodal information and dynamic incremental update of star map based on star map coordinates, in step S1, when different modal information is anchored to nodes or relationships in the star map coordinate model that have semantic conflicts, a conflict detection and arbitration mechanism is triggered. The arbitration mechanism makes a ruling based on the confidence level of the information source or a preset meta-rule.

[0014] The conflict detection unit compares anchoring results from different modalities in real time. When a semantic conflict is detected, such as visual joy versus audible sadness, the direct fusion process is immediately frozen, and an arbitration process is initiated. The arbitration can be based on preset rules, such as prioritizing visual over audio, or on dynamic calculations of source confidence, such as sensor accuracy and model prediction probability, to ensure that the input information is consistent before fusion.

[0015] In the above-mentioned method for multimodal information semantic anchoring and dynamic incremental update of star map based on star map coordinates, the assessment of structural impact in step S3 includes: predicting whether the introduction of the knowledge increment unit will lead to logical loops, node semantic overload, or relationship network imbalance in the local subgraph region.

[0016] Logical loops refer to the generation of circular reasoning chains; node semantic overload refers to a single node carrying too many contradictory relationships; relational network imbalance refers to abnormal local connection density. The evaluation engine uses graph theory algorithms for quantitative analysis, such as calculating the existence of loops and changes in node degree centrality, thereby transforming structural risks from qualitative descriptions into quantifiable indicators that can be used for decision-making.

[0017] In the above-mentioned method for multimodal information semantic anchoring and dynamic incremental update of star map based on star map coordinates, in step S4, the selection and execution of incremental operations follows the principle of minimum intervention, that is, priority is given to selecting the combination of operations that causes the least change to the topology of the star map coordinate model.

[0018] The principle of minimum intervention drives the system to prioritize the operation path with the fewest required operations, the smallest range of nodes and edges involved, and the least disturbance to the global topology among multiple possible operation paths. For example, when faced with updating the attribute of an existing node, it prioritizes strengthening / weakening operations rather than deletion and reconstruction. This ensures the stability and efficiency of knowledge updates from a decision-making logic perspective.

[0019] In the above-mentioned method for multimodal information semantic anchoring and dynamic incremental update of star map based on star map coordinates, the conflict resolution mechanism in step S5 includes: adding context-limiting nodes to conflicting knowledge so that it is valid under specific conditions, or conducting confidence competition adjudication based on the timeliness and authority of the information source.

[0020] Adding contextual constraints is a refined form of knowledge management. For example, adding a context node for "penguins cannot fly in a standard atmospheric environment" allows it to coexist with the common knowledge that birds can fly. Confidence competition establishes a dynamic evaluation system that comprehensively considers factors such as the freshness of information, the authority of the source (e.g., authoritative press releases versus personal blogs), and the original signal-to-noise ratio, assigning dynamic weights to information to achieve evidence-based adjudication.

[0021] In the above-mentioned method for multimodal information semantic anchoring and dynamic incremental update of star map based on star map coordinates, after step S5, the method further includes: recording the incremental update to the version log, and propagating the impact of the update along the relational edges in the star map coordinate model with a finite step length according to a preset strategy.

[0022] Each successful update is recorded in the version log, forming an auditable and rollbackable history of knowledge evolution. In addition, the semantic propagation mechanism allows the influence of knowledge to spread along the relational network in a limited range. For example, after updating that a species is extinct, the weight of its relationship with the current food chain is automatically weakened, enabling related knowledge to be updated collaboratively and maintaining the dynamic consistency of the knowledge base.

[0023] A system for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates, used to implement the aforementioned method for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates, includes: The multimodal semantic anchoring module is used to execute step S1. It includes a multimodal encoder array, a semantic anchoring engine, and a conflict detection unit. The multimodal encoder array is responsible for feature extraction. The semantic anchoring engine is the core processor that coordinates the two types of anchoring operations: retrieval and generation. The conflict detection unit serves as a quality monitoring point to ensure the internal consistency of the input information.

[0024] The StarMap dynamic incremental update module executes steps S2 to S5. It includes an incremental monitoring and encapsulation unit, a local impact assessment engine, an incremental operation executor, and a consistency maintainer. The incremental monitoring and encapsulation unit is responsible for capturing and formatting learning signals; the local impact assessment engine performs pre-update review; the incremental operation executor implements specific changes; and the consistency maintainer performs post-verification. These four components form a complete autonomous learning closed loop of assessment-decision-execution-audit.

[0025] The star map coordinate knowledge base stores structured star map coordinate models and interacts with the anchoring and updating modules. It employs graph database technologies to store nodes, edges, and their rich attributes, supporting not only efficient relational queries and graph traversal but also transaction support for atomic incremental operations, ensuring data consistency and atomicity of update operations under concurrent access.

[0026] In the aforementioned multimodal information semantic anchoring and dynamic incremental update system based on star map coordinates, the semantic anchoring engine further includes: The retrieval-based anchoring submodule is used to associate modal information with existing nodes in the star map coordinate model; The generative anchoring submodule is used to create new nodes and relationships for new concepts that do not exist in the star map coordinate model.

[0027] The retrieval submodule utilizes and enhances existing knowledge, while the generative submodule empowers the system to discover and define new knowledge. The combination of these two allows the system to both understand the known world and explore and integrate into the unknown.

[0028] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned method for multimodal information semantic anchoring and dynamic incremental update of star map based on star map coordinates.

[0029] A computer-readable storage medium storing a computer program thereon, characterized in that, when executed by a processor, the program implements the aforementioned method for multimodal information semantic anchoring and dynamic incremental update of star map based on star map coordinates.

[0030] Compared with existing technologies, the advantages of this invention are as follows: By anchoring the information of each modality to specific nodes and relationships in a unified, structured star map coordinate knowledge model, deep semantic alignment based on symbols and logic is achieved; a local impact assessment and atomized incremental operation set are introduced, where any new knowledge first assesses its impact on the local subgraph, predicts its logical consistency, complementarity, and structural risks with existing knowledge, and uses atomic-level graph operations such as creation, splitting, merging, and adding contextual constraints to integrate knowledge, maximizing the protection of the stability of the original knowledge structure, fundamentally avoiding catastrophic forgetting, and achieving safe and controllable lifelong learning; through conflict detection and arbitration mechanisms and version logs, all knowledge updates are recorded, documented, and traceable, making the knowledge evolution process of the entire system transparent and auditable, providing a basis for debugging, optimization, and ethical review; a loosely coupled modular design is adopted, where adding a new modality only requires training or connecting the corresponding encoder and defining its mapping method with star map nodes in the anchoring engine, without touching the core fusion and inference logic or realigning with other modalities, which greatly reduces the complexity and cost of system expansion. Attached Figure Description

[0031] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a flowchart of the multimodal semantic anchoring process of the present invention; Figure 3 This is a diagram illustrating the core steps of the dynamic incremental update of this invention. Detailed Implementation

[0032] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0033] like Figure 1-3 As shown, a multimodal information semantic anchoring and dynamic incremental update system for star map coordinates consists of three core parts that form a collaborative pipeline: Multimodal semantic anchoring module: As the system's perception front end, it is responsible for receiving and understanding heterogeneous raw inputs. Internally, it includes: a multimodal encoder array, composed of pre-trained or dedicated encoders, responsible for converting raw signals from different modalities such as images, text, speech, and sensor data into high-level feature vectors and preliminary semantic label sets; a semantic anchoring engine, the core of understanding, embedding retrieval-based anchoring submodules and generative anchoring submodules. It uses a star map coordinate knowledge base as a "semantic dictionary," calculating similarity to associate input information with existing concept nodes, or creating new nodes and establishing relationships at appropriate locations in the knowledge base when encountering entirely new concepts; and a conflict detection unit, which compares anchoring results from different modalities in real time. When a semantic conflict is detected, it immediately initiates front-end arbitration, making a preliminary ruling based on source confidence or pre-defined meta-rules to ensure the inherent consistency of the input information.

[0034] The Star Map Coordinates Knowledge Base, serving as the system's central memory and cognitive foundation, employs a graph structure to persistently store the "Star Map Coordinates Model." This model represents world knowledge as a complex network composed of nodes, edges, and their attributes. It is not only a database for queries but also the object of all reasoning and updating operations.

[0035] The StarMap dynamic incremental update module, serving as the system's learning and growth hub, is responsible for securely integrating the results of front-end anchoring into the knowledge base. It's a miniature decision-making and execution closed loop, comprising: an incremental monitoring and encapsulation unit that listens to the output of the semantic anchoring module and formats it into standardized knowledge incremental units; a local impact assessment engine that, for each incremental unit, quickly locates its directly related local subgraphs in the knowledge base and performs a sandbox-style assessment, quantifying its consistency, complementarity, and structural risks to the subgraph topology; an incremental operation executor with a complete set of atomic-level graph structure operation primitives that, based on the assessment results and following the principle of minimal intervention, selects and executes the most concise and safest operation or set of operations; and a consistency maintainer that acts as a quality inspector after each operation, logically scanning the affected areas and, if new conflicts are found, invoking mechanisms such as "adding context constraints" or "confidence competition" to resolve them, ensuring the knowledge base is always logically self-consistent.

[0036] A method for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates mainly includes the following steps: S1: Receive input information from at least two different modalities, map each modal information to a corresponding semantic node or relationship in a unified star map coordinate model, and complete semantic anchoring. The semantic anchoring includes retrieval-based anchoring, which associates modal information with existing nodes in the star map coordinate model based on semantic similarity; and / or generative anchoring, which creates new nodes and relationships for new concepts or instances that do not exist in the star map coordinate model. Each anchor operation records the source modality, timestamp, confidence level, and extraction algorithm version information.

[0037] S2: In the semantic anchoring process, perform multimodal consistency verification and conflict detection: The system compares descriptions of the same entity or event from different modalities in real time and calculates a multimodal consistency score. The consistency score is calculated as follows: consistency_score = α × semantic_similarity + β × temporal_alignment + γ × spatial_coherence, where α, β, and γ are weighting coefficients that are dynamically adjusted according to the task context.

[0038] When a conflict is detected, it is categorized based on consistency scores, historical credibility records, and contextual information. Conflict types include at least: Type 1 (Technical Conflict): caused by sensor errors or environmental noise; Type 2 (Misunderstanding Conflict): caused by cognitive limitations or differences in perspective; and Type 3 (Malicious Conflict): caused by intentional deception or information manipulation.

[0039] Based on the conflict classification results and consistency check results, the credibility weight of each modality information is dynamically adjusted.

[0040] S3: Based on the current context, task objective, and credibility evaluation results of each modality in the star map coordinate model, dynamically fuse the anchored and validated multimodal information. According to the current task type, the mental state of AGI in the star map coordinate model, and the real-time credibility score of each modality, assign dynamic weights to the anchored information of each modality. The weight calculation formula is: FinalWeight_m = BaseWeight_m × CredibilityScore_m × ContextRelevance_m.

[0041] S4: Perform credibility-aware collaborative reasoning within the star map coordinate model to obtain a fusion understanding or decision result and a corresponding confidence score. Specifically, this includes traversing and inferring along the semantic relationship edges between anchored multimodal information nodes in the star map, and labeling each reasoning path with a credibility score. When a logical contradiction is detected between a "behavior node" and a "declaration node," the "deception detection subgraph" is automatically activated for pattern matching and risk assessment.

[0042] S5: Incremental Update Triggered. The generated, credibility-assessed fusion understanding or decision results are encapsulated into structured knowledge increment units. This unit contains information such as new nodes, new relationships, and weight adjustments, along with metadata such as information source, confidence level, and timestamp.

[0043] S6: Local Impact Assessment. In the star map coordinate model, determine the local subgraph regions that the knowledge increment unit intends to influence, and assess its consistency, complementarity, and structural impact with the existing knowledge in that region, such as whether it will lead to logical loops or node semantic overload.

[0044] S7: Atomized Incremental Operations. Based on the evaluation results, one or more operations are selected and executed from a predefined set of atomic incremental operations to update the star map coordinate model. The operation set includes operations for creating, strengthening, weakening, merging, and splitting nodes, as well as operations for adding, adjusting weights, changing types, and deleting relationships. Selection follows the principle of minimal intervention.

[0045] S8: Consistency Maintenance and Intelligent Arbitration. After the incremental operation is executed, logical consistency verification is performed on the local subgraph region. If a conflict is detected, a conflict resolution and intelligent arbitration mechanism is triggered. This mechanism is a refinement of conflict detection and includes: Multi-level arbitration: Based on the level of conflict risk, initiate automatic arbitration, metacognitive arbitration, or request external arbitration.

[0046] Adaptive Response: Take appropriate measures based on the type of conflict (technical, misunderstanding, malicious), such as re-collecting data, requesting clarification, issuing a security alert, or initiating an in-depth investigation.

[0047] Credibility propagation and updating: Positively propagate reliable inference results to enhance the credibility of relevant nodes; negatively propagate identified deception or misinformation to reduce the credibility of relevant sources. Establish and maintain an "entity-credibility" archive.

[0048] In addition, the system will record this incremental update in the version log and, according to the preset strategy, propagate the impact of the update along the relational edges in the star map coordinate model with a finite step length. Example 1

[0049] This embodiment is used for real-time environmental understanding and obstacle avoidance knowledge updates in robot guide scenarios. It integrates information from multiple sensors such as cameras, LiDAR, and microphones to understand the dynamic environment and update its knowledge base after encountering new obstacles. The specific implementation steps are as follows: S1: Multimodal semantic anchoring: The robot "sees" a cylindrical, red, stationary object through the camera, perceives its precise outline and distance through the lidar, and at the same time the microphone does not detect any sound emitted by the object.

[0050] After being encoded, visual and radar features are used for retrieval-based anchoring against a star map coordinate knowledge base. The visual features "cylindrical" and "red" show high similarity and match to the "fire hydrant" node; the radar features "stationary" and "specific contour" further support this match. The microphone's "silent" feature is added as an attribute.

[0051] S2: Incremental Update Triggered: The "fire hydrant" node sensed this time, along with its "location" and "stationary" attributes, is encapsulated into a knowledge increment unit. The unit records the information sources and confidence levels for each modality.

[0052] S3: Local Impact Assessment: The system locates local subgraphs composed of nodes such as "fire hydrants," "street facilities," and "obstacles" in the star map coordinate model. The assessment concludes that this perception is consistent with existing knowledge that "fire hydrants are street facilities," and provides specific instance attributes, constituting complementary information. Structural impact assessment: Confirmed that it will not cause logical loops or relational imbalances.

[0053] S4: Atomized Incremental Operations: Following the principle of minimal intervention, the system selects and executes atomic operations: strengthens the association weight between the "fire hydrant" node and the current specific instance; adds a "belongs" relationship edge from the instance node to the "current path obstacle" node.

[0054] S5: Consistency Maintenance and Follow-up: After the operation, verify that there are no logical conflicts in the local subgraph. The system records this update to the version log. Subsequently, according to the preset strategy, the location information of the "stationary fire hydrant" is used as a key obstacle avoidance basis, and its impact is semantically propagated in the star map path planning subgraph with a limited step size, directly leading to the real-time adjustment of the subsequent navigation path. Example 2

[0055] The AGI system in this embodiment continuously monitors multi-source information such as news, financial reports, and social media to update its understanding of the listed company's business layout. The specific implementation steps are as follows: S1: Multimodal semantic anchoring. The system receives an authoritative financial news text: Company A announced a major strategic transformation, focusing on investing in the field of new energy batteries. At the same time, the analysis of its latest annual report charts shows that R&D expenses are tilted towards the energy storage technology sector.

[0056] The text encoder anchors news retrieval to strategic transformation and new energy battery nodes. The chart encoder anchors data retrieval to R&D investment and energy storage technology nodes. These nodes are connected to Company A node through relationships such as involvement and investment direction.

[0057] S2: Incremental update triggered, the anchoring results of the newly added relationships involving Company A, new energy batteries, and energy storage technology are encapsulated into knowledge increment units, and the information source is marked as highly credible authoritative news and the company's official annual report.

[0058] S3: Local Impact Assessment. In the star map, a local subgraph centered on Company A is located. This subgraph contains edges representing the original Company A's relationship from "belonging" to "traditional automotive." The assessment reveals a potential conflict between new information and old knowledge regarding the understanding of the core business. However, the new information source has high confidence, and strategic transformation and belonging are not mutually exclusive, potentially reflecting business diversification.

[0059] S4: Atomized incremental operations. Based on the evaluation, the system executes a series of atomic operations: creating a new node for new energy batteries; adding a new edge company A - strategic focus shift -> new energy batteries; weakening the weight of the old edge company A - belonging to -> traditional automobiles; adding a historical major business attribute to the old edge.

[0060] S5: Consistency maintenance and follow-up, conflict resolution is initiated after execution. The system adopts a mechanism of adding context constraints: the old edge belongs to a context condition node before the time transition. At this point, the conflict is resolved, and AGI uses the new edge when answering the current main business, and can refer to the old edge when answering the historical background. This update is fully logged. Example 3

[0061] This embodiment describes rare case learning in an intelligent medical image-assisted diagnosis scenario. The AGI system assists doctors in analyzing medical images and can learn from rare cases to enrich their diagnostic knowledge base. The specific implementation steps are as follows: S1: Multimodal semantic anchoring: Systematic analysis of a set of lung CT images revealed a unique and previously unseen pattern of diffuse nodules. Simultaneously, the pathology report text concluded: "A rare subtype of idiopathic pulmonary fibrosis."

[0062] The abnormal features extracted by the visual encoder could not find a high similarity match in the existing nodes such as "pneumonia", "tumor" and "common pulmonary fibrosis" in the star map, so generative anchoring was triggered: under the "lung disease" node, a new instance node "rare subtype of pulmonary fibrosis X" was created, and the CT feature vector was stored as the modal attribute attachment of this node.

[0063] The text encoder anchors the pathology report search to the "idiopathic pulmonary fibrosis" node.

[0064] S2: Incremental Update Trigger: The generated new node "Rare Subtype of Pulmonary Fibrosis X", its "is a type" relationship with "Idiopathic Pulmonary Fibrosis", and the associated CT feature attributes are encapsulated into a knowledge increment unit. The unit information source is marked as "verified pathology report".

[0065] S3: Local Impact Assessment: Evaluated within the "Lung Diseases" subgraph of the Star Map. The new node is logically consistent with existing disease classification knowledge and represents entirely new complementary information. The structural impact assessment concludes that creating new nodes and establishing classification relationships is clear, safe, and will not lead to semantic overload.

[0066] S4: Atomization Incremental Operation: Perform atomic operations: ① Strengthen the association between the "idiopathic pulmonary fibrosis" node and the "rare subtype" concept; ② Establish the "having imaging manifestation" relationship edge between the "rare subtype pulmonary fibrosis X" node and related CT imaging feature nodes.

[0067] S5: Consistency Maintenance and Follow-up: Verify the logical consistency of the new subgraph. The system records this discovery in the version log as an important extension of medical knowledge. Subsequently, when similar imaging features reappear, the system can perform collaborative reasoning based on the star map and propose an auxiliary diagnostic suggestion of "suspected rare subtype of pulmonary fibrosis X" along the newly established relation edges, realizing a closed loop of knowledge application and value.

[0068] In summary, the principle of this embodiment is as follows: using a structured star map coordinate model as a unified semantic representation framework, the original perceptual information from different modalities is first mapped to interpretable nodes and relationships within this framework, and then each anchoring result is regarded as an evaluable knowledge increment. The system then follows the principle of minimal intervention, and after strictly evaluating the logical consistency, complementarity, and topological risks of this increment to the local structure of the knowledge graph, it executes a series of predefined atomic graph operations to complete knowledge fusion and updating. Finally, through conflict resolution mechanisms and versioned log management, the entire process from multimodal perception to the autonomous evolution of the knowledge system is ensured to be safe, controllable, and traceable, thereby realizing continuous, stable, and interpretable lifelong learning of the artificial intelligence system in a complex and open environment.

[0069] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

[0070] Although this article uses a lot of... Multimodal semantic anchoring module, star map dynamic incremental update module, star map coordinate knowledge base The invention uses terms such as "inventory" and "extensions," but does not preclude the use of other terms. These terms are used merely for the convenience of describing and explaining the essence of the invention; interpreting them as any additional limitation would be contrary to the spirit of the invention.

Claims

1. A method for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates, characterized in that, Includes the following steps: S1: Multimodal semantic anchoring, receiving input information from at least two different modalities, mapping each modal information to corresponding semantic nodes or relationships in a unified star map coordinate model, and completing semantic anchoring; the semantic anchoring includes similarity-based retrieval anchoring and / or generative anchoring that creates new nodes; S2: Incremental update triggered, the semantic anchoring result is encapsulated into a structured knowledge increment unit; S3: Local impact assessment: In the star map coordinate model, determine the local sub-graph region that the knowledge increment unit intends to influence, and assess its consistency, complementarity, and structural impact with the existing knowledge in that region. S4: Atomized incremental operation, based on the evaluation results, selects and executes one or more operations from a predefined set of atomic incremental operations to update the star map coordinate model; The atomic incremental operation set includes operations for creating, strengthening, weakening, merging, and splitting nodes, as well as operations for adding, adjusting weights, changing types, and deleting relationships between nodes. S5: Consistency maintenance. After the incremental operation is executed, the logical consistency of the local subgraph region is verified. If a conflict is detected, the conflict resolution mechanism is triggered.

2. The method for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates according to claim 1, characterized in that, In step S1, when different modal information is anchored to nodes or relationships in the star map coordinate model that have semantic conflicts, a conflict detection and arbitration mechanism is triggered. The arbitration mechanism makes a ruling based on the confidence level of the information source or a preset meta-rule.

3. The method for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates according to claim 1, characterized in that, In step S3, the assessment of structural impact includes: predicting whether the introduction of the knowledge increment unit will lead to logical loops, node semantic overload, or relational network imbalance in the local subgraph region.

4. The method for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates according to claim 1, characterized in that, In step S4, the selection and execution of incremental operations follows the principle of minimum intervention, that is, priority is given to the combination of operations that causes the least change to the topology of the star map coordinate model.

5. The method for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates according to claim 1, characterized in that, In step S5, the conflict resolution mechanism includes: adding context-limiting nodes to conflicting knowledge to make it valid under specific conditions, or conducting confidence competition adjudication based on the timeliness and authority of the information source.

6. The method for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates according to claim 1, characterized in that, After step S5, the method further includes: recording the incremental update to the version log, and propagating the impact of the update along the relational edges in the star map coordinate model with a finite step length according to a preset strategy.

7. A system for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates, used to implement the method for multimodal information semantic anchoring and dynamic incremental update of star maps based on star map coordinates as described in any one of claims 1-6, characterized in that, include: A multimodal semantic anchoring module is used to perform step S1, and it includes a multimodal encoder array, a semantic anchoring engine, and a collision detection unit; The Star Chart Dynamic Incremental Update Module is used to execute steps S2 to S5, and includes an incremental monitoring and encapsulation unit, a local impact assessment engine, an incremental operation executor, and a consistency maintainer. A star map coordinate knowledge base is used to store structured star map coordinate models and to interact with the anchoring module and the update module.

8. The multimodal information semantic anchoring and dynamic incremental update system for star map based on star map coordinates according to claim 7, characterized in that, The semantic anchoring engine further includes: The retrieval-based anchoring submodule is used to associate modal information with existing nodes in the star map coordinate model; The generative anchoring submodule is used to create new nodes and relationships for new concepts that do not exist in the star map coordinate model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for multimodal information semantic anchoring and dynamic incremental update of star map based on star map coordinates as described in any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements a method for multimodal information semantic anchoring and dynamic incremental update of star map based on star map coordinates as described in any one of claims 1-6.

Images